The brain is hard-wired to detect and respond to threats. Catching sight of a spider or a snake, or hearing an unfamiliar sound in the house at night, can make you freeze momentarily. Multiple regions of the brain contribute to this process. According to the textbook view of threat detection, the amygdala and prefrontal cortex estimate the size of the threat. They then send this information to an area called the ventrolateral periaqueductal gray (vlPAG). The vlPAG responds by triggering a fear response such as freezing.

But some studies suggest that the vlPAG may do more than just trigger a fear response. To test this idea, Wright and McDannald trained rats to associate three different 10-second tones with different probabilities of receiving a mild electric shock to the foot. One of the tones was followed by a foot shock 100% of the time. Another tone was followed by a foot shock 37.5% of the time, while the third was never followed by a foot shock. The first tone thus signaled danger, the second uncertainty, and the third safety.

Wright and McDannald recorded vlPAG activity as the rats heard each of the tones. The recordings revealed two distinct groups of vlPAG neurons. Both groups responded most to the tone that signaled danger, less to the tone that signaled uncertainty, and least to the tone that signaled safety. However, they responded at different times. One group of neurons was most active at the start of the tone. The activity of this group depended upon the degree of threat, and not upon whether the rat showed a fear response. The second group of neurons increased its activity over the course of each tone. The activity of this group mainly reflected the degree of threat, but also represented the rat's fear response to a lesser extent.

The vlPAG thus helps to signal the size of a threat, rather than simply generating a fear response. This distinction is important because people with anxiety disorders tend to overestimate threats, and many treatments for anxiety target the brain regions involved in threat estimation. Future studies should examine how the vlPAG works together with other areas, including the amygdala and the prefrontal cortex, to evaluate threats. Understanding this circuit in full could ultimately lead to better treatments for phobias and anxiety.

When confronted with potential harm, an estimate of threat probability must be made and followed by an appropriate fear response. The ventrolateral periaqueductal gray (vlPAG) has long been implicated in this fear response (Fanselow, 1993; Kim et al., 1993; Liebman et al., 1970; Carrive et al., 1997; Bandler et al., 1985; Vianna et al., 2001; Arico et al., 2017; Assareh et al., 2017). In the canonical fear circuit, threat probability estimates (the stored associative strength of cue and foot shock, for example a danger cue in Pavlovian fear conditioning) originate in amygdalar nuclei (Duvarci and Pare, 2014; Fanselow and LeDoux, 1999; Davis, 2006; Maren et al., 2013). Amygdalar threat estimates are sent to the vlPAG, which in turn organizes behavioral components of fear output, most notably freezing (Perusini and Fanselow, 2015; Tovote et al., 2015; Dejean et al., 2015; Koutsikou et al., 2014; Walker et al., 1997). While the vlPAG contributes to diverse fear and pain-related processes (Assareh et al., 2017; Bellgowan and Helmstetter, 1998; Parsons et al., 2010; McNally et al., 2011; Johansen et al., 2010), the view of the vlPAG as a node for fear output is so prevalent, it is commonly presented in introductory neuroscience textbooks (Bear et al., 2016; Carlson and Birkett, 2017).

A series of studies have uncovered a vlPAG population showing short-latency increases in firing to a danger cue. This characteristic would be expected of neurons organizing fear output. Yet in these studies, robust relationships between vlPAG single-unit activity and freezing were observed in a minority of neurons (Tovote et al., 2016), weakly observed at danger cue onset (Watson et al., 2016), or mixed with activity that purely reflected the danger cue (Ozawa et al., 2017). At best, freezing only partially accounted for vlPAG activity. If not freezing, then what aspect of fear do vlPAG neurons signal? Here we test the hypothesis that vlPAG neurons signal threat probability.

We drew from learning theory (Rescorla, 1968) to devise a fear discrimination procedure in which three cues predict unique foot shock probabilities: danger (p=1.00), uncertainty (p=0.375) and safety (p=0.00). This behavioral approach was essential as previous studies utilized procedures in which a single cue predicted foot shock with certainty (Tovote et al., 2016; Watson et al., 2016; Ozawa et al., 2017), precluding the ability to observe neural activity reflecting threat probability. We measured fear output using conditioned suppression of reward seeking (Rescorla, 1968; Bouton and Bolles, 1980) over the entire cue and in one-second cue intervals. Using this procedure, we have found that suppression non-linearly scales to shock probability. So while suppression is strong to danger, intermediate to uncertainty and weak to safety; uncertainty produces more suppression than would be expected given its shock probability (Walker et al., 2018; Ray et al., 2018; DiLeo et al., 2016; Wright et al., 2015; Berg et al., 2014). Concurrent with fear discrimination, we recorded vlPAG single-unit activity. Indeed, the vlPAG could potentially signal fear output via conditioned suppression of reward seeking (Arico et al., 2017), a long-established measure of fear (Estes and Skinner, 1941) that correlates with freezing (Bouton and Bolles, 1980). The non-linear relationship between behavior and shock probability permitted us to determine whether vlPAG single-unit activity was better captured by fear output or threat probability.

Six male, Long Evans rats were trained to nose poke in a central port in order to receive a food pellet from a cup below. During fear discrimination (Figure 1A), three distinct auditory cues predicted unique foot shock probabilities: danger (p=1.00), uncertainty (p=0.375) and safety (p=0.00). Trial order was randomized for each rat, each session. Fear was measured with suppression ratio and was calculated by comparing nose poke rates during baseline and cue periods (see Materials and methods). After eight discrimination sessions, rats were implanted with 16-wire, drivable microelectrode bundles dorsal to the vlPAG (Figure 1B). Following recovery, rats were returned to fear discrimination. Single-units were isolated and held for the duration of each recording session. The electrode bundle was advanced ~40–80 µm between sessions to record from new single-units in subsequent sessions.

Figure 1 with 6 supplements see all

Download asset Open asset

Across all 88 recording sessions, these six rats showed excellent discriminative fear: high to danger, intermediate to uncertainty, and low to safety (Figure 1C). ANOVA for suppression ratios for the total 10 s cue [factor: cue (danger vs. uncertainty vs. safety)] found a significant effect of cue (F_2,174 = 592.00, p=2.32×10⁻⁷⁸, η_p² = 0.87, observed power (op) = 1.00). Paired t-tests confirmed differing ratios for each cue (danger vs. uncertainty, t₈₇ = 12.36, p=7.44×10⁻²¹; uncertainty vs. safety, t₈₇ = 20.85, p=3.50×10⁻³⁵). Visualizing nose poke rates to each cue revealed a discrimination pattern matching that for suppression ratio (Figure 1—figure supplement 1). While the foot shock probability associated with uncertainty was closer to safety: danger (p=1.00) >> uncertainty (p=0.375) > safety (p=0.00); the mean suppression ratio to uncertainty was closer to danger: danger (ratio = 0.80) > uncertainty (ratio = 0.53) >> safety (ratio = 0.03). The non-linear relationship between shock probability and behavior will be critical for regression analyses that seek to determine if single-unit firing is better captured by threat probability or fear output.

When the total cue period was divided into 10, 1 s intervals, discrimination was observed in each interval and its pattern showed subtle change over cue presentation (Figure 1D). ANOVA for suppression ratios [factors: cue (danger vs. uncertainty vs. safety) and interval (1-10)] found a significant effect of cue (F_2,174 = 489.40, p=3.59×10⁻⁷², η_p² = 0.85, op = 1.00) and a cue x interval interaction (F_18,1566 = 6.21, p=5.90×10⁻¹⁵, η_p² = 0.07, op = 1.00). Suppression ratios are artificially high when calculated in short intervals, making for a poor measure of absolute fear (Figure 1—figure supplement 2). However, this artificial skewing is observed equally to all cues, making suppression ratios in short intervals a valid, relative measure of fear.

vlPAG neurons show differential firing that is maximal to danger

Despite identifying neurons without regard for relative firing to the three cues, differential firing was observed in Onset neurons at the single-unit (Figure 1F) and population (Figure 2A) levels. Onset neurons (n = 29) sharply increased activity during the first 1 s cue interval, with greatest firing to danger, lesser firing to uncertainty, and least firing to safety (Figure 2B, Left). The differential firing pattern diminished as cue presentation proceeded and was absent by the last 1 s cue interval (Figure 2B, Right). The temporal firing pattern (onset → offset) was observed for all trials in the session (Figure 2—figure supplement 1A–D). ANOVA for normalized firing rate (Z-score transformation) for the 29 Onset neurons [data from Figure 2A; factors: cue (danger vs. uncertainty vs. safety) and bin (100 ms: 1 s prior to cue onset through 10 s cue)] revealed main effects of cue and bin (Fs > 9, ps<0.01, η_p² > 0.20, op > 0.95) but most critically, a cue x bin interaction (F_218,6104 = 1.94, p<0.01, η_p² = 0.06, op = 1.00). Consistent with the ANOVA interaction, Onset neurons showed significantly greater firing to danger compared to uncertainty in the first 1 s cue interval (t₂₈ = 4.54, p=9.70×10⁻⁵). While numerically greater firing to uncertainty over safety failed to reach significance in the first 1 s interval (t₂₈ = 1.37, p=0.18), ANOVA restricted to uncertainty and safety (100 ms: 1 s prior to cue onset through first 5 s of the cue) revealed a significant cue x bin interaction (F_59,1652 = 1.50, p=0.01, η_p² = 0.051, op = 1.00). Differential firing was not observed to danger vs. uncertainty (t₂₈ = 1.69, p=0.10) or to uncertainty vs. safety (t₂₈ = 0.60, p=0.55) in the last 1 s cue interval.

Figure 2 with 2 supplements see all

Download asset Open asset

Selective firing was observed in Ramping neurons at the single-unit (Figure 1G) and population (Figure 2C) levels. Ramping neurons (n = 14) did not increase firing to any cue during the first 1 s cue interval (Figure 2D, Left). Instead, activity ramped over cue presentation with greatest firing observed during the last 1 s cue interval (Figure 2D, Right). Ramping activity was most apparent to danger, intermediate to uncertainty, and absent to safety. The temporal firing pattern (onset → offset) was consistent across trials (Figure 2—figure supplement 1E–H). ANOVA for normalized firing rate for the 14 Ramping neurons [data from Figure 2C; factors: cue (danger vs. uncertainty vs. safety) and bin (100 ms: 1 s prior to cue onset through 10 s cue)] revealed main effects of cue and bin (Fs >60, ps <0.01, η_p² > 0.40, op = 1.00) and a cue x bin interaction (F_218,2834 = 4.33, p<0.01, η_p² = 0.24, op = 1.00). Illustrative of the ANOVA interaction, Ramping neurons showed no significant differences in firing to danger vs. uncertainty (t₁₃ = 0.62, p=0.55) or uncertainty vs. safety (t₁₃ = 0.24, p=0.81) in the first 1 s cue interval. However, differential firing to danger vs. uncertainty (t₁₃ = 3.17, p=7.41×10⁻³), and uncertainty vs. safety (t₁₃ = 8.26, p=2.00×10⁻⁶), was observed in the last 1 s cue interval.

Ramping activity to danger and uncertainty could be the product of the time at which activity began to increase, or the rate of increase. We performed a two-tailed t-test for population firing to danger vs. safety (Figure 2E, red line) and uncertainty vs. safety (Figure 2E, purple line) in a 1 s window, starting with cue onset. We slid the 1 s window across the 10 s cue in 100 ms increments, to reveal the time in which danger and uncertainty population firing departed from safety. We then calculated the rate of firing increase from the departure window to the last 1 s interval for danger and uncertainty. Differential firing was determined by the time of departure from safety, as opposed to the rate of increase. Ramping activity to danger emerged earlier (Figure 2E; 2.8 s following cue onset for p<0.05; 5.8 s for full Bonferroni correction, p<0.00025 [0.05/200]) than ramping activity to uncertainty (5.7 s following cue onset; 6.5 s for full Bonferroni correction). Change in firing rate did not differ between danger and uncertainty (Figure 2E, Inset).

It is possible that while Onset and Ramping neurons are distinct, activity of one population drives the other. We identified four Onset-Ramping pairs recorded in the same session. We then asked if trial-by-trial variations in firing over each 1 s cue interval were negatively correlated for any of the four pairs. We failed to uncover such a relationship (Figure 3), demonstrating that Onset and Ramping populations are distinct and independent.

Figure 3

Download asset Open asset

Population biases are evident in vlPAG single-units

To demonstrate that Onset population activity was the result of a consistent bias across neurons, we directly compared single-unit firing to cue pairs. Danger and uncertainty firing were correlated, and single-units were biased towards greater firing to danger (Figure 4A). Uncertainty and safety firing were also correlated; however, the single-unit bias towards greater firing to uncertainty was not significant (Figure 4B). Underscoring their specificity to cue onset, single-units were biased towards greater firing to danger in the first 1 s cue interval compared to the last interval, and there was no correlation between firing in the two epochs (Figure 4C). Examining relative cue firing for each Onset neuron in the first 1 s interval revealed the most common pattern to be: danger > uncertainty > safety (n = 14). This was the only pattern to contain more units than would be expected than chance (Figure 4D).

Figure 4

Download asset Open asset

We performed the same analysis for the Ramping population, only for the last 1 s interval. Ramping neurons showed a selective firing pattern. A significant correlation between firing to danger and uncertainty was observed, along with a single-unit bias towards greater firing to danger (Figure 4E). Only now, there was no correlation between uncertainty and safety firing, but a consistent bias towards greater uncertainty firing (Figure 4F). Ramping single-units were biased towards danger activity in the last 1 s cue interval, and there was no correlation between firing in the two epochs (Figure 4G). The most common firing pattern in the last 1 s interval was: danger > uncertainty > safety (n = 11). This was the only pattern to contain more units than expected by chance (Figure 4H).

VlPAG activity is greatest to danger, the cue most strongly suppressing rewarded nose poking. It is therefore possible that Onset and Ramping neurons are simply responsive to nose poke cessation. To examine this possibility, we identified naturally occurring periods of nose poke cessation in inter-trial intervals, when no cues were presented. This analysis found no meaningful changes in Onset or Ramping activity during periods of nose poke cessation (Figure 2—figure supplement 2), demonstrating activity patterns are specific to cue-induced suppression of nose poking.

At first glance, the firing patterns of Onset and Ramping neurons appear to support the prevailing hypothesis that vlPAG neurons signal fear output. Differential fear (Figure 1C) and differential firing (Figure 2B,D) show the same general pattern: danger > uncertainty > safety. However, closer inspection reveals that relative differences in fear do not match relative differences in firing. Rats showed robust discrimination between uncertainty and safety, regardless of the temporal resolution with which fear was measured (Figure 1C,D). Yet, robust differential firing to uncertainty and safety was modest in the Onset population (Figure 2B, left; Figure 4B). The Ramping population showed stronger differential firing between uncertainty and safety (Figure 2D, right; Figure 4F), but this pattern did not emerge in until the end of the cue. Fear discrimination was reliably detected in the first 1 s interval (Figure 1D), indicating that Ramping neurons cannot organize fear output early in cue presentation. While inconsistencies between fear output and neural activity are evident for the Onset and Ramping populations, the analyses conducted so far cannot conclusively test the relative contribution of threat probability and fear output to vlPAG single-unit activity.

Onset neurons signal threat probability

To formally test the degree to which vlPAG activity is captured by fear output and threat probability, we used simultaneous linear regression for single-unit firing (Figure 5; see Materials and methods for example regression input). 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 10 s cue. For each trial, we calculated fear on two time scales: total fear (suppression ratio for the entire 10 s cue) and interval fear (suppression ratio for the specific 1 s interval). The corresponding shock probability was assigned for each trial (danger = 1.00, uncertainty = 0.375 and safety = 0.00). Total fear, interval fear and threat probability were used as regressors to explain trial-by-trial variance in single-unit firing. Statistical output was a beta coefficient quantifying the strength (|>0| = stronger) and direction (>0 = positive) of the predictive relationship between each regressor and single-unit firing. Beta coefficients for single-units comprising the Onset and Ramping populations were subjected to ANOVA with regressor (total fear vs. interval fear vs. probability) and interval (1 s intervals for 10 s cue) as factors. This allowed us to determine the relative contribution of total fear, interval fear and probability to single-unit firing over the course of cue presentation.

Figure 5 with 2 supplements see all

Download asset Open asset

The results of primary interest for the Onset population came from the first 1 s cue interval, when activity was highest and differential firing was observed. Linear regression unequivocally revealed that Onset single-unit activity was captured by threat probability (Figure 5A). The beta coefficient for the probability regressor was positive and significant, exceeding the beta coefficient for either measure of fear output – neither of which differed from zero. The population bias was observed across Onset neurons. Single-unit beta coefficients were positively biased for threat probability, but not for either measure of fear output (Figure 5B,C).

Examining the entirety of cue presentation, threat probability signaling was highest in the first interval, persisted several more seconds and diminished (Figure 5D). Total fear or interval fear did not account for variance in single-unit firing at any interval. Consistent with this description, ANOVA for beta coefficient with factors of regressor and interval (10 total) revealed a main effect of regressor (F_2,56 = 7.16, p<0.01, η_p² = 0.20, op = 0.92) and a regressor x interval interaction (F_18,504 = 2.38, p<0.01, η_p² = 0.08, op = 0.99). Identical results were obtained when probability was compared to total fear or interval fear separately. In fact, significance for fear output could only be found if total fear was the only regressor used in the analysis – producing a result very similar to that of a previous study (Watson et al., 2016). Even then, the predictive relationship was weaker than that of probability (Figure 5—figure supplement 1, A–E).

The threat probability regressor in the above analyses utilized the actual shock probability assigned to each cue. Of course, the subjects had no a priori knowledge of shock probability assignments. It is then possible that vlPAG activity is ‘tuned’ to an alternative shock probability. To examine this, we performed single-unit linear regression for normalized firing in the first 1 s cue interval 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). The mean beta coefficient for each of the nine increments is plotted as a threat-tuning curve for the Onset population (Figure 5E). The beta coefficient resulting from regression using the actual shock probability (uncertainty = 0.375), was the ‘peak’ of the tuning curve. The probabilities with the next highest beta coefficients were those flanking 0.375. Beta coefficients dropped off rapidly as the uncertainty assignment moved to the extremes.

This result is particularly revealing for the analysis in which the uncertainty assignment was 0.000 (first data point on the curve Figure 5E). Onset neurons showed high firing to danger but lower and more similar firing to uncertainty and safety, leaving open the possibility that Onset neurons signal a more binary output (danger = 1.000) > (uncertainty and safety = 0.000). However, the actual uncertainty assignment (0.375) captured single-unit activity better than the binary assignment (0.000) in the first 1 s interval and across the remainder of cue presentation (Figure 5—figure supplement 2).

Ramping neurons signal threat probability through delay until shock receipt

Onset neurons are tuned to initial cue presentation, providing a rapid estimate of threat probability. By contrast, Ramping neurons are initially unresponsive, but gradually increase activity over cue presentation. Ramping neurons may continue to signal threat probability through the delay period, up until foot shock receipt. We first examined population activity during the five seconds following cue offset (2 s delay, 0.5 s shock and 2.5 s post-shock). Activity during the 500 ms shock period was not analyzed because it may have been contaminated by electrical artifacts. Onset neurons showed negligible delay activity and little to no post-shock activity (Figure 6A). Ramping neurons continued firing to danger and uncertainty throughout the delay period, but this firing diminished shortly after shock presentation (Figure 6B). ANOVA for normalized firing rate [factors: trial type (danger vs. uncertainty shock vs. uncertainty omission vs. safety) and bin (100 ms: 5 s post cue)] revealed no main effects or interactions for Onset neurons (F_s <2.3, ps >0.09). By contrast, identical ANOVA revealed main effects of trial and bin (Fs >3, ps <0.01), as well as a trial x bin interaction for Ramping neurons (F_132,1716 = 3.14, p<0.01, η_p² = 0.19, op >0.99). The Onset and Ramping firing patterns differed from one another. ANOVA with neuron type as a factor (Onset vs. Ramping) found significant interactions for trial type x neuron type, bin x neuron type, and trial type x bin x neuron type (Fs >2.50, ps <0.01, η_p² > 0.19, op >0.94).

Figure 6

Download asset Open asset

Observed differences in neural activity suggest that Ramping neurons maintain threat probability signaling throughout the delay period and this signal abruptly decreases following foot shock presentation. Single-unit regression was used to determine whether trial-by-trial, post-cue firing was best described by interval fear or probability (total fear sampled only the prior 10 s cue period and was omitted). Regression was performed in 500 ms intervals for the 5 s post-cue period minus the shock interval (nine total intervals). Onset neurons did not signal interval fear or probability at any time following cue offset (Figure 6C). ANOVA revealed no main effect of regressor (F_1,28 = 1.02, p=0.32, η_p² = 0.04, op = 0.16) or regressor x interval interaction (F_8,224 = 0.53, p=0.83, η_p² = 0.02, op = 0.24). Illustrative of the lack of information contained in Onset neurons, post hoc comparisons found that no beta coefficient, for any regressor differed from zero. By contrast, Ramping neurons signaled probability throughout the delay period (Figure 6D, first four intervals), which diminished following shock delivery. In support, ANOVA revealed a significant regressor x interval interaction (F_8,104 = 5.20, p<0.01, η_p² = 0.29, op = 0.99).

Threat probability signaling prior to shock delivery was abruptly halted following shock delivery in Ramping neurons, but not Onset neurons. To reveal the degree to which this occurred, we constructed correlation matrices using the probability beta coefficient from each of the nine relevant intervals (4 delay and five post-shock). Of greatest interest were the twenty comparisons between the probability beta coefficient for the four delay and five post-shock intervals (Figure 6E & F; bottom-left quadrant). Significant between-interval correlations for the probability beta coefficient were observed for 13/20 comparisons in Onset neurons (Figure 6E), indicating that threat probability signaled during the delay tended to persist following foot shock. By contrast, significant correlations were found for only 4/20 comparisons in Ramping neurons (Figure 6F); and the proportion of intervals showing a significant correlation differed for the Onset and Ramping neurons (χ² = 8.08, p=4.5×10⁻³). Bonferroni correction (0.05/5 = 0.01, five tests per interval) found significant correlations for 6/20 Onset comparisons, 1/20 Ramping comparisons, and these proportions significantly differed (χ² = 4.22, p=0.039).

We recorded vlPAG single-unit activity while rats discriminated between danger, uncertainty and safety. Consistent with previous reports (Tovote et al., 2016; Watson et al., 2016; Ozawa et al., 2017), we found a population of Onset neurons with short-latency excitation to danger. Consistent with the most recent report (Ozawa et al., 2017), we found a Ramping population that increased activity over danger presentation. Onset activity reflected an estimate of threat probability, invariant of fear output. Ramping activity reflected threat probability and fear output, though probability emerged earlier and was stronger overall. While vlPAG signals for fear output could potentially emerge at the ensemble level (Jones et al., 2007; Zhou et al., 2018), these multi-unit codes would be composed of single-units primarily signaling threat probability. Activity reflecting fear output may be found in other vlPAG populations, such as neurons showing inhibition of firing to cues (Tovote et al., 2015), or in non-cue-responsive single-units (Insanally et al., 2019). Yet, this would still mean that signals for threat probability and fear output co-exist in the vlPAG.

It is important to note that these results are correlative and that Onset neuron activity may not play a causal role in fear output. Previous work has found that short-latency, excitatory responses to danger are largely observed in vlPAG VGlut2 +neurons, and that excitation of this population is sufficient to produce freezing (Tovote et al., 2016). Though all Onset neurons fell into the low firing cluster, we cannot conclude these were VGlut2 +neurons. Moreover, inhibition of vlPAG GABA neurons also promotes freezing (Tovote et al., 2016). These neurons have comparatively higher baseline firing rates and respond to danger cues through inhibition of neural activity. A causal, vlPAG signal for fear output may be observed in a GABAergic/cue-inhibited population, in concert with or independent of a glutamatergic/cue-excited population. At the same time, it is clear how the observed Onset signal could play a causal role in fear output. Blocking vlPAG Onset activity to danger and uncertainty would equate neural activity to that for safety, removing the threat impetus for suppression of reward seeking, freezing and related defensive behaviors.

Before further discussing implications, we consider some alternative accounts for the observed firing patterns. Perhaps vlPAG neurons signaling fear output are anatomically distinct from those recorded here. We intentionally recorded from caudal vlPAG, the subregion preferentially activated by fearful contexts in rats (Carrive et al., 1997). VlPAG manipulations that disrupt fear-related behaviors typically include this more caudal region (De Oca et al., 1998), and high-resolution functional magnetic resonance imaging reveals caudal vlPAG activation specific to aversive stimuli in humans (Satpute et al., 2013). We observed threat probability signaling in all recording locations (bregma −7.62 → −7.98). The vlPAG stretches ~0.7 mm beyond our most caudal recording site. It is therefore possible that neurons signaling fear output are restricted to the extreme caudal vlPAG.

Maybe the vlPAG signals fear output, but we did not measure the relevant output. Previous studies have failed to find robust relationships between vlPAG activity and cued freezing. Here we used conditioned suppression of rewarded nose poking to provide an objective measure of fear on two timescales and to perhaps better capture vlPAG activity. This measure of fear did not capture Onset neuron firing, and only partially captured Ramping neuron firing at the end of cue presentation. Further, Onset and Ramping activity were not merely driven by nose poke cessation or withdrawal from the port.

If not freezing, conditioned suppression, or nose poke cessation then perhaps another measure of fear? Danger cues elicit active fear responses: escape-like behaviors such as darting (Greiner et al., 2018; Gruene et al., 2015). However, darting is prevalent in females, but less so in males. Further, the males used in this study had extensive experience with fear discrimination, and at no point was escape from the foot shock possible. Danger cues increase arterial blood pressure and reduce heart rate, however, neither of these abilities has been linked to vlPAG function (Helmstetter and Tershner, 1994; LeDoux et al., 1988; Wilson and Kapp, 1994). Danger cues also enhance startle, perhaps more in line with Onset population firing. Yet, dorsal, rather than ventrolateral, PAG subregions have been implicated in startle behavior (Walker et al., 1997; Zhao et al., 2009). Of course, many other fear behaviors are possible: piloerection, hyperventilation, changes in body temperature, vocalization, etc. (Kim et al., 2010; Iwata and LeDoux, 1988; Gallego et al., 2001; Vianna and Carrive, 2005). Problematic is that most fear behaviors are initiated at cue onset and maintained until the aversive event occurs. We did not observe a substantial population of neurons with these temporal firing characteristics, making the vlPAG a poor candidate for sustained fear output. Above all, any potential behavior signaled by Onset neurons must closely match shock probability, confounding this behavior signal with probability itself.

The present findings are best understood through comparison to the account of vlPAG function outlined in the predatory imminence continuum (PIC), a highly influential theory of defensive behavior (Fanselow and Lester, 1988). Organizing features of the PIC are time and degree of threat. As predation becomes more imminent (pre-encounter → post-encounter → circa-strike), the form and intensity of defensive behaviors change. Cued fear is argued to capture post-encounter defenses: immobility elicited when predators are nearby. In the neural instantiation of PIC, the amygdala integrates information about environmental stimuli (auditory cues here), nociceptive information (foot shock) and time to produce a signal for degree of threat (Fanselow and Lester, 1988). This amygdala-derived signal is relayed to the vlPAG to organize fear output (Fanselow, 1991; Fanselow, 1994). Implicit in the PIC model, is that the vlPAG does not contain information about degree of threat – only the resultant fear output. Yet, we find that single vlPAG neurons contain detailed information about time and degree of threat.

These results require more careful consideration of the role of the vlPAG in the fear circuit and the PIC. Rather than signaling fear output, vlPAG Onset neurons signal threat probability. This information could be used to organize a variety of fear responses, but these neurons do not intrinsically signal fear output. For example, vlPAG projections to the central amygdala (CeA) and rostral ventromedial medulla (RVM) could inform fear output via freezing (Vianna et al., 2008), while projections to midline/intralaminar thalamus could rapidly relay threat probability estimates to a larger fear network (basolateral amygdala, prelimibic cortex, infralimbic cortex, insular cortex, etc.) (Vertes et al., 2015; Krout and Loewy, 2000; Buchanan and Thompson, 1994; Sengupta and McNally, 2014), promoting a variety of threat-related processes (Faull et al., 2016). In this way, rapid threat probability estimates generated by vlPAG Onset neurons may be more akin to brainstem-derived signals for rapid detection of visual threats, which are distributed to a wider brain network (Liddell et al., 2005).

Neurons responsive toward the end of cue presentation were more heterogeneous, in terms of their baseline firing rate and their signaling. Ramping neurons prioritized threat probability but also signaled fear output. However, Ramping activity could not drive fear output in full. Differential fear to safety, uncertainty and danger was observed even in the first second of cue presentation, when these neurons were unresponsive. Unlike Onset neurons, Ramping neurons showed selective firing throughout the delay period, only diminishing once foot shock had been delivered. Ramping neurons may provide a threat probability estimate that increases as threat draws nearer and peaks when threat is imminent. Ramping neurons may help sustain threat estimates in the absence of explicit stimuli, such as in trace conditioning (McEchron et al., 1998; Büchel et al., 1999), or estimate more precisely when the noxious event will occur. Shifts toward PAG-centric activity are apparent in humans, as capture becomes imminent (Mobbs et al., 2007) or natural threats draw closer Mobbs et al., 2010; with the caveat that neither of these studies could specify the PAG subregion activated.

Ramping neurons may also provide two forms of feedback: the estimated probability of foot shock and a readout of fear output on that trial. Either type of information could be compared to that trial’s outcome, particularly for uncertainty trials on which shock can be present or absent. This could be used to adjust estimates of threat probability and fear output on future encounters with the cue. This is broadly consistent with a central role for the vlPAG in feedback processes (McNally et al., 2011; Ozawa et al., 2017; Yeh et al., 2018). Even more, this Ramping signal could alter or reduce foot shock processing through descending control of dorsal horn nociceptive inputs via endogenous opioid circuits. This behavioral phenomenon, conditioned analgesia (MacLennan et al., 1980; Fanselow and Bolles, 1979; Chance et al., 1978), may be related to a more general phenomenon of placebo analgesia. VlPAG activation is consistently observed in studies of placebo analgesia (Tracey, 2010; Eippert et al., 2009; Wager and Atlas, 2015; Petrovic et al., 2002) and our observation of increasing neural activity toward presentation of a noxious event may provide a suitable neural substrate for this process. Perhaps most remarkable is that although independent, between the Onset and Ramping populations, the vlPAG contains an estimate of threat probability from the time of first encounter up through the noxious event itself. The vlPAG may not signal fear output per se, but is rich with information that would inform a variety of fear processes and behaviors.

If fear output via conditioned suppression is non-linear, and vlPAG activity scales linearly to threat probability, how does the vlPAG fit into the fear circuit? An ultimate explanation for non-linear fear output may be that threat systems evolved to avoid predation, not to precisely match degree of defensive behavior to threat probability. Erring on the side of greater fear to uncertain threats may promote survival. A proximate explanation may be that fear output is the summed product of multiple threat signals. VlPAG output to the RVM may instruct fear output to match threat probability. If this were the only threat signal, then vlPAG activity would in fact reflect fear output. We speculate that additional threat signals govern fear output. For example, neurons in the retrorubral field (RRF) project to the RVM, as well as the CeA (Zahm and Trimble, 2008; Deutch et al., 1988; Von Krosigk and Smith, 1991). Preliminary data from our laboratory suggest that RRF neurons are primarily responsive to danger and uncertainty, but weight uncertainty similarly to danger (danger = uncertainty > safety). RRF neurons may signal threat probability plus uncertainty-induced stress, or may favor cue-shock contiguity over contingency (Rescorla, 1967). Summation of vlPAG-derived and RRF-derived threat signals by RVM neurons would produce a non-linear fear output like that observed in our discrimination procedure.

The results pose questions about the specific relationship between the vlPAG and the CeA. VlPAG threat probability signals may be trained up by the CeA, but become CeA-independent with sufficient training (Ozawa et al., 2017). Consistent with this interpretation, the CeA is essential to the acquisition of conditioned suppression with limited training, but that more extended training mitigates the effects of CeA lesions (Lee et al., 2005; McDannald, 2010). This is not to say that we would expect the CeA to become inessential following extensive fear discrimination training. Updating threat probability should occur when environmental stimuli become more or less predictive of noxious events. We anticipate the CeA is essential to updating vlPAG threat probability signaling (McNally et al., 2011; Ozawa et al., 2017).

It is near universally accepted that the amygdala is a key node of dysfunction in stress (Rauch et al., 2000) and anxiety disorders (Etkin and Wager, 2007). This may be driven in part by technical considerations: whole-brain fMRI can detect amygdala BOLD signals (Johnstone et al., 2005), while detecting subregion-specific PAG BOLD signals requires a more deliberate approach (Satpute et al., 2013). Perhaps the primary intellectual driver is that the amygdala is theorized to be a privileged cite of integration/learning in the fear circuit (Mahan and Ressler, 2012; Admon et al., 2013). The present findings illustrate that the amygdala is not privileged in this regard, and mark the vlPAG as likely node of dysfunction in psychiatric disorders of stress and anxiety. Appreciation for the vlPAG as a site of integration will hasten mapping of a more complete fear circuit. Deliberate study of vlPAG function (Arico et al., 2017; Assareh et al., 2017; Rozeske et al., 2018) and dysfunction in psychiatric disease (Yeh et al., 2018), will be essential to developing effective therapies for disorders characterized by exaggerated threat estimation and aberrant fear.

Single-unit data are publicly available on CRCNS (http://crcns.org/data-sets/brainstem/pag-1), under the doi: 10.6080/K0R49P0V. Users must first create a free account (https://crcns.org/register) before they can download the datasets from the site.

1. 1. Wright KM
   2. McDannald MA

   (2019) CRCNS

   Data from: Ventrolateral periaqueductal gray neurons prioritize threat probability over fear output.

   https://doi.org/10.6080/K0R49P0V

1. 1. Admon R
   2. Milad MR
   3. Hendler T

   (2013) A causal model of post-traumatic stress disorder: disentangling predisposed from acquired neural abnormalities

   Trends in Cognitive Sciences 17:337–347.

   https://doi.org/10.1016/j.tics.2013.05.005

   • PubMed
   • Google Scholar
2. 1. Anglada-Figueroa D
   2. Quirk GJ

   (2005) Lesions of the basal amygdala block expression of conditioned fear but not extinction

   Journal of Neuroscience 25:9680–9685.

   https://doi.org/10.1523/JNEUROSCI.2600-05.2005

   • PubMed
   • Google Scholar
3. 1. Arico C
   2. Bagley EE
   3. Carrive P
   4. Assareh N
   5. McNally GP

   (2017) Effects of chemogenetic excitation or inhibition of the ventrolateral periaqueductal gray on the acquisition and extinction of pavlovian fear conditioning

   Neurobiology of Learning and Memory 144:186–197.

   https://doi.org/10.1016/j.nlm.2017.07.006

   • PubMed
   • Google Scholar
4. 1. Arico C
   2. McNally GP

   (2014) Opioid receptors regulate blocking and overexpectation of fear learning in conditioned suppression

   Behavioral Neuroscience 128:199–206.

   https://doi.org/10.1037/a0036133

   • PubMed
   • Google Scholar
5. 1. Assareh N
   2. Sarrami M
   3. Carrive P
   4. McNally GP

   (2016) The organization of defensive behavior elicited by optogenetic excitation of rat lateral or ventrolateral periaqueductal gray

   Behavioral Neuroscience 130:406–414.

   https://doi.org/10.1037/bne0000151

   • PubMed
   • Google Scholar
6. 1. Assareh N
   2. Bagley EE
   3. Carrive P
   4. McNally GP

   (2017) Brief optogenetic inhibition of rat lateral or ventrolateral periaqueductal gray augments the acquisition of pavlovian fear conditioning

   Behavioral Neuroscience 131:454–459.

   https://doi.org/10.1037/bne0000217

   • PubMed
   • Google Scholar
7. 1. Bandler R
   2. Depaulis A
   3. Vergnes M

   (1985) Identification of midbrain neurones mediating defensive behaviour in the rat by microinjections of excitatory amino acids

   Behavioural Brain Research 15:107–119.

   https://doi.org/10.1016/0166-4328(85)90058-0

   • PubMed
   • Google Scholar
8. Book

   1. Bear MF
   2. Connors BW
   3. Paradiso MA

   (2016)

   Neuroscience: Exploring the Brain

   Philadelphia: Wolters Kluwer.

   • Google Scholar
9. 1. Bellgowan PS
   2. Helmstetter FJ

   (1998) The role of mu and kappa opioid receptors within the periaqueductal gray in the expression of conditional hypoalgesia

   Brain Research 791:83–89.

   https://doi.org/10.1016/S0006-8993(98)00057-2

   • PubMed
   • Google Scholar
10. 1. Berg BA
    2. Schoenbaum G
    3. McDannald MA

    (2014) The dorsal raphe nucleus is integral to negative prediction errors in pavlovian fear

    European Journal of Neuroscience 40:3096–3101.

    https://doi.org/10.1111/ejn.12676

    • PubMed
    • Google Scholar
11. 1. Bouton ME
    2. Bolles RC

    (1980) Conditioned fear assessed by freezing and by the suppression of three different baselines

    Animal Learning & Behavior 8:429–434.

    https://doi.org/10.3758/BF03199629

    • Google Scholar
12. 1. Buchanan SL
    2. Thompson RH

    (1994) Neuronal activity in the midline thalamic nuclei during pavlovian heart rate conditioning

    Brain Research Bulletin 35:237–240.

    https://doi.org/10.1016/0361-9230(94)90128-7

    • PubMed
    • Google Scholar
13. 1. Büchel C
    2. Dolan RJ
    3. Armony JL
    4. Friston KJ

    (1999) Amygdala-hippocampal involvement in human aversive trace conditioning revealed through event-related functional magnetic resonance imaging

    The Journal of Neuroscience 19:10869–10876.

    https://doi.org/10.1523/JNEUROSCI.19-24-10869.1999

    • PubMed
    • Google Scholar
14. Book

    1. Carlson NR
    2. Birkett MA

    (2017)

    Physiology of Behavior

    Pearson.

    • Google Scholar
15. 1. Carrive P
    2. Leung P
    3. Harris J
    4. Paxinos G

    (1997) Conditioned fear to context is associated with increased fos expression in the caudal ventrolateral region of the midbrain periaqueductal gray

    Neuroscience 78:165–177.

    https://doi.org/10.1016/S0306-4522(97)83047-3

    • PubMed
    • Google Scholar
16. 1. Chance WT
    2. White AC
    3. Krynock GM
    4. Rosencrans JA

    (1978) Conditional fear-induced antinociception and decreased binding of [3H]N-leu-enkephalin to rat brain

    Brain Research 141:371–374.

    https://doi.org/10.1016/0006-8993(78)90208-1

    • PubMed
    • Google Scholar
17. 1. Davis M

    (2006) Neural systems involved in fear and anxiety measured with fear-potentiated startle

    American Psychologist 61:741–756.

    https://doi.org/10.1037/0003-066X.61.8.741

    • PubMed
    • Google Scholar
18. 1. De Oca BM
    2. DeCola JP
    3. Maren S
    4. Fanselow MS

    (1998) Distinct regions of the periaqueductal gray are involved in the acquisition and expression of defensive responses

    The Journal of Neuroscience 18:3426–3432.

    https://doi.org/10.1523/JNEUROSCI.18-09-03426.1998

    • PubMed
    • Google Scholar
19. 1. Dejean C
    2. Courtin J
    3. Rozeske RR
    4. Bonnet MC
    5. Dousset V
    6. Michelet T
    7. Herry C

    (2015) Neuronal circuits for fear expression and recovery: recent advances and potential therapeutic strategies

    Biological Psychiatry 78:298–306.

    https://doi.org/10.1016/j.biopsych.2015.03.017

    • PubMed
    • Google Scholar
20. 1. Deutch AY
    2. Goldstein M
    3. Baldino F
    4. Roth RH

    (1988) Telencephalic projections of the A8 dopamine cell group

    Annals of the New York Academy of Sciences 537:27–50.

    https://doi.org/10.1111/j.1749-6632.1988.tb42095.x

    • PubMed
    • Google Scholar
21. 1. DiLeo A
    2. Wright KM
    3. McDannald MA

    (2016) Subsecond fear discrimination in rats: adult impairment in adolescent heavy alcohol drinkers

    Learning & Memory 23:618–622.

    https://doi.org/10.1101/lm.043257.116

    • PubMed
    • Google Scholar
22. 1. Duvarci S
    2. Pare D

    (2014) Amygdala microcircuits controlling learned fear

    Neuron 82:966–980.

    https://doi.org/10.1016/j.neuron.2014.04.042

    • PubMed
    • Google Scholar
23. 1. Eippert F
    2. Bingel U
    3. Schoell ED
    4. Yacubian J
    5. Klinger R
    6. Lorenz J
    7. Büchel C

    (2009) Activation of the opioidergic descending pain control system underlies placebo analgesia

    Neuron 63:533–543.

    https://doi.org/10.1016/j.neuron.2009.07.014

    • PubMed
    • Google Scholar
24. 1. Estes WK
    2. Skinner BF

    (1941) Some quantitative properties of anxiety

    Journal of Experimental Psychology 29:390–400.

    https://doi.org/10.1037/h0062283

    • Google Scholar
25. 1. Etkin A
    2. Wager TD

    (2007) Functional neuroimaging of anxiety: a meta-analysis of emotional processing in PTSD, social anxiety disorder, and specific phobia

    American Journal of Psychiatry 164:1476–1488.

    https://doi.org/10.1176/appi.ajp.2007.07030504

    • PubMed
    • Google Scholar
26. Book

    1. Fanselow MS
    2. Lester LS

    (1988)

    A functional behavioristic approach to aversively motivated behavior: predatory imminence as a determinant of the topography of defensive behavior

    In: Bolles R. C, Beecher M. D, editors. In Evolution and Learning. New Jersey: Erlbaum. pp. 185–212.

    • Google Scholar
27. 1. Fanselow MS

    (1991)

    The midbrain periaqueductal gray as a coordinator of action in response to fear and anxiety

    Midbrain Periaqueductal Gray Matter 213:151–173.

    • Google Scholar
28. 1. Fanselow MS

    (1993)

    The periaqueductal gray and the organization of defensive behavior

    Aggressive Behavior 19:18–19.

    • Google Scholar
29. 1. Fanselow MS

    (1994) Neural organization of the defensive behavior system responsible for fear

    Psychonomic Bulletin & Review 1:429–438.

    https://doi.org/10.3758/BF03210947

    • PubMed
    • Google Scholar
30. 1. Fanselow MS
    2. Bolles RC

    (1979) Triggering of the endorphin analgesic reaction by a cue previously associated with shock: reversal by naloxone

    Bulletin of the Psychonomic Society 14:88–90.

    https://doi.org/10.3758/BF03329408

    • Google Scholar
31. 1. Fanselow MS
    2. LeDoux JE

    (1999) Why we think plasticity underlying pavlovian fear conditioning occurs in the basolateral amygdala

    Neuron 23:229–232.

    https://doi.org/10.1016/S0896-6273(00)80775-8

    • PubMed
    • Google Scholar
32. 1. Faull OK
    2. Jenkinson M
    3. Ezra M
    4. Pattinson KTs

    (2016) Conditioned respiratory threat in the subdivisions of the human periaqueductal gray

    eLife 5:e12047.

    https://doi.org/10.7554/eLife.12047

    • PubMed
    • Google Scholar
33. 1. Gallego J
    2. Nsegbe E
    3. Durand E

    (2001) Learning in respiratory control

    Behavior Modification 25:495–512.

    https://doi.org/10.1177/0145445501254002

    • PubMed
    • Google Scholar
34. Preprint

    1. Greiner EM
    2. Norris MR
    3. Müller I
    4. Ng KH
    5. Sangha S

    (2018) Sex differences in fear regulation and reward seeking behaviors in a fear-safety-reward discrimination task

    bioRxiv.

    https://doi.org/10.1101/390377

    • Google Scholar
35. 1. Gruene TM
    2. Flick K
    3. Stefano A
    4. Shea SD
    5. Shansky RM

    (2015) Sexually divergent expression of active and passive conditioned fear responses in rats

    eLife 4:e11352.

    https://doi.org/10.7554/eLife.11352

    • PubMed
    • Google Scholar
36. 1. Helmstetter FJ
    2. Tershner SA

    (1994) Lesions of the periaqueductal gray and rostral ventromedial medulla disrupt antinociceptive but not cardiovascular aversive conditional responses

    The Journal of Neuroscience 14:7099–7108.

    https://doi.org/10.1523/JNEUROSCI.14-11-07099.1994

    • PubMed
    • Google Scholar
37. 1. Insanally MN
    2. Carcea I
    3. Field RE
    4. Rodgers CC
    5. DePasquale B
    6. Rajan K
    7. DeWeese MR
    8. Albanna BF
    9. Froemke RC

    (2019) Spike-timing-dependent ensemble encoding by non-classically responsive cortical neurons

    eLife 8:e42409.

    https://doi.org/10.7554/eLife.42409

    • PubMed
    • Google Scholar
38. 1. Iwata J
    2. LeDoux JE

    (1988) Dissociation of associative and nonassociative concomitants of classical fear conditioning in the freely behaving rat

    Behavioral Neuroscience 102:66–76.

    https://doi.org/10.1037/0735-7044.102.1.66

    • PubMed
    • Google Scholar
39. 1. Johansen JP
    2. Tarpley JW
    3. LeDoux JE
    4. Blair HT

    (2010) Neural substrates for expectation-modulated fear learning in the amygdala and periaqueductal gray

    Nature Neuroscience 13:979–986.

    https://doi.org/10.1038/nn.2594

    • PubMed
    • Google Scholar
40. 1. Johnstone T
    2. Somerville LH
    3. Alexander AL
    4. Oakes TR
    5. Davidson RJ
    6. Kalin NH
    7. Whalen PJ

    (2005) Stability of amygdala BOLD response to fearful faces over multiple scan sessions

    NeuroImage 25:1112–1123.

    https://doi.org/10.1016/j.neuroimage.2004.12.016

    • PubMed
    • Google Scholar
41. 1. Jones LM
    2. Fontanini A
    3. Sadacca BF
    4. Miller P
    5. Katz DB

    (2007) Natural stimuli evoke dynamic sequences of states in sensory cortical ensembles

    PNAS 104:18772–18777.

    https://doi.org/10.1073/pnas.0705546104

    • PubMed
    • Google Scholar
42. 1. Kim JJ
    2. Rison RA
    3. Fanselow MS

    (1993) Effects of Amygdala, Hippocampus, and periaqueductal gray lesions on short- and long-term contextual fear

    Behavioral Neuroscience 107:1093–1098.

    https://doi.org/10.1037/0735-7044.107.6.1093

    • PubMed
    • Google Scholar
43. 1. Kim EJ
    2. Kim ES
    3. Covey E
    4. Kim JJ

    (2010) Social transmission of fear in rats: the role of 22-kHz ultrasonic distress vocalization

    PLOS ONE 5:e15077.

    https://doi.org/10.1371/journal.pone.0015077

    • PubMed
    • Google Scholar
44. 1. Koutsikou S
    2. Crook JJ
    3. Earl EV
    4. Leith JL
    5. Watson TC
    6. Lumb BM
    7. Apps R

    (2014) Neural substrates underlying fear-evoked freezing: the periaqueductal grey-cerebellar link

    The Journal of Physiology 592:2197–2213.

    https://doi.org/10.1113/jphysiol.2013.268714

    • PubMed
    • Google Scholar
45. 1. Krout KE
    2. Loewy AD

    (2000) Periaqueductal gray matter projections to midline and intralaminar thalamic nuclei of the rat

    The Journal of Comparative Neurology 424:111–141.

    https://doi.org/10.1002/1096-9861(20000814)424:1<111::AID-CNE9>3.0.CO;2-3

    • PubMed
    • Google Scholar
46. 1. LeDoux JE
    2. Iwata J
    3. Cicchetti P
    4. Reis DJ

    (1988) Different projections of the central amygdaloid nucleus mediate autonomic and behavioral correlates of conditioned fear

    The Journal of Neuroscience 8:2517–2529.

    https://doi.org/10.1523/JNEUROSCI.08-07-02517.1988

    • PubMed
    • Google Scholar
47. 1. Lee JL
    2. Dickinson A
    3. Everitt BJ

    (2005) Conditioned suppression and freezing as measures of aversive pavlovian conditioning: effects of discrete amygdala lesions and overtraining

    Behavioural Brain Research 159:221–233.

    https://doi.org/10.1016/j.bbr.2004.11.003

    • PubMed
    • Google Scholar
48. 1. Liddell BJ
    2. Brown KJ
    3. Kemp AH
    4. Barton MJ
    5. Das P
    6. Peduto A
    7. Gordon E
    8. Williams LM

    (2005) A direct brainstem-amygdala-cortical 'alarm' system for subliminal signals of fear

    NeuroImage 24:235–243.

    https://doi.org/10.1016/j.neuroimage.2004.08.016

    • PubMed
    • Google Scholar
49. 1. Liebman JM
    2. Mayer DJ
    3. Liebeskind JC

    (1970) Mesencephalic central gray lesions and fear-motivated behavior in rats

    Brain Research 23:353–370.

    https://doi.org/10.1016/0006-8993(70)90062-4

    • PubMed
    • Google Scholar
50. 1. MacLennan AJ
    2. Jackson RL
    3. Maier SF

    (1980) Conditioned analgesia in the rat

    Bulletin of the Psychonomic Society 15:387–390.

    https://doi.org/10.3758/BF03334567

    • Google Scholar
51. 1. Mahan AL
    2. Ressler KJ

    (2012) Fear conditioning, synaptic plasticity and the amygdala: implications for posttraumatic stress disorder

    Trends in Neurosciences 35:24–35.

    https://doi.org/10.1016/j.tins.2011.06.007

    • PubMed
    • Google Scholar
52. 1. Maren S
    2. Phan KL
    3. Liberzon I

    (2013) The contextual brain: implications for fear conditioning, extinction and psychopathology

    Nature Reviews Neuroscience 14:417–428.

    https://doi.org/10.1038/nrn3492

    • PubMed
    • Google Scholar
53. 1. McDannald MA

    (2010) Contributions of the amygdala central nucleus and ventrolateral periaqueductal grey to freezing and instrumental suppression in pavlovian fear conditioning

    Behavioural Brain Research 211:111–117.

    https://doi.org/10.1016/j.bbr.2010.03.020

    • PubMed
    • Google Scholar
54. 1. McDannald MA
    2. Galarce EM

    (2011) Measuring pavlovian fear with conditioned freezing and conditioned suppression reveals different roles for the basolateral amygdala

    Brain Research 1374:82–89.

    https://doi.org/10.1016/j.brainres.2010.12.050

    • PubMed
    • Google Scholar
55. 1. McEchron MD
    2. Bouwmeester H
    3. Tseng W
    4. Weiss C
    5. Disterhoft JF

    (1998) Hippocampectomy disrupts auditory trace fear conditioning and contextual fear conditioning in the rat

    Hippocampus 8:638–646.

    https://doi.org/10.1002/(SICI)1098-1063(1998)8:6<638::AID-HIPO6>3.0.CO;2-Q

    • PubMed
    • Google Scholar
56. 1. McNally GP
    2. Johansen JP
    3. Blair HT

    (2011) Placing prediction into the fear circuit

    Trends in Neurosciences 34:283–292.

    https://doi.org/10.1016/j.tins.2011.03.005

    • PubMed
    • Google Scholar
57. 1. Mobbs D
    2. Petrovic P
    3. Marchant JL
    4. Hassabis D
    5. Weiskopf N
    6. Seymour B
    7. Dolan RJ
    8. Frith CD

    (2007) When fear is near: threat imminence elicits prefrontal-periaqueductal gray shifts in humans

    Science 317:1079–1083.

    https://doi.org/10.1126/science.1144298

    • PubMed
    • Google Scholar
58. 1. Mobbs D
    2. Yu R
    3. Rowe JB
    4. Eich H
    5. FeldmanHall O
    6. Dalgleish T

    (2010) Neural activity associated with monitoring the oscillating threat value of a tarantula

    PNAS 107:20582–20586.

    https://doi.org/10.1073/pnas.1009076107

    • PubMed
    • Google Scholar
59. 1. Ozawa T
    2. Ycu EA
    3. Kumar A
    4. Yeh LF
    5. Ahmed T
    6. Koivumaa J
    7. Johansen JP

    (2017) A feedback neural circuit for calibrating aversive memory strength

    Nature Neuroscience 20:90–97.

    https://doi.org/10.1038/nn.4439

    • PubMed
    • Google Scholar
60. 1. Parsons RG
    2. Gafford GM
    3. Helmstetter FJ

    (2010) Regulation of extinction-related plasticity by opioid receptors in the ventrolateral periaqueductal gray matter

    Frontiers in Behavioral Neuroscience 4:44.

    https://doi.org/10.3389/fnbeh.2010.00044

    • PubMed
    • Google Scholar
61. Book

    1. Paxinos G
    2. Watson C

    (2007)

    The Rat Brain in Stereotaxic Coordinates (6th Edition)

    Boston, Amsterdam: Academic Press/Elsevier.

    • Google Scholar
62. 1. Perusini JN
    2. Fanselow MS

    (2015) Neurobehavioral perspectives on the distinction between fear and anxiety

    Learning & Memory 22:417–425.

    https://doi.org/10.1101/lm.039180.115

    • PubMed
    • Google Scholar
63. 1. Petrovic P
    2. Kalso E
    3. Petersson KM
    4. Ingvar M

    (2002) Placebo and opioid analgesia-- imaging a shared neuronal network

    Science 295:1737–1740.

    https://doi.org/10.1126/science.1067176

    • PubMed
    • Google Scholar
64. 1. Pickens CL
    2. Golden SA
    3. Adams-Deutsch T
    4. Nair SG
    5. Shaham Y

    (2009) Long-lasting incubation of conditioned fear in rats

    Biological Psychiatry 65:881–886.

    https://doi.org/10.1016/j.biopsych.2008.12.010

    • PubMed
    • Google Scholar
65. 1. Rauch SL
    2. Whalen PJ
    3. Shin LM
    4. McInerney SC
    5. Macklin ML
    6. Lasko NB
    7. Orr SP
    8. Pitman RK

    (2000) Exaggerated amygdala response to masked facial stimuli in posttraumatic stress disorder: a functional MRI study

    Biological Psychiatry 47:769–776.

    https://doi.org/10.1016/S0006-3223(00)00828-3

    • Google Scholar
66. 1. Ray MH
    2. Hanlon E
    3. McDannald MA

    (2018) Lateral orbitofrontal cortex partitions mechanisms for fear regulation and alcohol consumption

    PLOS ONE 13:e0198043.

    https://doi.org/10.1371/journal.pone.0198043

    • PubMed
    • Google Scholar
67. 1. Rescorla RA

    (1967) Pavlovian conditioning and its proper control procedures

    Psychological Review 74:71–80.

    https://doi.org/10.1037/h0024109

    • PubMed
    • Google Scholar
68. 1. Rescorla RA

    (1968) Probability of shock in the presence and absence of CS in fear conditioning

    Journal of Comparative and Physiological Psychology 66:1–5.

    https://doi.org/10.1037/h0025984

    • PubMed
    • Google Scholar
69. 1. Roesch MR
    2. Calu DJ
    3. Schoenbaum G

    (2007) Dopamine neurons encode the better option in rats deciding between differently delayed or sized rewards

    Nature Neuroscience 10:1615–1624.

    https://doi.org/10.1038/nn2013

    • PubMed
    • Google Scholar
70. 1. Rozeske RR
    2. Jercog D
    3. Karalis N
    4. Chaudun F
    5. Khoder S
    6. Girard D
    7. Winke N
    8. Herry C

    (2018) Prefrontal-Periaqueductal Gray-Projecting neurons mediate context fear discrimination

    Neuron 97:898–910.

    https://doi.org/10.1016/j.neuron.2017.12.044

    • PubMed
    • Google Scholar
71. 1. Satpute AB
    2. Wager TD
    3. Cohen-Adad J
    4. Bianciardi M
    5. Choi JK
    6. Buhle JT
    7. Wald LL
    8. Barrett LF

    (2013) Identification of discrete functional subregions of the human periaqueductal gray

    PNAS 110:17101–17106.

    https://doi.org/10.1073/pnas.1306095110

    • PubMed
    • Google Scholar
72. 1. Sengupta A
    2. McNally GP

    (2014) A role for midline and intralaminar thalamus in the associative blocking of pavlovian fear conditioning

    Frontiers in Behavioral Neuroscience 8:148.

    https://doi.org/10.3389/fnbeh.2014.00148

    • PubMed
    • Google Scholar
73. 1. Tovote P
    2. Fadok JP
    3. Lüthi A

    (2015) Neuronal circuits for fear and anxiety

    Nature Reviews Neuroscience 16:317–331.

    https://doi.org/10.1038/nrn3945

    • PubMed
    • Google Scholar
74. 1. Tovote P
    2. Esposito MS
    3. Botta P
    4. Chaudun F
    5. Fadok JP
    6. Markovic M
    7. Wolff SB
    8. Ramakrishnan C
    9. Fenno L
    10. Deisseroth K
    11. Herry C
    12. Arber S
    13. Lüthi A

    (2016) Midbrain circuits for defensive behaviour

    Nature 534:206–212.

    https://doi.org/10.1038/nature17996

    • PubMed
    • Google Scholar
75. 1. Tracey I

    (2010) Getting the pain you expect: mechanisms of placebo, nocebo and reappraisal effects in humans

    Nature Medicine 16:1277–1283.

    https://doi.org/10.1038/nm.2229

    • PubMed
    • Google Scholar
76. 1. Vertes RP
    2. Linley SB
    3. Hoover WB

    (2015) Limbic circuitry of the midline thalamus

    Neuroscience & Biobehavioral Reviews 54:89–107.

    https://doi.org/10.1016/j.neubiorev.2015.01.014

    • PubMed
    • Google Scholar
77. 1. Vianna DM
    2. Graeff FG
    3. Brandão ML
    4. Landeira-Fernandez J

    (2001) Defensive freezing evoked by electrical stimulation of the periaqueductal gray: comparison between dorsolateral and ventrolateral regions

    Neuroreport 12:4109–4112.

    https://doi.org/10.1097/00001756-200112210-00049

    • PubMed
    • Google Scholar
78. 1. Vianna DM
    2. Allen C
    3. Carrive P

    (2008) Cardiovascular and behavioral responses to conditioned fear after medullary raphe neuronal blockade

    Neuroscience 153:1344–1353.

    https://doi.org/10.1016/j.neuroscience.2008.03.033

    • PubMed
    • Google Scholar
79. 1. Vianna DM
    2. Carrive P

    (2005) Changes in cutaneous and body temperature during and after conditioned fear to context in the rat

    European Journal of Neuroscience 21:2505–2512.

    https://doi.org/10.1111/j.1460-9568.2005.04073.x

    • PubMed
    • Google Scholar
80. 1. Von Krosigk M
    2. Smith AD

    (1991) Descending projections from the substantia nigra and retrorubral field to the medullary and pontomedullary reticular formation

    European Journal of Neuroscience 3:260–273.

    https://doi.org/10.1111/j.1460-9568.1991.tb00088.x

    • PubMed
    • Google Scholar
81. 1. Wager TD
    2. Atlas LY

    (2015) The neuroscience of placebo effects: connecting context, learning and health

    Nature Reviews Neuroscience 16:403–418.

    https://doi.org/10.1038/nrn3976

    • PubMed
    • Google Scholar
82. 1. Walker DL
    2. Cassella JV
    3. Lee Y
    4. De Lima TC
    5. Davis M

    (1997) Opposing roles of the amygdala and dorsolateral periaqueductal gray in fear-potentiated startle

    Neuroscience & Biobehavioral Reviews 21:743–753.

    https://doi.org/10.1016/S0149-7634(96)00061-9

    • PubMed
    • Google Scholar
83. 1. Walker RA
    2. Andreansky C
    3. Ray MH
    4. McDannald MA

    (2018) Early adolescent adversity inflates threat estimation in females and promotes alcohol use initiation in both sexes

    Behavioral Neuroscience 132:171–182.

    https://doi.org/10.1037/bne0000239

    • PubMed
    • Google Scholar
84. 1. Watson TC
    2. Cerminara NL
    3. Lumb BM
    4. Apps R

    (2016) Neural correlates of fear in the periaqueductal gray

    The Journal of Neuroscience 36:12707–12719.

    https://doi.org/10.1523/JNEUROSCI.1100-16.2016

    • PubMed
    • Google Scholar
85. 1. Wilson A
    2. Kapp BS

    (1994) Effect of lesions of the ventrolateral periaqueductal gray on the pavlovian conditioned heart rate response in the rabbit

    Behavioral and Neural Biology 62:73–76.

    https://doi.org/10.1016/S0163-1047(05)80061-5

    • PubMed
    • Google Scholar
86. 1. Wright KM
    2. DiLeo A
    3. McDannald MA

    (2015) Early adversity disrupts the adult use of aversive prediction errors to reduce fear in uncertainty

    Frontiers in Behavioral Neuroscience 9:227.

    https://doi.org/10.3389/fnbeh.2015.00227

    • PubMed
    • Google Scholar
87. 1. Yeh LF
    2. Watanabe M
    3. Sulkes-Cuevas J
    4. Johansen JP

    (2018) Dysregulation of aversive signaling pathways: a novel circuit endophenotype for pain and anxiety disorders

    Current Opinion in Neurobiology 48:37–44.

    https://doi.org/10.1016/j.conb.2017.09.006

    • PubMed
    • Google Scholar
88. 1. Zahm DS
    2. Trimble M

    (2008) The dopaminergic projection system, basal forebrain macrosystems, and conditioned stimuli

    CNS Spectrums 13:32–40.

    https://doi.org/10.1017/S1092852900016138

    • PubMed
    • Google Scholar
89. 1. Zhao Z
    2. Yang Y
    3. Walker DL
    4. Davis M

    (2009) Effects of substance P in the Amygdala, Ventromedial Hypothalamus, and periaqueductal gray on fear-potentiated startle

    Neuropsychopharmacology 34:331–340.

    https://doi.org/10.1038/npp.2008.55

    • PubMed
    • Google Scholar
90. Preprint

    1. Zhou J
    2. Gardner MPH
    3. Stalnaker TA
    4. Ramus SJ
    5. Wikenheiser AM
    6. Niv Y
    7. Schoenbaum G

    (2018) Rat orbitofrontal ensemble activity contains a multiplexed but value-invariant representation of task structure in an odor sequence task

    bioRxiv.

    https://doi.org/10.1101/507376

    • Google Scholar

Author details

1. Kristina M Wright

   Psychology Department, Boston College, Chestnut Hill, United States

   Contribution

   Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Methodology, Writing—original draft, Writing—review and editing

   For correspondence

   wrightko@bc.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-1446-3009

2. Michael A McDannald

   Psychology Department, Boston College, Chestnut Hill, United States

   Contribution

   Conceptualization, Resources, Data curation, Software, Formal analysis, Supervision, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing—original draft, Project administration, Writing—review and editing

   For correspondence

   michael.mcdannald@bc.edu

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-8525-1260

• 3,332

  views

• 394

  downloads

• 42

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.