Vigilance-Associated Gamma Oscillations Coordinate the ... - Cell Press

Article

Vigilance-Associated Gamma Oscillations Coordinate the Ensemble Activity of Basolateral Amygdala Neurons Highlights d

In BL, high-gamma is most pronounced when rats are vigilant

d

High-gamma entrains BL firing more strongly than all other rhythms

d

High-gamma causes minor changes in firing rates but large increases in synchrony

d

The entrainment of principal cells varies depending on their projection site

Amir et al., 2018, Neuron 97, 656–669 February 7, 2018 ª 2018 Elsevier Inc. https://doi.org/10.1016/j.neuron.2017.12.035

Authors Alon Amir, Drew B. Headley, Seung-Chan Lee, Darrell Haufler, Denis Pare´

Correspondence [email protected]

In Brief Few principal basolateral amygdala (BL) neurons are activated by emotionally valenced stimuli, yet they profoundly influence behavior. Amir et al. show that the solution to this paradox resides in the synchronization of individual BL neurons by high-gamma.

Neuron

Article Vigilance-Associated Gamma Oscillations Coordinate the Ensemble Activity of Basolateral Amygdala Neurons Alon Amir,1 Drew B. Headley,1 Seung-Chan Lee,1,2 Darrell Haufler,1 and Denis Pare´1,3,* 1Center

for Molecular and Behavioral Neuroscience, Rutgers University–Newark, 197 University Avenue, Newark, NJ 07102, USA address: Department of Neuroscience, Johns Hopkins University School of Medicine, 725 North Wolfe Street, Baltimore, MD 21205, USA 3Lead Contact *Correspondence: [email protected] https://doi.org/10.1016/j.neuron.2017.12.035 2Present

SUMMARY

Principal basolateral amygdala (BL) neurons profoundly influence motivated behaviors, yet few of them are activated by emotionally valenced stimuli. Here, we show that a likely explanation for this paradox is the synchronizing influence of the highgamma rhythm. High-gamma (75–95 Hz) entrained BL firing more strongly than all other rhythms. It was most pronounced during states of increased vigilance, when rats were apprehensive. Relative to behavioral states, high-gamma produced minor changes in firing rates yet dramatic increases in synchrony. Moreover, connected pairs of cells showed similarly high levels of entrainment and synchronization. Unexpectedly, prefrontal- and accumbens-projecting cells, respectively, showed high and low entrainment by high-gamma, indicating that this rhythm differentially synchronizes the activity of BL neurons projecting to specific sites. Overall, our findings suggest that individual BL neurons encode information not only by changing their firing rates, but also by synchronizing their collective activity, amplifying their impact on target structures.

INTRODUCTION Like the cerebral cortex, the basolateral nucleus of the amygdala (BL) contains a dominant group of glutamatergic projection cells and diverse subtypes of GABAergic interneurons (Muller et al., 2003, 2005; Spampanato et al., 2011). Although BL exerts a profound influence over motivated behaviors (Amano et al., 2011; Ambroggi et al., 2008; Herry et al., 2008; Stuber et al., 2011; Tye et al., 2011), principal BL neurons fire at much lower rates than their cortical counterparts (Amir et al., 2015; Jacobs and McGinty, 1971; Pare´ and Gaudreau, 1996; Steriade and Hobson, 1976), and very few of them are activated by emotionally valenced stimuli (Lee et al., 2016). Thus, BL signaling may not depend solely on the firing rates of individual cells but on syn656 Neuron 97, 656–669, February 7, 2018 ª 2018 Elsevier Inc.

chronizing their discharges, thereby enhancing their impact on target networks. A potential mechanism for such synchronization is gamma oscillations (Fries et al., 2007; Fries, 2015). First identified as an electroencephalogram (EEG) correlate of behavioral arousal (Magoun, 1958), gamma oscillations (30–100 Hz) have since been observed in local field potentials (LFPs) throughout the brain. In keeping with the diverse roles of the networks exhibiting them, gamma oscillations have been implicated in multiple functions (Engel and Singer, 2001). Depending on the brain region, state of vigilance, experimental context, and behavior, different gamma subtypes with distinct frequencies predominate (Ainsworth et al., 2011; Colgin et al., 2009; van der Meer et al., 2010). In the human amygdala, emotional stimuli, particularly negative ones, elicit robust gamma oscillations (Luo et al., 2007; Oya et al., 2002), even when presented subliminally (Luo et al., 2009). In animals, gamma is enhanced during periods of arousal (Pagano and Gault, 1964). Moreover, stimuli that predict affective events can elicit coherent gamma in BL and related networks (Bauer et al., 2007; Popescu et al., 2009). Last, BL gamma also shows differential coupling to hippocampal or medial prefrontal theta during fear and safety (Stujenske et al., 2014). The cellular basis of gamma oscillations has been most extensively characterized in cortical networks where they depend on reciprocal interactions between pyramidal and fast-spiking cells (Buzsa´ki and Wang, 2012). Indeed, of the various interneuron subtypes, fast-spiking cells show the greatest gamma entrainment (Bragin et al., 1995; Csicsvari et al., 2003; Gloveli et al., 2005; Otte et al., 2010). They receive strong excitatory inputs from pyramidal cells (Levy and Reyes, 2012), are interconnected by chemical and electrical synapses (Galarreta and Hestrin, 1999; Gibson et al., 1999), and have divergent axons, each contributing many inhibitory synapses in the perisomatic region of pyramidal cells (Buhl et al., 1994). These properties promote activity synchronization of large subsets of principal cells. Although similar principles of connectivity have been reported in BL (Smith et al., 2000; Spampanato et al., 2011; Woodruff and Sah, 2007), far less data are available on the genesis of gamma. Gamma emerges in amygdala slices in the presence of glutamate (Randall et al., 2011) or cholinergic agonists (Sinfield and Collins, 2006). In vivo, gamma occurs spontaneously and entrains principal cell firing (Bauer et al., 2007; Popescu et al., 2009;

Stujenske et al., 2014). However, no data are available regarding the gamma-related activity of fast-spiking cells. Nor is it clear whether gamma synchronizes amygdala neurons as in cortex. At present, these and many aspects of gamma genesis and function remain uncharacterized in BL. The present study was undertaken to shed light on these questions. RESULTS Behavioral Correlates To characterize the behavioral dependence of gamma in BL, we obtained simultaneous unit and LFP recordings in rats engaged in two tasks characterized by different threat levels: a semi-naturalistic foraging task (Figure 1A) and a food-seeking shuttle task (Figure 1B). Foraging Task In this task (Choi and Kim, 2010; Amir et al., 2015), hungry rats (n = 5) were confronted with a mechanical predator when they left their nest to retrieve food pellets in a long and brightly lit arena (Figure 1A1). The nest and predator were located at opposite ends of the apparatus and the food pellets at varying distances between them. When rats approached the predator to attain food, it surged forward, causing rats to flee to their nest. Rats were first trained to retrieve food in the absence of the predator. Trials started when the door to the foraging arena opened. Following a delay during which rats seemed to hesitate at the door threshold, they ran into the arena, took the food pellet in their mouth, and returned to the nest to consume it, at which point the door was shut. The next trial started R1 min later. On later days, alternating blocks of 10–20 trials were conducted with the predator absent or present. We first summarize our earlier behavioral findings in the foraging task (Amir et al., 2015) and then describe task-related changes in gamma activity. Introduction of the predator changed the rats’ behavior in many ways, all consistent with a state of apprehension and increased vigilance. After door opening and waiting at the door threshold, rats sometimes retreated to the nest instead of initiating foraging. Such aborted trials (marked ‘‘A’’ in Figure 1A2) were
10 times more frequent in the presence than absence of predator. Also, on trials with foraging initiation, the proportion of successful trials (marked ‘‘S’’ in Figure 1A2) decreased by up to 70% and the food retrieval interval almost doubled as rats foraged more tentatively. Last, behavior on a given predator trial depended on the outcome of the prior trial: rats spent twice longer retrieving the food on trials that followed failed trials (marked ‘‘F’’ in Figure 1A2) than successful ones. See Table S1 for a breakdown of these various types of trials in each animal. In BL, we observed task-related variations of LFP power in the theta (6–10 Hz), beta (15–25 Hz), and three gamma bands: low(35–55 Hz), mid- (55–75 Hz), and high-gamma (75–95 Hz) (repeated-measures Friedman ANOVAs, c2s = 40.83–1,564, df = 3, p < 0.001; see Figure 1 legend for post hoc tests). Figure 1C1 shows average Z-scored LFP power fluctuations as a function of relative time for all available rats (n = 5) and trials with foraging initiation (n = 1,238). Similar results were obtained with raw power and individual rats yielded consistent results (Figure S1). The spectrogram starts 30 s prior to door opening.

When rats were in the nest (baseline), beta power was high while theta- and gamma-power were low, except for a transient rise of mid-gamma when rats approached the door in anticipation of its opening, consistent with prior findings in the corticospinal system (Schoffelen et al., 2005, 2011). When rats started waiting at the door threshold, a weak elevation of low- and high-gamma occurred. Upon initiation of foraging, the power of low- and highgamma increased further, in parallel with a rise in theta power. Upon escape, low- and high-gamma power abruptly decreased while theta power remained elevated. During escape, an increase in the power of frequencies above 100 Hz was also observed but this change likely resulted from movement artifacts. The elevated low- and high-gamma power observed during foraging is not simply related to movement since it was not seen during escape (Figure 1C1), when rats ran the fastest. To analyze the behavioral correlates of these gamma bands, we compared different trial types. Figure 1C2 depicts gamma power on trials where rats, after waiting at the door, retreated back into the nest (134 trials) instead of initiating foraging (Figure 1C1). A Kruskal-Wallis ANOVA comparing the baseline and waiting periods in the two trial types revealed a significant effect of trial type and task phase on low- (c2 = 96.37, df = 3, p < 0.001) and highgamma power (c2 = 124.97, df = 3, p < 0.001). During the waiting period, when rats were immobile, low- and (particularly) highgamma power was higher when rats eventually aborted (Figure 1C2) rather than initiated foraging (Figure 1C1; Dunn tests, p < 0.05). Similarly, low- and high-gamma power during the baseline and waiting periods was higher on trials that followed failed (Figure 1D1) than successful trials (Figure 1D2; 231 and 992 trials, respectively; see Figure 1 legend for statistics). Since waiting periods prior to aborted foraging or after failed trials are devoid of movement, the associated enhancement of low- and high-gamma is likely related to an inner state akin to apprehension or a process associated with this state such as scanning vigilance. If so, why is low- and high-gamma power increased during foraging? To address this question, we next compared gamma power on trials with low or high foraging speeds as we previously found that low foraging velocities betray a state of apprehension in this task (Amir et al., 2015). Consistent with this interpretation, low- and, particularly highgamma power during foraging was higher on trials when rats moved slowly (Figure 2A1) than fast (Figure 2A2), and the opposite relationship was seen in the theta band. Moreover, a separate analysis of 151 trials where rats showed signs of hesitation during foraging, pausing one or more times during approach of the food pellet, revealed that low- and high-gamma power were higher during foraging pauses than during forward movements (Figures 2B1 and 2B2). To dissociate the influence of movement speed, presence or absence of the predator, and whether the prior trial had been a success or failure, we computed a generalized linear model (GLM) on LFP power. The GLM confirmed a significant negative association between movement speed and low- and highgamma as well as a positive association between movement speed and theta (p < 0.01; Figures 2C1–2C4). The predator and prior trial variables had a smaller, positive influence on low- and high-gamma (see Figure 2 legend for statistics).

Neuron 97, 656–669, February 7, 2018 657

Rat

B1 Shuttle task

A2 Predator

A F

25 cm

Food retrieval area

B2

20 cm

All trials with foraging

C1

Door

10 cm

Food retrieval area

A S

Food pellet

Door

S

Trials

A1 Foraging task

Waiting

C2

120

No foraging

120 0.4

Frequency (Hz)

100

80 60 40

0.2

80

z-score

Frequency (Hz)

100

60

0

40

-0.2

20

20

0 3

0 3

Baseline

Waiting

Foraging

Escape Firing rate

Firing rate

-0.4

2 1 0

Frequency (Hz)

2 1

D2

Failed prior trial

120

Firing rate

Waiting

0

D1

100

80

80

60

60

40

40

20

20 Baseline

Waiting

Successful prior trial

120

100

0

Baseline

Foraging

0

3

3

2

2

1

1

0

0

Baseline

Waiting

Foraging

Figure 1. Behavioral Tasks and Changes in Theta, Beta, and Gamma Power Seen during the Foraging Task (A and B) Behavioral tasks. (A1–2) Foraging task. (A1) Apparatus. (A2) Five example trials where red dots indicate rat position and the distance between the dots is proportional to the rat’s velocity. S, A, and F mark successful, aborted, and failed trials, respectively. Empty black squares on the right indicate position of food pellets. Vertical dashed lines mark 25-cm intervals. (B) Shuttle task. (B1) Apparatus. (B2) Three example trials where red and blue dots indicate rat position and the distance between the dots is proportional to the rat’s velocity. (C and D) Spectrograms showing Z-scored power fluctuations in various frequency bands as a function of relative time. All spectrograms shown here and in the following figures were normalized using the same data (STAR Methods) and thus can be directly compared. Warmer colors indicate higher power. Spectrograms (top) and average unit activity (bottom; normalized to baseline firing rates in C1) in various conditions (black, principal cells; red, interneurons; dashed and solid lines, respectively, represent cells with significantly decreased (type 1) and increased (type 2) firing rates during foraging (Kruskal-Wallis ANOVA followed by Dunn test, p < 0.05). (C1) All trials with foraging (1,238 trials). (C2) Subset of trials where rats, after hesitating at the door retreated in the nest (134 trials). Post hoc Dunn tests on the data shown in C1 revealed that theta power was significantly (p < 0.05) different in the four phases (baseline < waiting < foraging < escape). The same approach revealed that beta power was significantly different in the four phases (foraging < waiting < escape < baseline). For mid-gamma, power was higher during baseline relative to the other three phases with no difference between waiting, foraging, and escape. For low- and high-gamma, power differed significantly in the four phases (foraging > waiting = escape > baseline). (D) Subsets of trials that followed failed (D1) or successful (D2) trials. Kruskal-Wallis ANOVAs; df = 5; low-gamma c2 = 430.8; p < 0.001; high-gamma c2 = 552.2; p < 0.001. Post hoc Dunn tests revealed the following significant (p < 0.05) differences. For lowgamma, (foraging [n  1, fail] = foraging [n  1, success]) > (baseline [n  1, fail] = waiting [n  1, fail]) > waiting [n  1, success] > baseline [n  1, success]. For high-gamma, (foraging [n  1, fail] = waiting [n  1, fail] = baseline [n  1, fail]) > foraging [n  1, success] > waiting [n  1, success] > baseline [n  1, success]. Number of units in the various panels: (C1, D1, D2: principal cells, 271 type 1 and 25 type 2; interneurons, 11 type 1 and 35 type 2; C2 principal cells, 226 type 1 and 22 type 2; interneurons, 8 type 1 and 29 type 2). Average baseline firing rates: principal cells type 1 (0.21 ± 0.03 Hz) and type 2 (1.55 ± 0.33 Hz); interneurons type 1 (27.47 ± 3.45 Hz) and type 2 (22.44 ± 2.19 Hz). See also Figure S1 and Tables S1 and S2.

658 Neuron 97, 656–669, February 7, 2018

Foraging speed ≤ 30 %

A1

Waiting

B1 Hesitation during foraging Action starts

Foraging

100

100

80 60

0

40 20

-1

3 -2

-1

60 40

C2

120

120

100

100

2

3

0 0.3

(m/s) 0 -1.6

-0.8

120 100

sign. (–)

80

80

80

60

60

40

40

40

20

20

20

Relative time

sign. +

0 Start

Beta velocity

Relative time

60

0 End % significant 0

End

ns

50

0 Start

0.8 -0.8

0 Time (s)

C3 Foraging (significant bins) C4

100

Frequency (Hz)

0 1 Time (s)

0

-0.4

2

Foraging (beta velocity)

0.4

-0.2

0 1 Time (s)

-0.3

-2

0

80

D Frequency (Hz)

(m/s)

Speed

-0.5

0

0.2

B2

Action stops

20

0 0.4

0.4

Speed

A2 Foraging speed ≥ 70 %

0.5 120

C1

Beta

Foraging

z-score

Frequency (Hz)

Waiting

120

0

0.8 Time (s)

1.6

Shuttle task 120 100 80 60 40 20 0

Baseline

Foraging

Relative time

Figure 2. Relation between Running Speed and Gamma Power (A and B) Spectrograms showing Z-scored power fluctuations in various frequency bands as a function of when the rats’ velocity during foraging was low (A1; 30% trials with lowest velocity; n = 371), when it was high (A2; top 30% trials; n = 371), or when rats hesitated during foraging, exhibiting one or more starts (B1) and pauses (B2). (C1) Relation between rat velocity and LFP power of different frequencies as estimated by a generalized linear model. Beta velocity values are color coded (scale on left) during foraging. (C2) Average beta velocity values (y axis) computed from entire foraging period for different power in different frequencies. (C3) Bins of C1 with statistically significant positive (red) and negative (blue) beta values during foraging (relative time, x axis). (C4) Proportion of bins with significant beta velocity values during entire foraging period. Repeated-measures Friedman ANOVAs relating absolute beta values associated with the speed, predator, and prior trial variables for low and high-gamma: c2 s = 417 and 582, df = 2, p < 0.001. Post hoc Dunn tests for GLM results revealed the following significant differences (p < 0.05): absolute beta speed > absolute beta prior trial > absolute beta robot for both low- and high-gamma. (D) Spectrogram showing Z-scored power fluctuations in various frequency bands as a function of relative time in the shuttle task (277 trials).

Food-Seeking Shuttle Task The above indicates that different gamma bands with distinct behavioral correlates can be distinguished in BL. Mid-gamma is most prominent when rats anticipate door opening. In contrast, power in the low- and high-gamma bands is higher during behaviors consistent with a state of apprehension and increased vigilance. However, the foraging task involves conflicting threats and rewards, complicating interpretation of the data. Thus, we next studied fluctuations in gamma activity in a task devoid of explicit threats where rats ran back-and-forth between two nest-like areas to retrieve food pellets. Retractable doors separated the nests from a dimly lit central compartment (Figure 1B1). Prior to the recordings, rats were habituated extensively to the apparatus. During the recordings, rats were placed in one of the nests and a food pellet in the other. After opening of the doors, rats ran across the central compartment to the other nest to consume the food (Figure 1B2). While rats ate the food, the doors were closed and another trial was initiated 1 min later. In contrast with the foraging task, there were no signs of apprehension in the shuttle task. Upon door opening, rats did not hesitate but initiated foraging at once. Also, they never skipped a trial and their foraging speed was uniformly high

(
0.35 m/s). During the baseline period, power in the frequency bands of interest did not differ between the shuttle (277 trials) and foraging tasks (1,238 trials; rank-sum test, p R 0.077). However, in contrast with the foraging task, high- and low-gamma power did not increase when rats sought the food pellets in the shuttle task (signed-rank tests, p R 0.64; Figure 2D). Since there are no explicit threats in the shuttle task, these results again support the idea that low- and high-gamma are linked to a state of apprehension or a process associated with this state, such as scanning vigilance. Entrainment of Unit Activity by Various LFP Rhythms Next, we compared the entrainment of BL units by the rhythms described above. BL units (n = 436) were classified as putative principal cells (n = 381) or interneurons (n = 50) on the basis of spike width and firing rates in the nest. Cells with firing rates 0.6 ms were classified as principal cells; units with firing rates R6 Hz and peak-totrough spike times %0.6 ms as interneurons. Units that only met one of the two criteria (n = 5) were not considered further. A subset of 56 units were positively identified as projection cells by antidromic invasion from the medial prefrontal cortex (mPFC)

Neuron 97, 656–669, February 7, 2018 659

or nucleus accumbens (nAc), and they all met the above criteria for principal cells. Figures 1C and 1D (bottom panels) plot the average firing rates of principal cells (black) and interneurons (red) with significantly decreased (dashed lines) or increased (solid lines) firing rates during foraging (Kruskal-Wallis ANOVA followed by Dunn test, p < 0.05). As evident in this figure, there was no systematic relation between firing rates and fluctuations in gamma power, a point investigated in detail below. To avoid spectral leakage of spike waveforms into LFPs (Scheffer-Teixeira et al., 2013; Zanos et al., 2011), we referenced unit activity to LFPs recorded by a shank located 200 mm away from the unit under consideration. For each unit, the LFP was band-pass filtered for the frequency band of interest, the magnitude of entrainment (R) was ascertained, and a Rayleigh test was performed to determine whether the cell was significantly modulated. As detailed in the STAR Methods, various approaches were used to bin the frequency space (4–250 Hz), but they all gave qualitatively identical results. Although the physiological properties of principal cells and fast-spiking interneurons differ markedly (reviewed in Spampanato et al., 2011), their frequency profile of firing entrainment was nearly identical. Figure 3 provides examples of peri-event histograms (PEHs) of spiking around high-gamma, low-gamma, beta, and theta peaks R2.0 SD for a representative principal cell (Figure 3A) and an interneuron (Figure 3B). As in these examples, the average firing entrainment of both cell types varied significantly with frequency (ANOVA, F(1,68) = 319.1, p < 0.001) and was maximal in the high-gamma range (Figures 3C and 3D). As the distance between the reference LFP and the recorded unit increased from 200 to 800 mm, both the entrainment (Figures 3C and 3D; ANOVA, F(1,3) = 4736.2, p < 0.001) and the proportion of cells significantly entrained by high-gamma (Figure 3E) decreased (c2 = 70.648, p < 0.001). Similar results were obtained with mid- and low-gamma (Figures 3C and 3E). Whereas there was no relationship between the firing rates of principal cells and their entrainment by high-gamma (Spearman r = 0.025, p = 0.625), a significant positive correlation was found in fast-spiking cells (Spearman r = 0.306, p = 0.031). Unexpectedly, frequency distributions of firing entrainment by highgamma were skewed to the left in principal (Figure 3F) and fast-spiking (Figure 3G) cells, suggesting that high-gamma does not influence BL neurons homogenously, a point we will return to below. Among the significantly entrained units, principal and fast-spiking cells preferentially fired at different phases of high-gamma (Figures 3H–3J). Principal cells mainly fired during the trough of high-gamma (Figures 3H and 3J) and fast-spiking cells, during the rise of its positive phase (Figures 3I and 3J). Relation between Behavioral States and Entrainment by High-Gamma In the previous section, we considered the entrainment of BL firing by high-gamma, irrespective of behavioral state. Here, we examine the impact of this factor. We distinguished four different waking states, three of which occurred during foraging trials (baseline, waiting, and foraging) and the last, in between trial blocks (quiet wakefulness). These states were associated with significant changes in the incidence of gamma bursts and in the firing rates of BL neurons (Table S2). Of note, the frequency

660 Neuron 97, 656–669, February 7, 2018

of gamma bursts was higher than the average firing rate of principal cells in all states, but not of interneurons. We first tested whether the neurons’ preferred firing phase changed as a function of behavioral state. Because principal cells generally have low firing rates (Table S2) and this analysis required data segmentation by states, to compensate for the loss of statistical power, we restricted our attention to a subset of 65 principal cells selected because they were more active (R100 spikes in each of the conditions under consideration; see Figure 4 legend). A non-parametric multi-sample test for equal medians revealed that the preferred firing phase of 10% of principal cells (Figure 4A1) and 25% of fast-spiking neurons (Figure 4A2) differed significantly between states (p < 0.05). However, even in these cells, the effect was minute (
6 degrees difference), and the direction of the change was not consistent. Next, we studied the influence of behavioral states on firing entrainment by high-gamma (Figures 4B–4D). As a first step, we asked whether the relative magnitude of entrainment across our samples was consistent between states. To this end, we sorted the same subset of spontaneously active principal cells (Figure 4C) and all interneurons (Figure 4D) by their entrainment in quiet waking (Figures 4C1 and 4D1; warmer colors indicate stronger entrainment) and plotted the data obtained in the other states without changing the order of the cells (Figures 4C2–4C4 and 4D2–4D4). Also, to control for state differences in gamma levels, we stratified the entrainment by gamma burst amplitude (y axis; STAR Methods). These analyses revealed that cell-to-cell variations in firing entrainment by high-gamma are largely consistent across states (Figure 4B; see figure legend for statistics). To compare the influence of task-related behavioral states and gamma on firing rates, for each unit we used a GLM to fit the binned spiking (12 ms windows) with behavioral state (baseline, waiting, and foraging) and Z-scored gamma amplitude. Figure 5 plots the beta coefficients for each state with respect to the gamma amplitude beta, revealing that states have a much larger influence on firing rates than high-gamma but that this effect is much more pronounced in principal cells (Figure 5, blue) than interneurons (Figure 5, red). In this analysis, the absolute ratio between state and gamma beta coefficients indicates how many SDs in gamma amplitude are required to cause a change in firing rate similar to that induced by a change in task-related states. For instance, in principal cells, gamma amplitude would have to change by 21.6 SDs to cause the same change in firing rate induced by a switch from quiet waking to foraging (Figure 5A). Changes from quiet waking to waiting (Figure 5B) or baseline (Figure 5C) had a similarly outsized impact on the firing rates of principal cells relative to gamma amplitude. Yet there was much heterogeneity in our samples of principal cells and interneurons (Figure 6). In both cell types, entrainment by high-gamma (resultant vector – R) varied widely (principal cells, 0–0.36, Figure 6A; interneurons, 0–0.3; Figure 6B) and it correlated positively with the influence of high-gamma on firing rates (Figures 6A5, 6B2, and S2A). Also, a minority of neurons (Figure 6B4) showed little firing entrainment to high-gamma (R < 0.1), yet displayed a clear reduction in firing rate during high-gamma bursts, reminiscent of prior observations in cortex (Puig et al., 2008).

A1

A2

A3

A4

B1

B2

B3

B4

C

E

F

H

I

D

G J

(legend on next page)

Neuron 97, 656–669, February 7, 2018 661

A1

B1

C1

A2

B2

C2

D1

D2

D3

C3

D4

C4

Impact of High-Gamma on Firing Synchrony in BL Even though high-gamma has minor effects on firing rates, it might still shape the ensemble activity of BL neurons. To test this possibility, we examined the cross-correlation in spiking between pairs of simultaneously recorded neurons during periods with low- or high-gamma power (STAR Methods; Figures S2B and S2C). We then compared the amount of synchrony in the two conditions, after normalizing for differences in baseline activity (synchronization index [SI]; STAR Methods). These analyses were carried out for couples of principal cells, of fastspiking interneurons, or of principal cells and interneurons (top to bottom rows of Figure 7, respectively). In these three types of cell couples, normalized cross-correlograms (CCGs; Figure 7A) and SIs (Figure 7B) revealed that firing synchrony was higher when high-gamma power was elevated (blue) than low (red; inter-cell distance of 200 mm; signed-rank tests; the number of pairs and p values are indicated in Figure 7). While this effect held irrespective of how far apart the cells were from each other, distance attenuated it significantly (Figure 7C; P-P, ANOVA F(2,3661) = 4.55, p = 0.011; INT-INT, ANOVA F(2,305) = 10.3, p < 0.001; INT-P, ANOVA F(2,3121) = 60.1, p < 0.001).

Figure 4. Impact of Behavioral States on Entrainment of Unit Activity by High-Gamma (A and B) Average ± SEM firing phase (A) and entrainment (B) by high-gamma for principal cells (A1 and B1) and interneurons (A2 and B2) during different behavioral states (x axis; QW, quiet waking; B, baseline; W, waiting; F, foraging). (C and D) Principal cells (C) and interneurons (D) were sorted by their entrainment in quiet waking (C1 and D1) and high-gamma bursts were stratified by amplitude (y axes). Data obtained in the other states (C2–C4 and D2–D4) were plotted without changing the ordering of the cells. Warmer colors indicate stronger entrainment. Note that the number of cells included in (A) (54 principal cells and 45 interneurons) versus (B)–(D) differs because (A) does not include cells that were not significantly entrained, whereas (B)–(D) do (65 principal cells and 50 interneurons). Nonsignificantly entrained cells were excluded from (A) since this panel reports on phase, which is meaningless when cells are not entrained. Regarding firing entrainment, a repeated-measures Friedman ANOVA revealed minor but significant differences in entrainment magnitude as a function of state in interneurons (c2(3,199) = 10.49; p = 0.015), but not in principal cells (c2(3,259) = 4.99; p = 0.17). The maximal significant difference in entrainment between states was 4% in interneurons (post hoc Dunn tests; p < 0.05).

However, the above analyses did not take into account how much gamma entrained each cell in a couple to begin with. To address this question, we correlated SIs with the joint gamma entrainment of the cells in each couple. This was operationally defined as the product of the R variable computed individually for each cell in a couple. The results of this analysis are shown in Figure 7D where couples are stratified by percentile of their R product. We found a significant correlation between R product and SI, which was strongest for couples of interneurons (Spearman r = 0.77, p < 0.001; Figure 7D2) and mixed couples (Spearman r = 0.53, p < 0.001; Figure 7D3) but lower in pairs of principal cells (Spearman r = 0.16, p < 0.001; Figure 7D1), likely because of their low firing rates. Together, these findings suggest that while high-gamma exerts minor effects on firing rates, it is associated with a marked increase in neuronal synchrony, particularly among cells that are highly entrained by high-gamma. In the foraging task, since high-gamma levels increased during foraging (Figure 1C1), the above results predict that firing synchrony in the three types of cell couples should be elevated at that time relative to baseline. To test this idea, we computed

Figure 3. Entrainment of BL Firing by Different LFP Rhythms (A and B) Peri-event histograms of neuronal discharges for a representative principal cell (A) and fast-spiking interneuron (B) around large amplitude (R2 SD) oscillatory peaks in different frequency bands (1, high-gamma; 2, low-gamma; 3, beta; 4, theta). In all cases, the bin width was set to 0.1 of the period under consideration. (C and D) Entrainment of unit activity (y axis; median of the entire sample) as a function of frequency (x axis) for principal cells (C) and fast-spiking interneurons (D). (E) Proportion of principal cells significantly entrained by different rhythms (y axis) as a function of distance between the recorded cell and reference LFP (x axis). (F and G) Frequency distributions of entrainment by high-gamma among principal cells (F) and interneurons (G). (H and I) Frequency distribution of preferred firing phase in relation to high-gamma for principal cells (H) and interneurons (I). (J) Average firing phase in relation to high-gamma for principal cells (blue) and interneurons (red).

662 Neuron 97, 656–669, February 7, 2018

A

Projection cells Interneurons

Quiet waking to Foraging

0.5

Principal cells ratio 21.6 Interneurons ratio 4.4

0

Gamma (Beta coefficient)

-0.5-5

-4

-3

-2

B

-1

0

1

State (Beta coefficient)

2

3

Principal cells ratio 13.4 Interneurons ratio 3.8

0

-4

-3

-2

C

-1

0

1

State (Beta coefficient)

2

3

4

5

Quiet waking to Baseline

0.5

Principal cells ratio 11.5 Interneurons ratio 2.7

0 -0.5 -5

5

Quiet waking to Waiting

0.5

-0.5 -5

4

-4

-3

-2

-1

0

1

State (Beta coefficient)

2

3

4

5

Figure 5. State-Related Variations in Firing Rates Are Much Larger Than Those Produced by High-Gamma (A–C) Comparisons of GLM beta coefficients for gamma amplitude and foraging (A), waiting (B), and baseline (C). Red, 50 interneurons; blue, 381 principal cells. The data were analyzed in 12-ms windows, but longer windows (50 or 100 ms) yielded qualitatively identical results. On the right of each panel, we provide the average absolute ratio of the beta coefficients associated with state and gamma.

CCGs of unit activity irrespective of high-gamma levels during the baseline period and, separately, foraging. We then computed SIs for both. Consistent with our prediction, these analyses revealed that firing synchrony was significantly enhanced during foraging for the three types of cell couples (signedrank tests; 559 couple of projection cells, 41.1% ± 17.6% increase; p = 0.005; 84 couples of interneurons, 13.6% ± 5.1% increase, p = 0.03; 587 mixed couples, 14.2% ± 6.2% increase, p = 0.01). Relation between Connectivity and Entrainment by High-Gamma BL contains many subsets of principal cells, as defined by their €nen, 2000). different cortical and subcortical projections (Pitka High-gamma might similarly entrain all principal cells irrespective of their connectivity or selectively entrain particular subsets that belong to specific networks. Thus, we examined how gamma entrainment relates to the cells’ intrinsic (Figure 8A) and extrinsic (Figure 8B) connectivity. To test whether high-gamma synchronizes monosynaptically connected cells more strongly than uncoupled neurons, we segregated cell pairs based on whether their CCGs displayed evidence of monosynaptic connections (see STAR Methods for criteria). To avoid confounding monosynaptic connections and gamma-induced synchronization, we first determined the range of high-gamma power that caused a rise in synchrony and then restricted our CCG analyses to epochs with lower gamma power (bottom 50th percentile; Figure S3A). Examples of CCGs with putative monosynaptic connections are illustrated in Figure S3B. We then compared the gamma entrainment of cells belonging to monosynaptically connected (red) or unconnected (blue) cell pairs (connections from principal cells to interneurons, Figures 8A and S4A; inter-

neurons to principal cells, Figure S4B; between principal cells, Figure S5). For couples of principal cells and interneurons, we found that high-gamma entrainment was associated with an increased incidence of monosynaptic connections. That is, principal cells with a high resultant vector were more likely to be connected to interneurons with a high resultant vector (Figure 8A1). However, this effect was marginal in the opposite direction (from interneurons to principal cells; Figure S4). To evaluate the statistical significance of this observation, we correlated the connection status of each cell couple (unconnected = 0; connected = 1) with the product of their resultant vector. The Spearman correlation coefficient was 0.49 (p < 0.001) for connections from principal cells to interneurons and 0.17 (p < 0.001) for interneurons to principal cells. However, a potential confound in the latter analysis is the possibility that the higher synchronization by high-gamma seen in monosynaptically coupled cells is due to the fact that these cells are more strongly modulated by gamma to begin with. To eliminate the influence of each cells’ gamma rhythmicity from synchronization estimates, we computed unit-unit coherence, an approach that controls for the autocorrelation structure of each cell’s activity (Figure S6). Comparing coherence estimates between connected and unconnected cell couples confirmed that relative to all other frequencies, highgamma synchronizes monosynaptically connected cells more strongly than non-connected neurons (Figures 8A2 and S7; rank-sum test, p < 0.001). So far, we have considered the impact of intrinsic connectivity on firing synchrony among pairs of principal cells or couples of interneurons and principal cells. For couples of interneurons, monosynaptic connectivity could not be assessed because their CCGs often had a peak at time lag zero such that subsequent

Neuron 97, 656–669, February 7, 2018 663

A1

A4

A2

A5

(A) Principal cells. (B) Interneurons. Examples of PEHs of neuronal discharges computed around high-gamma peaks (R2 SD) for principal cells (A1–A4 and A6) and interneurons (B1 and B3–B6). Firing entrainment (y axis) as a function of change in firing rate (% of baseline) produced by high-gamma for principal cells (A5) and interneurons (B2). Firing rate (FR) changes are expressed as (FR(g) – FR(no g)) / FR(no g). See also Figure S2.

A6

B1

B2

B3

B4

B5

B6

reductions in spike counts might have been caused by spike afterhyperpolarizations, synaptic inhibition, or both. Nevertheless, we noticed that there was a strong correlation between, on the one hand, the height of the peak at time zero in the absence of gamma and, on the other, their R product (r = 0.76; p < 0.0001) or the increase in synchrony observed during highgamma (r = 0.50; p < 0.0001). Last, we examined whether gamma entrainment varies depending on the cells’ projection site(s) (Figure 8B). To this end, we computed separate frequency distributions of gamma entrainment for the subsets of cells antidromically responsive to electrical stimulation of nAc (Figure 8B3) or the mPFC (Figure 8B4). For reference, the gamma entrainment of all putative principal cells as well as of all principal cells positively identified as such by antidromic invasion are depicted in Figures 8B1 and 8B2, respectively. We found that gamma entrainment varied significantly as a function of the cells’ projections site(s). It was significantly lower among nAc-projecting cells (0.092 ± 0.032) than neurons projecting to mPFC (0.174 ± 0.033; rank-sum test, p < 0.001). Some neurons projected to both sites, but there were too few of them (n = 6)

664 Neuron 97, 656–669, February 7, 2018

Figure 6. Impact of High-Gamma on Firing Rates

A3

for meaningful statistical comparisons with the other two subsets. Potentially related to these results, we also found that adjacent pairs of principal cells, that is, cells recorded from the same shank, had more similar levels of gamma entrainment compared to pairs recorded from different shanks. To test the statistical significance of this observation, we used two approaches (Figure S8). We first correlated the R of cells in each pair of simultaneously recorded principal cells. The correlation between Rs was significant (r = 0.52, p < 0.0001) for cell pairs recorded from the same shank (n = 1615), but not for cell pairs recorded from different shanks (n = 6247; r = 0.003, p = 0.82). In the second approach, we compared the absolute difference in the Rs for cell pairs recorded from the same or different shanks. A significant difference was observed (ranksum test, p < 0.0001). This result likely reflects regional differences in the connectivity of BL neurons that impact on their gamma-related activity. DISCUSSION This study reports on the behavioral correlates and cellular mechanisms of gamma oscillations in BL. To address these questions, we recorded unit and LFP activity in rats performing tasks characterized by different threat levels. Our main findings are as follows. First, different gamma bands with distinct behavioral correlates occur in BL. Second, compared to all other frequencies, high-gamma most strongly entrains BL firing, and it is particularly pronounced when the rats’ behavior betrays scanning vigilance. Third, principal cells and fast-spiking interneurons preferentially fire at different phases of high-gamma. Fourth, phase preference and entrainment of BL units by high-gamma do not change appreciably with behavioral states. Fifth, relative to behavioral states, high-gamma only produces minor changes in firing rates. Sixth, the entrainment of principal cells by high-gamma varies as a function of their projection site. Last, high-gamma

A1

B1

C1

D1

A2

B2

C2

D2

A3

B3

C3

D3

Figure 7. Impact of High-Gamma on Firing Synchrony Couples of principal cells (1), interneurons (2), or principal cells and interneurons (3). (A) Cross-correlation of unit activity when high-gamma power was high (blue) or low (red). Average ± SEM of normalized CCGs for all available cell couples. Note that in contrast with couples of principal cells (A1) and interneurons (A2), mixed couples generally fired with a time lag between them, interneuronal discharges following principal cells’ by
2 ms (A3). Thick lines, averages; thin lines, SEM. (B) Synchronization indices for individual cell couples where both neurons generated at least one spike in at least one of the analyzed time windows. The n’s indicate the number of couples included in each analysis, but we only show in (B1) and (B3) the 200 most active cell pairs for clarity. Because principal cells fire rarely, only
40% of cell couples met this criterion. In contrast, it was met by all interneuron pairs as well as most mixed couples. In (B1)–(B3), the black horizontal lines in the blue rectangles indicate the medians. The black vertical lines indicate the range of values including 95% of the observations. The top and bottom edges of the blue rectangles indicate the 75th and 25th percentile, respectively. In (A) and (B), the distance between recorded units was 200 mm. (C) Relation between synchronization index (y axis; average ± SEM) and distance between units (x axis) when high-gamma power was high (blue) or low (red). In (B) and (C), significance levels are indicated, based on signed-rank tests. Provided we controlled for gamma level, we found no effect of states (QW, baseline, waiting, foraging) on SIs in pairs of interneurons (ANOVA F(3,1223) = 0.87, p = 0.45), and couples of principal cells and interneurons (ANOVA F(3,6057) = 1.0, p = 0.39). This analysis could not be performed in pairs of principal cells, due to their low firing rates. (D) Relation between R product and gamma-related firing synchrony. Cell couples were stratified by their R product (numbers on the right are percentiles; thick lines are averages; thin lines are SEM). See also Figure S2.

dramatically increases firing synchrony, particularly among synaptically coupled cells. Behavioral Correlates of Different Gamma Bands Many brain regions display multiple gamma bands with different input dependencies and behavioral or cognitive correlates (Ainsworth et al., 2011; Colgin et al., 2009; van der Meer et al., 2010).

We found that a similar situation exists in BL. In the foraging and shuttle tasks, mid-gamma was most prominent when rats anticipated door opening. In contrast, power in the low- and, especially, high-gamma bands was generally higher when rats were apprehensive. For instance, in the foraging task, low- and highgamma were pronounced during waiting periods that preceded aborted trials or followed a failed trial. On trials where rats

Neuron 97, 656–669, February 7, 2018 665

Connection probability

200 microns

Interneuron R2

0.6 0.4 0.2 0 0.3

0.1

All principal cells

B1 Number of cells

P to INT Excitatory Connection

A1

30

n = 381

20 10 0

0

Number of cells

0.2

Number of cells

0.2

0.3

Principal Cell

0.006

0.004

0.002

0 10

connected (n=260) not connected (n=350) 50

100 150 200 Frequency (Hz)

250

2 0

0

0.1

0.2

0.3

0.4

Backfired from nAc n = 23

6 4 2 0

0

0.1

0.2

0.3

0.4

Backfired from mPFC n = 27

6 4

(A) Connections from principal cells to interneurons separated by 200 mm. (A1) Red and blue are used to represent mono-synaptically connected versus non-connected cell couples, respectively. The x and y axes represent the resultant vector of the two cells in each couple. The graphs on the left and at the bottom plot the probability of connections as a function of the resultant vector. To generate these graphs, we computed the probability of connections in bins containing 20% of the cells. The black lines indicate the borders of the bins. (A2) Average unit-unit coherence (±SEM, dotted lines) plotted as a function of frequency for connected (red) and unconnected (blue) couples of principal cells and interneurons. (B) Entrainment of principal cells by high-gamma varies depending on the cells’ projection site(s). (B) shows frequency distributions of the resultant vector for all presumed principal cells (n = 381; B1), all cells that were positively identified as projection cells by antidromic invasion (n = 56; B2), cells that were backfired from nAc (n = 23; B3), and cells that were backfired from the mPFC (n = 27, B4). Six neurons that were backfired from nAc and mPFC are not included in (B3) and (B4). See also Figures S4–S8.

2 0

0 0.1 0.2 0.3 0.4 Gamma entrainment (Resultant vector)

initiated foraging, low- and high-gamma power increased more during waiting periods that followed failed than successful trials. Moreover, it was also higher on trials where rats foraged tentatively, hesitating as they approached the reward. Underscoring the importance of apprehension, no significant increase in low- and high-gamma power was observed in a shuttle task, which is devoid of explicit threats. While these results suggest that low- and high-gamma are linked to a state of apprehension, the negative correlation between low- and high-gamma power with movement speed, including when rats escaped to the nest in response to the charging predator, suggests that fear per se is not the critical correlate. An associated process such as scanning vigilance seems more likely. This interpretation is consistent with a recently revised model of amygdala function, which asserts that the amygdala mediates an evaluative process that regulates behavioral engagement (Pare´ and Quirk, 2017). High-Gamma Strongly Entrains BL Firing By comparing the entrainment of units to the LFP in different frequency bands, we found that BL cells were more strongly entrained by high-gamma than all other rhythms. Critically, this result does not mean that the main LFP rhythm in BL is highgamma but that when high-gamma occurs, it entrains BL neurons more robustly. In principal cells, gamma-related activity occurred on a mostly silent background and consisted of occasional single spikes concentrated during negative gamma

666 Neuron 97, 656–669, February 7, 2018

0.4

4

B4 Number of cells

Connection probability Coherence

A2

0.1

R1

0.3

n = 56

B3 0

0.2

6

0 0.6 0.4 0.2 0

0.1

All backfired cells

B2

Figure 8. Relation between Connectivity and Entrainment by High-Gamma

phases. By contrast, fast-spiking cells discharged tonically and gamma produced alternating periods of decreased and increased firing rates, the latter occurring mainly after the preferred firing phase of principal cells. While the propensity of principal cells and interneurons to fire at different gamma phases is ubiquitous in cortical networks (Bragin et al., 1995; Atallah and Scanziani, 2009; Hasenstaub et al., 2005; Vinck et al., 2013; reviewed in Buzsa´ki and Wang, 2012), preferential entrainment by high-gamma is not (Newman and Hasselmo, 2014). At present, the origin of this property is unclear. One possibility is that coherent gamma inputs to BL preferentially occur in the high-gamma range. Arguing against this idea, however, entrainment by highgamma (magnitude and phase preference) did not vary as a function of behavioral states, even though such shifts are expected to correlate with changing input patterns to BL. The state-invariant entrainment of BL neurons by high-gamma raises a second possibility, namely, that the intrinsic BL network is such that it gives rise to a preferential resonance at highgamma. The length of the gamma cycle is determined by multiple interacting properties such as the kinetics and reversal potentials of synaptic currents as well as the neurons’ intrinsic properties and connectivity (Buzsa´ki and Wang, 2012). While these properties fall in a similar range in BL and cortical networks (Martina et al., 2001; Spampanato et al., 2011; Wang, 2010; Woodruff and Sah, 2007), this possibility cannot be excluded at this time.

Entrainment by High-Gamma Varies as a Function of the Cells’ Projection Site Entrainment by high-gamma was not expressed homogeneously in projection cells. Whereas most were significantly entrained by high-gamma, a small subset was not, even though both groups had similar firing rates. By comparing how high-gamma entrained neurons backfired from the mPFC or nAc, we found that this heterogeneity was related to the neurons’ projection sites. Neurons projecting to mPFC were strongly entrained whereas most of those targeting nAc were weakly to negligibly so. While we cannot rule out that nAc-projecting neurons would be more strongly entrained by highgamma under different circumstances, our observation that the entrainment of BL cells did not change appreciably across many different states and behaviors argues against this possibility. Rather, it seems more likely that BL cells projecting to distinct sites form different connections with fast-spiking interneurons. Interestingly, the contrasting expression of gamma entrainment as a function of projection sites is paralleled by differences in the connectivity of mPFC neurons that target the amygdala or nAc. Indeed, McGarry and Carter (2016) reported that BL neurons excite corticoamygdala neurons more strongly than those targeting nAc. Moreover, they found that, via PV cells, BL inputs inhibit corticoamygdala neurons more strongly than corticostriatal neurons. In light of these findings, the target-specific expression of high-gamma by principal BL cells emerges as but one manifestation of a larger system design that segregates mPFC and amygdala influences on nAc from those exerted by the mPFC and amygdala on each other.

BL cells by high-gamma may explain how BL neurons impact motivated behaviors despite the fact that emotionally valenced stimuli activate very few of them. Indeed, by clustering the discharges of multiple cells in short recurring time windows, high-gamma should enhance their impact on common postsynaptic targets (Fries et al., 2007; Fries, 2015). Also, depending on whether they coincide with the de- or hyperpolarizing phases of the BL driven gamma oscillation, convergent synaptic inputs should be biased to undergo associative potentiation or depression (Wespatat et al., 2004). An important challenge for future studies will be to test these hypotheses by selectively manipulating high-gamma in real time, which would require novel techniques such as closed-loop optogenetic manipulations. STAR+METHODS Detailed methods are provided in the online version of this paper and include the following: d d d d

d

Impact of High-Gamma on Firing Rates versus Synchrony Whereas high-gamma and behavioral states caused similarly modest changes in the firing rates of interneurons, in projection cells the impact of behavioral states was much larger. Among the significantly entrained cells, the effect of high-gamma ranged from an
25% increase in firing rate to an
10% reduction in activity. While high-gamma had a minor impact on discharge rates, it caused a major increase in firing synchrony, particularly among monosynaptically connected cells. This effect was not ubiquitous however. Gamma-related synchrony decreased rapidly as the distance between the recorded cells increased, much like entrainment by high-gamma decayed with distance between the recorded cell and reference LFP. Also, even with nearby cells, the effect of gamma on synchrony was large when both cells were highly entrained by gamma and low when one or both cells were poorly entrained by gamma. Conclusions Overall, our results indicate that high-gamma synchronizes particular subsets of monosynaptically connected principal cells with specific projection sites and preferential connectivity with fast-spiking interneurons. Critically, gamma-related changes in firing rates are much smaller than those associated with shifts in behavioral states. As a result, modulation of BL activity by high-gamma can easily go undetected unless multiple cells are recorded simultaneously. The synchronization of

d

KEY RESOURCES TABLE CONTACT FOR REAGENT AND RESOURCE SHARING EXPERIMENTAL MODEL AND SUBJECT DETAILS METHOD DETAILS B Overview of the experimental timeline B Surgery B Foraging task B Shuttle task B Analysis of behavior B Histology QUANTIFICATION AND STATISTICAL ANALYSIS B Behavioral states B LFP processing, unit recording and clustering DATA AND SOFTWARE AVAILABILITY

SUPPLEMENTAL INFORMATION Supplemental Information includes eight figures and two tables and can be found with this article online at https://doi.org/10.1016/j.neuron.2017.12.035. ACKNOWLEDGMENTS This material is based upon work supported by NIMH R01 grants MH107239 and MH112505 to D.P. AUTHOR CONTRIBUTIONS A.A., D.B.H., and D.P. contributed to the design of the experiments as well as to the data and statistical analyses. A.A. conducted most of the experiments. S.-C.L. and D.H. conducted some the experiments and contributed to the data analyses. D.P. wrote the first draft of the manuscript and made the figures. All authors contributed to refining the manuscript. DECLARATION OF INTERESTS The authors declare no competing interests. Received: September 12, 2017 Revised: November 30, 2017 Accepted: December 21, 2017 Published: February 7, 2018

Neuron 97, 656–669, February 7, 2018 667

REFERENCES

€ller, C., and Lu €thi, A. (2008). Herry, C., Ciocchi, S., Senn, V., Demmou, L., Mu Switching on and off fear by distinct neuronal circuits. Nature 454, 600–606.

Ainsworth, M., Lee, S., Cunningham, M.O., Roopun, A.K., Traub, R.D., Kopell, N.J., and Whittington, M.A. (2011). Dual g rhythm generators control interlaminar synchrony in auditory cortex. J. Neurosci. 31, 17040–17051.

Jacobs, B.L., and McGinty, D.J. (1971). Amygdala unit activity during sleep and waking. Exp. Neurol. 33, 1–15.

Amano, T., Duvarci, S., Popa, D., and Pare´, D. (2011). The fear circuit revisited: contributions of the basal amygdala nuclei to conditioned fear. J. Neurosci. 31, 15481–15489. Ambroggi, F., Ishikawa, A., Fields, H.L., and Nicola, S.M. (2008). Basolateral amygdala neurons facilitate reward-seeking behavior by exciting nucleus accumbens neurons. Neuron 59, 648–661. Amir, A., Lee, S.C., Headley, D.B., Herzallah, M.M., and Pare, D. (2015). Amygdala signaling during foraging in a hazardous environment. J. Neurosci. 35, 12994–13005. Atallah, B.V., and Scanziani, M. (2009). Instantaneous modulation of gamma oscillation frequency by balancing excitation with inhibition. Neuron 62, 566–577. Bartho´, P., Hirase, H., Monconduit, L., Zugaro, M., Harris, K.D., and Buzsa´ki, G. (2004). Characterization of neocortical principal cells and interneurons by network interactions and extracellular features. J. Neurophysiol. 92, 600–608. Bauer, E.P., Paz, R., and Pare´, D. (2007). Gamma oscillations coordinate amygdalo-rhinal interactions during learning. J. Neurosci. 27, 9369–9379. Bragin, A., Jando´, G., Na´dasdy, Z., Hetke, J., Wise, K., and Buzsa´ki, G. (1995). Gamma (40-100 Hz) oscillation in the hippocampus of the behaving rat. J. Neurosci. 15, 47–60. Buhl, E.H., Halasy, K., and Somogyi, P. (1994). Diverse sources of hippocampal unitary inhibitory postsynaptic potentials and the number of synaptic release sites. Nature 368, 823–828. Buzsa´ki, G., and Wang, X.J. (2012). Mechanisms of gamma oscillations. Annu. Rev. Neurosci. 35, 203–225. Choi, J.S., and Kim, J.J. (2010). Amygdala regulates risk of predation in rats foraging in a dynamic fear environment. Proc. Natl. Acad. Sci. USA 107, 21773–21777. Colgin, L.L., Denninger, T., Fyhn, M., Hafting, T., Bonnevie, T., Jensen, O., Moser, M.B., and Moser, E.I. (2009). Frequency of gamma oscillations routes flow of information in the hippocampus. Nature 462, 353–357. Csicsvari, J., Jamieson, B., Wise, K.D., and Buzsa´ki, G. (2003). Mechanisms of gamma oscillations in the hippocampus of the behaving rat. Neuron 37, 311–322. Engel, A.K., and Singer, W. (2001). Temporal binding and the neural correlates of sensory awareness. Trends Cogn. Sci. 5, 16–25. Fries, P. (2015). Rhythms for cognition: communication through coherence. Neuron 88, 220–235. , D., and Singer, W. (2007). The gamma cycle. Trends Fries, P., Nikolic Neurosci. 30, 309–316. Fujisawa, S., Amarasingham, A., Harrison, M.T., and Buzsa´ki, G. (2008). Behavior-dependent short-term assembly dynamics in the medial prefrontal cortex. Nat. Neurosci. 11, 823–833. Galarreta, M., and Hestrin, S. (1999). A network of fast-spiking cells in the neocortex connected by electrical synapses. Nature 402, 72–75. Gibson, J.R., Beierlein, M., and Connors, B.W. (1999). Two networks of electrically coupled inhibitory neurons in neocortex. Nature 402, 75–79. Gloveli, T., Dugladze, T., Saha, S., Monyer, H., Heinemann, U., Traub, R.D., Whittington, M.A., and Buhl, E.H. (2005). Differential involvement of oriens/pyramidale interneurones in hippocampal network oscillations in vitro. J. Physiol. 562, 131–147. Hasenstaub, A., Shu, Y., Haider, B., Kraushaar, U., Duque, A., and McCormick, D.A. (2005). Inhibitory postsynaptic potentials carry synchronized frequency information in active cortical networks. Neuron 47, 423–435. Hazan, L., Zugaro, M., and Buzsa´ki, G. (2006). Klusters, NeuroScope, NDManager: A free software suite for neurophysiological data processing and visualization. J. Neurosci. Methods 155, 207–216.

668 Neuron 97, 656–669, February 7, 2018

Lee, S.C., Amir, A., Headley, D.B., Haufler, D., and Pare, D. (2016). Basolateral amygdala nucleus responses to appetitive conditioned stimuli correlate with variations in conditioned behaviour. Nat. Commun. 7, 12275. Levy, R.B., and Reyes, A.D. (2012). Spatial profile of excitatory and inhibitory synaptic connectivity in mouse primary auditory cortex. J. Neurosci. 32, 5609–5619. Luo, Q., Holroyd, T., Jones, M., Hendler, T., and Blair, J. (2007). Neural dynamics for facial threat processing as revealed by gamma band synchronization using MEG. Neuroimage 34, 839–847. Luo, Q., Mitchell, D., Cheng, X., Mondillo, K., Mccaffrey, D., Holroyd, T., Carver, F., Coppola, R., and Blair, J. (2009). Visual awareness, emotion, and gamma band synchronization. Cereb. Cortex 19, 1896–1904. Magoun, H.W. (1958). The Waking Brain (Charles C. Thomas). Martina, M., Royer, S., and Pare´, D. (2001). Cell-type-specific GABA responses and chloride homeostasis in the cortex and amygdala. J. Neurophysiol. 86, 2887–2895. McGarry, L.M., and Carter, A.G. (2016). Inhibitory gating of basolateral amygdala inputs to the prefrontal cortex. J. Neurosci. 36, 9391–9406. Muller, J.F., Mascagni, F., and McDonald, A.J. (2003). Synaptic connections of distinct interneuronal subpopulations in the rat basolateral amygdalar nucleus. J. Comp. Neurol. 456, 217–236. Muller, J.F., Mascagni, F., and McDonald, A.J. (2005). Coupled networks of parvalbumin-immunoreactive interneurons in the rat basolateral amygdala. J. Neurosci. 25, 7366–7376. Newman, E.L., and Hasselmo, M.E. (2014). Grid cell firing properties vary as a function of theta phase locking preferences in the rat medial entorhinal cortex. Front. Syst. Neurosci. 8, 193. Otte, S., Hasenstaub, A., and Callaway, E.M. (2010). Cell type-specific control of neuronal responsiveness by gamma-band oscillatory inhibition. J. Neurosci. 30, 2150–2159. Oya, H., Kawasaki, H., Howard, M.A., 3rd, and Adolphs, R. (2002). Electrophysiological responses in the human amygdala discriminate emotion categories of complex visual stimuli. J. Neurosci. 22, 9502–9512. Pagano, R.R., and Gault, F.P. (1964). Amygdala activity: a central measure of arousal. Electroencephalogr. Clin. Neurophysiol. 17, 255–260. Pare´, D., and Gaudreau, H. (1996). Projection cells and interneurons of the lateral and basolateral amygdala: Distinct firing patterns and differential relation to theta and delta rhythms in conscious cats. J. Neurosci. 16, 3334–3350. Pare´, D., and Quirk, G.J. (2017). When scientific paradigms lead to tunnel vision: lessons from the study of fear. npj Science of Learning 2. Published online March 27, 2017. https://doi.org/10.1038/s41539-017-0007-4. €nen, A. (2000). Connectivity of the rat amygdaloid complex. In The Pitka Amygdala: A Functional Analysis, J.P. Aggleton, ed. (Oxford University Press), pp. 31–115. Popescu, A.T., Popa, D., and Pare´, D. (2009). Coherent gamma oscillations couple the amygdala and striatum during learning. Nat. Neurosci. 12, 801–807. Puig, M.V., Ushimaru, M., and Kawaguchi, Y. (2008). Two distinct activity patterns of fast-spiking interneurons during neocortical UP states. Proc. Natl. Acad. Sci. USA 105, 8428–8433. Randall, F.E., Whittington, M.A., and Cunningham, M.O. (2011). Fast oscillatory activity induced by kainate receptor activation in the rat basolateral amygdala in vitro. Eur. J. Neurosci. 33, 914–922. Scheffer-Teixeira, R., Belchior, H., Lea˜o, R.N., Ribeiro, S., and Tort, A.B. (2013). On high-frequency field oscillations (>100 Hz) and the spectral leakage of spiking activity. J. Neurosci. 33, 1535–1539. Schoffelen, J.M., Oostenveld, R., and Fries, P. (2005). Neuronal coherence as a mechanism of effective corticospinal interaction. Science 308, 111–113.

Schoffelen, J.M., Poort, J., Oostenveld, R., and Fries, P. (2011). Selective movement preparation is subserved by selective increases in corticomuscular gamma-band coherence. J. Neurosci. 31, 6750–6758. Sinfield, J.L., and Collins, D.R. (2006). Induction of synchronous oscillatory activity in the rat lateral amygdala in vitro is dependent on gap junction activity. Eur. J. Neurosci. 24, 3091–3095. Smith, Y., Pare´, J.F., and Pare´, D. (2000). Differential innervation of parvalbumin-immunoreactive interneurons of the basolateral amygdaloid complex by cortical and intrinsic inputs. J. Comp. Neurol. 416, 496–508. Spampanato, J., Polepalli, J., and Sah, P. (2011). Interneurons in the basolateral amygdala. Neuropharmacology 60, 765–773. Steriade, M., and Hobson, J. (1976). Neuronal activity during the sleep-waking cycle. Prog. Neurobiol. 6, 155–376. Stuber, G.D., Sparta, D.R., Stamatakis, A.M., van Leeuwen, W.A., Hardjoprajitno, J.E., Cho, S., Tye, K.M., Kempadoo, K.A., Zhang, F., Deisseroth, K., and Bonci, A. (2011). Excitatory transmission from the amygdala to nucleus accumbens facilitates reward seeking. Nature 475, 377–380. Stujenske, J.M., Likhtik, E., Topiwala, M.A., and Gordon, J.A. (2014). Fear and safety engage competing patterns of theta-gamma coupling in the basolateral amygdala. Neuron 83, 919–933. Tye, K.M., Prakash, R., Kim, S.-Y., Fenno, L.E., Grosenick, L., Zarabi, H., Thompson, K.R., Gradinaru, V., Ramakrishnan, C., and Deisseroth, K. (2011).

Amygdala circuitry mediating reversible and bidirectional control of anxiety. Nature 471, 358–362. van der Meer, M.A., Kalenscher, T., Lansink, C.S., Pennartz, C.M., Berke, J.D., and Redish, A.D. (2010). Integrating early results on ventral striatal gamma oscillations in the rat. Front. Neurosci. 4, 300. Vandecasteele, M., M, S., Royer, S., Belluscio, M., Bere´nyi, A., Diba, K., Fujisawa, S., Grosmark, A., Mao, D., Mizuseki, K., et al. (2012). Large-scale recording of neurons by movable silicon probes in behaving rodents. J. Vis. Exp. 4, e3568. Vinck, M., Womelsdorf, T., Buffalo, E.A., Desimone, R., and Fries, P. (2013). Attentional modulation of cell-class-specific gamma-band synchronization in awake monkey area v4. Neuron 80, 1077–1089. Wang, X.J. (2010). Neurophysiological and computational principles of cortical rhythms in cognition. Physiol. Rev. 90, 1195–1268. Wespatat, V., Tennigkeit, F., and Singer, W. (2004). Phase sensitivity of synaptic modifications in oscillating cells of rat visual cortex. J. Neurosci. 24, 9067–9075. Woodruff, A.R., and Sah, P. (2007). Networks of parvalbumin-positive interneurons in the basolateral amygdala. J. Neurosci. 27, 553–563. Zanos, T.P., Mineault, P.J., and Pack, C.C. (2011). Removal of spurious correlations between spikes and local field potentials. J. Neurophysiol. 105, 474–486.

Neuron 97, 656–669, February 7, 2018 669

STAR+METHODS KEY RESOURCES TABLE

REAGENT or RESOURCE

SOURCE

IDENTIFIER

N/A

N/A

Charles River Laboratories

RGD_734476

Deposited Data The full dataset and custom MATLAB code will be made available upon request. Experimental Models: Organisms/Strains Sprague Dawley rat Software and Algorithms MATLAB

MathWorks

MATLAB R2014b

KlustaKwik

http://klustakwik.sourceforge.net/

N/A

Klusters

Hazan et al., 2006

N/A

Chronux

http://chronux.org

N/A

Other Silicon probe Buzsaki64L

NeuoNexus

N/A

Plexon amplifier

Plexon

N/A

CONTACT FOR REAGENT AND RESOURCE SHARING Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Denis Pare´ ([email protected]). EXPERIMENTAL MODEL AND SUBJECT DETAILS Procedures were approved by the Institutional Animal Care and Use Committee of Rutgers University, in compliance with the Guide for the Care and Use of Laboratory Animals (DHHS). We used naive male Sprague-Dawley rats (10-11 weeks old and 310-360 g at the beginning of experiments, Charles River Laboratories, New Field, NJ). METHOD DETAILS Overview of the experimental timeline Rats were maintained on a 12 h light/dark cycle. Experiments were performed during the light phase. After habituation to the animal facility and handling, rats underwent a stereotaxic surgery during which silicon probes were inserted in the amygdala. Following the surgery, rats were housed individually with ad libitum access to water. The behavioral experiments began R 7 days after the surgery. Surgery Rats were anesthetized with isoflurane and administered atropine sulfate (0.05 mg/kg, i.m.) to aid breathing. In aseptic conditions, rats were mounted in a stereotaxic apparatus. A local anesthetic (bupivacaine, sc) was injected in the scalp and 15 min later, the scalp was incised and a craniotomy was performed above the amygdala. Then, silicon probes (Neuronexus, Ann Arbor, MI) attached to Buzsaki-style microdrives (Vandecasteele et al., 2012) were aimed at BL under stereotaxic guidance. Silicon probes consisted of either eight (3 rats) or four (3 rats) shanks (inter-shank distance of 200 mm), each with 8 recording leads (de-insulated area, 144 mm2) separated by
20 mm dorsoventrally. One of the six rats had to be excluded because movements caused artifacts in local field potentials (LFPs). In three of the five remaining rats, a second craniotomy was performed above nAc and mPFC. Then, pairs of tungsten stimulating electrodes (inter-tip spacing, 1-1.7 mm) were inserted into these two structures. We used the following stereotaxic coordinates (all expressed in mm relative to bregma, and where AP, ML, and DV stand for anteroposterior, mediolateral, and dorsoventral, respectively). BL, AP 2.2 to 3.6, ML 5 to 5.3, DV 8.8; mPFC, AP 2.7 to 3.7, ML 0.5, DV 3.6 to 5.2; nAc, AP 1.5, ML 1.35, DV 6.7. Rats were allowed two to three weeks to recover from the surgery. Foraging task To ensure proper motivation in the behavioral tasks, daily access to food was restricted in time so that the rats’ bodyweight was maintained at about 90% of age-matched subjects with continuous access to food.

e1 Neuron 97, 656–669.e1–e7, February 7, 2018

Foraging apparatus The foraging apparatus was a long rectangular alley (length, 245 cm width, 60 cm) with high walls (60 cm) and no ceiling. A door (height, 50 cm; width, 10 cm) divided it into two compartments. One compartment (length, 30 cm) was dimly lit (10 Lux) and served as a nest area with a water bottle. The other compartment was a much longer (215 cm) and brighter (200 Lux) foraging arena. Mechanical predator On some trials, a mechanical predator (Mindstorms, LEGO systems, Billund, Denmark; length, 34 cm; width, 17 cm; height, 14 cm) on wheels was positioned at the end of the foraging arena, facing the nest. It was equipped with a sensor that detected the rats’ approach, triggering a sudden forward movement (80 cm at 60 cm/sec) and repeated jaw openings and closings, followed by return to its original position. Habituation to nest (Days 0-1) Rats were first habituated to the nest for 2 daily consecutive sessions of 7 h with the door shut at all times. On habituation days, rats could consume up to 6 g of food (sweet cereal pellets). Foraging in the absence of predator (Days 2-3) On the next two days, in the absence of the predator, rats were trained to retrieve sweetened food pellets (80-100 mg) in the foraging arena. No food was available in the nest. One each trial, a single food pellet was placed at various distances from the door. The door was then opened. After a period of hesitation at the doorway, rats retrieved the food pellet and returned to the nest to consume it. Upon return to the nest, the door was closed. Initially, the distance between the nest and food pellet was progressively increased in steps of 25 cm (from 25 to150 cm), after three successful trials at each distance. Later on, the distance was varied randomly from trial to trial. Foraging in the presence of the predator (Days 4-5) On the next two days, rats were again given the opportunity to retrieve food pellets in the foraging arena. However,
60% of trials were conducted with the predator present. Alternating blocks of trials with (n = 10-20) or without (n = 10-15) the predator were conducted. All our analyses are based on activity recorded in Days 3-5. Shuttle task In this task, rats ran back and forth between two nest-like areas (length, 50 cm; width, 20 cm) through a central compartment (length, 50 cm) to retrieve food pellets at the end of the other nest. The apparatus was made of black Plexiglas, had walls 45 cm in height, and was dimly illuminated (10 Lux). Retractable doors separated the nests and central compartment. Prior to the recordings, rats received extensive habituation to the apparatus with the doors open. During the recordings, rats were placed in one of the nests and a food pellet in the other. After opening of the doors, rats immediately ran to the other nest to retrieve and consume the food. While rats consumed the food, the door was closed. The inter-trial interval was R 1min. Three of the five rats used for the foraging task also performed the shuttle task. In these three rats, we alternated between the two tasks during different recording sessions. Analysis of behavior Behavior was recorded by an overhead videocamera at a frame rate of 29.97 Hz and analyzed using the following two approaches. First, the position and velocity of the rats was automatically detected using a MATLAB script that considered the shifting distribution of light intensity across frames. In addition, an experienced observer conducted a frame-by-frame analysis of the video file to ascertain when rats started waiting at the door threshold (defined as when the rat’s snout extended beyond the door into the foraging arena), when they initiated foraging (defined as the last frame of immobility prior to completely moving out of the nest), retrieved the food pellet, escaped, and retreated into the nest. The observer also noted the outcome of each trial (failure or success). Escape onset was defined as when rats, after approaching the food pellet, abruptly turned around and ran all the way to the nest. When detecting transitions between immobility and movement, the following conditions had to be met. In order not to mistake waiting for interrupted foraging, rats had to be at least 30 cm away from the nest. Also, rats had to be completely immobile for R 1 s, followed by R 1 s of forward movement with a peak speed R 15 cm/sec, followed by another pause for R 1 s. In the second pause, rat had to be R 30 cm from the food pellet. The last condition was imposed to avoid confusing hesitation with reaching the food. After automatically detecting potential transitions, we visually inspected each and eliminated cases where rats did not face the predator. Histology At the end of the experiments, the animals were deeply anesthetized. On each shank, one of the recording sites was marked with a small electrolytic lesion (10 mA between a channel and the animals’ tail for 10 s). Rats were then perfused-fixed through the heart, their brains extracted, cut on a vibrating microtome and the sections counterstained with cresyl violet. We only considered neurons that were histologically-determined to have been recorded in BL. QUANTIFICATION AND STATISTICAL ANALYSIS All grouped data are reported as averages ± SEM. All statistical tests were two-sided. When comparing only two conditions, we used rank-sum or signed-rank tests, as appropriate. To assess the statistical significance of differences in LFP power between different phases of the same trial type, we computed non-parametric repeated-measures Friedman ANOVAs. For between trial type

Neuron 97, 656–669.e1–e7, February 7, 2018 e2

comparisons, we used a non-parametric Kruskal-Willis ANOVA for independent samples. In both cases, each trial was an observation of the mean power in a particular frequency band across the behavioral epoch of interest. Post hoc comparisons were performed with Dunn tests using a p value of 0.05. In a few instances, when assessing the statistical significance of multiple factors (e.g., frequency, distance between recording sites), we had to use a parametric ANOVA as there are no non-parametric alternatives. MATLAB was used for statistical analyses. Behavioral states Periods of slow-wave sleep were excluded from all the analyses reported in this study. States of vigilance were distinguished using spectral analyses of LFPs and behavioral observations. Spontaneous LFP activity was segmented in five-second windows and frequency distributions of LFP power in different frequency bands computed. Active waking could be distinguished from all other states because it was associated with a broadband increase in the power of high frequencies (200-240 Hz), reflecting electromyographic activity. After eliminating active waking, we could easily distinguish slow-wave sleep from quiet waking because total power at frequencies < 20 Hz was distributed bimodally between the two states: epochs of high power at low frequencies corresponded to periods of slow-wave sleep. LFP processing, unit recording and clustering BL unit and LFP recordings were performed during all the phases of the behavioral protocols described above with the exception of habituation. However, the first training day of the foraging task was not included in the analyses because the rats’ performance was poor on that day. In rats that had been implanted with stimulating electrodes, at the end of each behavioral session, electrical stimuli (300-600 mA, 0.1 ms) were delivered at 1 Hz to determine whether recorded cells could be antidromically invaded from one or more of the stimulated sites. Then, the silicon probe was lowered R 30 mm (40 ± 10 mm), R 8 hours ahead of the next recording session, to ensure mechanical stability. In our experience, when recordings are performed shortly after moving the probes, spike shapes are not stable. Presumably, this is the case because the surrounding brain tissue has not completely adapted to the probe’s movement. In contrast, if we perform recordings R 8 hours after moving the probes, spike shapes are stable. Note that it is possible that the same cells were sampled from on different days. However, the incidence of such recordings is impossible to assess because moving the probes changes spike shapes. The signals were sampled at 25 kHz and stored on a hard drive. The data was first high-pass filtered using a median filter (window size of 1.1 ms), then thresholded to extract spikes. We then ran PCA on the spikes and the first three components were clustered using KlustaKwik (http://klustakwik.sourceforge.net/). Spike clusters were then refined manually using Klusters (Hazan et al., 2006). The reliability of cluster separation was verified by inspecting auto- and cross-correlograms (CCGs). Autocorrelograms had to display a refractory period of at least 2 ms. CCGs should not show evidence of a refractory period, as this feature betrays overlap between clusters. Units with unstable spike shapes during a given recording session were excluded. Identification of principal cell and interneurons Cells with firing rates < 6 Hz and spike peak-to-through times > 0.6 ms were classified as principal cells; units with firing rates R 6 Hz and peak-to-through spike times % 0.6 ms, as interneurons. To determine spike duration, we first selected the channel where, for a given cell, action potentials had the largest peak to trough amplitude. We then measured spike duration as the time between spike trough and peak (Bartho´ et al., 2004). To test the reliability of our classification criteria, using CCGs, we looked for evidence of monosynaptic inhibition from putative principal cells to other cells (which would be an instance where an interneuron was misclassified as a projection cell) or of excitatory connections from putative interneurons to other cells (which would be an instance where a principal cell was misclassified as an interneuron). A total of 15,514 cell pairs were examined for the first case, and 2046 cell pairs for the second. Of these, respectively 5.82% (or 904) and 5.37% (or 110) formed a putative connection. Supporting the validity of our classification criteria, the incidence of misclassified connections was extremely low (Principal cells: 0.99% or 9 of 904; Interneurons: 3.6% or 4 of 110). Furthermore, repeating the analyses shown in Figures 3, 4, 5, 6, 7, and 8 on these restricted samples of BL cells yielded qualitatively identical results. Antidromic action potentials were identified as such when they had a fixed latency (%0.1 ms jitter) and collided with spontaneously occurring spikes. Nomenclature used for gamma Based on their different frequencies and contrasting behavioral correlates, we distinguished three different gamma bands, which we termed low- (35-55 Hz), mid- (55-75 Hz) and high-gamma (75-95 Hz). We wish to emphasize that high-gamma is not non-specific high frequency activity that shows up non-rhythmically in the higher frequencies, but a genuine LFP rhythm in the band identified. Processing of LFPs Shanks that were located outside BL were not considered. LFP analyses involved many steps. We first enumerate these steps and then discuss each step in detail. The LFP processing steps were (1) down-sampling, (2) extraction of one LFP per shank and recording session, (3) calculation of spectrogram for all foraging trials on each day and for each shank, (4) calculation of the average and standard deviation of the spectrogram for each day and shank, (5) normalization of each trial by the above average and standard deviation, (6) time normalization, (7) average of normalized spectrogram across shanks, and finally, (8) average across trials of a given type.

e3 Neuron 97, 656–669.e1–e7, February 7, 2018

(1) Down-sampling: signals were first down-sampled to 1250 Hz. (2) Extraction of one LFP per shank and recording session: We extracted a single LFP for each shank by taking the median LFP signals of the eight leads on the corresponding shank. (3) Calculation of spectrogram for all foraging trials on each day and for each shank: We used the Chronux spectral analysis MATLAB toolkit. We used the function mtspecgramc to calculate a multi-taper time-frequency spectrogram. The number of tapers was set to 5 and the time-bandwidth product to 3. To compensate for power attenuation with frequency, power was calculated as its natural logarithm. We used a 0.8 s window sliding in 0.2 s steps. The only exception to this rule is when we analyzed trials where rats showed signs of hesitation during foraging, in which case 0.1 s steps were used. For each shank on a given day, we calculated the power spectrogram during all available foraging trials, excluding the inter-trial periods. Below, we refer to this spectrogram as S(shank, freq, time). (4) Calculation of the average and standard deviation of the spectrogram for each day and shank: The average and SD was calculated as: N P

meanSðshank; freqÞ = time

Sðshank; freq; timeÞ N

vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 u0 P u N 2 u ðSðshank; freq; timeÞ  meanSðshank; freqÞÞ C uBtime C: B stdSðshank; freqÞ = u A t@ N1 These averages and standard deviations were later used to normalize (or in other words z-score) all the trials, including the shuttle trials. This approach allowed us to compare different trial types using the same metric. (5) Normalization of each trial by the above average and standard deviation: To normalize each trial, we used the following equation zStrial ðshank; freq; timeÞ =

Strial ðshank; freq; timeÞ  meanSðshank; freqÞ : stdðshank; freqÞ

The rationale for z-scoring LFP power is the following. First, due to variations in the properties of electrodes, signal amplitudes can vary between shanks. As a result, not z-scoring would give more weight to data obtained with some shanks. Second, low frequencies have a much higher power than high frequencies. As a result, when power is not z-scored, it is difficult to visually detect changes in the power of high frequencies in raw spectrograms. (6) Time normalization: In order to average trials of different durations, we computed relative time, using the linear interpolation function interp1 of MATLAB. Effectively, this function distributes a fixed number of power samples across the various phases (20 for baseline, waiting, and foraging; 5 for escape). So for example, if a given waiting period lasted 40 s, the power data was pooled into 20 consecutive samples. When the duration of the phase under consideration was too short (< 0.8 s), the trial was excluded. Of course, when using actual time, we skipped this step. (7) Average of normalized spectrogram across shanks: For each trial, we then computed the average z-scored spectrogram across shanks using the following formula. Ns P

zStrial ðfreq; normTimeÞ =

zStrial ðshank; freq; normTimeÞ

shank

Ns

:

Neuron 97, 656–669.e1–e7, February 7, 2018 e4

(8) Average across trials of a given type: Last, we computed a grand average of all trials of a given sub-type using the following equation. Ntrial P

zSðfreq; normTimeÞ =

zStrial ðfreq; normTimeÞ

trial

Ntrial

:

Generalized linear model We computed a generalized linear model (GLM) relating LFP power in different frequency bands to movement speed, presence or absence of the predator, and outcome of the previous trials. After calculating the z-scored power spectrum of each individual trial across shanks, trials were labeled for the presence or absence of the predator (respectively 1 or 0) and whether the prior trial was a failure or success (respectively 1 or 0). In addition, instantaneous velocity was measured. Next, we combined the data from all available trials and z-scored the velocity, robot presence/absence and prior trial outcome. Last, we fit a GLM as follows: zSpectrumðfreq; timeÞ = BetaVelðfreq; timeÞ  zVelðtimeÞ + BetaRobotðfreq; timeÞ  zRobotðtimeÞ + : + BetaTrialOutcomeðfreq; timeÞ  zTrialOutcomeðtimeÞ + constant Similar results were obtained when the trial outcome and Robot variables were not z-scored. Relationship between unit activity and field potentials We referenced unit activity to LFPs picked up by a shank located 200, 400, 600 and 800 mm away from the unit of interest. When two shanks at the same distance from the recorded cell were available, we selected the shank nearest to the center of the silicon probe. For calculating the entrainment and preferred phase of cells, we first bandpass filtered the LFPs in the frequency band of interest using the 2 pole butterworth filter using the filtfilt MATLAB function and applied a Hilbert transform to the resulting signal to determine the phase and amplitude at each point. Then, each spike was assigned a phase. Last, using the MATLAB Toolbox for Circular Statistics, we computed the mean resultant vector length for circular data (circ_r), a measure of entrainment, the mean phase (circ_mean), and determined whether the cell was significantly entrained using a Rayleigh test. However, the resultant vector metric is positively biased with low spike counts. That is, even if cells are not entrained by an oscillation, the resultant vector does not approach zero when few spikes are considered. Thus, before conducting the analyses shown in our paper, we examined the reliability of the resultant vector. Specifically, we randomly selected gradually increasing subsets of spikes from the same spike trains to determine the minimum number of spikes needed for the resultant vector to become reliable. These tests revealed that resultant vector values stabilize when the spike train includes > 100 spikes. To place these figures in context, note that in our samples, the average spike count was 9803 for principal cells and 618,616 for interneurons. The principal with the lowest spike count we considered included 119 spikes and only 13 principal cells (out of 381) had trains of less than 500 spikes. We estimated that, on average, the R vector was biased by less than
1% for principal cells and less than
0.1% for interneurons. Note that only in Figure 4 did we have to exclude cells because they had less than 100 spikes in some of the conditions considered. When comparing preferred phase across states, statistical significance of the change was assessed using the non-parametric multi-sample test for equal medians (circ_cmtest) with a significance threshold of 0.05. Cells that fired < 100 spikes in the period under consideration were excluded. When computing peri-event histograms of neuronal discharges around oscillatory bursts, we used the following approach. The signal was first bandpass filtered in the frequency band of interest as described above. Next, the standard deviation of the signal was calculated and bursts of oscillations that included at least three consecutive peaks R 2 standard deviations were detected. The largest trough was used as temporal reference of the peri-event histograms. Two approaches were used to bin the frequency space (4-250 Hz). In all cases, the frequency space was divided in 69 bins that partially overlapped and shifted in different increments, as described below. The approach illustrated in the Results consisted of gradually increasing the bandwidth as the frequency under consideration (from 4 Hz in the theta range to 20 in the gamma range), generally matching the bandwidth of the oscillations that varied in a behavior-dependent manner in the foraging task. The specific binning scheme is detailed below, using the following abbreviations: Freq, center frequency; freq_high, high cutoff frequency; freq_low, low cutoff frequency; deltaFreq, bandwidth. Freq < 8 8 < Freq < 20 20 < Freq < 45 Freq > 45

deltaFreq = 4 deltaFreq = 4 + ðFreq  8Þ  1=2 : deltaFreq = 10 + ðFreq  20Þ  2=5 deltaFreq = 20

The above method of frequency binning was modified in the following way. The bandwidth was always 10 Hz and the center frequencies shifted logarithmically in 0.2 Hz increments in the theta range to 15 Hz in the gamma range. All these binning methods gave qualitatively identical results. Impact of behavioral states on the entrainment of unit activity by gamma Because gamma power varied with behavior, when comparing entrainment of unit activity across states, we stratified the data by gamma magnitude in the baseline period and then applied the same stratification scheme to the data obtained in other states. As

e5 Neuron 97, 656–669.e1–e7, February 7, 2018

a result, the percent time spent in the different strata varied with states but the unit activity in each stratum can be compared directly across states. Using a GLM to compare the influence of states versus gamma on firing rates To compare changes in firing rates associated with states versus gamma power, we used a GLM with a Poisson distribution. We used non-overlapping time windows of 12 ms, but longer windows (50 or 100 ms) yielded qualitatively identical results. For each cell and time window, the number of spikes, z-scored gamma amplitude, and a dummy-coded state vector (quiet waking, baseline, waiting, foraging) were tallied. We then computed a GLM, with the intercept set to the quiet waking period, to find for each cell, beta values associated with variations in task-related behavioral states and gamma power. The GLM fitted the number of spikes as follows: NumSpike = NumSpikeQW  expðbetaForaging  Sf + betaWaiting  Sw + betaBaseline  Sb + betaGamma  ZscoreGammaÞ where Sf = 1 during foraging state, otherwise Sf = 0; Sw = 1 during waiting state, otherwise Sw = 0; Sb = 1 during baseline state, otherwise Sb = 0. Quantifying the influence of high-gamma on firing rates To account for the variable duration of gamma cycles and the dependence of firing rates on gamma phase, we used the following approach. After detecting gamma bursts as described above, for each burst, the largest gamma cycle was identified and we counted the number of spikes generated during this cycle and ± 55-80 ms from the center of the gamma burst. After dividing the number of spikes by duration of the windows, we then computed (FR(gamma)-FR(no gamma)) / FR(no gamma). Quantifying the influence of gamma on the synchronization of unit activity To quantify the impact of gamma on firing synchrony, we computed a synchronization index (SI) when gamma power was high or low. This was done for all available cell couples, using the LFP derived from the shank where the reference cell was located. After detecting high amplitude gamma bursts as described above, the highest gamma peak within each burst was located. In addition, for each gamma burst, we identified the lowest gamma peak within ± 0.5 s of the highest peak. Then, we computed two CCGs, using the activity within ± 18 ms of the highest and lowest gamma peaks, respectively. Last, in both CCGs, we measured the number of counts within ± 3 ms of the CCG’s time zero (‘‘on window’’) as well as 6 to 3 ms and 3 to 6 ms from the CCG’s time zero (‘‘off window’’). For both the high- and low-amplitude gamma peaks, we computed a SI using the spike counts in the on and off windows as follows: (on windows–off windows)/(on windows+off windows). The rationale for taking into account spike counts in the off windows when computing the SI is to control for the influence of slow changes in firing rates on synchrony estimates. Importantly, the following changes in the definition of the on and off windows yielded qualitatively identical results: we tested off windows of 47 to 50 and 47 to 50 ms as well as on windows of 1.5 to 1.5 ms and off windows of 3 to 1.5 and 1.5 to 3 ms. In order to successfully calculate the SI of a cell pair, at least one pair of spikes in one of the windows had to be detected. To be included in the statistical analyses (signed-rank test), a cell couple had to satisfy this condition during both the high and low gamma periods. When carrying out this analysis for cell couples that included a principal cell and an interneuron, since principal cells generally fire
2 ms before interneurons, the on and off windows were shifted by 2 ms. Finally, to test whether synchronization and entrainment by high gamma are related, for each cell couple, we computed the product of their resultant vectors. We then correlated these products to their SI during periods of high amplitude gamma (Spearman r). Comparing gamma-related unit synchrony as a function of behavior For each pair of cells separated by 200 mm, we calculated two SIs as above for the baseline and foraging periods, but regardless of gamma amplitude. We only considered cell pairs in which we could compute the two SIs (for the baseline and foraging periods) and used signed-rank tests to assess significance. Identifying putative monosynaptic connections The spikes of all simultaneously recorded cells were cross-correlated, separately considering couples that included a principal cell and an interneuron or two principal cells, and further stratifying the data as a function of the distance between the two cells (200, 400, or 600 mm). Each cross-correlogram (CCG) was computed with a bin size of 1 ms in windows of ± 29.5 ms. To determine whether peaks or troughs in the CCGs were significant, we used the procedure described in Fujisawa et al. (2008). First, for each couple, the spikes of the reference cell were jittered randomly within an interval of ± 5 ms. Then, the jittered spike train from the reference cell and the original spike train from the target cell were cross-correlated. This procedure was repeated 1000 times to compute 1000 surrogate jittered CCGs. The minimum and maximum values (from 29.5 to 29.5 ms) of each of the surrogate CCGs were extracted to compute a distribution of 1000 global minimum and 1000 global maximum values. From these, the 99% of the maximum distribution was considered the global maximum band and the lower 1% of the minimum distribution was considered as the global minimum band. In order to be considered a significant excitatory monosynaptic connection, the peak of the actual CCG had to exceed the global maximum band. In addition, the peak of the actual CCG had to be located within 0.5 to 4.5 ms of the zero time. For inhibitory connections, the trough of the actual CCG had to be lower than the global minimum band for at least two consecutive bins. In addition, the trough of the actual CCG had to be located within 0.5 to 5.5 ms of the zero time. Finally, in order to be classified as a putative monosynaptic connection, the CCG at time 0 (0.5 to 0.5 ms) should not exceed the global maximum band or be below the global minimum band. See Figure S3.

Neuron 97, 656–669.e1–e7, February 7, 2018 e6

Analyzing the frequency dependence of unit-unit coherence For a given pair of neurons, we extracted the autocorrelation of the reference cell (ACG1), the autocorrelation of the target cell (ACG2) as well as their cross correlation (CCG). The coherence between these cells was derived using a Fast Fourier Transform (fft) as follows:   coherence spectrum of a given pair = fftðCCGÞ ðfftðACG1Þ  fftðACG2ÞÞ^ 0:5 : For each type of cell couple, we then averaged absolute coherence values across all available pairs. To obtain CCGs corrected for the autocorrelation structure of each unit, we computed an inverse Fast Fourier Transform on the coherence spectrum. DATA AND SOFTWARE AVAILABILITY The full dataset will be made available upon request.

e7 Neuron 97, 656–669.e1–e7, February 7, 2018

Neuron, Volume 97

Supplemental Information

Vigilance-Associated Gamma Oscillations Coordinate the Ensemble Activity of Basolateral Amygdala Neurons Alon Amir, Drew B. Headley, Seung-Chan Lee, Darrell Haufler, and Denis Paré

Supplementary figure 1. Analyses based on raw or z-scored power support the same conclusions regarding changes in gamma activity during the foraging task. (A1) Average raw power spectrogram for the same subjects and trials as in Figure 1C1. (A2) Average ± SEM time-dependent fluctuations in the raw power of low (blue) and high-gamma (red). (B1) Average raw power spectrum ± SEM during baseline (blue) and foraging (red). Note that the error bars are not visible due to low variability. (B2) Difference between traces shown in B1 (Foraging minus baseline) ± SEM. (C) Z-scored spectrograms of five different rats. Related to Figure 1.

1

Supplementary figure 2. (A) Method used to quantify the influence of high-gamma on firing rates. Top traces, raw data. Second trace, bandpass filtered LFP between 75-95 Hz. We first detected gamma bursts greater than two standard deviations of the signals for three consecutive cycles (cyan lines) and for each burst identified the largest gamma cycle (black lines). Next, we counted the number of spikes generated by each cell during this cycle and ± 55 to 80 ms from the center of the gamma burst (red lines below). After dividing the number of spikes by the duration of the windows, we then computed (FR(gamma)-FR(no gamma)) / FR(no gamma). (B) Dark blue trace, bandpass filtered LFP between 75-95 Hz. First, we detected high-gamma events (blue bars; 36 ms windows) using a threshold of three consecutive peaks above two standard deviations (cyan line). For each gamma event, we detect a no-gamma event (red bars; 36 ms windows), as describe in the Methods. (C) We identified the spikes associated to the gamma (C1) and, separately, no-gamma events (C2) and constructed a CCG for each. Black and red bars indicate on-windows and off-windows, respectively. Last, we computed synchronization indices (SIs) for the gamma and nogamma events (equations at bottom). Related to figures 6 and 7.

2

Supplementary figure 3. (A) Relation between gamma power and neuronal synchrony. For 610 couples of principal cells and interneurons, we computed normalized CCGs, averaged them in separate groups based on gamma power. Note that synchronization increased markedly past the 60th percentile. Thus, when looking for monosynaptic connections, we used periods in the lower half of the gamma power distribution so as not to misclassify gamma-related increases in synchrony as monosynaptic connections. (B) Examples of CCGs showing putative monosynaptic connections between principal cells and interneurons. (B1) Principal cells contacting an interneuron. (B2) Interneuron contacting a principal cell. As a control, we randomly shifted spike times by 0 to ± 5 ms a thousand times, recomputed the CCG each time and averaged the results (dashed red line). The solid red lines show the jittered CCGs with the 1% (top line) or 99% (bottom line) value. The top and bottom horizontal blue lines represent the maximum and minimum global bands (see Methods). Related to Figure 8.

3

Supplementary figure 4. Relation between connectivity and entrainment by high-gamma. Connections from (A) principal cells to interneurons or (B) interneurons to principal cells. In A and B, the top and bottom rows show the same data but with individual cells (circles) in the top row vs. bins (squares) in the bottom row. Red and blue are used to represent mono-synaptically connected vs. non-connected cell couples, respectively. The x- and y-axes represent the resultant vector of the two cells in each couple. From left to right, the columns indicate couples where the cells are separated by 200, 400, or 600 µm. The red and blue bars indicate the probability of observing connected (red) and non-connected (blue) cell couples. Green and black symbols indicate the mean and SEM of the connected and non-connected pairs, respectively. Related to Figure 8.

4

Supplementary figure 5. Relation between connectivity and entrainment by high-gamma. Connections between principal cells. The top and bottom rows show the same data but with individual cells (circles) in the bottom row vs. bins (squares) in the two row. Red and blue are used to represent mono-synaptically connected vs. non-connected cell couples, respectively. The x- and y-axes represent the resultant vector of the two cells in each couple. From left to right, the columns indicate couples where the cells are separated by 200, 400, or 600 µm. The red and blue bars indicate the probability of observing connected (red) and non-connected (blue) cell couples. Green and black symbols indicate the mean and SEM of the connected and non-connected pairs, respectively. Note that the denominators above each of the top graphs refer to the number of possible connections, which correspond to double the number of cell pairs. Related to Figure 8.

5

Supplementary figure 6. Computing unit-unit coherence. Two mixed cell couples, one connected (A-D), the second unconnected (E-H). We computed the autocorrelogram (ACG) of each cell (A1, B1, E1, F1), the corresponding FFT (A2, B2, E2, F2), the cross-correlogram (CCG) of each cell pair (C1, G1) and its FFT (C2, G2). The FFT of the CCG was then divided by the square root of the product of the ACG’s FFT (D2, H2). The corrected CCG was obtained by running an inverse FFT (D1, H1). For details, see Methods. Related to Figure 8.

6

Supplementary figure 7. Frequency dependence of unit-unit coherence. Pairs of interneurons (A) or couples of principal cells and interneurons (B) separated by 200 µm. For both kinds of couples, the graphs show average unit-unit coherence stratified by high-gamma power (red lines) or for all the data (black). Coherence values for pairs of interneurons appear higher than for pairs including a principal cell and an interneuron. However, after controlling for the number of spikes, this difference vanished (not shown). Related to Figure 8.

7

Supplementary figure 8. Closely spaced principal cells have more similar levels of entrainment than distant cells. (A) Correlation between R vector in pairs of principal cells recorded from the same shank (A1, 1615 pairs) or different shanks (A2, 6247 pairs). (B) Normalized frequency distributions of the absolute difference between the R vectors in pairs of principal cells recorded from the same (blue) or different (red) shanks. In B2, the red group is divided further as a function of the number of shanks between the cell pairs (dashed and solid red lines of different thicknesses). Related to Figure 8.

8

Supplementary Table 1. Daily distribution of trial types in the foraging task. Foraging trials with predator

Foraging trials without predator

All aborted trials

Aborted trials with predator

Aborted trials without predator

Foraging trial with prior trial success

Foraging trial with prior trial failed

Excluded trials

Rat

Day

All trials

Foraging trials

1

3 4 5

58 77 149

58 77 120

0 46 92

58 31 28

0 0 29

0 0 26

0 0 3

53 54 87

4 22 32

26 22 14

2

3 4 5

81 70 153

78 65 145

0 41 90

78 24 55

3 5 8

0 5 8

3 0 0

71 52 127

6 12 17

18 26 26

3

3 4 5

72 93 91

69 72 75

0 43 43

69 29 32

3 21 16

0 21 16

3 0 0

67 46 49

1 25 25

12 23 34

4

3 4 5

53 100 124

53 98 111

0 75 90

53 23 21

0 2 13

0 2 13

0 0 0

52 82 75

0 15 35

8 27 25

5

3 4 5

66 73 112

66 65 86

0 37 57

66 28 29

0 8 26

0 8 18

0 0 8

59 45 73

6 19 12

17 36 27

Note that due to inconsistency in the rats’ behavior, we only considered data obtained on days 3-5. In all rats, day 3 only included no-predator trials. Foraging trials listed only include those trials where all four phases of the task (baseline, waiting, foraging, escape) lasted at least 0.8 sec. The number of trials that did not meet this criterion is listed in the right-most column. Related to Figure 1.

Supplementary Table 2. Firing rates of principal cells and interneurons compared to high-gamma burst rates. Principal cells firing rate (Hz) Average ± SEM Median Interneurons firing rate (Hz) Average ± SEM Median High-gamma bursts rate (Hz) Average ± SEM Median

Quiet waking

Baseline

Waiting

Foraging

0.47 ± 0.04 0.193

0.39 ± 0.05 0.08

0.44 ± 0.06 0.06

0.41 ± 0.06 0.04

24.1 ± 1.7 21.41

24.0 ± 1.71 21.66

22.95 ± 1.71 20.05

26.67 ± 1.75 26.84

1.22 ± 0.07 1.38

1.22 ± 0.04 1.28

1.61 ± 0.09 1.68

2.0 ± 0.11 2.14

n = 381 n= 50

Friedman repeated measures ANOVAs on the firing rates of principal cells and interneurons revealed a significant effect of states for principal cells (X2(3,1523)=399, p