Animals

Male C57BL/6J were purchased from Guangdong Medical Laboratory Animal Center (China). They were group housed (5–6 mice/cage) under a 12 hr light/dark cycle (8:00 AM to 8:00 PM) and were provided with food and water ad libitum. Bedding, water, and food were changed every week. Mice of 3–5 months of age were used. For in vivo electrophysiology experiments, mice were singly housed before the surgery to implant multi-wire electrodes. They were habituated by gentle handling for 3 days before behavioral and recording experiments. All animal experiments were performed in accordance with the ARRIVE (Animal Research: Reporting of In Vivo Experiments) guidelines, approved (Approval Number: 11530) by the Peking University Shenzhen Graduate School Animal Care and Use Committee.

In vivo recording

Mice were anesthetized with isoflurane (induction 3%, maintenance 1.5%). Two screws were implanted to secure electrode array implants. Electrodes consisted of 16 individually insulated nichrome wires (35 μm inner diameter, impedance 300–900 KΩ; Stablohm 675, California Fine Wire) arranged in a 3 × 5 × 5 × 3 pattern (approximately 200 μm spacing between wires). Electrodes were attached to an 18-pin connector (Mil-Max) and secured with dental cement. Optrode (opto-electrode) arrays are composed of an optic fiber (250 μm) and multiwire electrodes (the end of the optical fiber is approximately 200–300 μm above the tips of the recording electrodes).

After surgery, mice were allowed to recover for 7–14 days. Broadband (0.3 Hz to 7.5 kHz) neural signals were simultaneously recorded (16 bits at 30 kHz) from implanted 16-channel arrays using a 64-channel data acquisition system (Zeus, Bio-Signal Technologies). At completion, recorded extracellular spikes were aligned and sorted (offline sorter software, Plexon), and further analyzed using NeuroExplorer (Nex Technologies) and MATLAB [14]. Responses during CS were normalized to pre-tone spike rate by calculating a Ζ-score [14]. For statistical analysis, comparisons were performed using the averaged Ζ-score values calculated during the 500 ms period after CS onset.

Neurons were considered CS responsive if they showed robust, time-locked changes in their spiking upon CS presentation. Neuron spiking within −1–0 s as an average basal level, when spiking above the average level adds 3 × S.E.M will be used as CS responsive.

The following are the coordinate positions of the brain regions involved in this experiment: dmPFC (1.98 mm anterior to Bregma; ±0.3 mm lateral to midline, and 2.15 mm below the Bregma plane); LA (1.75 mm posterior to Bregma; ±3.3 mm lateral to midline and 4.65 mm below the Bregma plane); BLA (1.31 mm posterior to Bregma, ±3.05 mm lateral to midline and 4.85 mm below the Bregma plane); LC (5.4 mm posterior to Bregma; ±0.85 mm lateral to midline and 4.5 mm below the Bregma plane); temporal association cortex (TeA, 3.52 mm posterior to Bregma; ±4.15 mm lateral to midline and 4.1 mm below the Bregma plane). These coordinates were used for the implantation of multiwire electrodes, injection of viruses, and placement of optical/imaging fibers. Notably, the target sites for optical fiber implantation were adjusted to be 0.1 mm above the specified coordinates.

In vivo imaging

To record Ca²⁺ responses in the dmPFC neurons receiving LA inputs, mice were injected with the rAAV2/9-Ef1α-DIO-GCaMP7s virus in dmPFC, and rAAV1-CaMKII-Cre virus in LA. To record Ca²⁺ responses in the dmPFC neurons projecting to BLA, mice were injected with the rAAV2/9-Ef1α-DIO-GCaMP7s virus in dmPFC, and rAAV2/retro-hSyn-Cre virus in BLA. To record Ca²⁺ responses in the LC neurons, mice were injected with the rAAV2/9-CaMKII-GCaMP7s virus in LC. To record Ca²⁺ responses in the LC neurons (receiving input from or projecting to dmPFC), mice were injected with the rAAV2/9-Ef1α-DIO-GCaMP7s virus in LC, and rAAV1-CaMKII-Cre or rAAV2/retro-hSyn-Cre viruses in dmPFC. To record NE responses in the dmPFC and BLA, mice were injected with the rAAV-hSyn-NE2h or rAAV-hSyn-NE2m viruses in dmPFC and BLA. To record Ca²⁺ responses in the LC NE-neurons, mice were injected with the rAAV2/9-TH-GCaMP6s virus in LC.

Following virus injection, an optical fiber (230 μm O.D., 1.25 mm Ferrule Size) (Inper LLC) was placed in a ceramic ferrule and inserted towards the BLA, dmPFC or LC through a craniotomy. The ceramic ferrule was supported with a skull-penetrating M1 screw and dental acrylic. Fluorescence signal of GCaMP/NE2h was recorded with a multichannel fiber photometry system (Nanjing Thinker Tech, China). Excitation light from a 470 nm LED was reflected via a dichroic mirror (MD498; Thorlabs) and focused through a 20×/0.4 NA objective lens. An optical fiber cable (230 μm O.D., NA = 0.37, 2 m long) transmitted light from the objective through the implanted optical fiber. The laser power was adjusted at the tip of optical fiber to 20–25 μW to minimize bleaching. The GCaMP/NE2h fluorescence was band-pass filtered (MF525-39, Thorlabs) and collected using CMOS (DCC3240M, Thorlabs). A commercial software (Nanjing Thinker Tech, China) was used to transform images into time-series fluorescence signals and data were stored on a disk. The sampling rate was set to 50 frames per second.

For NE uptake inhibitor, 10 mg/kg duloxetine (Macklin, dissolved in sterile 0.9% saline) was administered (i.p.) 30 min before behavioral tests.

Data were analyzed using MATLAB. Quantification of fluorescence values (ΔF/F) were calculated and expressed as (F − F0)/F0, where F0 is the baseline fluorescence averaged between −10 and −2 s before CS presentation. The magnitudes of responses were measured by calculating the area under curve (AUC) throughout the entire CS duration.

Fluorescence tracing and histology

To label dmPFC neurons that receive LA inputs and project to BLA, mice were injected with rAAV1-CaMKII-Cre virus in LA, rAAV2/9-Ef1α-DIO-BFP in dmPFC and retrobeads (red) in BLA. To label dmPFC neurons receiving TeA inputs, mice were injected with retrobeads (Red) in dmPFC. To determine whether dmPFC neurons receive TeA inputs, mice were injected with retrobeads (Red) in dmPFC. To determine whether LC neurons receive dmPFC, TeA or LA/BLA inputs, mice were injected with retrobeads (Red) in LC. To label the LC neurons that receive dmPFC inputs and send their projections back to dmPFC or to LA/BLA, mice were injected with the rAAV1-CaMKII-Cre virus in dmPFC and rAAV2/9-Ef1α-DIO-mCherry virus in LC. To label the dmPFC neurons that receive LA inputs and project to LC, mice were injected with the rAAV1-CaMKII-Cre virus in LA, rAAV2/9-Ef1α-DIO-mCherry virus in dmPFC and CTB 488 in LC.

To confirm the efficiency of Cre⁺ virus expression, mice were injected with rAAV2/retro-hSyn-Cre-mCherry virus in BLA, and rAAV2/9-hSyn-EGFP virus in dmPFC (S1A Fig). For Cre-DIO co-transduction assessment, mice were injected with rAAV2/retro-hSyn-Cre virus in BLA, rAAV2/9-hSyn-EGFP virus and rAAV2/9-Ef1α-DIO-mCherry virus in dmPFC (S1B Fig). Both Cre-dependent recombination (overlapping of EGFP⁺ and Cre-mCherry⁺ cells) and Cre-DIO co-transduction efficiency (overlapping of EGFP⁺ and DIO-mCherry⁺ cells) satisfied the experimental requirement (S1 Fig).

Histological procedures were performed. An overdose of sodium pentobarbital was used to anesthetize the mice, after which they were perfused through the heart with 1× PBS followed by a fixative solution containing 4% paraformaldehyde in PBS. Mouse brains were then removed and post-fixed in the aforementioned fixative solution at 4 °C overnight. The samples were subsequently transferred to a sucrose solution consisting of 30% sucrose in PBS and stored at 4 °C for 48 hr. Mouse brains were then embedded in O.C.T. compound (Optimal cutting temperature compound) and sectioned coronally at a thickness of 30 μm using a cryostat at −20 °C. Sections containing the regions of interest were collected and preserved. Coronal sections with a thickness of 30 μm were analyzed to confirm the expression of the virus in the brain.

Slice recording and optogenetic stimulation

Mouse brains were rapidly removed and placed in an ice-cold cutting solution containing (in mM): 110 choline chloride, 7 MgSO₄, 2.5 KCl, 1.25 NaH₂PO₄, 25 NaHCO₃, 25 D-glucose, 11.6 sodium ascorbate, 3.1 sodium pyruvate, and 0.5 CaCl₂ gassed with 95% O₂ and 5% CO₂. Slices of 350 μm thickness were cut with a VT-1200S Vibratome tissue slicer (Leica Biosystems, Germany) in cutting solution. Slices were transferred to a holding chamber with artificial cerebrospinal fluid ACSF containing (in mM): 127 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 25 NaHCO₃, 25 D-glucose, 2 CaCl₂, and 1 MgSO₄, and allowed to recover for 30 min at 32 °C, then kept at room temperature for at least 1 hr before recording. Individual slices were transferred to the recording chamber on an Olympus microscope (BX51WI) with a ×40 water-immersion differential interference contrast objective. Slices were constantly perfused at room temperature (23–26 °C) with oxygenated aCSF (4–5 mL/min). Recordings were made from layer 2/3 of the dmPFC, in a depth of about 50–100 μm in the slices.

Data were acquired using HEKA EPC10 double patch clamp amplifier (HEKA). Signals were acquired at a sampling rate of 10 kHz and filtered at 2 kHz. Series resistance of the recording pipette was between 10 and 25 MΩ. Neurons with holding current > −200 pA (at −70 mV) were excluded from data analysis. To record spontaneous excitatory post-synaptic currents (sEPSCs) in T- and S-neurons, somatic whole-cell voltage clamp recordings (−70 mV) were obtained from layer II/III excitatory neurons in dmPFC. Recording electrodes were filled with (in mM): 125 CsMeSO4, 5 NaCl, 1.1 EGTA, 10 HEPES, 0.3 Na₂GTP, 4 Mg-ATP, and 5 QX-314. To record spontaneous inhibitory post-synaptic currents (sIPSCs), somatic whole-cell voltage clamp recordings (+5 mV) were obtained from layer II/III excitatory neurons in the dmPFC. Current clamp recordings were used to examine spikes evoked by a series of 500 ms depolarizing current pulses with 4 s intervals, and each injection step with an increase of 25 pA (from 0 to 375 pA). Recording pipette solution contains (in mM): 128 K+-gluconate, 10 NaCl, 2 MgCl₂, 10 Hepes, 0.5 EGTA, 4 Na₂ATP, and 0.4 NaGTP. All neurons were recorded for at least 5 min. Resting membrane potentials (RMPs) were measured at I = 0 pA, and neurons with RMPs above −50 mV were excluded from the analysis. The sEPSCs were recorded in the LC neurons in response to stimulation of the TeA axonal terminals (ChrimsonR), or combined opto-stimulation of PFC axonal terminals (ChR2) and TeA axonal terminals, at a holding membrane potential of −70 mV.

To identify T-neurons and S-neurons in brain slice, we injected rAAV1-CaMKII-Cre anterior transport virus in the LA and rAAV2/9-EF1α-DIO-ChR2-EGFP virus in the dmPFC, and injected retrobeads (red) in BLA. The dmPFC T-neurons showed EGFP fluorescence (green) and S-neurons showed retrobeads fluorescence (red). To activate TeA/dmPFC axonal terminals in LC neurons, we injected rAAV2/9-CaMKII-ChrimsonR-mCherry virus in TeA and rAAV2/9-CaMKII-ChR2-mCherry virus in dmPFC. To activate dmPFC axonal terminals in LC neurons (projecting to dmPFC), we injected rAAV2/9-CaMKII-ChR2-mCherry virus and retrobeads (green) in the dmPFC. All optogenetic stimulations were conducted with the same parameters based on the specific opsin used: a 473-nm laser (20 ms pulse width, 20 Hz, approximately 12 mW) for ChR2 activation whereas a 564-nm laser (20 ms pulse width, 20 Hz, approximately 12 mW) for ChrimsonR activation.

Behavioral assays

Threat conditioning and retrieval test were performed in two contexts (threat conditioning in context A; retrieval test in context B). Context A and context B were different in their shapes and odors. Chamber floors were cleaned with 1% acetic acid or 75% ethanol before testing. Scoring freezing behavior was done using a video recording system (Coulbourn Instruments). Mice were considered freezing if no movement was detected for at least 2 s. On day 0, mice were habituated in context A and received 4 CS (conditioned stimulus; 30 s, 60 dB, 50 ms pips tone, 3 kHz). Threat conditioning was conducted on day 1, with CS co-terminated with a US (unconditioned stimulus; 2 s footshock, 0.75 mA), for 3 CS-US pairing trials. Training trials were separated by 90 s intervals. For threat memory retrieval on day 2, conditioned mice received 4 trials of various combinations of CS+ and CS− (30 s, 60 dB; CS+: 50 ms pips tone (3 kHz); CS−: white noise) in context B. For experiments with Calcium/NE signal imaging, conditioned mice received 3 trials of 30 s CS+, 10 s CS+, 2 s CS+ or 2 s CS+/28 s CS− in context B at 24 hr post-conditioning.

The selection of CS+ durations (30, 10, 2, and 2 s CS+/28 s CS−) were based on the following rationale: 30 s CS+ and 10 s CS+ were used to examine whether calcium/NE signals were temporally matched to CS duration; 2 s CS+ was used to mimic the activation of T-neurons, whereas 2 s CS+/28 s CS− was used to probe post-activation effects following T-neurons.

In vivo optogenetic manipulations

Mice were injected with rAAV2/9-CaMKII-eNpHR3.0-mCherry virus or rAAV2/9-CaMKII-ChR2-EGFP virus bilaterally in LA and opto-fiber was placed in the dmPFC to inhibit or activate LA axonal terminals in dmPFC. Mice were injected with rAAV2/retro-hSyn-Cre virus bilaterally in the BLA, rAAV2/9-Ef1α-DIO-eNpHR3.0-mCherry virus in dmPFC, and opto-fiber was placed in the dmPFC to inhibit BLA-projecting dmPFC neurons [40,41]. Mice were injected with rAAV2/retro-hSyn-Cre virus in dmPFC, rAAV2/9-Ef1α-DIO-eNpHR3.0-mCherry virus in LC and opto-fiber was placed in the LC to inhibit dmPFC-projecting LC neurons. Mice were injected with rAAV1-CaMKII-Cre virus in dmPFC, rAAV2/9-Ef1α-DIO-ChR2-mCherry virus in the LC and optical fiber was placed in the LC to activate LC neurons (receiving dmPFC inputs). To activate TeA excitatory neurons, rAAV2/9-CaMKII-ChR2-mCherry virus was injected to TeA bilaterally, and opto-fiber was placed in the TeA. For optogenetic inhibition experiments involving eNpHR3.0, a 593-nm laser system was employed with continuous illumination (approximately 10 mW at the end of the optical fibers).

In vivo pharmacogenetic manipulations

Mice were injected with the rAAV2/9-Ef1α-DIO-hM4D (Gi)-mCherry virus in the dmPFC and rAAV2/retro-hSyn-Cre virus in BLA to inhibit dmPFC neurons that project to BLA. Four weeks later, mice were injected intraperitoneally with CNO (3.3 mg/kg, MedChemExpress) 30 min before threat memory retrieval.

Intra-cerebroventricular injection

Mice were anesthetized with isoflurane and placed in a stereotaxic apparatus. A guide cannula (2.5 mm in diameter) was implanted into the brain at coordinates relative to bregma: anteroposterior (AP) −0.20 mm, mediolateral (ML) −1.0 mm, and dorsoventral (DV) −2.0 mm. Following recovery, 5 μL of saline or 5 μL of propranolol (1 μg/μL, total 5 μg) was administered intracerebroventricularly (i.c.v.) through the cannula 30 min prior to the experiments [42].

Cell harvesting

The T- and S-neurons were identified in brain slices using their selective fluorescence markers, respectively. Weak negative pressure was applied to aspirate the entire neuron into the glass patch electrodes, which were treated with DEPC water (ThermoFisher) in advance [43]. The pipette tip was broken onto the wall of a 0.2 mL tight-lock tube (TubeOne) to allow the entire neuron to be immersed in a 1 μL drop of RNase-free lysis buffer (provided by Beijing Genomics Institution (BGI)) placed on the side of the tube. The tube was kept on ice and two neurons were collected in one tube within 5 min. The tube was then placed in dry ice until 20 neurons were harvested from each mouse. Samples were rapidly spun down (5–10 s) and stored at −80 °C before reverse transcription. Reverse transcription, PCR amplification and sequencing were performed by BGI.

RNA sequencing

A total of 6 samples were sequenced separately and were performed as before [43]. The expression amount of identified RNAs was analyzed using TPM (transcripts per million). Differential expression analysis was conducted using DESeq2 package in R software (1.2.22). Differentially expressed genes (DEGs) were defined as those with |fold change| ≥ 2 and q-value (a corrected P-value using Benjamini and Hochberg multiple testing correction) <0.05.

Statistical analysis and graphing

RNA Sequencing figures were generated using Dr. Tom online system and Adobe Illustrator CC 2018. Other figures were generated by GraphPad Prism. Statistical significance was calculated using a two-tailed paired/unpaired t test, one-way/two-way repeated measures (RM) ANOVA (GraphPad Prism), as noted. Data are reported as mean ± SEM. Significance levels are noted as * p < 0.05; ** p < 0.01; *** p < 0.001.

Conditioned threat cues induce transient and sustained neuronal responses in the dmPFC

We simultaneously examined spiking in dmPFC neurons and freezing levels in mice [14,15,44]. Although both transient and sustained neuronal responses have been shown in prelimbic/dmPFC neurons in response to CS+ (conditioned CS, paired with US) after auditory threat conditioning, they have not been demonstrated in the same mice. We first recorded dmPFC neuronal responses to CS presentation in conditioned mice (Fig 1A–1C). CS+ elicited two spiking patterns in two nonoverlapping neuronal populations recorded in the dmPFC in the same mice, with either transient (T-neurons) (Fig 1B) or sustained (S-neurons) responses (Fig 1C). Consistent with previous reports, T-neuron responses lasted approximately 1 s (Fig 1B), whereas S-neuron responses roughly matched the CS+ duration (Fig 1C). Both increased and decreased sustained responses were observed, but only increased transient responses were observed (S2 Fig). No obvious responses to CS− were observed, indicating the selectivity of these responses to the conditioned stimulus (S2A Fig). Among the 692 dmPFC neurons recorded, 22% (151/692) exhibited transient responses, while 6% (41/692) exhibited sustained responses, and 6% (40/692) were inhibitory neurons with increased responses (based on their spike waveforms, S2C Fig) (Fig 1D). The remaining recorded dmPFC neurons did not display obvious CS-induced changes in spiking. The ratio of total T-neurons to S-neurons was 3.68. For the remainder of the study, we present only the results from the dmPFC excitatory neurons that responded to the CS.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Activity of dmPFC T-neurons and S-neurons during retrieval of threat memory.

(A) (Upper) Experimental procedure for threat conditioning. (Lower) Freezing levels before, during, and after auditory threat conditioning. N = 12 mice. (B) (Upper) Raster plots of spiking in representative T-neurons during habituation (Hab) and CS retrieval after conditioning (Ret). (Lower) Peri-event histograms showing spiking in the T-neurons. Bars indicate CS presentation. Insert: spike waveform. Bin width, 0.5 s. n = 23 units/5 mice. (C) (Upper) Raster plots of spiking in representative S-neurons during habituation and CS retrieval after conditioning. (Lower left) Peri-event histograms of S-neuron spiking. (Lower right) Averaged z-scores vales for S-neurons during habituation and retrieval. Insert: spike waveform and Avg of Z-score (0–30 s). Two-tailed unpaired t test, Hab vs. Ret, P < 0.001. n = 17 units/5 mice. (D) Distribution of recorded dmPFC neurons based on their responses to CS in conditioned mice. About 22% (151/692) showed a transient increase, 6% (41/692) sustained increase, 66% (460/692) no response and 6% (40/692) were inhibitory neurons and. n = 692 units/42 mice. (E) Spiking in S-neurons elicited by three CSs with different durations. Dotted lines indicate z-score = 1. (F) (Upper) The AUC (z-score) of CS-elicited S-neuron response. CSs correspond to those used in (E). One-way RM ANOVA, F (2, 45) = 9.671, Bonferroni’s posttest; CS+ (10 s) vs. CS+ (30 s), P < 0.01; CS+ (10 s) vs. CS+ (40 s), P < 0.001; n = 16 units/8 mice. (Lower) The AUC (z-score) of CS-elicited responses in T-neurons (n = 29 units/8 mice). (G) (Upper) Freezing levels and time courses elicited by three CSs. One-way RM ANOVA, F (2, 18) = 15.98, Bonferroni’s posttest; CS+ (10 s) vs. CS+ (30 s), P < 0.01; CS+ (10 s) vs. CS+ (40 s), P < 0.001. (Lower) Freezing levels elicited by three CSs. N = 7 mice. Unless specified, statistical comparisons were performed using two-tailed unpaired t test; *, P < 0.05; **, P < 0.01; ***, P < 0.001. Data represented as mean ± SEM. Numerical data can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003272.g001

In our experiments, a 30 s long CS+ was used during both the conditioning and retrieval tests. To determine whether the durations of the conditioned freezing responses and S-neuron spiking match those of the CSs with different durations, we used three different CS+ durations (10, 30 and 40 s). We observed that the durations of S-neuron spiking were well matched with those of the corresponding CSs (Fig 1E, 1F), whereas T-neurons responded with similar amplitudes and durations regardless of the duration of CS+ (Fig 1F). In addition, freezing durations matched the CS+ durations, whereas freezing levels were comparable (Fig 1G). In comparison, the majority of LA excitatory neurons only showed transient spiking to CS+ (S2B Fig) [18,19,45]. In summary, transient and sustained responses were observed in two nonoverlapping dmPFC neuron populations with distinct response profiles, suggesting differential contributions to threat memory.

Distinct connection patterns of dmPFC T-neurons and S-neurons

The above differences in the CS+-elicited responses in the dmPFC T- and S-neurons suggest that they may have distinct connection patterns. Previous studies have indicated that dmPFC neurons receive synaptic inputs from the LA [5,14,15,18] and that some dmPFC neurons project to the BLA [16,17]. To confirm these findings under our experimental conditions, we labeled the dmPFC neurons receiving LA inputs with BFP (BFP⁺, rAAV1-CaMKII-Cre virus injected in the LA and rAAV2/9-Ef1α-DIO-BFP virus in the dmPFC) and BLA-projecting dmPFC neurons with retrobeads (Retro⁺, injected in the BLA; Fig 2A). The majority of labeled neurons (BFP⁺, Retro⁺) were located in layers II/III of the dmPFC, with approximately 92.54% of LA projections terminating in layers 2/3 and 7.46% in layer 5 (Fig 2B). The ratio of BFP⁺ to Retro⁺ neurons was 2.94 (Fig 2A), which was consistent with the ratio of T-neurons to S-neurons from in vivo recordings. About 5.35% of BFP⁺ neurons and 15.84% of Retro⁺ neurons were positive for both BFP⁺ and Retro⁺, suggesting minimal overlap between these two neuronal populations (Fig 2C).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Distinct responses in dmPFC T-neurons and S-neurons, their inputs, outputs and connections with each other.

(A) (Top) Diagram showing the virus injection sites. The dmPFC neurons receiving LA inputs were labeled with BFP, BLA-projecting dmPFC neurons were labeled with retrobeads (red). (Bottom) Ratio of BFP⁺ vs. Retro⁺ neurons (each dot represents averaged values from a single mouse). N = 47 sections/10 mice. (B) Representative images showing the spatial localization of labeled dmPFC neurons with BFP (blue) or retrobeads (red), taken using 10 X (upper) or 40 X (lower) objectives. The majority of labelled neurons were located in layer II/III of dmPFC. Scale bars, 100 μm (10 X) and 25 μm (40 X). (C) Numbers of neurons positive for both BFP⁺ and Retro⁺ over BFP⁺ neurons (blue) and numbers of neurons positive for both BFP⁺ and Retro⁺ over Retro⁺ neurons (red). N = 10 mice. (D) (Left, top) Sites for virus injections and Ca²⁺ response recording; (Left, bottom), Heat maps showing Ca²⁺ signals elicited by 30 s CS+ (dotted lines) in dmPFC neurons receiving LA inputs. (Right) Averaged Ca²⁺ signals from individual mouse (red traces) and the averaged responses from all mice (black), elicited by 30 s CS+ in dmPFC neurons receiving LA inputs. Scale bars, 1% ΔF/F and 10 s. N = 5 mice. (E) (Left, top) Sites for virus injections and Ca²⁺ response recording; (Left, bottom) Heat maps showing Ca²⁺ signals elicited by 30 s CS+ in dmPFC neurons projecting to BLA. (Right) Averaged Ca²⁺ signals from individual mouse (red traces) and averaged responses from all mice (black), elicited by 30 s CS+ in dmPFC neurons projecting to BLA. Scale bars, 1% ΔF/F and 10 s. N = 5 mice. (F) (Left, top) Sites of virus injections and optogenetic inhibition; (Left, bottom) Opto-inhibition of LA axonal terminals in the dmPFC on freezing levels during CS+ retrieval. Two-tailed paired t test, Laser ON vs. Laser OFF, P < 0.01; N = 8 mice. (Right) Opto-inhibition of LA axonal terminals in dmPFC on spike rates in T-neurons (top) and S-neurons (bottom), during CS+ retrieval. T-neurons, n = 35 units/8 mice; S-neurons, n = 20 units/8 mice. (G) (Left, top) Sites of virus injections and optogenetic inhibition; (Left, bottom) Opto-inhibition of BLA-projecting dmPFC neurons on freezing levels during CS+ retrieval. Two-tailed paired t test, Light ON vs. Light OFF, P < 0.001; N = 8 mice. (Right) Opto-inhibition of BLA-projecting dmPFC neurons on spike rates in dmPFC T-neurons (top) and S-neurons (bottom), during CS+ retrieval. T-neurons, n = 41 units/8 mice; S-neurons, n = 24 units/8 mice. (H) Response latency of dmPFC T-neurons and S-neurons elicited by CS+. Dotted lines represent the time points when slope (dy/dx) of responses were equal to 1 (points of intersection were 0.122 and 0.298 s, respectively). T-neurons, n = 54 units/12 mice; S-neurons, n = 29 units/12 mice. (I) (Left) Sites of virus injections and dmPFC recording. (Right) Responses in dmPFC T-neurons and S-neurons to opto-stimulation of LA neuron axonal terminals in dmPFC (bar). Dotted lines represented the time points when slope (dy/dx) of responses were equal to 1 (points of intersection were 0.008 and 0.159 s, respectively). T-neurons, n = 32 units/6 mice; S-neurons, n = 16 units/6 mice. (J) Schematic diagram showing opto-activation of T-neurons and recording in S-neurons in PFC slices. (K) Representative EPSC traces in the S-neurons (projecting to BLA) elicited by opto-stimulation of dmPFC neurons receiving LA inputs (5 ms, bar), in sequential presence of aCSF, TTX, and TTX + 4-AP. The EPSC trace during TTX bath application (black) was rightward shifted for better visualization. Scale bars, 100 ms and 80 pA. (L) EPSC amplitudes in the S-neurons from H₂. n = 6 cells/3 mice, paired t test, P < 0.05. Numerical data can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003272.g002

The above findings suggest that dmPFC T-neurons and S-neurons may be differentially tagged and studied. We first examined the CS+-elicited Ca²⁺ responses in dmPFC neurons receiving LA inputs by injecting rAAV1-CaMKII-Cre virus into the LA and rAAV2/9-Ef1α-DIO-GCaMP7s virus into the dmPFC (Fig 2D). These Ca²⁺ responses were clearly transient (approximately 1 s) (Fig 2D) and therefore indicate that dmPFC neurons receiving LA inputs respond to CS+ in a transient manner, similar to the response pattern of T-neurons from our in vivo recordings. We then examined Ca²⁺ responses in dmPFC neurons projecting to the BLA by injecting rAAV2/retro-hSyn-Cre virus into the BLA and rAAV2/9-Ef1α-DIO-GCaMP7s virus into the dmPFC (Fig 2E). These neurons showed sustained responses to CS+, roughly matching the CS+ duration (Fig 2E), similar to the S-neurons recorded in vivo. Taken together, these results indicate that the two dmPFC neuron populations show distinct responses to CS+ on the basis of their inputs or projections. This finding provides the foundation for further analysis of their properties and selective manipulation of their activities.

The implication of the above finding is that T-neurons and S-neurons can be selectively targeted to manipulate their activities. This manipulation will allow further investigation of their contributions to the generation of freezing behavior. To this end, we used opto-inhibition of the LA neuron axonal terminals in the dmPFC by expressing the rAAV2/9-CaMKII-eNpHR3.0-mCherry virus in the LA (Fig 2F). Compared with no-light conditions, yellow light illumination during CS+ led to reduced freezing levels and spiking in both T-neurons and S-neurons (Fig 2F). Opto-inhibition of BLA-projecting dmPFC neurons using retrograde virus (rAAV2/retro-hSyn-Cre virus in the BLA and rAAV2/9-Ef1α-DIO-eNpHR3.0-mCherry virus in the dmPFC) significantly reduced freezing levels (Fig 2G). Importantly, an obviously lower spiking rate was observed in the S-neurons but not in the T-neurons (Fig 2G).

The above findings indicate that T-neurons are upstream of S-neurons in the propagation of threat memory-related information from the LA to the BLA. Consistent with this, CS+-elicited spiking in the S-neurons lagged spiking in the T-neurons by approximately 200 ms (Fig 2H). To further test whether S-neurons receive inputs from T-neurons, we injected the rAAV2/9-CaMKII-ChR2-EGFP virus into the LA (Fig 2I). Opto-stimulation of the axonal terminals in the dmPFC resulted in spiking in the T-neurons followed by spiking in the S-neurons by approximately 200 ms (Fig 2I). Interestingly, opto-stimulation of the T-neurons evoked transient responses in the S-neurons, suggesting that the sustained spiking in the S-neurons is likely caused by the long CS.

As shown in both Fig 2H and 2I, we observed a 200 ms delay between spiking in T-neurons and S-neurons. One explanation for this long delay is that T-neuron projections onto S-neurons are polysynaptic. To investigate this possibility, we injected the rAAV1-CaMKII-Cre virus into the LA and the rAAV2/9-Ef1α-DIO-ChR2-EGFP virus into the dmPFC to activate T-neurons via anterograde transport and then injected the retrobeads (red) into the BLA to label S-neurons (Fig 2J). The dmPFC S-neurons were identified by fluorescence (Fig 2J). Light-evoked EPSCs were recorded in S-neurons, and these EPSCs were eliminated by bath perfusion with tetrodotoxin (TTX) but rescued by further addition of 4-aminopyridine (4-AP) in the perfusion medium (Fig 2K; 2L). Since monosynaptic but not polysynaptic connections are preserved in TTX + 4-AP [46], this result indicates that T-neurons project to S-neurons monosynaptically. The 200 ms delay in Fig 2H and 2I is likely caused by the amount of time required for synaptic integration to reach the spiking threshold.

Distinct neuronal properties of T-neurons and S-neurons

We next asked whether T-neurons and S-neurons possess different properties in addition to the distinct connection patterns. To answer this question, we first conducted an RNA-seq analysis. We aspirated RNAs from labeled T-neurons or S-neurons from brain slices using patch electrodes and examined their RNA profiles [43]. Here, T-neurons were defined as those dmPFC neurons receiving LA inputs, whereas S-neurons were defined as those dmPFC neurons projecting to the BLA. A total of 14,609 genes were detected (S2 Table), with 650 showing significantly higher expression and 728 with lower expression in the S-neurons than in the T-neurons (Fig 3A and S2 Table). Principal component analysis of the 6 samples distinguished between T-neurons and S-neurons (Fig 3B). We first found that for ion channels, the expression of 8 genes (Kctd4, Kctd10, Kctd12, Kcne3, Kcng2, Kcna5, Kcnt2 and Kcnd2) was significantly greater and that of 4 genes (Kctd18, Kcnab3, Kcna4 and Kcns1) was significantly lower in the S-neurons (Fig 3C and S3 Table), whereas the expression of Cacnb2 was significantly greater in the S-neurons, with no difference in the expression of sodium channel genes (S3 Table). These results suggest differences in neuronal excitability between T- and S-neurons. Second, no difference was detected for GABA receptors or glutamate receptors between T- and S-neurons, except for Gabrr2 and Grik1, which were lower and higher in the S-neurons, respectively (S3 Table). Third, Adrb1 (adrenergic receptor, β-1) was expressed at higher levels in S-neurons than in T-neurons (Fig 3D and S3 Table). Fourth, T- and S-neurons presented distinct cell-adhesion molecule (CAM) expression profiles, suggesting different cell-surface assembly codes (Fig 3E and S3 Table). Taken together, the above RNA expression profile suggests that T-neurons and S-neurons are two distinct neuronal populations.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Characterizations of T-neurons and S-neurons.

(A) Volcano plots showing significantly differentially expressed genes (DEGs) in the T-neurons and S-neurons. (B) Principal components analysis (PCA) of an RNA-seq data set of 6 samples was sufficient to distinguish between T-neurons and S-neurons. (C) Heat map showing average transcripts per million (TPM) of differentially expressed genes encoding potassium channels with significance. Heatmap was standardized by rows. (D) Heat map showing average TPM of differentially expressed genes related to norepinephrine signaling with significance. (E) Heat map showing average TPM of differentially expressed genes encoding cell adhesion molecules with significance. In all heatmaps, the redder the color, the higher the expression level. Standardized method: log (average TPM + 1). n = 20 cells/mouse, 3 mice each group. (F) Input resistance of T-neurons and S-neurons from non-conditioned (NC) mice. Two-tailed unpaired t test, NC-T vs. NC-S, P = 0.0179; n = 11 cells/3 mice (NC-T), 13 cells/3 mice (NC-S). (G) (Left) Representative sEPSC traces and (right) sEPSC frequency in dmPFC T-neurons and S-neurons, from conditioned (Cond) and non-conditioned (NC) mice. One-way RM ANOVA, F (2, 53) = 7.804, Bonferroni’s posttest, NC-T vs. Cond-T, P < 0.01; NC-S vs. Cond-S, P < 0.01; n = 15 cells/4 mice (NC-T), 15 cells/4 mice (NC-S), 20 cells/4 mice (Cond-T), 24 cells/6 mice (Cond-S). Scale bars, 20 pA and 500 ms. (H) (Left) Representative traces of action potentials in dmPFC T-neurons and S-neurons elicited by injection of current through the recording electrodes. (Right) Spike frequency of action potentials plotted against injected currents. Two-way RM ANOVA, F (30, 990) = 6.894, Bonferroni’s posttest, P < 0.001; NC-T vs. NC-S, P < 0.01; NC-T vs. Cond-T, P < 0.001; n = 10 cells/4 mice (NC-T), 11 cells/4 mice (NC-S), 11 cells/4 mice (Cond-T), 21 cells/6 mice (Cond-S). Scale bars, 50 mV and 200 ms. (I) Half-width of action potentials (AP) in the T-neurons and S-neurons. One-way RM ANOVA, F (3, 42) = 13.57, P < 0.001, Bonferroni’s posttest; NC-T vs. NC-S, P < 0.01; Cond-T vs. Cond-S, P < 0.001; n = 12 cells/4 mice (NC-T), 10 cells/4 mice (NC-S), 12 cells/4 mice (Cond-T), 12 cells/6 mice (Cond-S). (J) Amplitude of AP in the T-neurons and S-neurons. One-way RM ANOVA, F (3, 41) = 5.530, P < 0.01, Bonferroni’s posttest; NC-T vs. Cond-T, P < 0.05; n = 12 cells/4 mice (NC-T), 9 cells/4 mice (NC-S), 12 cells/4 mice (Cond-T), 12 cells/6 mice (Cond-S).

https://doi.org/10.1371/journal.pbio.3003272.g003

The above differences in the mRNA profiles of T- and S-neurons predict differences in their electrophysiological properties. We thus made patch-clamp recordings from identified T- and S-neurons in the dmPFC slices and compared their electrophysiological properties. We observed significantly greater input resistance in the S-neurons (Fig 3F). Excitatory inputs, measured using sEPSCs, were not different between T- and S-neurons (Figs 3G and S3). Interestingly, T-neurons presented a conditioning-dependent increase in sEPSC frequency (Fig 3G), which likely reflects an increased probability of presynaptic release since the sEPSC amplitude was unaltered (S3A Fig). The intrinsic excitability was not different between T- and S-neurons (Fig 3H). The half-width of spikes was significantly greater in the S-neurons (Fig 3I), and spike amplitudes were significantly lower in the S-neurons (Figs 3J and S3F), whereas most other properties of spikes were not different (S3B–S3E Fig). The spike amplitude in the S-neurons also showed a conditioning-dependent increase.

Sustained S-neuron responses and freezing require TeA inputs and dmPFC NE elevation

The observations that S-neuron responses match well with the CS durations (Fig 1E) and that opto-activation of axonal terminals onto T-neurons resulted in only a transient response in the S-neurons (Fig 2I) suggest that sensory (sound) inputs contribute to the CS-elicited sustained responses in the S-neurons. Previous studies have shown that the temporal association cortex (TeA) mediates the specific type of auditory conditioning stimuli used in our experiments [47–49]. To examine whether TeA represents the primary auditory input to the dmPFC, we first injected retrobeads into the dmPFC and observed that the majority of the beads were present in TeA but not in the primary auditory cortex (Fig 4A). Optogenetic stimulation of TeA inputs resulted in strong spiking in the dmPFC S-neurons and smaller responses in the T-neurons in vivo (Fig 4B). In addition, opto-stimulation of TeA inputs in PFC slices elicited strong responses in the identified S-neurons (on the basis of their projection to the BLA, green, Retro⁺ neurons) (Fig 4C). Both observations support strong TeA inputs onto the dmPFC S-neurons.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Contribution ofβ-norepinephrine receptors to dmPFC neuronal responses and freezing levels.

(A) Experimental paradigm for examining inputs to dmPFC using injection of retrobeads in dmPFC (left) and visualization of retrobeads in the auditory cortex (right). A1, primary auditory cortex; AuV, auditory cortex; TeA, temporal association cortex. Scale bar, 200 μm. (B) Responses (z-score) of T-neurons and S-neurons to opto-activation of TeA inputs (0.9 Hz, 250 ms). T-neurons, n = 20 units/5 mice; S-neurons, n = 12 units/5 mice. rAAV2/9-CaMKII-ChR2-mCherry viruses were injected in TeA and optical fiber implanted in the TeA to activate TeA neurons. (C) Injection of rAAV2/9-CaMKII-ChR2-mCherry virus in the TeA (Left). Representative EPSC trace in the dmPFC neurons to opto-stimulation of TeA inputs (Right). Bar indicates blue light illumination. Scale bars, 200 pA and 25 ms. (D) Experimental procedure for examining intra-cerebroventricular injection of propranolol or saline on neuronal responses and freezing levels. (E) Freezing levels during retrieval with (Prop) or without (Saline) propranolol injection in conditioned mice. Two-tailed paired t test, Prop vs. Saline, P < 0.05; N = 6 mice. (F) Effect of intra-cerebroventricular injection of Prop on spike rates in the S-neurons. Two-tailed paired t test, Prop vs. Saline, P < 0.05. N = 9 units/6 mice. (G) Effect of intra-cerebroventricular injection of Prop on spike rates in the T-neurons. N = 22 units/6 mice. (H) (Top) Representative sEPSCs traces in the S-neurons during bath application of Isop (50 μM). Scale bars, 20 pA and 50 ms. (Bottom) Effects of Isop on sEPSC frequency (Two-tailed paired t test, t = 4.246, df = 10, Pre-Isop vs. Isop, P < 0.01; n = 11 cells/4 mice), and sEPSC amplitude in the S-neurons (Two-tailed paired t test, t = 0.976, df = 20, Pre-Isop vs. Isop, P > 0.05; n = 11 cells/4 mice). (I) (Top) Representative sIPSCs traces in S-neurons during bath application of Isop. Scale bars, 50 pA and 50 ms. (Bottom) Effects of Isop on sIPSC frequency (Two-tailed paired t test, t = 4.961, df = 7, Pre-Isop vs. Isop, P < 0.01; n = 8 cells/3 mice) and sIPSC amplitude in the S-neurons (Two-tailed paired t test, t = 2.088, df = 7, Pre-Isop vs. Isop, P > 0.05; n = 8 cells/3 mice). (J) (Top) Representative action potential traces in the S-neurons before and after bath application of Isop. Scale bars, 20 mV and 200 ms. (Lower left) Effects of Isop on the resting membrane potential (RMP) in the S-neurons (Two-tailed paired t test, t = 6.143, df = 9, Pre-Isop vs. Isop, P < 0.01; n = 10 cells/4 mice) and intrinsic excitability of S-neurons (Two-way RM ANOVA, F (15, 288) = 2.294, Bonferroni’s posttest, P < 0.01; Pre-Isop vs. Isop, P < 0.01; n = 12 cells/3 mice). (K) Effects of Isop on sEPSC frequency (Two-tailed paired t test, t = 2.333, df = 9, Pre-Isop vs. Isop, P < 0.05; n = 10 cells/4 mice) and sEPSC amplitude in the T-neurons (Two-tailed paired t test, t = 1.758, df = 9, Pre-Isop vs. Isop, P < 0.05; n = 10 cells/4 mice). (L) Effects of Isop on sIPSC frequency (Two-tailed paired t test, t = 1.265, df = 7, Pre-Isop vs. Isop, P > 0.05; n = 8 cells/3 mice) and sIPSC amplitude in the T-neurons (Two-tailed paired t test, t = 1.778, df = 7, Pre-Isop vs. Isop, P > 0.05; n = 8 cells/3 mice). (M) Effects of Isop on RMP (Two-tailed paired t test, Pre-Isop vs. Isop, P > 0.05; n = 8 cells/3 mice) and intrinsic excitability of T-neurons (Two-way RM ANOVA, F (15, 336) = 0.4195, Bonferroni’s posttest, P > 0.05; Pre-Isop vs. Isop, P > 0.05; n = 12 cells/3 mice). Numerical data can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003272.g004

We next investigated the mechanism underlying the transition from transient responses to sustained responses in dmPFC neurons, or in other words, the mechanism required to sustain the responses in S-neurons to match the CS+ duration. Previous studies have shown that NE signaling via β-adrenergic receptors maintains S-neuron responses and freezing [16,17]. The intracerebroventricular injection of the β-noradrenergic antagonist propranolol (Prop, Fig 4D) resulted in a significant reduction in both freezing levels (Fig 4E) and S-neuron spiking (Fig 4F) elicited by CS+. Importantly, T-neuron spiking was not significantly affected by Prop (Fig 4G), indicating that NE modulation is specific to S-neurons and also highlighting the necessity for S-neuron responses in the generation of freezing behavior.

NE is known to increase the activity/excitability of postsynaptic neurons and neurotransmitter release [50–53]. Our RNA-seq results also suggested increased expression of β-adrenergic receptors in S-neurons (Fig 3D). Therefore, we tested whether NE, especially the activation of β-ARs, preferentially modulates the properties of S-neurons. To do so, we bath applied the β-AR agonist isoprenaline (ISOP, 50 μM), which led to a significant increase in sEPSC frequency but not amplitude (Fig 4H) in the identified S-neurons in the PFC slices, suggesting increased presynaptic glutamate release. ISOP also induced a significant reduction in sIPSC frequency but not amplitude (Fig 4I), significant membrane depolarization and a significant increase in intrinsic excitability (Fig 4J), in S-neurons. Collectively, these results indicate that NE enhances S-neuron activity via increased excitatory inputs, decreased inhibitory inputs, and increased intrinsic neuronal excitability. Thus, CS-elicited NE release in the dmPFC is likely to promote greater responsivity in S-neurons during threat memory retrieval. In contrast, the main impact of ISOP on T-neurons was increased sEPSC frequency (Figs 4K–4M and S4). Collectively, these results indicate a preferential enhancement of S-neuron activity by NE via increased excitatory inputs, decreased inhibitory inputs, and increased intrinsic neuronal excitability. Thus, CS-elicited NE release in the dmPFC is likely to promote greater responsivity in S-neurons during threat memory retrieval.

Contribution of dmPFC NE elevation to freezing behavior

Our results thus far suggest that NE plays an important role in the transition from memory to behavior. Previous studies have indicated that conditioned threat cues increase dmPFC NE levels [54]. To directly monitor changes in dmPFC NE levels, we injected rAAV-hSyn-NE2h virus (NE fluorescence sensor) [55–57] into the dmPFC and monitored extracellular NE levels using fiber photometry (Fig 5A). In the conditioned mice, 30 s or 10 s CS+ led to a rapid and significant increase in NE sensor fluorescence, which decayed slowly toward the pre-CS level after CS termination (Fig 5A). In contrast, a 30 s CS− (which did not elicit significant freezing) did not elicit a significant change in fluorescence in the same mice (Fig 5A), indicating that PFC NE elevation is specific to the CS+. In comparison, a 2 s CS+ (mimicking T-neuron activation; Fig 1B) or 10 s CS+ led to a shorter increase in NE sensor fluorescence, with a duration elicited by a 2 s CS+ longer than the CS+ duration (Fig 5A). We observed similar response dynamics using a sensor with medium NE affinity (rAAV-hSyn-NE2m) [57] (S5 Fig), indicating that the affinity of the NE sensor does not significantly affect the dynamics of the observed NE sensor fluorescence. Although our NE fluorescence signals cannot be readily converted into NE concentrations, the extracellular NE concentration has been shown to be linearly related to the LC output across a range of tonic LC activities [58,59].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. The dmPFC NE levels, LC neuron activity and BLA NE levels associated with CS retrieval.

(A) (Left) Sites of virus injections and NE signal recording (top), and plots of NE signals with stimuli (middle and bottom). Scale bars, 0.5% ΔF/F and 10 s. Inserts: AUC of ΔF/F of dmPFC NE signals. Two-tailed unpaired t test, 30 s CS+ vs. 30 s CS−, P < 0.001; 10 s CS+ vs. 2 s CS+, P < 0.05. (Right) Heat maps showing dmPFC NE signals elicited by 30 s CS+, 30 s CS−, 10 s CS+ and 2 s CS+. N = 5 mice per group. (B) (Left) Sites of virus injections and Ca²⁺ signal recording (top), and plots of Ca²⁺ signals with stimuli (middle and bottom). Scale bars, 2% ΔF/F and 10 s. Inserts: AUC of ΔF/F of LC Ca²⁺ signals. Two-tailed unpaired t test, 30 s CS+ vs. 30 s CS−, P < 0.01; 10 s CS+ vs. 2 s CS+, P < 0.05. (Right) Heat maps of Ca²⁺ responses in the LC-neurons in response to 30 s CS+, 30 s CS−, 10 s CS+ and 2 s CS+. N = 7 mice per group. (C) (Left) Sites of virus injections and optical fiber recording (top), and plots of Ca²⁺ signals in the LC TH⁺ neurons with stimuli (middle and bottom). Scale bars, 2% ΔF/F and 10 s. Insert: AUC of ΔF/F of Ca²⁺ signals in the LC TH⁺ neurons. Two-tailed unpaired t test, 30 s CS+ vs. 30 s CS−, P < 0.01; 10 s CS+ vs. 2 s CS+, P < 0.05. (Right) Heat maps for Ca²⁺ signals in the LC TH⁺ neurons in response to 30 s CS+, 30 s CS−, 10 s CS+ and 2 s CS+. N = 7 mice per group. (D) Sites of virus injections and optical fiber recording (left), and impact of opto-inhibiting dmPFC-projecting LC neurons on freezing levels (right). Two-tailed paired t test, CS+ vs. CS+ (NpHR), P < 0.001; CS+ (NpHR) vs. CS+ (mCherry), P < 0.001. N = 5 mice, each group. Numerical data can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003272.g005

A major source of dmPFC NE comes from the LC [60–63]. In support of this, we observed a significant increase in Ca²⁺ signals at 30 and 10 s CS+ but not at 30 s CS− in the LC neurons expressing the rAAV2/9-CaMKII-GCaMP7s virus (Fig 5B) [58]. In addition, significant increases in the Ca²⁺ signals were elicited by 2 s CS+ or 10 s CS+ (Fig 5B). Since the majority of NE-releasing LC neurons express tyrosine hydroxylase (TH) [64], we monitored Ca²⁺ activity in the LC TH⁺ neurons using injection of rAAV2/9-TH-GCaMP6s virus in the LC and observed similar response patterns (Fig 5C). These results indicate that CS+ elicits a strong, selective NE elevation in the dmPFC and that this elevation roughly matches the CS+ duration. This dmPFC NE elevation is likely mediated by the activation of dmPFC-projecting LC neurons. To determine whether dmPFC NE elevation is required for freezing behavior, we inhibited LC-to-dmPFC projections by injecting rAAV2/retro-hSyn-Cre virus into the dmPFC and rAAV2/9-Ef1α-DIO-eNpHR3.0-mCherry virus or rAAV2/9-Ef1α-DIO-mCherry virus (control virus) into the LC (Fig 5D). We found that yellow light illumination resulted in freezing levels that were significantly lower than those elicited by the CS+ or freezing methods in mice expressing the control virus and subjected to the same light illumination (Fig 5D). This effect was similar to that induced by Prop injection (Fig 4E), supporting the critical contribution of LC to dmPFC projections to freezing behavior. LC neurons that project to the BLA are also activated by CS+ after threat-conditioning [23,61], and we observed strongly elevated NE signals in the BLA in response to 30 s CS+ but not CS− (S6 Fig).

LC–NE neurons receive inputs from the dmPFC and TeA

As dmPFC neurons are activated during CS+ presentation, this activation may in turn activate LC-neurons [65–68]. In addition, the time-locked responses of the LC-neurons suggest that they may be activated by sound inputs. In support of this idea, we observed prominent fluorescence in the dmPFC and TeA but not in the LA or BLA in the mice injected with retrobeads (red) in the LC (Figs 6A and S7). By expressing the rAAV2/9-CaMKII-ChR2-mCherry virus in TeA, we observed robust responses in LC neurons to opto-activation of TeA axonal terminals in the LC slices (Fig 6B). Since both T-neurons and S-neurons are present in the dmPFC, we asked whether their inputs to LC neurons may differ. To this end, we labeled the dmPFC neurons receiving LA inputs with mCherry (mCherry⁺, rAAV1-CaMKII-Cre virus injected in the LA and rAAV2/9-Ef1α-DIO-mCherry virus in the dmPFC) and LC-projecting dmPFC neurons with CTB 488 (CTB 488⁺, injected in the LC). Most of the mCherry⁺ and CTB 488⁺ neurons overlapped in the dmPFC, and the ratio of mCherry⁺ neurons to CTB 488⁺ neurons was 0.835. This observation confirms an overlapping population within the dmPFC that both receives LA inputs and projects to the LC (S8 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. LC/NE neurons are activated by inputs from dmPFC and TeA, and project to dmPFC.

(A) (Left) Experimental paradigm for examining major brain regions that project to LC, using retrobeads (red) injected in the LC. (Right) Brain regions with red fluorescence, dmPFC and TeA. Scale bar, 200 μm. (B) Responses of the LC neurons in LC slice to opto-activation of axonal terminals from TeA neurons (red bar, 10 s). rAAV2/9-CaMKII-ChR2-mCherry virus injected in the TeA, and laser illumination in the slices. Scale bars, 20 pA and 5 s. (C) (Left) Representative responses of LC neurons to opto-activation (bar, 50 ms) of axonal terminals from either T-neurons or S-neurons. Scale bars, 20 pA and 50 ms. (Right) All LC neurons responded to opto-activation of axonal terminals from the T-neurons, but none responded to opto-activation of axonal terminals from the S-neurons. n = 10 cells. (D) (Left) Experimental paradigm for examining outputs from the LC neurons receiving dmPFC projections. (Right) Presence of the mCherry fluorescence (red) in the dmPFC but not LA or BLA. Scale bars, 200 μm. (E) (Left) Schematic diagram showing opto-activation of dmPFC neurons axon terminals and recording in the dmPFC-projecting LC neurons in LC slices. (Right) 71% of dmPFC-projecting LC neurons responded to opto-activation of dmPFC neuron axon terminals and 29% showed no response to the same stimuli. (F) (Left) Diagram for virus injections, (middle and right) Ca²⁺ responses elicited by 30 s CS+, 30 s CS−, 10 s CS+ and 2 s CS+ in the LC neurons receiving dmPFC inputs. Scale bars, 0.5% ΔF/F and 10 s. N = 7 mice. Insert: AUC of ΔF/F of Ca²⁺ signals. Two-tailed unpaired t test, 30 s CS+ vs. 30 s CS−, P < 0.001; 10 s CS+ vs. 2 s CS+, P < 0.05. (G) (Left) Diagram for virus injections, (middle and right) Ca²⁺ responses to 30 s CS+, 30 s CS−, 10 s CS+ and 2 s CS+ in the LC neurons projecting to dmPFC. Scale bars, 0.5% ΔF/F and 10 s. N = 7 mice. Insert: AUC of ΔF/F of Ca²⁺ signals. Two-tailed unpaired t test, 30 s CS+ vs. 30 s CS−, P < 0.001; 10 s CS+ vs. 2 s CS+, P < 0.05.

https://doi.org/10.1371/journal.pbio.3003272.g006

To confirm this finding using electrophysiological recording, we recorded LC neuron responses to selective opto-stimulation of inputs from either T-neurons (their receiving LA inputs; rAAV1-CaMKII-Cre virus was injected into the LA, and rAAV2/9-Ef1α-DIO-ChR2-mCherry virus was injected into the dmPFC to label T-neurons) or S-neurons (their projections to the BLA; rAAV2/retro-hSyn-Cre virus was injected into the BLA, and the rAAV2/9-Ef1α-DIO-ChR2-mCherry virus was injected into the dmPFC to label S-neurons). Strong light-evoked EPSCs in the LC neurons stimulated T-neuron inputs (10/10), but no responses were observed with the stimulation of S-neuron inputs (0/10) (Fig 6C). This result indicates that the dmPFC T-neuron inputs constitute the major PFC inputs to LC neurons in the context of threat memory.

As the dmPFC projects to the LC and the LC projects back to the dmPFC, we examined whether the same LC-neurons receive dmPFC inputs and project back to the dmPFC. For this analysis, mice were injected with the rAAV1-CaMKII-Cre virus in the dmPFC and the rAAV2/9-Ef1α-DIO-mCherry virus in the LC, and the presence of fluorescently labeled axon terminals indicated projections from the LC to the dmPFC (Fig 6D). In contrast, no fluorescence was observed in the LA or BLA, indicating that no strong projections to the LA or BLA from LC neurons received dmPFC inputs (Fig 6D). To further confirm these observations via electrophysiological recordings, we examined the responses of dmPFC-projecting LC neurons to the opto-stimulation of dmPFC inputs (axonal terminals) in LC slices. For this analysis, the rAAV2/9-Ef1α-CaMKII-ChR2-mCherry virus and retrobeads (green) were injected into the dmPFC. Approximately 71% of dmPFC-projecting LC neurons showed strong responses to the opto-activation of dmPFC neuron axonal terminals, and 29% showed no responses (Figs 6E and S9). Given that some connections are likely severed during the slicing procedure, this result indicates that the majority of LC neurons are reciprocally connected to dmPFC neurons.

We next examined the Ca²⁺ responses of the LC neurons receiving dmPFC inputs and observed robust responses to 2, 10, and 30 s CS+ but not to 30 s CS− (Figs 6F and S10). We also observed strong CS+-elicited Ca²⁺ responses in the LC neurons that project to dmPFC inputs in vivo, which roughly matched the CS duration (Figs 6G and S10). In addition, inhibiting dmPFC-to-BLA projections significantly reduced freezing levels without affecting dmPFC NE signals, whereas freezing levels in mice expressing control virus were not affected by CNO (S11 Fig), which is consistent with these projections being downstream of dmPFC-NE signaling.

An increase in the dmPFC NE level opens a short window for the transition to behavior during memory retrieval

Our results thus far have demonstrated that dmPFC NE levels are significantly elevated during CS+-triggered threat memory retrieval, and this elevation enhances the activity in the dmPFC S-neurons to enable their sustained spiking during CS+.

In addition, increased LC–NE neuronal activity and the resulting increased dmPFC NE levels are required for sustained S-neuron responses and freezing. We therefore next asked whether higher dmPFC NE levels sufficient to drive these responses and behaviors.

The NE fluorescence signals elicited by 10 s CS+ took approximately 30 s to decay to the baseline level (Fig 7A), and the duration of freezing was approximately 10 s (Fig 1G). This observation suggests that CS+ but not NE levels control whether freezing occurs. To test this hypothesis, we presented a 10 s CS− starting either 10 s or 30 s after CS+ termination (sections C or E, respectively; Fig 7A). The rationale for using CS− was that it does not elicit significant freezing or NE elevation by itself, but S-neurons are expected to respond to CS− since they show vigorous responses to the activation of TeA neurons (Fig 4B). Interestingly, CS− presented significant freezing at 10 s but not at 30 s after CS+ termination (Fig 7A and 7B). This finding suggests that the dmPFC NE levels at 10 s after CS+ termination (or 20 s from the start of the CS+, section C in Fig 7A) can enhance S-neuron responses only if a CS is present. In contrast, the NE level at 30 s after CS+ termination (section E, Fig 7B) was not sufficient. If a certain dmPFC NE level can elicit freezing with the presence of a CS, we hypothesized that it may be possible to alter the duration of the freezing response window by manipulating the NE levels. The extracellular NE levels are modulated by NE transporters, and injection of the NE uptake blocker duloxetine (Dulo) resulted in higher NE levels and slower decay to baseline (Fig 7B). This enhancement by Dulo was limited to the CS+ as the CS− did not affect NE levels (Fig 7B), likely because the CS− does not induce significant activation of LC neurons (Fig 5B and 5C). We then tested whether significant freezing could be elicited by the CS− at 30 s after CS+ termination in the presence of Dulo (section E, Fig 7A) and significant freezing was observed (Fig 7B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. An increase in dmPFC NE level opens a short window for transition to behavior during memory retrieval.

(A) (Left, top) Representative dmPFC NE responses to 10 s CS+ in the absence of Dulo. (Left bottom) Diagram for testing the impact of CS− on freezing levels when they were given at different time points after termination of 10 s CS+. Blue regions mark the period when CS− was given. (Right) Freezing levels during the corresponding time periods. Two-tailed paired t test, A vs. C, P < 0.001; B vs. C, P < 0.001; C vs. C/ CS−, P < 0.01. The same set of mice was used, N = 10 mice. (B) (Left) dmPFC NE signals elicited by 10 s CS+ and 30 s CS− in the absence or presence of NE uptake inhibitor Dulo. Insert: AUC of ΔF/F of dmPFC NE signals. Two-tailed unpaired t test, Dulo vs. No Dulo, P < 0.01. (Middle) Freezing levels without Dulo injection. Two-tailed paired t test, A vs. E, P < 0.001; A vs. E/CS−, P < 0.001; D vs. E, P < 0.001; D vs. E/CS−, P < 0.001. (Right) Freezing levels with Dulo injection. Two-tailed paired t test, A vs. E, P < 0.001; D vs. E, P < 0.001; E vs. E/CS−, P < 0.001. Scale bars, 0.5% ΔF/F and 10 s. The same set of mice was used, N = 10 mice. (C) Experimental protocol and stimulus patterns used (left) and corresponding freezing levels (right). One-way RM ANOVA, F (2, 8) = 0.163, Bonferroni’s posttest; CS− vs. CS+/CS−, P < 0.01; CS− vs. CS+, P < 0.001; N = 10 mice. (D) Spike rates in the T-neurons during CS retrieval (left) and AUC (z-score, 2 s) (right). One-way RM ANOVA, F (2, 72) = 4.699, Bonferroni’s posttest; CS− vs. CS+/CS−, P < 0.01; CS− vs. CS+, P < 0.05; n = 25 units/10 mice. (E) Spike rates in the S-neurons during CS retrieval (left) and AUC (z-score, 30 s) (right). One-way RM ANOVA, F (2, 33) = 2.092, Bonferroni’s posttest; CS− vs. CS+/CS−, P < 0.01; CS− vs. CS+, P < 0.01; n = 12 units/10 mice. (F) Experimental protocol for CS retrieval test, by CS+, CS−, by a hybrid stimulus (L/CS−) with opto-activation of LA axonal terminals in dmPFC for 2 s followed immediately by CS− (left) and corresponding freezing levels during retrieval (right). One-way RM ANOVA, F (2, 18) = 3.807, Bonferroni’s posttest; CS− vs. L/CS−, P < 0.05; N = 6 mice. (G) Spike rates in the T-neurons (left) and AUC (z-score, 2 s) (right). One-way RM ANOVA, F (2, 42) = 6.67, Bonferroni’s posttest; CS− vs. L/CS−, P < 0.001; CS− vs. CS+, P < 0.05; n = 13 units/6 mice. (H) Spike rates in the T-neurons (left) and AUC (z-score, 30 sec) (right). One-way RM ANOVA, F (2, 18) = 5.099, Bonferroni’s posttest; CS− vs. L/CS−, P < 0.001; CS− vs. CS+, P < 0.001; n = 16 units/6 mice. Numerical data can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003272.g007

The above observations indicate that to sustain S-neuron responses and enable freezing behavior, both TeA (CS) inputs and sufficiently high dmPFC NE levels are needed. These observations also suggest that transient activation of dmPFC neurons may be sufficient to elicit significant elevation of dmPFC NE levels, which can be sustained by an ensuing CS (including the CS−). If this is true, a brief CS+ followed by a CS− should be sufficient to elicit significant freezing levels. To test this possibility, we used a hybrid CS signal: a brief (2 s) CS+ signal followed immediately by a CS− signal (28 s) (Fig 7C). This hybrid CS elicited freezing levels similar to those elicited by 30 s CS+ (Fig 7C) and spiking in both T-neurons (Fig 7D) and S-neurons (Fig 7E), with similar amplitudes and durations to those elicited by 30 s CS+ in the same set of mice. To exclude the possibility that the non-dmPFC neurons activated by the 2 s CS+ may have a large contribution to the above effects of the hybrid CS, we used opto-activation of LA axonal terminals in the dmPFC. Replacing the 2 s CS+ with a 2 s opto-activation of LA axonal terminals in the dmPFC (L/CS− hybrid, Fig 7F) resulted in freezing levels (Fig 7F) and spiking in T-neurons (Fig 7G) and S-neurons (Fig 7H) that were indistinguishable from those elicited by the CS+ in the same set of mice. Notably the frequency and pattern used for the CS− in these experiments were very different from those used for the CS+ experiments, and the CS− alone did not elicit significant freezing. Control experiments using light or 2 s CS+ alone were not conducted since 2 s is too short to measure freezing levels accurately. Taken together, these findings strongly indicate that brief CS+ presentation initiates dmPFC neuron responses and freezing responses to nonconditioned CS.

Short-term plasticity of the dmPFC-LC connections enables CS generalization

We next investigated the mechanism underlying the CS generalization we observed with hybrid CS. Since dmPFC NE levels play a critical role in the above processes, we first asked whether LC neuron activity is high and sustained during the 2 s CS+/28 s CS−. Indeed, dmPFC NE fluorescence signals elicited by the 2 s CS+/28 s CS− remained high throughout the entire CS presentation (Figs 8A and S12). The dmPFC NE signals were similar between 2 s CS+/28 s CS− and 10 s CS+, but the freezing levels were much longer for the former, which is consistent with the presence of CS− in the former case, and the CS is required to sustain S-neuron responses. We also observed high and sustained LC-neuron responses to 2 s CS+/28 s CS− (Figs 8B and S12). Consistently, sustained Ca²⁺ signals in the TH⁺-LC neurons were also observed during 2 s CS+/28 s CS− (Figs 8C and S12) and in the LC neurons receiving dmPFC inputs (Figs 8D and S10) or projecting to the dmPFC (Figs 8E and S10). A comparison of the Ca²⁺ responses to 2 s CS+ and hybrid CS suggested that the initial responses were similar between these two CSs and that the sustained phase of hybrid CS likely sustains S-neuron responses (Fig 8B–8E). Thus, we mimicked this initial 2 s CS+ activation using a 2 s opto-activation of LC neurons that receive dmPFC inputs followed by 28 s of CS− and observed freezing levels similar to those following 30 s CS+ but no elevated freezing levels in mice expressing the control virus (Fig 8F). This finding suggests a critical role, likely gating in nature, of dmPFC inputs in enabling high and sustained responses to CS− in LC neurons.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Short-term plasticity of the dmPFC-LC connections enables the CS generalization during threat memory retrieval.

(A) NE signals in the dmPFC elicited by 10 s CS+ or 2 s CS+/28 s CS−. Scale bars, 0.5% ΔF/F and 10 s. N = 5 mice. Insert: AUC of ΔF/F of dmPFC NE signals. (B) LC Ca²⁺ signals elicited by 2 s CS+ or 2 s CS+/28 s CS−. Scale bars, 2% ΔF/F and 10 s. N = 7 mice. Insert: AUC of ΔF/F of Ca²⁺ signals. Two-tailed unpaired t test, 2 s CS+/28 s CS− vs. 2 s CS+, P < 0.05. (C) Ca²⁺ signals in the LC TH⁺ neurons elicited by 2 s CS+ and 2 s CS+/28 s CS−. Scale bars, 2% ΔF/F and 10 s. N = 7 mice. Insert: AUC of ΔF/F of Ca²⁺ signals. Two-tailed unpaired t test, 2 s CS+/28 s CS− vs. 2 s CS+, P < 0.001. (D) Ca²⁺ signals elicited by 2 s CS+/28 s CS− and 2 s CS+ in the LC neurons receiving dmPFC projections. N = 7 mice, Two-tailed unpaired t test, 2 s CS+/28 s CS− vs. 2 s CS+, P < 0.01. (E) Ca²⁺ signals elicited by 2 s CS+/28 s CS− and 2 s CS+ in the LC neurons projecting to dmPFC. N = 7 mice, Two-tailed unpaired t test, 2 s CS+/28 s CS− vs. 2 s CS+, P < 0.01. (F) (Left) Sites of virus injections and optical fiber implantation. (Right) Freezing levels elicited by 30 s CS+, 2 s blue light stimulation followed by 28 s CS−. N = 5 mice, each group. (G) (Left) (Top) Experimental procedure for opto-stimulation. (Bottom) Normalized sEPSC frequency in the LC neurons to the sEPSCs during baseline period (period A). (Right) Representative traces of recorded EPSCs in the LC neurons during baseline (TeA stimulation only), during TeA + PFC stim and after stimulation (TeA stimulation only). Scale bars, 2 s and 20 pA. N = 6 cells/2 mice. (H) (Left) (Top) Experimental procedure for opto-stimulation. (Bottom) Normalized sEPSC frequency in the LC neurons to the sEPSC during baseline period (period A). (Right) Representative traces of recorded EPSCs in the LC neurons to opto-stimulation of TeA inputs, for the same durations as in (G). Scale bars, 2 s and 20 pA. N = 9 cells/2 mice. Numerical data can be found in S1 Data.

https://doi.org/10.1371/journal.pbio.3003272.g008

Since LC neurons show experience-driven plasticity [21], we tested whether brief dmPFC inputs may enhance the responses of LC neurons to TeA (sound) inputs. For this analysis, we recorded sEPSCs in LC neurons that respond to the opto-activation of both dmPFC and TeA inputs. After establishing a stable baseline response to TeA opto-stimulation, we administered either 20 s of combined dmPFC + TeA stimulation (Fig 8G) or TeA stimulation alone as a control (Fig 8H). Only in the case of stimulating both dmPFC and TeA inputs were significantly greater responses to TeA inputs observed, and this increase lasted for at least 10 s in the absence of dmPFC stimulation. Thus, the short-term plasticity of the dmPFC-LC system may enable CS generalization by enhancing LC neuron responses to sounds/CSs (including nonconditioned CSs).

Summary

Our findings demonstrate that the transition from memory to behavior is an elaborate process that requires the integration of memory, sensory inputs (CS), and valence/internal states (mediated by NE and likely other modulators) for best fitting to the CS encountered and the bodily states at the time of memory retrieval. This fine-tuning during the transition stage maintains the memory fidelity and dynamic adaptation of the generated behavior.