Nine male Long-Evans rats (Janvier Labs, Le Genest-Saint Isle France; weight, 280–400 g) were maintained on a 12 h:12 h light-dark cycle (lights on at 7 a.m.). The rats were handled on each workday. To motivate animals for behavioral training on the T-maze, water was restricted except for a 10–30 min period daily to maintain body weight at 85% of normal values according to age. Rats were rehydrated during weekends. Food was restricted to 14 g of rat chow daily (the normal daily requirement) to prevent the animals from becoming overweight. Recording properties and behavioral correlates of the LC neurons of four of the rats were reported in Xiang et al. (2019), but no LFP data are presented there. Recordings of LC neurons during task performance was not possible in the remaining five rats included here.

The behavioral task took place in an elevated automated T-maze (see Figure 3A) consisting of a start area, a central arm, two reward arms and two return arms, which connected the reward arms to the start area. Small wells at the end of each reward arm delivered liquid reward (30 μl of 0.25% saccharin solution in water) via solenoid valves controlled by a CED Power1401 system (Cambridge Electronic Design, Cambridge, UK) with a custom-written script. As the rats crossed a central photo-detector, visual cues (VCs) were displayed in pseudo-random sequence on video monitors positioned behind, and parallel to the two reward arms. This is the “VC,” or “central arm PD” event. The VCs were either lit or dim uniform fields. The rat then selected the left or right arm and crossed the Reward arm (Rew) photodetector (PD), triggering reward release from an audible solenoid valve. Crossing the photodetector in the middle of the return arm of the T-maze (Return arm PD; Ret) triggered the visual cue to be turned off. Photodetectors detected task events and triggered cues and rewards via the CED Spike2 script.

The Canine Adenoviral vector (CAV2-PRS-ChR2-mCherry) was produced at the University of Bristol. Details about it and injections procedures appear in Xiang et al. (2019). Also see Xiang et al. (2019) for histological and immunohistochemical procedures and neuron characterization.

Following VC task pre-training, at least 1 day before surgery, rats were returned to ad libitum water and food. Surgical procedures and electrode construction are the same as in Xiang et al. (2019). Moveable tungsten microelectrodes (insulated with epoxylite^®, impedance = 2–4 MΩ, FHC Inc., Bowdoin, ME, USA) were used for LC recordings. A single microelectrode, or two or three such electrodes glued together, was implanted at AP −3.8 to –4 mm relative to lambda, and ML 1.1–1.2 mm, with a 15° rostral tilt. A stainless steel wire (Teflon coated, diameter = 178 μm, A-M Systems Inc., Sequim, WA, USA) implanted in the midbrain area about 1–2 mm anterior to the LC electrode tip served as a fixed LC reference electrode, permitting differential recording. The rat with the virus injection (R328) was implanted with an optrode composed of a tungsten microelectrode (insulated with epoxylite, impedance = 2–4 MΩ, FHC Inc., Bowdoin, ME, USA) glued to a 200 μm optic fiber implant with a ferrule (0.37 numerical aperture, hard polymer clad, silica core, multimode, Thorlabs, Bergkirchen, Germany), with tip distances 1 mm apart (the electrode was deeper). The optic fiber implant and optic fiber cables were constructed at the NeuroFabLab (CPN, Ste. Anne Hospital, Paris). Two screws (diameter = 1 mm, Phymep, Paris) with wire leads were placed in the skull above the cerebellum to serve as ground. LC electrodes were progressively lowered under electrophysiological control until characteristic LC spikes were identified (located ∼ 5–6 mm below the cerebellar surface, see Xiang et al., 2019 for details). For the virus-injected rat, LC spikes could also be identified by responses to laser stimulations (described below). Following implantation, the microelectrode was fixed to a micro-drive allowing for adjustments along the dorsal-ventral axis. The headstage was fixed to the skull with dental cement, and surrounded by wire mesh stabilized with dental cement for protection and shielding. After the surgery, animals were returned to their home cages for at least 1week recovery with ad libitum water and food, and regular observation.

Rats were then returned to dietary restriction. The movable electrodes were gradually advanced until a well-discriminated LC unit was encountered, and then all channels were recorded simultaneously while the rat performed in the T-maze. If no cells could be discriminated, the electrodes were advanced and there was at least a 2 h delay before the next recording session.

For daily online monitoring of LC spikes, pre-amplified signals were filtered between 300–3,000 Hz for verification on the computer screen (Lynx-8, Neuralynx, Bozeman, MT, USA) and also transmitted to an audio monitor (audio analyzer, FHC, Bowdoin, ME). For recordings, brain signals were pre-amplified at unity gain (Preamp 32, Noted Bt, Pecs, Hungary) and then led through a flexible cable to amplifiers (x500, Lynx-8, Neuralynx, Bozeman, MT, USA) and filters (0.1–9 kHz, Lynx-8, Neuralynx, Bozeman, MT, USA). Brain signals were digitized at ∼20 kHz using CED Power1401 converter and Spike2 data acquisition software (CED, Cambridge, UK). The LC unit activity was identified as in Xiang et al. (2019) by: (1) spike waveform durations ≥0.6 ms; (2) low average firing rate (1–2 Hz) during quiet immobility; (3) brief responses to unexpected acoustic stimuli followed by prolonged (around 1 s) inhibition; (4) for the virus-injected rat (R328), LC units were verified by responses to laser stimulation. A laser driver (Laserglow Technologies, North York, ON, Canada, wavelength 473 nm) was controlled by signals from a stimulator (Grass Technologies, West Warwick, RI, USA, Model SD9). Light intensity from the tip of optic fiber was measured by a power meter (Thorlabs, Bergkirchen, Germany, Model PM100D). If unit firing was entrained to the pulses with an increased rate (to at least twice the baseline firing rate) averaged over all the stimulations, they were considered to be noradrenergic LC units.

A light emitting diode (LED) was mounted on the cable that was plugged into the headstage. This was detected by a video camera mounted above the T-maze and transmitted to the data acquisition system at a sampling rate of ∼30 Hz for the purpose of position tracking.

For off-line spike detection of LC activity in three of the rats, the wide-band signals were converted and digitally high-pass filtered (non-linear median-based filter). Waveforms with amplitudes passing a threshold were extracted, and then subjected to principal component analysis (PCA). All of these processes were performed with NDManager (Hazan et al., 2006). Spikes were sorted with a semi-automatic cluster cutting procedure combining KlustaKwik (KD Harris) and Klusters (Hazan et al., 2006).¹ Spikes with durations less than 0.6 ms were rejected. In one rat (R311), the LC signal was filtered from 300–3,000 Hz during recording, and the spike sorting was performed with Spike2 software (which employs a waveform template-matching algorithm). Most data analyses were performed using Matlab (R2010a) with the statistical toolbox FMAToolbox (developed by M. Zugaro)² and scripts developed in the laboratory, as well as some statistical analyses performed with Microsoft© Excel©. Phase was computed with the “Phase” function of the FMAToolbox, employing the Hilbert transform. To characterize periods with infra-slow rhythms, a criterion for salient phase-locking to task events was established as when the SEM range of LFP phase was less than 0.75*π radians (cf., Figure 4, middle column). This is termed “regular phase-locking.” Sessions tallied for phase-locking of LC neurons to infra-slow rhythms were included only if they had at least 1,000 LC neuron spikes.

Infra-slow rhythms were readily apparent in visual inspections of hippocampal (Hip) and prefrontal cortical (Pfc) local field potentials (LFPs) (Figure 1A). These were rendered more salient by filtering the signal in a 0.1–1.0 Hz window (Figure 1B). We applied an amplitude threshold to examine data from those periods when the infra-slow rhythm amplitude was elevated (Figure 1C). This was intended to limit analyses to those periods when the infra-slow oscillation was sufficiently robust, thus avoiding possibly spurious computations of phase angle from low amplitude oscillations. In each of four rats, LC neurons were phase-locked to Pfc (e.g., Figures 1D, 2), as well as Hip infra-slow LFP rhythms. The incidence of phase-locking of the LC neurons in sessions with regular phase-locking of the infra-slow rhythms to task events was 20 out 23 for Hip LFP and 15/20 for Pfc LFP (Rayleigh test, p < 0.05; for histology, see Figure 2 of Xiang et al., 2019). In the animal where noradrenergic LC neurons were identified optogenetically (see Materials and methods), all were phase-locked to the infra-slow rhythms (n = 8 for both Pfc and Hip; Rayleigh test, p < 0.05).

The modal preferred infra-slow phase among these neurons was 0.35*π radians for Hip infra-slow and 0.15*π radians for Pfc infra-slow oscillations (p < 0.05, Rayleigh test; not shown).

The infra-slow rhythms were phase-locked to task-relevant positions on the maze (Figure 3B and Supplementary Figure 1B). To quantify this phase-locking, the mean (± SEM) phase of the rhythm was plotted in peri-event time color plots (see Figure 4 and Supplementary Figure 1A for examples) over all trials in 57 sessions from nine rats (including the four with LC recordings). Infra-slow rhythms were phase-locked to the reward arm photodetector crossing (Rwd) in 51 of the recording sessions for Hip, and 46 sessions for Pfc (see Table 1). The other maze events had fewer incidences of regular phase-locking (Pfc return arm photodetector crossing, or Rtn: 11; Pfc central arm visual cue onset PD, or VC: 18; Hip Rtn: 18; Hip VC: 20). The mean phases at the respective PD crossings (in those cases when SEM ≤ 0.75*π radians there) were 0.70*π and 0.24*π radians for Pfc and Hip Rtn, 0.25*π and 0.22*π radians for Pfc and Hip Rwd, and 0.19*π and 0.01*π radians for Pfc and Hip VC. The root-mean-square differences between Pfc and Hip mean phase (calculated pairwise by session) at the respective PD crossings were 0.14*π, 0.13*π, and 0.12*π rad. The regular phase-locking could last from less than one to over 2.5 successive rhythmic cycles (Supplementary Figures 1, 2 and Table 1) and could continue from one event to the next (Figure 4 and Supplementary Figure 1). For PL Rwd and Hip Rwd, 30 and 37 sessions had durations of regular phase-locking lasting one or more cycles, respectively. These permitted quantification of the temporal duration of the cycles, which ranged from 2.0 to 2.6 s, the equivalent of 0.4–0.5 Hz. In the six cases of Rwd PD phase-locking which had a second complete cycle, the mean of the first was 2.3 s, while the second was lower, 2.0 s (pairwise t-test, p = 0.0009, df = 5). Thus, these are not regular periodic oscillations, but, rather, this is consistent with phase-locking to task events. Pfc and Hip infra-slow rhythms sometimes resembled one another (e.g., Figure 2). To compare them, sessions were classified as having Pfc and Hip regular phase-locking in several ranges of cycles (see Table 1). In 17 of the 57 sessions, these numbers of cycles were different between Pfc and Hip for VC, Rwd and/or Rtn (e.g., Supplementary Figure 2). This indicates that it is unlikely that Pfc and Hip infra-slow rhythms are related by volume conduction, and suggests that they could be independently generated.

The infra-slow rhythms were regularly phase-locked to two (in 24 sessions), or even all three (in 8 sessions) different task events. Thus, they were not linked to any specific task-related behavior. To test whether infra-slow rhythms were triggered by rapid head movements, regression analysis compared the onset of regular phase-locking and times of peak acceleration, or deceleration around the Rwd PD crossings, and were not significant (r² = 0.034, p = 0.49 and r² = 0.0056, p = 0.80, respectively; df = 15; e.g., see Supplementary Figure 3). In Xiang et al. (2019), we showed that LC neurons fire more during accelerations. Indeed, the periods with the greatest increase in LC activity were not those most frequent for the start of regular phase-locking (i.e., reset) of the infra-slow rhythm; rather phase-locking occurred most frequently to Rwd PD crossing (see above), where no consistent accelerations occurred (see Supplementary Figures 3, 4). These results indicate it is unlikely that Pfc and Hip infra-slow rhythms are due to a biomechanical artifact, for example, from locomotion or head rocking.

In the four sessions where Hip and Pfc neurons could be discriminated from the LFP electrodes, most were also modulated by infra-slow rhythms (Pfc LFP modulated 6/12 Pfc units and 8/11 Hip units; Hip LFP modulated 8/12 Pfc units and 8/11 Hip units; Rayleigh test p < 0.05). The LC neurons could have relatively consistent phases with respect to the two infra-slow rhythms (not shown). LC neurons could be phase-locked to oscillations in the delta frequency range (1–4 Hz) in Pfc (n = 15/37) and Hip (11/37) as well as theta (5–10 Hz; 7/37 and 5/37, respectively) for Pfc and Hip (see Supplementary Figure 5). While phase-locking of LC neurons to gamma (40–80 Hz) was rare (n = 2 for both structures’ LFPs), the infra-slow rhythm did modulate the amplitude of gamma oscillations at 35–45 Hz in hippocampus and prefrontal cortex (Figure 5).