We recorded 245 neurons in six male Long Evans rats over 88 fear discrimination sessions. We identified 91 neurons (~37% of all neurons recorded) showing significant decreases in firing rate to danger or uncertainty. Visualization of all cue-inhibited neurons revealed heterogeneous inhibition of danger firing during late cue presentation (Figure 1D). To determine whether this heterogeneity reflected the activity of two separate populations, we performed k-means clustering for all cue-inhibited neurons on early (first 5 s) and late (last 5 s) firing to danger. The first cluster (n = 45) consisted of neurons that were danger-inhibited early, but danger-excited late. These neurons are referred to as the Flip population. The second cluster (n = 46) consisted of neurons that were danger-inhibited early and late, and are referred to as the Sustain population. Independent samples t-tests for waveform properties revealed no differences between Flip and Sustain neurons, indicating these populations could only be distinguished by their function (Figure 1E–G): baseline firing, t₈₉ = 0.95, p = 0.343; half duration, t₈₉ = 0.77, p = 0.444; amplitude ratio t₈₉ = 0.10, p = 0.918.

A vlPAG signal for fear output should begin at cue onset, continue throughout cue presentation, and discriminate danger, uncertainty and safety. To determine if cue-inhibited vlPAG neurons complied with these requirements, we examined mean population activity over cue presentation for Flip and Sustain neurons. Flip neurons were initially inhibited to uncertainty and danger, but lesser to safety (Figure 2A). As cue presentation continued, inhibition to uncertainty weakened toward safety and firing to danger switched to excitation (Figure 2A). ANOVA for normalized firing rate (Z-score) for the 45 Flip neurons [Figure 2A; factors: trial-type (danger, uncertainty and safety) and bin (250 ms bins encompassing: 2 s baseline, 10 s cue, and 2 s delay)] revealed main effects of cue (F_2,88 = 16.58, p = 7.74×10⁻⁷, η_p² = 0.27, op = 1.00) and bin (F_55,2420 = 14.83, p = 1.03×10⁻¹¹⁴, η_p² = 0.25, op = 1.00), but most critically a cue x bin interaction (F_110,4840 = 7.85, p = 8.89×10⁻¹⁰⁶, η_p² = 0.15, op = 1.00). The population pattern was consistent across individual trials, though late danger excitation was least on the first and last trials (Figure 2—figure supplement 1A–D).

Figure 2 with 1 supplement see all

Download asset Open asset

We constructed a 95% bootstrap confidence interval to determine if differential firing was observed early and late in cue presentation for Flip neurons. Differential firing was not observed to danger vs. uncertainty early (M = 0.04, 95% CI [−0.05, 0.11]), but was observed late when danger flipped to excitation (M = 0.35, 95% CI [0.26, 0.42]) (Figure 2B, plus signs). In contrast, differential firing was observed to uncertainty vs. safety early (M = −0.18, 95% CI [−0.26,–0.11]), but not late (M = −0.05, 95% CI [−0.14, 0.05]) (Figure 2B). Furthermore, the 95% bootstrap confidence interval for normalized firing did not contain zero for any cues in either period (Figure 1B, pound signs), indicating the Flip population was responsive to all cues.

Sustain neurons showed cue-selective inhibition of firing: danger < uncertainty < safety ≈ 0. This firing pattern was observed throughout cue presentation (Figure 2C). ANOVA for normalized firing rate [Figure 2C; factors maintained from above] revealed main effects of cue (F_2,86 = 72.25, p = 3.88×10⁻¹⁹, η_p² = 0.63, op = 1.00) and bin (F_55,2365 = 14.91, p = 6.13×10⁻¹¹⁵, η_p² = 0.26, op = 1.00), as well as a cue x bin interaction (F_110,4730 = 5.24, p = 1.17×10⁻⁵⁹, η_p² = 0.11, op = 1.00). The population pattern was consistent across all trials (Figure 2—figure supplement 1E–H). Selective firing was observed early and late in cue presentation. In support, the 95% bootstrap confidence interval for differential firing did not contain zero for danger vs. uncertainty (Early: M = −0.24, 95% CI [−0.35,–0.14], Late: M = −0.18, 95% CI [−0.25,–0.11]) and uncertainty vs. safety (Early: M = −0.23, 95% CI [−0.32,–0.15], Late: M = −0.27, 95% CI [−0.35, 0.18]) (Figure 2D, plus signs). Even more, the 95% bootstrap confidence interval for normalized firing did not contain zero for danger and uncertainty during both periods, but did contain zero for safety during both periods (Figure 1D, pound signs). Not only was differential firing observed, but Sustain neurons were selectively responsive to danger and uncertainty.

If the vlPAG signals fear output, one would expect population-level signals to be observed at the single-unit level. To examine this, we used sign tests to identify whether single-unit firing was biased away from zero during early and late cue presentation. Flip single-units were biased towards decreased firing to danger [Early: (p(sign) = 9.33×10⁻⁹)] and uncertainty [Early: (p(s) = 5.89×10⁻¹¹)] during early cue presentation. Strikingly, and consistent with the population response, Flip neurons were biased towards increased firing to danger [Late: (p(s) = 8.24×10⁻⁴)], but decreased firing to uncertainty [Late: (p(s) = 2.47×10⁻⁴)] during late cue presentation. Contrary to the population result, there was no bias away from zero in single-unit firing to safety early or late. Single-unit biases of Sustain neurons mirrored those observed in the population. Sustain single-units showed a consistent bias toward decreased firing to danger [Early: p(s) = 3.08×10⁻¹¹, Late: p(s) = 2.84×10⁻¹⁴)] and uncertainty [(Early: p(s) = 3.10×10⁻⁷, Late: (p(s) = 5.10×10⁻⁹)] throughout cue presentation. Further, Sustain single-units showed no bias in firing to safety in either cue period. Observing fully differential firing by single-units throughout cue presentation further marks Sustain neurons as a candidate for signaling fear output.

Descriptive analyses reveal two cue-inhibited populations with distinct temporal activity patterns. However, these analyses do not reveal the information signaled by each population. We used linear regression for single-unit firing to formally test the degree to which Flip and Sustain neurons signaled fear output and threat probability (Figure 3). For each single-unit, we calculated the normalized firing rate for each trial (32 total: six danger, six uncertainty shock, 10 uncertainty omission, and 10 safety) in 1 s bins over the course of cue presentation (14 s total: 2 s pre-cue, 10 s cue, 2 s post-cue). Fear output was the suppression ratio on that trial. Threat probability was the shock probability associated with the cue: danger: 1.00, uncertainty: 0.375 and safety: 0.00. Regression output for each single-unit was a beta coefficient quantifying the strength (|>0| = stronger) and direction (>0 = positive and <0 = negative) of the predictive relationship between the regressor and single-unit firing. Beta coefficients for single-units were subjected to ANOVA with regressor (fear output vs. threat probability) and interval (1 s cue intervals) as factors.

Figure 3

Download asset Open asset

Single-unit regression revealed an early-to-late switch in threat probability signaling by Flip neurons (Figure 3A). ANOVA for beta coefficients with factors of regressor (fear output vs. threat probability) and interval was performed for three periods: baseline (two intervals), cue (10 intervals) and delay (two intervals). The baseline and delay ANOVAs returned no main effects or interaction (all F < 0.6, all p > 0.4). In contrast, the cue ANOVA found significant main effects, but most critically a regressor x interval interaction (F_9,396 = 3.56, p = 2.85×10⁻⁴, η_p² = 0.075, op = 0.990). The interaction was driven by negative beta coefficients for fear output and threat probability in two, early cue intervals (95% bootstrap confidence interval did not contain zero, pound signs), that gave way to positive beta coefficients specific to threat probability in all late cue intervals (95% bootstrap confidence interval did not contain zero, pound signs; Figure 3A). Further supporting the interaction, beta coefficients for Flip single-units were not biased away from zero for fear output and threat probability during the first 5 s cue period [Probability Early: p(s) = 0.37, Fear Output Early: p(s) = 0.14] (Figure 3B). However, there was positive bias toward threat probability, but not fear output, during the last 5 s cue period [Probability Late: p(s) = 2.47×10⁻⁴, Fear Output Late: p(s) = 0.77] (Figure 3C). Fear responses are sustained for the cue duration, yet Flip neurons do not consistently signal threat probability or fear output in early cue presentation. The inconsistency in signaling reveals that Flip neurons are not a suitable neural substrate for governing fear output throughout cue presentation.

Linear regression revealed consistent signals for fear output and threat probability in Sustain neurons. Beta coefficients were negative at cue onset for each regressor, and maintained this negativity throughout cue presentation (Figure 3D). ANOVA for beta coefficients with factors of regressor and interval was performed as before for baseline, cue and delay. The baseline ANOVA returned no main effects or interaction (all F < 1, all p > 0.3). The cue ANOVA found only a main effect of bin (F_9,405 = 4.23, p = 2.90×10⁻⁵, η_p² = 0.086, op = 0.997), indicating similar signaling of fear output and threat probability. The delay ANOVA found only a main effect of regressor (F_1,45 = 7.27, p = 0.01, η_p² = 0.14, op = 0.751), indicating a difference in signaling of fear output and threat probability during the delay period. For each regressor over the 10, 1 s cue intervals, the 95% bootstrap confidence interval did not contain zero, indicating that fear output and threat probability signaling were observed throughout cue presentation. Consistent with equivalent signaling of fear output and probability throughout cue presentation, single-unit beta coefficients for each regressor were biased away from zero for fear output and threat probability during early and late cue presentation [Probability (Early: p(s) = 1.83×10⁻⁶ and Late: p(s) = 4.06×10⁻⁵), Fear Output (Early: p(s) = 0.002 and Late: p(s) = 1.56×10⁻⁴)] (Figure 3E & F). Further, single-unit beta coefficients for threat probability and fear output were correlated early and late (Early: R² = 0.41, p = 2.38×10⁻⁶ and Late: R² = 0.37, p = 9.55×10⁻⁶). The majority of Sustain single-units showed negative beta coefficients for both regressors. However, even the extremes of the distribution showed signaling for both regressors, albeit in opposing directions. Sustain neurons signal fear output and threat probability throughout cue presentation.

The threat probability regressor in the above analyses utilized the actual shock probability assigned to uncertainty (0.375). Of course, the subjects, and by extension their neurons, had no a priori knowledge of the actual shock probability. Thus, it is possible that Flip and Sustain single-units are ‘tuned’ to alternative shock probabilities. To test this, we performed single-unit linear regression for normalized firing in each 1 s interval as before, maintaining the probabilities for danger (1.00) and safety (0.00), but incrementing the probability assigned to uncertainty from 0 to 1 in 0.125 steps (0.000, 0.125, 0.250, 0.375, 0.500, 0.625, 0.750, 0.875, and 1.000). Threat probability beta coefficients were averaged over early and late cue presentation. The mean beta coefficient for each uncertainty assignment is plotted as a threat-tuning curve, early and late, for each population (Figure 4).

Figure 4

Download asset Open asset

Flip neurons were tuned to alternative foot shock probabilities and this tuning changed from early to late cue presentation. Early threat overestimation (equating uncertainty to danger) gave way to late underestimation (equating uncertainty to safety). The tuning curve trough for early cue presentation occurred at 0.750 (Figure 4A, light pink); overshooting the actual probability of 0.375 (Figure 4A, dashed black line) and exceeding mean fear output (Figure 4A, dashed green line). The tuning curve peak for late cue presentation occurred at 0.250 (Figure 4A, dark pink); undershooting the actual probability (Figure 4A, dashed black line). ANOVA for beta coefficient with factors of time (early vs. late) and uncertainty assignment (9) found both main effects and the interaction to be significant (all F > 13, all p < 0.001). The relative firing patterns of Flip neurons do not approximate the actual probability of foot shock or the pattern of fear output, and further indicate that these neurons are unlikely to govern fear output.

In contrast, Sustain neuron tuning consistently fell between the bounds of the actual foot shock probability and the mean fear output, changing only subtly over cue presentation (Figure 4B). The trough of the tuning curve occurred at an assignment of 0.500 early and at an assignment of 0.625 late. ANOVA found a main effect of assignment (F_8,360 = 5.66, p = 8.85×10⁻⁷, η_p² = 0.112, op = 1.00) and a time x assignment interaction (F_8,360 = 3.25, p = 0.001, η_p² = 0.067, op = 0.971). The stability of Sustain tuning and the bias toward mean fear output further suggests this population as a candidate for fear output.

Ten adult male rats at postnatal day 55 (P55) were obtained from Charles River Laboratories in Raleigh, NC. On arrival, rats were single-housed on a 12 hr light cycle (lights off at 6:00pm) and allowed three acclimation days with ad libitum access to water and standard chow (18% Protein Rodent Diet #2018, Harlan Teklad Global Diets, Madison, WI) prior to surgery. Rats were implanted with drivable, sixteen-wire microelectrode bundles. Each animal received between eleven and sixteen days to recover from surgery with ad libitum access to water and standard chow. Throughout the experiment, rats had ad libitum access to water; however, to generate motivation for a food-reward, standard chow was restricted to maintain rats at 85% of their free-feeding body weight. Three rats were eliminated from the study because electrodes failed to register single-unit activity and one rat was eliminated due to incorrect electrode placement. Reported data are from remaining six individuals. All protocols were approved by the Boston College Animal Care and Use Committee and all experiments were carried out in accordance with the NIH guidelines regarding the care and use of rats for experimental procedures.

Microelectrodes were constructed on site and consisted of a drivable bundle of sixteen Formvar-Insulated Nichrome wires (25.4 µm diameter: 761500, A-M Systems, Carlsborg, WA) within a 27-gauge cannula (B000FN3M7K, Amazon Supply). The cannula bundle was attached to a manually operated microdrive calibrated to permit ~0.042 mm advancement increments. Two free-hanging 127 µm diameter PFA-coated stainless-steel ground wires were also part of the assembly (791400, A-M Systems, Carlsborg, WA). All wires were electrically connected to a Nano Strip omnetics connector (A79042-001, Omnetics Connector Corp., Minneapolis, MN) on a custom 24-contact, individually-routed and gold-immersed circuit board (San Francisco Circuits, San Mateo, CA).

Aseptic stereotaxic surgery was performed under isoflurane anesthesia (1% to 5% in oxygen). Prior to incision, Rimadyl/Carprofen (024751, Henry Schein Animal Health, s.c. 5 mg/kg) and Ringer’s lactate solution (014792, Henry Schein Animal Health, s.c. 2 to 5 mL) were administered sub-cutaneously to the back, and 2% lidocaine (002468, Henry Schein Animal Health, s.c. 0.25 mL) was administered sub-cutaneously above the skull. Post-incision, the skull was scoured in a crosshatch pattern with a scalpel blade to strengthen implant adhesion. Five screws (two anterior to Bregma, two between Bregma and lambda: 3 mm medial to the lateral ridges of the skull, and one on the midline: 5 mm posterior of lambda) were installed in the skull to further stabilize the bond between the skull, electrode assembly and protective head cap. A 1.4 mm diameter burr hole was drilled through the skull, centered on the implant site and the underlying dura was removed to expose the cortex. Nichrome recording wires were freshly cut with surgical scissors to extend approximately 2.0 mm beyond the cannula at a 15° angle. Just before implant, current was delivered to each recording wire in a saline bath, stripping each tip of its formvar insulation. Each omnetics connector contact was stimulated for 2 s using a resistor-equipped lead; current was supplied by a 12 V lantern battery. Machine grease was placed by the cannula and on the microdrive to prevent orthodontic resin from seizing moveable components.

The electrode assembly was slowly advanced at a 20° angle for implantation dorsal to the vlPAG. Coordinates from cortex: anterior-posterior (AP) −8.00 mm, medial-lateral (ML) −2.45 mm, and dorsal-ventral (DV) −5.52 mm. Once in place, stripped ends of both ground wires were wrapped around the posterior midline screw inserted previously. The microdrive base and a protective head cap surrounding the electrode assembly were cemented in place on the skull with orthodontic resin (C 22-05-98, Pearson Dental Supply, Sylmar, CA). At the end of the procedure, the omnetics connector was affixed to the head cap.

The apparatus for Pavlovian fear conditioning consisted of two individual behavior chambers with clear acrylic walls and top, and a grid floor with an acrylic waste pan below. Each grid floor bar was electrically connected to an aversive shock generator (Med Associates, St. Albans, VT) through a custom grounding device, which permitted the floor to be grounded at all times except during shock delivery. A nose poke opening equipped with infrared photocells was mounted on a central, acrylic wall panel and an acrylic external food cup was mounted on the same wall panel three inches below. Each behavior chamber was enclosed in a separate sound-attenuating shell. Auditory stimuli were presented through two speakers mounted on the ceiling of the shell, above the behavior chamber.

On P56 and P57, all rats received 5 g of test pellets in their home cage. On P58, all rats received 30 test pellets released (one per minute) in the behavior chamber food cup (F0021, Bio-Serv, Flemington, NJ). On P59, all rats were shaped to nose poke for pellet delivery in the behavior chamber using a fixed ratio (FR1) schedule in which one nose poke yielded one pellet. Shaping sessions lasted 30 min or until approximately 50 nose pokes were completed. On P60, all rats received one variable interval (VI30) session in which nose pokes were reinforced on average every 30 s. On P61-P64 (inclusive) all rats received four variable interval (VI60) sessions in which nose pokes were reinforced on average every 60 s. For the remainder of behavioral testing, nose pokes were reinforced on a VI60 schedule independent of all Pavlovian contingencies.

On P65 and P66, all rats received one, 42 min session of pre-exposure to the three cues to be used in Pavlovian discrimination. Pre-exposure consisted of four presentations of each cue (12 total presentations) with mean inter-trial intervals (ITIs) of 3.5 min. The order of trial type presentation was randomly determined by the behavioral program and differed for each rat during each session. Auditory cues were 10 s in duration and consisted of repeating motifs of a broadband click, phaser, or trumpet (listen or download: http://mcdannaldlab.org/resources/ardbark).

Prior to single-unit recording sessions, each rat received eight, 93 min sessions (one per day) of fear discrimination, consisting of 32 cue trials with mean ITIs of 3.5 min. Each 10 s auditory cue was associated with a unique probability of foot shock (0.5 mA, 0.5 s): danger, p = 1.00; uncertainty, p = 0.375; and safety, p = 0.00. Cue identity was counterbalanced across rats. Foot shock was administered 2 s following the termination of the auditory cue on danger and uncertainty shock trials. This was done in order to observe possible neural activity during the delay period not driven by an explicit cue. A single session consisted of six danger trials, ten uncertainty no-shock trials, six uncertainty shock trials, and ten safety trials. The order of trial type presentation was randomly determined by the behavioral program, and differed for each rat during each session. After the eighth discrimination session, rats were given ad libitum access to standard rat chow for at least 24 hr, followed by stereotaxic surgery. Following recovery, discrimination (identical to that described above) resumed with single-unit recording. Animals received discrimination with recording every other day. After each discrimination session with recording, electrodes were advanced either 0.042 mm or 0.084 mm to record from new units during the following session.

Rats were deeply anesthetized using isoflurane and final electrode coordinates were marked by passing current from a 6 V battery through 4 of the 16 nichrome electrode wires. Rats were perfused with 0.9% biological saline and 4% paraformaldehyde in a 0.2 M potassium phosphate buffered solution. Brains were extracted and post-fixed in a 10% neutral-buffered formalin solution for 24 hr, and a 10% sucrose/formalin solution for an additional 24 hr before microtome sectioning. All brains were processed for light microscopy with immunohistochemistry for anti-tryptophan hydroxylase (T8575, Sigma-Aldrich, St. Louis, MO) coupled with a NovaRed chromagen reaction (SK-4800, Vector Laboratories, Burlingame, CA). Sections were mounted and imaged using a light microscope to confirm electrode placement.

During recording sessions, a 1x amplifying head stage connected the Omnetics connector to the commutator via a shielded recording cable (Head stage: 40684–020, Cable: 91809–017 and Commutator: 50122, Plexon Inc, Dallas TX). Analog neural activity was digitized and high-pass filtered via amplifier to remove low-frequency artifacts and sent to an Ominplex D acquisition system (Plexon Inc, Dallas TX). Behavioral events (cues, shocks, nose pokes) were controlled or recorded by a computer running Med PC-IV software (Med Associates, St. Albans, VT). Timestamped events from Med PC-IV were sent to the Ominplex D acquisition system via a dedicated interface module (DIG-716B). This acquisition process resulted in a single file (.pl2) containing all time stamps for all spikes and events. Single-units were sorted offline with a template-based spike-sorting algorithm (Offline Sorter V3, Plexon Inc, Dallas TX). Timestamped spikes and events (cues, shocks, nose pokes) were extracted and analyzed with statistical routines in MATLAB (Natick, MA). Neural activity was recorded throughout the 500 ms shock delivery period. Data are not presented from this period because we cannot be certain that shock artifacts did not disrupt spike collection.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank Dr. Hiram Brownell for statistical advice on linear regression, Dr. Donald B Katz for guidance on electrode construction, Dr. Thomas Stalnaker for guidance on waveform analysis, and Blake Zimmerman for guidance on Matlab scripts. Perceptually uniform colour-maps are used in this study to prevent visual distortion of the data (Crameri, 2018a; Crameri, 2018b).

Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols (#2018-002) of Boston College. The protocol was approved by the Institutional Animal Care and Use Committee of Boston College. All surgery was performed under isoflurane anesthesia, and every effort was made to minimize suffering.

1. Received: July 10, 2019
2. Accepted: September 28, 2019
3. Accepted Manuscript published: September 30, 2019 (version 1)
4. Version of Record published: October 30, 2019 (version 2)

© 2019, Wright et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.