All experiments were performed in accordance with EU guidelines (directive 86/609/EEC). Subjects were four transgenic adult (2–9 months) male mice expressing a channelrhodopsin2- yellow fluorescent protein fusion protein (ChR2 (H134R)-eYFP). These mice were obtained by mating mice expressing the CRE recombinase under the control of the dopamine transporter (DAT-IRES-CRE mice, Stock 006660, Jackson Laboratory, Bar Harbor, ME, USA) with mice bearing a CRE-dependent ChR2(H134R)-eYFP gene (Ai32 mice, Stock 012569, Jackson Laboratory, Bar Harbor, ME, USA). Animals were housed in temperature-controlled rooms with standard 12-h light/dark cycles and food and water were available ad libitum. Each workday, animals were handled to habituate to human contact, and weighed. During pre-training and experimental periods, food intake, including pellets provided in the experimental apparatus, was restricted to a maximum of 3 g/day. Water access remained ad lib. Supplemental food was provided if weight fell below 85% of the normal weight.

In order to compare activity along the mediolateral axis of the dopaminergic nuclei, an array of 4 recording probes was constructed attached to 2 optical stimulation probes, with each serving the medial or lateral pair of recording probes (see Supplementary Figure 1). Twisted bundles of 8 formvar coated 12 micron diameter nichrome wires (“octrodes”) were inserted in polyimide tubes. The wires were gold-plated to an impedance of about 350 kohm. The main body of the headstage is composed of a resin (MadeSolid, Oakland, CA, USA) molded in a 3D printer. The drives were adapted from the “lambda drive” of Headley et al. (2015). Two optic fibres angled rostrally by 15° in the headstage were positioned to be implanted at depths of 3.9 mm (VTA) and 3.5 mm (SNc). The medial two octrodes shared a microdrive, and were positioned for implantation at a depth of 3.9 mm, while the lateral two had independent drives, and their implanted depths were 3.7 and 3.5 mm. The entire assembly is shielded from electrical noise with a fine copper mesh spot-glued with dental cement. This was connected to ground, serving as a miniature Faraday cage. Recordings were made with an Ampliplex system (sampling rate: 20 kHz; cutoff frequencies of the analog low-pass and high-pass filters were 0.3 and 10 kHz, respectively). After surgical implantation, electrodes were advanced as often as twice daily until neurons appeared with responses to optical stimulation. Electrodes were further advanced when discriminable units were no longer present.

Anesthesia was induced with a mixture of ketamine (66 mg/kg) and xylazine (13 mg/kg) and sustained with isoflurane (0.5–1.0%). Mice were placed in a stereotaxic device (David Kopf Instruments) and maintained at 37°C. The scalp was exposed and cleaned with hydrogen peroxide solution. Miniature jeweler’s screws were screwed into trephines, and attached with dental cement. The electrode assembly described above was implanted into the left VTA and SNc. The stainless-steel screws implanted on the left and right cerebellum as ground and reference electrodes, respectively.

In recordings from microelectrodes confirmed optogenetically to be in the midbrain dopaminergic nuclei (Supplementary Figure 3), several rhythms were readily apparent even in raw unfiltered LFP traces (Figure 1). To verify this, spectra were analyzed to determine the principal frequency bands most prominent in these nuclei, without making any a priori assumptions of their ranges (Figure 2). Peak frequencies were identified as inflection points in this spectrum, visualized as local minima in the second derivatives of these traces. Their frequencies were 0.7, 4, 10, 22, and 40 Hz (Figure 2). To verify this, a FOOOF analysis (Donoghue et al., 2020) was performed, and curve-fitting yielded bands centered at 4, 8, 41, and 50 Hz (not shown). These results derive exclusively from LFP traces, which carry the risk of contamination with volume-conducted signals. Thus, single unit recordings were analyzed for evidence of rhythmic activity intrinsic to the dopaminergic nuclei.

We recorded in mice training and performing sensory discrimination tasks (described further below). The neurons reported here were recorded from octrodes with at least one positive dopaminergic response (SALT+), or on days between days with dopaminergic responses, and when the electrodes had not been advanced. There were 171 dopaminergic neurons (DA), 113 FSI neurons, and 375 Other neurons (a total of 659 neurons). Neurons were excluded when there was evidence that they had been repeatedly recorded on successive days (see section “2.5.1 Single unit discrimination”). After this, there remained 85 dopaminergic neurons (DA, SALT+), 41 fast spiking inter neurons (FSI), and 254 SALT- non-FSI (Other). Note that the latter group could include dopaminergic neurons whose optogenetic responses were not significant (because of possible technical issues such as insufficient illumination or expression of opsins).

Rayleigh and pairwise phase consistency (PPC; Vinck et al., 2012) analyses assessed neuronal activity phase-locked to LFP oscillations. Frequency bands with narrow windows were examined, and the sliding windows passed successively through the range from 0.1 to 100 Hz. The incidences of significant results were tallied, and several frequencies appeared repeatedly in both tests (Figure 3). These were primarily centered around bands similar to those previously reported elsewhere in the brain: 1–2.5 (slow), 2.5–6.5 (4 Hz), 6.5–13 (theta), 13–30 (alpha), 25–55 (low gamma), and 40–75 Hz (high gamma). These bands are also largely consistent with the LFP spectrum analyses, and thus are selected for further analyses below.

Figure 4 shows examples of neurons phase-locked to the respective frequency bands. For the example in the slow band, the spike-triggered LFP averages (STA) completed a full cycle 0.6–0.8 s after the spike time, corresponding to frequencies of 1.2–1.7 Hz. STA cycle lengths for the other frequency bands are highlighted by arrows above the plots, and these correspond to the respective frequency bands. For 4 Hz, the auto-correlogram of the example on the left in Figure 4 shows a prominent peak corresponding to this frequency band. For the theta band, both examples have auto-correlograms and spike-triggered average with delays on the order of 100 ms, corresponding to 10 Hz.

Pairwise Phase Consistency analyses of whole session data examined the incidence of phase-locking of DA and FSI to the respective bands. In FSIs, the incidences of phase-locked neurons exceeded 25% in the slow and 4 Hz bands, but these were not significantly greater than the other bands for FSI’s (Figure 5A). However, in DA neurons, there was a greater incidence of phase-locking in the theta band than slow or beta bands (binomial test, p < 0.05 with Bonferroni correction).

We next tested whether neurons were phase-locked to multiple bands in PPC analyses (e.g., two adjacent frequency intervals). Supplementary Figure 4 shows that low proportions of neurons were phase-locked to more than one band, indicating that the selected frequency bands are discrete and independent.

Pairwise Phase Consistency analyses of whole session data examined the incidence of phase-locking of DA and FSI to the respective bands. This was significantly greater in FSI than DA neurons in the slow and 4 Hz bands (Figure 5A; binomial test p < 0.05 with Bonferroni correction).

We then examined the preferred phases of FSI and DA for the respective bands for all neurons (excluding repeat recordings; Figure 5B). Rayleigh tests revealed significant (p < 0.05 with Bonferroni correction) preferred phases for both neuron types for all bands except for the beta band, and DA in the slow band. The preferred phases of FSI and DA were significantly different at 4 Hz, theta and high gamma (Kuiper test, p < 0.05) but not low gamma (which had a relatively low sample).

Next, we examined the evolution LFP amplitude and incidence of phase-locking during acquisition of visual and odor discrimination tasks (described in section “2.4. Behavioral task”; see Figure 6A). Since the respective phases of task execution would call upon different cognitive processes, requiring synchronization of various circuits, analyses compare event-related oscillations and phase-locking of neurons at different task phases. As the mice performed a trial, they first entered the opposite compartment, crossing a photo-detector, which triggered onset of visual or olfactory cues, or both. This is the “cue” event, and the 500 ms periods prior and after this involved approach behavior (“pre-cue”) and cue detection and analysis (“post-cue”). The next event was the nose-poke response to the left or right odor ports, again crossing a photo-detector. The 500 ms events prior and after corresponded to decision (“pre-choice”) and anticipation of reward or punishment (“post-choice”). There was a variable delay between the nose-poke and either the click of the food dispenser for rewarded trials, or the onset of the punishment signals. These delimited the 500 ms “pre-outcome” and “post-outcome” periods.

Since the task involved acquisition of new cue-reward associations, for which the dopaminergic nuclei are strongly implicated, we assessed if phase-locking changed as a function of performance level of the animals. We computed the proportion of phase-locked neurons in sessions where the animal reached criterion performance in at least one of the tasks, triggering the rule to be changed (n = 96). These were compared to other sessions where mice did not achieve criterion performance (n = 33). Significant differences were rare, and were not concentrated on any particular frequency band or trial event.

Other tests showed no differences in incidences of phase-locking of cells or of PPC values in comparisons of rewarded vs. punished trials or between cues presented (visual vs. odor vs. both; p < 0.05 with Bonferroni correction; data not shown).

The frequency bands here were determined with several approaches based upon the actual data. This led to distinguishing separate bands for slow and 4 Hz. In some work, the delta band is considered to range from 1 to 5 Hz, and this could lead to confounds. For this reason, we did not use the term “delta” to describe the 1 to 2.5 Hz band here.

Overall, reports in the literature of slow oscillations during waking behavior are relatively sparse. However, Orzeł-Gryglewska et al. (2014) observed prominent 1.5 Hz oscillations in VTA and hippocampal LFP in awake rats, and their power was cross-correlated. Here, the incidence of phase-locking of neurons to the slow rhythm here tended to be greater than the other bands, although this difference was not significant.

There are reports of cortical delta rhythms in awake rodents during stationary immobility in the absence of experimental sensory stimulation (Crochet and Petersen, 2006; Schultheiss et al., 2020). However, these results are not directly comparable here since the range of their delta frequency band included 4 Hz. With intracranial recordings, Halgren et al. (2018) found 2 Hz oscillations that were generated in the superficial layers of the cerebral cortex of awake epilepsy patients. These were coupled to higher frequency oscillations in other layers. We found slow rhythms in hippocampus and prefrontal cortex of rats performing in a T-maze, and noradrenergic locus coeruleus neurons were phase-locked to these slow oscillations (Xiang et al., 2023).

In chloral hydrate anesthetized rats, Peters et al. (2004) and Gao et al. (2007) observed infraslow (0.65 and 0.86 Hz, respectively) oscillations in DAergic neurons. They considered this as an experimental model for delta rhythms during sleep. Intracellular PFC UP states (Peters et al., 2004) and PFC LFP negative deflections (Gao et al., 2007) were synchronous with negative deflections of the VTA LFP at this frequency range. The VTA neurons fired in phase with high, but not low amplitude prefrontal cortical oscillations at this frequency. Non-DA neurons were also modulated, but were almost pi radians (180°) out of phase relative to the DA neurons (Gao et al., 2007). In contrast, in our awake behaving mice, the FSI and SALT- Other neurons had a phase difference of only 0.8 and 1.2 rad, respectively from DAergic neurons. TTX infusion in PFC suppressed the slow oscillations in VTA neurons (Gao et al., 2007). Interestingly, lidocaine infusion in VTA (but not substantia nigra) suppressed the UP-DOWN state oscillations in PFC.

Here, more FSIs than DA were phase-locked to the slow and 4 Hz rhythms. Interneurons are known to be instrumental in the generation of theta and gamma rhythms. Our experimental design was not appropriate for testing whether the rhythms are generated locally. While the higher firing rates of interneurons relative to principal neurons could render the PPC and Rayleigh tests more sensitive, this result remained robust since we down-sampled all spikes for PPC.

More neurons were phase-locked to the slow rhythm during the pre-outcome period than the pre-cue period. This could be related to synchronizing dopaminergic structures with associated structure during the period when reward is being anticipated.

van der Velden et al. (2020) optogenetically stimulated lateral VTA in mouse brain slices. Varying the frequency of stimulation, they found that individual neurons had optimal responses (“resonance”) at several values between 0.5 and 5 Hz. This corresponds to the slow band observed here, extending to the 4 Hz band as well. The authors suggest these variations in preferred infraslow and slow frequencies among individual neurons could support temporal processing of inputs and coding for outputs.

Fujisawa and Buzsáki (2011) showed that the power of 4 Hz oscillation (filtered from 2 to 5 Hz) in rat PFC and VTA, and its coherence between these two structures, are stronger during the choice phase of a working memory task. They found that 46% of putative dopaminergic and 38% of putative GABAergic VTA neurons were significantly phase-locked to the 4 Hz oscillation. Similarly, 4 Hz rhythmicity (with 3–6 Hz filters) occurs in mice during choice periods in another working memory task (Duvarci et al., 2018). 75% of DAergic, and 78% of GABAergic VTA cells were modulated by VTA 4 Hz. (In these papers, neuron types were identified according to responses to systemic apomorphine administration, baseline firing rates and spike width). In contrast, here, the incidence of neurons with significant 4 Hz PPC was significantly elevated during the post-choice period (during the delay prior to reward or punishment). Discrepancies with the cited results could be related to our task lacking an explicit trial-by-trial working memory component. Another difference is that here, twice as many FSI’s than DAergic neurons were modulated by 4 Hz oscillations. This difference could be related to different methods used to identify the two types of neurons, the different filter range used, or sampling properties of the different types of electrodes used. Also, these other papers used the Rayleigh test to detect phase-locking, while we reported PPC results, which tended to have fewer significant FSI’s in this band (Supplementary Figure 5).

In Duvarci et al. (2018), the median preferred phase of DAergic neurons was about −2 rad, while GABAergic neurons were about −0.75 rad. In Fujisawa and Buzsáki (2011), relative to PFC LFP, DAergic neurons had a median phase of about −0.5 rad and GABAergic neurons had a median phase of about −2.6 rad. Here the mean preferred phase of DA neurons in the corresponding frequency band was 0.8 rad, but for FSI’s the value was 2.2 rad.

Lowes et al. (2021) found increased amplitude of 4 Hz amplitude in VTA and nucleus accumbens (NAc) LFPs during stressful restraint in mice (which also reduces reward seeking). Furthermore, multi-unit activity in VTA was phase-locked to NAc 4 Hz. Interestingly, silencing VTA GABAergic neurons with muscimol suppressed NAc LFP 4 Hz in stressed animals.

One potential confound for 4 Hz LFP in mice is the correlation with breathing signals sent from olfactory bulb through piriform cortex and PFC (e.g., Bagur et al., 2021). Rhythmic firing in the DAergic neurons here provides evidence that this is not simply a volume-conducted signal. Thus, the DA nuclei are part of the widespread network of brain structures coordinated by the 4 Hz rhythm (Tort et al., 2018).

In behaving rodents, hippocampal theta is typically reported at 7–8 Hz. Here, the spectral data revealed an inflection point at 10 Hz, and thus the range studied was 8 to 13 Hz, and this helped to avoid overlap with the band centered on 4 Hz. Fujisawa and Buzsáki (2011) filtered for theta in their hippocampal LFP recordings at 7–11 Hz, but did not report VTA LFPs in this frequency range. However, they did observe VTA neurons modulated both by VTA 4 Hz (as replicated here) and by hippocampal theta. Future work should test for potential relations between hippocampal theta and VTA 10 Hz oscillations. Orzeł-Gryglewska et al. (2014) observed VTA LFP at 6–9 Hz, and its power was cross-correlated with that of hippocampal theta. Note that Park and Moghaddam (2017) filtered rat VTA LFPs for theta between 5 and 15 Hz, and found peaks at 10–12 Hz, consistent with the present work in mice. Fujisawa and Buzsáki (2011) found significant phase locking to the hippocampal theta rhythm in 44% of putative dopaminergic neurons and 39% of putative GABAergic cells. In the Park and Moghaddam (2017) study, 45% of DAergic neurons were modulated by VTA 5–15 Hz, while only 23% of non-DAergic neurons were. Here 26% of DA neurons, and 24% of FSI’s were phase-locked to 8–13 Hz, comparable to the latter work. In conditioned rats, Kim et al. (2012) found increased theta (4–8 Hz) power in VTA LFP for appetitive conditioned stimuli (CS) but not aversive CS. Following learning, more putative GABA neurons were phase-locked to the theta rhythm than putative dopamine neurons.

Here, there were no significant differences between incidence of theta phase-locked neurons during the respective trial periods. However, the failure of results to be significant cannot prove the absence of an effect, in particular, in light of our conservative selection criteria. Park and Moghaddam (2017) found coherence between LFPs in VTA and PFC at 5–15 Hz, and this decreased with greater risk of punishment. Thus, both the latter study and Lowes et al. (2021) support a role of these rhythmic oscillations in gating of communication between dopaminergic nuclei and striatum (4 Hz) and with neocortex (10 Hz). However, both Fujisawa and Buzsáki (2011) and Duvarci et al. (2018) found that VTA activity at hippocampal theta frequency was not associated with working memory performance, in contrast with the 4 Hz band. This is consistent with the proposal that the different frequency bands could coordinate communication of complementary types of information (Akam and Kullmann, 2014).

Relatively few neurons were phase-locked to these bands of DA nucleus LFP’s, and this likely contributed to the absence of significant differences between incidences between cell types and task events. In calcium imaging of awake mice, Eban-Rothschild et al. (2020) found population activity of VTA-GABAergic neurons, but not VTA-dopaminergic neurons, was positively correlated with EEG gamma power, and negatively correlated with theta power. Fujisawa and Buzsáki (2011) found high gamma (30–80 Hz) coherence between PFC and the VTA. The narrow 50 Hz band in the choice arm persisted in both their working memory and control tasks.