Abstract
Hippocampal theta oscillations (HTOs) during rapid eye movement (REM) sleep play an important role in mnemonic processes by coordinating hippocampal and cortical activities. However, it is not fully understood how HTOs are modulated by subcortical regions, including the median raphe nucleus (MnR). The MnR is thought to suppress HTO through its serotonergic outputs. Here, our study on male mice revealed a more complex framework indicating roles of nonserotonergic MnR outputs in regulating HTO. We found that nonselective optogenetic activation of MnR neurons at theta frequency increased HTO amplitude. Granger causality analysis indicated that MnR theta oscillations during REM sleep influence HTO. By using three transgenic mouse lines, we found that MnR serotonergic neurons exhibited little or no theta-correlated activity during HTO. Instead, most MnR GABAergic neurons and Vglut3 neurons respectively increased and decreased activities during HTO and exhibited hippocampal theta phase-locked activities. Although MnR GABAergic neurons do not directly project to the hippocampus, they could modulate HTO through local Vglut3 and serotonergic neurons as we found that MnR GABAergic neurons monosynaptically targeted Vglut3 and serotonergic neurons. Additionally, pontine wave recorded from the MnR during REM sleep accompanied nonserotonergic activity increase and HTO acceleration. These results suggest that MnR nonserotonergic neurons modulate hippocampal theta activity during REM sleep, which regulates memory processes.
SIGNIFICANCE STATEMENT The MnR is the major source of serotonergic inputs to multiple brain regions including the hippocampus and medial septal area. It has long been thought that those serotonergic outputs suppress HTOs. However, our results revealed that MnR serotoninergic neurons displayed little firing changes during HTO. Instead, MnR Vglut3 neurons were largely silent during HTO associated with REM sleep. Additionally, many MnR GABAergic neurons fired rhythmically phase-locked to HTO. These results indicate an important role of MnR nonserotonergic neurons in modulating HTO.
Introduction
Hippocampal theta oscillations (HTOs), characterized by a 6–10 Hz rhythm in rodents, have been shown to play an important role in a variety of functions, such as spatial navigation, learning, memory consolidation, memory retrieval, and action selection (Buzsáki, 2002; Hasselmo, 2005; Vertes, 2005; Wang et al., 2020). In particular, HTOs during rapid eye movement (REM) sleep coordinate hippocampal neural firing to replay the waking experience and contribute to the consolidation of relevant memories (Louie and Wilson, 2001; Boyce et al., 2016).
It is generally believed that HTOs are generated by the medial septum/diagonal band of Broca (MS/DBB), as GABAergic MS/DBB neurons fire rhythmically at the theta frequency, and lesion or inactivation of the MS/DBB abolishes HTO (Winson, 1978; Brazhnik and Fox, 1997; King et al., 1998; Buzsáki, 2002; Crooks et al., 2012). However, the complete neural circuitry that mediates HTOs is more complex and not fully understood because HTOs are also modulated by other subcortical regions, notably the median raphe nucleus (MnR) located along the midline of the ventral mesopontine area (Azmitia and Segal, 1978; Acsady et al., 1996; Vertes et al., 1999; Varga et al., 2009).
In contrast to the MS/DBB, the MnR is thought to suppress HTO because inhibition or lesion of the MnR increases HTO (Maru et al., 1979; Crooks et al., 2012). In particular, previous studies have shown that inactivation of putative serotonergic (5-HT) MnR neurons selectively increases HTO (Vertes et al., 1994; Nitz and McNaughton, 1999; Varga et al., 2002; Crooks et al., 2012), and this effect can be reversed by restoring serotonin (Yamamoto et al., 1979). These findings indicate a suppressive effect of MnR 5-HT neurons on HTO (Vertes et al., 1999; Crooks et al., 2012); however, the precise role of MnR 5-HT neurons remains unclear because most are active rather than silent during HTO (Viana Di Prisco et al., 2002; Kocsis et al., 2006; but see Jacobs and Azmitia, 1992).
The MnR contains at least three types of neurons, namely 5-HT, Vglut3, and GABAergic neurons (Domonkos et al., 2016; Szőnyi et al., 2016; Sos et al., 2017). It is important to examine whether these three types of MnR neurons play differential roles in affecting HTO. Indeed, a subset of non-5-HT MnR neurons fire rhythmically in a phase-locked manner with HTO (Kocsis and Vertes, 1996; Viana Di Prisco et al., 2002), indicating a non-5-HT mechanism in regulating HTO. In the current study, we sought to investigate how each of the three MnR neuronal types (i.e., 5-HT, Vglut3, GABAergic) was related to HTO during REM sleep in freely behaving mice. We also investigated a possible interaction among the three MnR neuronal types.
Materials and Methods
Mice
Male C57BL/6 (stock #000664), ePet-Cre mice (stock #012712), and Vglut3-Cre mice (Slc17a8-IRES2-Cre-D, stock #028534), all purchased from The Jackson Laboratory, were used at Drexel University. In addition, male Vglut3-Cre (Cheng et al., 2017), ePet-Cre (Scott et al., 2005), and Vgat-Cre (Vong et al., 2011) crossed with C57BL/6 mice, bred at the National Institute on Drug Abuse–Intramural Research Program (NIDA–IRP) animal facility, were used at NIDA–IRP. There was no noticeable difference between the two Vglut3 transgenic lines, although they were independently generated (Hongkui Zeng vs Bradford Lowell Lab). Mice were 2.5–5 months old at the time of surgery; after surgery, they were singly housed in cages (30 × 20 × 20, 40 × 20 × 20, or 36 × 24 × 20 cm) containing woodchips and cotton material and kept on a 12 h light/dark cycle with ad libitum access to food and water. Experimental procedures were approved by the Institutional Animal Care and Use Committees of NIDA–IRP and Drexel University and were in accordance with the National Research Council Guide for the Care and Use of Laboratory Animals.
Viral vectors
Adeno-associated viruses (AAVs), including AAV1-hSyn-hChR2 (H134R)-EYFP (catalog #26973, Addgene) and AAV1-EF1α-DIO-hChR2 (H134R)-EYFP-WPRE-hGHpA (catalog #20298, Addgene) were used at Drexel University. Other AAVs used at NIDA–IRP, including AAV1-Ef1a-DIO-hChR2 (H134R)-EYFP-WPRE-hGH and AAV1-EF1α-DIO-EYFP-WPRE-hGH, were the same as reported previously (Wang et al., 2015). The final viral concentration for microinjection was ∼5–10 × 1012 genome copies/ml.
Stereotaxic surgery
Surgery procedures were similar to those reported previously (Wang et al., 2015; Opalka et al., 2020). In brief, mice were anesthetized with ketamine/xylazine mixture (∼100/10 mg/kg, i.p.) and kept on a heating pad at ∼37°C. All C57BL/6 mice received implantation of one bundle of 4–8 tetrodes in the CA1; meanwhile, the same mice received either another bundle of 4–8 tetrodes or an AAV viral injection followed by an implantation of an optical fiber (diameter 200 μm) in the MnR. All transgenic mice received an AAV viral injection, followed by an implantation of either an optical fiber (diameter 200 μm) or an optrode in the MnR; meanwhile, most of these mice received another implantation of 4–8 tetrodes in the CA1. The MnR coordinates were anteroposterior (AP) –4.5 mm, ML 0.0 mm, DV 3.9–4.0 mm; the CA1 coordinates were AP –2.3 mm, ML 1.8 mm, and DV 1.1 mm. AAV (∼0.2 μl; in a few transgenic mice, 0.5 μl) was microinjected into the MnR with a syringe pump over 4–10 min (0.05 μl/min; World Precision Instruments), with an additional 5–10 min before removal of the injection needle (34 gauge, beveled). Optical fibers and electrode bundles were slowly lowered toward the MnR or CA1 and secured on the skull with stainless screws and dental cement. We started the experiments 2–3 weeks after surgery to allow gene expression.
Tetrode and optrode
Each tetrode consisted of four wires (90% platinum/10% iridium; 18 μm diameter with an impedance of ∼1–2 MΩ for each wire; California Fine Wire). Each optrode consisted of an optical fiber (diameter 105 μm) attached to 4–6 tetrodes; the distance between the tips of the optical fiber and tetrodes was 0.3–0.6 mm. A microdrive was used to couple with the electrode bundle or optrode, similar to that described previously (Wang et al., 2015; Opalka et al., 2020). Optostimulations (0.2–2 mW) were delivered with a variable interval schedule of 8–15 s.
In vivo electrophysiology
Neural signal was preamplified, digitized, and recorded using a Neuralynx Digital Lynx acquisition or Blackrock CerePlex system; the behaviors of the animals were simultaneously recorded. Neural signals for local field potential (LFP) were digitized at 2 kHz and filtered at 1–500 Hz (Neuralynx) or 500 Hz low cut (Blackrock), using ground as the reference. Spikes were digitized at 30 kHz and filtered at 600–6000 Hz, using one recording electrode that lacked obvious spike signals as the reference. The CA1 electrode bundle was lowered slowly until we recorded clear sharp-wave ripple events (i.e., CA1 oriens/pyramidal layers). The MnR electrode bundle was lowered by 50–100 μm after each recording session; each mouse received 3–6 recording sessions (up to 6 h per session).
Electrophysiological data
We obtained 111 MnR neurons and 83 CA1 neurons from C57BL/6 mice (n = 6), 154 MnR neurons from Vglut3-Cre mice (n = 10), 90 MnR neurons from ePet-Cre mice (n = 6), and 149 MnR neurons from Vgat-Cre mice (n = 3). We sorted neurons based on spike waveforms using multiple spike-sorting parameters (e.g., principal component analysis, energy analysis) in Offline Sorter (Plexon); sorted neurons were processed and analyzed in NeuroExplorer 5 (Nex Technologies) and MATLAB (MathWorks). Neurons with firing rates of 0.2 Hz or lower were excluded from further analysis.
Slow-Wave Sleep and REM sleep
Slow-wave sleep (SWS) and REM sleep stages were determined by the theta (6–10 Hz)/delta (1–4 Hz) ratio extracted from the power spectrograms when mice stayed immobile in their home cage cardboard box filled with cotton material (i.e., bed), where they exclusively slept (Wang et al., 2015). A ratio of 2 or greater was identified as REM sleep, whereas a ratio of 1 or lower was identified as SWS (Wang et al., 2015). Only REM sleep that lasted for 30 s or longer were used for further analysis (note that each REM episode immediately followed an SWS episode, and that at the end of each REM episode, mice typically moved a little or woke up). All C57BL/6, Vglut3-Cre, and ePet-Cre mice showed clear SWS-REM sleep cycles during the recording sessions. In contrast, the Vgat-Cre mice often stayed awake/hyperactive for many hours, thus limiting our acquiring sleep data.
MnR 5-HT neurons
We defined putative 5-HT neurons based on their firing rate and interspike interval (ISI), as previously described (Wang et al., 2015). Briefly, an MnR neuron with a spontaneous firing rate <5 Hz and an ISI longer than 200 ms was defined as a putative 5-HT neuron. We tested 12 of the classified putative 5-HT with intraperitoneal injection of the serotonin 1A receptor agonist 8-OH-DPAT (0.2 mg/kg), and 11 of them showed a clear decrease of firing as reported previously (Wang et al., 2015). In ePet-Cre mice, we confirmed the identity of putative 5-HT neurons if they responded to the optostimulation with a short latency between 2 and 10 ms (see Fig. 4). Any response with a latency shorter than 2 ms was confirmed an artifact (based on waveform); and any response with a latency longer than 10 ms was considered a network effect (Cohen et al., 2012; Wang et al., 2015).
MnR Vglut3 neurons
We defined putative Vglut3 neurons based on their firing pattern and relationship to hippocampal ripple events, as previously described (Wang et al., 2015). Briefly, an MnR neuron with decreased firing rate by 20% or greater from the baseline, lasting 1 s or longer before hippocampal ripple events during SWS, was defined as a putative Vglut3 neuron. Notably, most of these putative Vglut3 neurons exhibited a cyclic on-and-off firing pattern during SWS (see Fig. 2). Similarly, we confirmed the identity of putative Vglut3 neurons if they responded to the optostimulation with a short latency between 2 and 10 ms (see Fig. 3).
MnR GABAergic neurons
If an MnR neuron was not defined as a putative 5-HT or Vglut3 neuron, it was considered a putative GABAergic or unidentified neuron (collectively referred to as GABA & others). Similarly, we confirmed the identity of putative GABAergic neurons if they responded to the optostimulation with a short latency between 2 and 10 ms (see Fig. 5).
Granger causality analysis
Granger causality analysis of LFPs was implemented as reported (Brovelli et al., 2004), although we were aware of its limitations in data interpretation (Sheremet et al., 2016; Stokes and Purdon, 2017). Such analysis calculated the extent of prediction for one time series by the information of another time series compared with its own past information. In our case, LFPs were first normalized by z score and downsampled (2000 Hz/8 = 250 Hz); the MATLAB code was provided by Lu Zhang (Zhang et al., 2012). This method was based on the multivariate autoregressive model, and all autoregressive (AR) parameters were estimated via the Levinson Wiggins-Robinson method by Morf's modification. We applied 25 as the order of the AR models. In addition, Granger causality spectrum was computed only for stationary LFPs from both MnR and CA1.
Phase–amplitude coupling and power–power correlation
We used a modulation index (MI) based on tools developed at the Kopell lab (Tort et al., 2010). MI was defined as the normalized Kullback–Leibler distance of the amplitude distribution of a given high-frequency oscillation across all phases of a given low-frequency oscillation from a uniform distribution. In our case, low-frequency oscillations were theta band range, high-frequency oscillations were gamma for both CA1 and MnR. Comodulogram maps were made by calculating the MI between the phase of frequencies ranging from 2 to 20 Hz (1 Hz stepwise) and amplitudes ranging from 20 to 120 Hz (5 Hz stepwise). No overlap was used for phase or amplitude (see Fig. 8B). The comodulogram for power–power correlation between CA1 and MnR was modified from the tools developed at the Buzsáki lab (Buzsáki et al., 2003; Khodagholy et al., 2017; see Fig. 8C). We applied spectrogram estimate with five wavelet cycles and computed the correlation of log power for each pair of frequencies.
Spike–theta phase locking
Theta oscillations were obtained from CA1 and MnR LFPs filtered at 6–10 Hz using a zero-phase filter in MATLAB (IIR Butterworth; filter order = 3). Theta phase was estimated by Hilbert transformation of the filtered signal. Then the theta phase value for each corresponding spike of a given neuron was extracted and accumulated for distribution analysis and statistics (Berens, 2009). We used standard Rayleigh test to determine uniformity of phase distribution (Totah et al., 2013), in which the p value for circular uniformity is calculated as follows:
Histology for implantation verification
At the completion of the electrophysiology recordings, the final electrode position was marked by passing a 20 s 10-μA current using a linear constant current stimulus isolator (NeuroLog System) through two tetrodes. Mice were deeply anesthetized and intracardially perfused with ice-cold PBS or saline, followed by 4% paraformaldehyde (PFA) or 10% formalin. Brains were then removed and postfixed in the PFA or formalin for at least 24 h. Brains were rapidly frozen and sliced on a cryostat or vibratome (50 μm coronal sections). Sections from the dual-site recording of C57BL/6 mice were stained with cresyl violet for microscopic examination of electrode placements, whereas other sections were mounted with the Mowiol mounting medium mixed with DAPI (Vector Laboratories) for microscopic fluorescent examination of viral vector expression and optical fiber placements.
RNAscope in situ hybridization
To colocalize EYFP- and Vglut3-expressing neurons, RNAscope 2.0 assay (Advanced Cell Diagnostics) was performed in Vglut3-Cre mice (n = 2) that had received an injection of AAV1-DIO-EYFP into the MnR. Briefly, the fresh frozen brain was embedded in cryo-embedding medium and then coronally sectioned at 14 μm. Mounted sections were then fixed in chilled 10% neutral buffered formalin, dehydrated in graded ethanol (50%, 70%, and 100%), digested with protease, and followed by hybridizations with target-specific probes. Sections were counterstained with DAPI, and the images were examined with a confocal microscope.
Statistical analyses
Sample sizes were based on previous similar studies in our labs (Wang et al., 2015; Opalka et al., 2020). To determine firing-rate increase, the value that deviates from the baseline mean by a z score of 3.3 or more was considered to be significant. To determine firing-rate decrease, for all except 5-HT neurons, the z score of –2.0 or less was considered to be significant. For 5-HT neurons, because of sparse spikes during baseline, the value that deviates from the baseline by 40% or more was considered to be significant. All statistical tests were two-sided t tests.
Results
To determine whether the stimulation of MnR neurons modulates hippocampal theta activity, we injected AAV-Syn-ChR2 into the MnR and then implanted an optical fiber slightly above the injection site and a bundle of tetrodes into hippocampal CA1 of C57BL/6 mice (n = 5; Fig. 1A). After 2–3 weeks to allow viral expression, we recorded CA1 activity in freely behaving mice as a function of MnR optostimulation (20 s at various frequencies ranging from 5 to 11 Hz). Our results showed that stimulations at frequencies ∼8 Hz evoked robust corresponding HTO, whereas more distant frequencies slightly suppressed endogenous HTO (Fig. 1B–D). We analyzed the CA1 power spectral density between 4 and 12 Hz from the five mice and compared the peak power ratios (A ratio was defined as optostimulation-evoked peak power divided by baseline peak power.). Our results revealed that the 8 and 11 Hz stimulation significantly increased and decreased baseline HTO power, respectively (t(4) = 3.18, p < 0.05 and t(4) = −4.88, p < 0.01; paired t test), whereas other frequencies of stimulation did not evoke significant changes of HTO power (Fig. 1E). We also noticed that MnR stimulation-evoked HTOs were independent of the behavioral states of the animals across the sleep-wake cycle, though a prolonged MnR stimulation during sleep tended to wake up the mice. Together, these results provide direct evidence that activation of MnR neurons modulates HTO.
MnR optostimulation modulates HTO. A, Schematics of experimental design. B, Representative CA1 LFP traces (duration, 1 s) in response to 7, 8, and 9 Hz optostimulations (laser on indicated by blue bars; laser power, ∼0.5 mW). C, Spectrograms of the CA1 LFPs on MnR optostimulation at various frequencies (5–11 Hz; 20 s). Color bar, power spectral density (PSD). D, PSDs of CA1 LFP at various frequencies of MnR optostimulation. Solid curves with shadow indicate normalized PSD values (mean ± SE from 5 mice). E, Peak power ratios corresponding to D. The black line with shadow indicates the mean (± SE), whereas gray lines indicate individual animals (*p < 0.05, **p < 0.01, paired t test in comparison to baseline).
Next, we sought to investigate how individual MnR neurons interact with HTO, using a dual-site recording strategy in C57BL/6 mice (n = 6). We implanted two bundles of tetrodes into hippocampal CA1 and MnR, respectively (Fig. 2A, right), and subsequently recorded LFPs and neuronal spikes during sleep-wake cycles. First, we conducted power spectrogram analysis of the LFPs (Fig. 2A, left) and determined the stages of the sleep-wake cycle based on prominent oscillation bands, including delta (1–4 Hz) and theta (6–10 Hz). Our results revealed that, overall, the CA1 and MnR exhibited similar oscillation patterns (i.e., dominant delta and theta oscillations) across the sleep-wake cycle, albeit a lower power in the MnR than CA1. During both awake and REM sleep stages, the CA1 and MnR had prominent theta oscillations, whereas during SWS, the CA1 and MnR had prominent delta oscillations (Fig. 2A, left; note that higher-frequency oscillations including gamma and CA1 sharp-wave ripple were truncated for better visualization of the delta/theta bands).
Firing patterns of three types of MnR neurons during sleep–wake cycle. A, Left, Power spectrograms of simultaneously recorded CA1 (top) and MnR LFPs (bottom). Note strong power levels in the delta band (1–4 Hz) during SWS and in the theta band (6–10 Hz) during REM sleep. Right, Schematics of dual-site recording. B, Firing rate histograms of three representative MnR neurons recorded simultaneously as in A. C, Interspike intervals of the same MnR neurons as shown in B.
Furthermore, we classified MnR neurons into three types, namely putative 5-HT, Vglut3, and others based on their firing patterns (Fig. 2B; see above, Materials and Methods). Briefly, the putative 5-HT neurons were defined by a low firing rate (<5 Hz) with a long interspike interval. The putative Vglut3 neurons were defined by a decreased activity immediately before hippocampal ripple events, based on our recent work (Wang et al., 2015). The rest of the MnR neurons consisted of GABAergic and unidentified neurons (GABA & others). Notably, putative 5-HT neurons had clock-like activity across all stages (Fig. 2B, top). Putative Vglut3 neurons often had intermediate firing frequency (3–12 Hz) during the awake state, and most of them (∼60%) had a cyclic on-and-off firing pattern during SWS and were nearly silent during REM sleep (Fig. 2B, middle). Putative GABAergic neurons were characterized by a high firing frequency during both awake and REM sleep (Fig. 2B, bottom). A further analysis on ISIs revealed a large difference among the three MnR groups: putative 5-HT neurons had the longest ISI, putative Vglut3 neurons had intermediate ISI, and putative GABA neurons had the shortest ISI (Fig. 2C).
To confirm the neurochemical identity of our classified three types of MnR neurons, we used a tagging method (Cohen et al., 2012; Wang et al., 2015) to probe genetically labeled Vglut3, 5-HT, and GABAergic neurons using three corresponding transgenic mouse lines, including the Vglut3-Cre, ePet-Cre, and Vgat-Cre lines. The ePet-Cre and Vgat-Cre mouse lines have previously been verified in targeting 5-HT and GABAergic MnR neurons, respectively (Wang et al., 2015). In the current study, we sought to verify the Vglut3-Cre mouse line. We injected Cre-dependent AAV-DIO-EYFP into the MnR of Vglut3-Cre mice (n = 2) and subsequently conducted in situ hybridizations of EYFP and Vglut3 using an RNAscope assay (Wang et al., 2015). We found that ∼95% of EYFP-positive neurons expressed Vglut3 in the MnR (Fig. 3A), confirming the validity of using the Vglut3-Cre mice for tagging Vglut3 neurons.
Tagging of Vglut3 neurons in Vglut3-Cre mice. A, RNAscope in situ hybridization of Vglut3 (left), EYFP (middle), and merged (right). This Vglut3-Cre mouse was previously injected with AAV-DIO-EYFP into the MnR. Scale bar, 0.05 mm. B, Top, schematics of tagging; the MnR was injected with AAV-DIO-ChR2-EYFP plus an optrode implantation (an optical fiber attached to 4–6 tetrodes); the CA1 was implanted with 4 tetrodes. Bottom, Coronal brain section showing the expression of fluorescent ChR2-EYFP mainly in the MnR (indicated by yellow dashed lines; red arrow indicates the optrode track). Scale bar, 0.5 mm. C, Peristimulus rasters and peristimulus time histograms (PSTH) for Vglut3 neurons. Top, One example Vglut3 neuron showing reliable firings on optostimulation. Bottom, Heatmap of the PSTH for all putative Vglut3 neurons. D, PSTH for putative 5-HT neurons. Top, One example putative 5-HT neuron. Bottom, Heatmap of the PSTH for all putative 5-HT neurons. Response latencies ranged between 3–9 ms for neurons 1–11, and 11 ms for neuron 12. E, PSTH for GABA and unidentified MnR neurons. Note that none of these neurons responded to the optostimulation. Optostimulation: 4 or 5 pulses at 25 Hz; pulse width, 3 ms. Color bar represents z-scored firing rate.
To tag MnR Vglut3 neurons, we injected Cre-dependent AAV-DIO-ChR2-EYFP and then implanted a bundle of tetrodes attached to an optical fiber into the MnR of Vglut3-Cre mice; meanwhile, we implanted another bundle of tetrodes into the CA1 (Fig. 3B, top). After 2–3 weeks to allow viral expression, we conducted optostimulation while simultaneously recording neural activity from the MnR and CA1. Our results revealed that the optostimulation activated 86% (32/37) of the putative Vglut3 neurons (Fig. 3C). The response latency, defined as latency to the peak firing rate, was 3.5 ± 1.5 ms (mean ± SD; ranged between 2.0 and 8.0 ms; n = 32), indicating a direct activation via the ChR2 expressed on the neurons. These results validated our classification criteria for putative Vglut3 neurons.
Additionally, the optostimulation activated about half of the putative 5-HT neurons (12/23), although their response latency (5.8 ± 2.3 ms; ranged between 3.0 and 11.0 ms; n = 12; Fig. 3D) was significantly longer than that of Vglut3 neurons described above (t(14.53) = 3.20, p < 0.01, independent t test). These results suggest that about half of the 5-HT neurons likely coexpress Vglut3, a notion supported by previous findings (Jackson et al., 2009; Domonkos et al., 2016; Sos et al., 2017; Senft et al., 2021), hereafter, referred to as 5-HT+Vglut3 neurons. In contrast, the optostimulation did not activate any putative GABA or unidentified neurons (Fig. 3E). After completion of recordings, our histology verifications confirmed the expression of ChR2-EYFP largely restricted within the MnR (Fig. 3B, bottom).
In ePet-Cre mice, the majority of putative 5-HT neurons (6/10) responded to the optostimulation with a latency of 4.7 ± 1.4 ms (range between 3.0 and 7.0 ms; n = 6; Fig. 4A), comparable to that of putative 5-HT neurons recorded in Vglu3-Cre mice (Fig. 3D). In contrast, the optostimulation activated none of the putative Vglut3, GABA, or unidentified neurons (Fig. 4B,C). Note that we recorded a relatively small number of Vglut3 neurons (n = 7), and thus we cannot exclude the possibility of the existence of Vglut3 neurons that coexpress the ePet marker (which would be activated by the optostimulation). Nonetheless, it appears that the ePet-Cre system only labels a subset of 5-HT neurons (Cardozo Pinto et al., 2019), perhaps the 5-HT-only but not 5-HT+Vglut3 neurons.
Tagging of 5-HT neurons in ePet-Cre mice. A, Peristimulus rasters and peristimulus time histograms (PSTH) for 5-HT neurons. Top, One example 5-HT neuron showing activation on optostimulation (4 pulses at 25 Hz; pulse width, 3 ms). Bottom, Heatmap of the PSTH for all putative 5-HT neurons. B, C, PSTH for putative Vglut3 and GABA & others. Top, Two examples. Bottom, Heatmaps of the PSTH for all putative Vglut3 (B) and GABA & others (C). None of these neurons responded to the optostimulation.
In Vgat-Cre mice, the optostimulation activated a subset of MnR neurons with a mean response latency of 3.8 ± 1.7 ms (range between 2.0 and 7.0 ms; n = 17), confirming their GABAergic identity (Fig. 5A). In addition, the optostimulation silenced almost all putative Vglut3 neurons (20/22) with a mean latency of 8.3 ± 3.8 ms (ranged between 2.0 and 16.0 ms; n = 18; the latency of the other two neurons could not be determined because of sparse spikes; Fig. 5B) while having mixed effects on putative 5-HT neurons (Fig. 5C). Some were inhibited (18/28; mean firing rate between 0 and 200 ms had >40% decrease compared with baseline; latency could not be determined because of sparse spikes), whereas some others showed a rebound activation (9/28) with a response latency of 21.2 ± 4.4 ms (range between 16.0 and 28.0 ms; Fig. 5C).
Tagging of MnR GABAergic neurons in Vgat-Cre mice. A, Peristimulus rasters and peristimulus time histograms (PSTH) for GABAergic neurons. Top, One example neuron showing reliable firings on optostimulation (4 pulses at 25 Hz; pulse width, 3 ms). Bottom, Heatmap of the PSTH for all genetically verified GABAergic neurons. B, PSTH for putative Vglut3 neurons. Top, One example putative Vglut3 showing an inhibition on optostimulation. Bottom, Heatmap of the PSTHs for all putative Vglut3 neurons (20/22 neurons were inhibited). C, PSTH for putative 5-HT neurons. Top, One example putative 5-HT neuron showing delayed activation on optostimulation (latency, 20 ms). Bottom, Heatmap of the PSTHs for all putative 5-HT neurons. Note that 9/28 neurons had delayed activation, whereas 18/28 neurons were inhibited on optostimulation. D, PSTH for unidentified MnR neurons. Left and middle, Two example neurons showing an inhibition followed by rebound activation (latency, 25 ms; left) or a delayed activation (latency, 35 ms; middle) on optostimulation. Right, Heatmap of the PSTHs for all unidentified MnR neurons: a subset was inhibited (1–18), another subset had a delayed activation (19–50), and the rest did not respond (51–82).
Interestingly, some of the unidentified MnR neurons were inhibited (18/82; at least five continuous bins with z score lower than –2) with latency of 6.0 ± 3.8 ms (range between 2.0 and 13.0 ms; n = 18; Fig. 5D), whereas another group (35/82) was activated with a latency of 30.7 ± 11.6 ms (range between 13.0 and 48.0 ms; n = 35; 3/35 neurons were initially inhibited followed by a rebound activation; Fig. 5D), likely a multisynaptic effect. This neuronal population (Fig. 5D) likely includes Vglut2 neurons (Szőnyi et al., 2019; Xu et al., 2021) and GABAergic neurons that could have been depolarized by optostimulation (via ChR2) but then immediately hyperpolarized by nearby GABAergic neurons. The rest of the MnR neurons were not affected by the optostimulation (n = 32). Together, these results begin to suggest how the three types of neurons are organized within the MnR: The Vglut3 and 5-HT neurons do not appear to interact locally (likely project to other brain regions), whereas the GABAergic neurons directly inhibit local Vglut3 and 5-HT neuron as well as other GABAergic and unidentified MnR neurons.
In summary, across awake–SWS–REM sleep stages, 5-HT MnR neurons exhibited limited or no firing change (Fig. 6A), whereas Vglut3 and GABAergic MnR neurons exhibited robust firing changes (Fig. 6B,C). More specifically, most MnR Vglut3 neurons exhibited the lowest firing rate during REM sleep, with ∼70% of them silent or nearly silent (Fig. 6B). On the other hand, ∼50% of putative GABA and unidentified MnR neurons exhibited the highest firing rate during REM sleep (Fig. 6C). Notably, those features of each neuron type are consistent between tagged (from transgenic mice) and putatively classified neurons (from C57BL/6 mice). These results revealed distinct firing patterns of the three types of MnR neurons across the sleep–wake cycle.
Firing rates of the three types of MnR neurons across the sleep-wake cycle. A, Mean firing rates of individual tagged (n = 6, top) and putative 5-HT neurons (bottom, black lines indicate neurons recorded from C57BL/6 mice; green lines indicate 5-HT+Vglut3 neurons as shown in Figure 3D). B, Mean firing rates of individual tagged (n = 32, top) and putative Vglut3 neurons (n = 36, bottom) C, Mean firing rates of individual tagged GABAergic (n = 12, top) and GABA & others (n = 59, bottom). Tagged and putative MnR neurons were recorded from Cre and C57BL/6 lines, respectively. Note that most Vglut3 neurons exhibited the lowest firing rate during REM sleep (red lines in B), whereas ∼50% of GABAergic neurons exhibited the highest firing rate during REM sleep (dark blue lines in C).
In addition to characterizing MnR neuronal activities and their interactions, we also examined relationships between their activities and LFP oscillations. The pontine waves (P-waves), which prominently occur during REM sleep, may have important cognitive functions (Hobson et al., 2000; Siegel, 2001; Vertes, 2004; Stickgold, 2005; Montgomery et al., 2008). Previous studies have detected P-waves in the pedunculopontine tegmentum, laterodorsal tegmentum, dorsal raphe nucleus, rostral part raphe nucleus, and so on, close to the pontine area (Datta et al., 1998). Here we detected P-wave (low cut at 2 Hz) in the MnR during REM sleep. Interestingly, the three types of MnR neurons displayed different firing patterns at P-wave troughs (defined as amplitude exceeding 2 SD below mean; Fig. 7A,B). Most of GABA and unidentified neurons (51/65 including 5/6 tagged GABAergic neurons) and a minority of putative Vglut3 neurons (16/68 including 7/37 tagged Vglut3 neurons) increased firing rates at P-wave troughs, whereas putative 5-HT neurons decreased firing rates (8/15, including 1/1 tagged 5-HT neuron). Furthermore, to determine how the CA1 and MnR may interact during P-wave activity, we conducted peri-P-wave power spectrogram analysis of the CA1 and MnR LFPs. We found that there was an increase of theta frequency at P-wave troughs in both the MnR and CA1, suggesting that the CA1 and MnR are synchronized with P-wave (Fig. 7C). In addition, CA1 putative interneurons were also significantly activated at P-wave troughs (24/28), whereas putative pyramidal neurons did not show any firing change (n = 55; Fig. 7D). These results support the notion that CA1 interneurons are modulated by the pontine areas (and possibly MnR) during P-wave activity.
Phasic activation during REM-associated P-wave. A, One example recording showing MnR and CA1 LFPs and four representative neurons recorded simultaneously. Blue curve is the 0–2 Hz filtered P-wave of MnR LFP. Note that the putative GABAergic MnR neuron and CA1 interneuron (Int) increased firing at P-wave troughs. B, Peristimulus raster/peristimulus time histogram (PSTH) for the three types of MnR neurons. Top, Three representative neurons. The GABA & others and Vglut3 neurons had a strong firing increase, whereas the 5-HT neuron had a firing decrease at P-wave troughs. Bottom, Heatmaps of PSTH for all three types of neurons (neurons with sparse firing are not shown here). C, Peritrough power spectrograms of MnR and CA1 LFPs. The frequency of theta band increases at P-wave troughs. D, PSTH for two types of CA1 neurons. Pyramidal (Pyr) neurons had no change, whereas Int increased activity at P-wave troughs. Top, Representative PSTH of two neurons; Bottom, Heatmaps of PSTH for all CA1 Pyr and Int. Red dash lines indicate P-wave troughs.
Next, we examined the relationship between MnR and CA1 LFPs during REM. First, we conducted cross-frequency coupling analysis (Fig. 8A,B). Our results showed that there was a clear theta phase (centered at ∼7 Hz) coupling with the gamma amplitude (centered at ∼70 Hz) in both the CA1 and MnR (Fig. 8B; lower gamma at ∼30 Hz was not evident in the CA1, likely because of our recording sites slightly above the pyramidal layer). Second, we conducted a power–power correlation between CA1 and MnR LFPs. The comodulogram showed a high-power CA1/MnR coupling at both the theta and gamma bands (Fig. 8C). In addition, there was also a delta (MnR)/theta (CA1) coupling (Fig. 8C), consistent with the association between MnR P-wave and CA1 theta (Fig. 7C). Third, we conducted a coherence analysis, which also revealed a high coherence of LFPs between the MnR and CA1 in theta and gamma bands, peaked at 7.6 and 64 Hz, respectively (Fig. 8D). Last, we calculated the Granger causality of LFPs from the MnR to CA1 and vice versa. Our result showed that the CA1→MnR causality was centered ∼7.6 Hz, but the MnR→CA1 causality varied across animals. Three animals peaked at ∼7.5 Hz, two animals peaked at ∼6 Hz, and one animal peaked at 5.3 Hz. It is unclear what caused these differences. Nonetheless, there was significant difference of the peak causality value between the two directions (t(5) = 3.23, p < 0.05, paired t test; Fig. 8E). These results suggest that the MnR modulates the CA1 during theta activity and that theta-gamma cross-frequency coupling supports multiple time-scale control of neuronal spikes within and across structures.
Phase–amplitude coupling, power–power correlation, coherence, and Granger causality. A, One example showing simultaneously recorded CA1 and MnR LFPs and corresponding theta and gamma waves. B, Heatmap of phase–amplitude coupling for the CA1 (top) and MnR (bottom) LFPs shown in A. Color bar represents coupling value. Note that the theta phase couples with the gamma amplitude for both CA1 and MnR. C, Comodulogram for power–power correlation of LFPs between CA1 and MnR. Color bar represents average correlation value (n = 6). D, Coherence spectra between CA1 and MnR. The red solid line and pink shadow indicate the mean and SD of the values from six mice. The peak in the theta band is on 7.6 Hz, and the peak in the gamma band is on 64 Hz. E, Granger causality spectra. The red and blue solid curves with shadow indicate the granger causality value (mean ± SE from 6 mice) of MnR to CA1 and CA1 to MnR, respectively.
To determine the contribution of the three types of MnR neurons to REM theta activity, we analyzed the spike–LFP phase locking to theta. We found that very few 5-HT neurons (2/22) exhibited phase-locking activity, whereas a small subset of Vglut3 neurons (15/53 including 6/17 tagged Vglut3 neurons) and about half of GABA and unidentified MnR neurons (36/71 including 6/12 tagged GABAergic neurons) were significantly phase locked at various phase points (z > 4.605; log z > 1.527; p < 0.01, Rayleigh's test; Fig. 9). These results suggest that Vglut3, GABA, and unidentified MnR neurons, rather than 5-HT neurons, play major roles in modulating HTO during REM sleep.
Spike-phase locking for MnR neurons. A, Left, Histogram in the theta phase for three representative MnR neurons. Right, Angular plot of the histogram shown on the left. Note that the Vglut3 and GABAergic neurons fired phase locked to different theta phases, whereas the 5-HT neuron showed no phase locking. All three neurons were genetically verified via tagging. B, Cumulative percentage curve of log z value of spike-phase locking for the three types of MnR neurons. The dashed line indicates significance level, α = 0.01. C, Heatmaps of histograms for the significantly phase-locking neurons (p < 0.01, Rayleigh's test). Color bar represents z-scored histograms.
Discussion
Three types of MnR neurons
It is well established that the MnR consists of at least three neuronal populations, including 5-HT, Vglut3, and GABAergic neurons (Köhler and Steinbusch, 1982; Gras et al., 2002; Herzog et al., 2004; Jackson et al., 2009; Varga et al., 2009; Hioki et al., 2010; Domonkos et al., 2016; Sos et al., 2017; Senft et al., 2021). The latest studies also revealed the existence of a fourth population of Vglut2 neurons, which, however, constitutes a relatively small population and is more prevalent in the paramedian raphe rather than midline MnR (Hioki et al., 2010; Szőnyi et al., 2019; Xu et al., 2021; https://mouse.brain-map.org/experiment/show/73818754). In our current study, we focused on genetically verified 5-HT, Vglut3, and GABAergic MnR neurons and characterized their activity patterns during the sleep–wake cycle in freely behaving mice. All 5-HT neurons displayed a slow and clock-like activity regardless of the stages across the sleep–wake cycle, whereas Vglut3 and GABAergic neurons displayed robust firing changes depending on awake–SWS–REM stages. Notably, most MnR Vglut3 neurons displayed a characteristic on-and-off cyclic firing pattern during SWS and were almost silent during REM sleep; in contrast, MnR GABAergic neurons displayed much higher activity during REM sleep than during SWS. Although GABAergic neurons are most likely responsible for the higher MnR activity during REM sleep, future research should examine whether Vglut2 neurons play any role.
Our current study also provided data concerning how 5-HT, Vglut3, and GABAergic neurons are organized within the MnR. First, our tagging data suggest that GABAergic neurons monosynaptically inhibit most Vglut3 neurons (Fig. 5B). Second, GABAergic neurons appear to monosynaptically inhibit some 5-HT neurons while polysynaptically exciting other 5-HT neurons (Fig. 5C). Similarly, GABAergic neurons appear to monosynaptically inhibit some while polysynaptically exciting other unidentified MnR neurons (Fig. 5D). It remains to be investigated how GABAergic neurons activate some of the 5-HT and unidentified MnR neurons. Finally, Vglut3 or 5-HT neurons do not appear to regulate local neurons, as optostimulation of either had no effect on other MnR neurons (Figs. 3,4).
Previous studies have shown that a subset of MnR neurons coexpress 5-HT and Vglut3 markers (Jackson et al., 2009; Okaty et al., 2015; Domonkos et al., 2016; Sos et al., 2017; Ren et al., 2019; Senft et al., 2021). Consistently, our results revealed that ∼50% putative 5-HT neurons likely coexpress the Vglut3 marker (Fig. 3D). Overall, 5-HT+Vglut3 neurons seem to be more similar to 5-HT-only than Vglut3-only neurons for two reasons. First, 5-HT+Vglut3 and 5-HT-only neurons exhibited similar firing patterns across the sleep–wake cycle. Second, 5-HT+Vglut3 and Vglut3-only neurons responded to the optogenetic stimulation with different latencies (6.3 vs 3.7 ms), for which a greater latency could indicate a lower expression level of ChR2 and, thereby, a lower expression level of Vglut3 in 5-HT+Vglut3 than Vglut3-only neurons. Nonetheless, the existence of 5-HT+Vglut3 neurons supports the idea that glutamate and serotonin may work synergistically in influencing postsynaptic targets. One possibility is that glutamate released from MnR Vglut3 terminals first prepotentiate targeted neurons, which are subsequently further potentiated by cotransmitted 5-HT via ionotropic 5-HT3a receptors (Jackson et al., 2009; Senft et al., 2021). Additionally, 5-HT released from 5-HT-only neurons may exert an added synergistic effect based on our previous finding that putative 5-HT neurons fire lagged putative Vglut3 neurons by ∼200 ms under physiologic conditions (Wang et al., 2015).
MnR non-5-HT neurons modulate hippocampal theta activity
The present study provided direct evidence that optostimulation of the MnR can evoke robust HTO. Moreover, the Granger causality analysis indicated that spontaneously occurring MnR theta oscillations influence HTO. However, the exact mechanism on how different MnR neuronal types modulate HTO remains not fully understood.
Previous studies suggested that 5-HT suppresses HTO (Vertes and Kocsis, 1997; Vertes et al., 1999; Crooks et al., 2012; Hsiao et al., 2019). Unexpectedly, we found that MnR 5-HT neurons showed neither firing rate change across the sleep–wake cycle nor theta-locked activity during REM sleep. Therefore, MnR 5-HT neurons likely play a limited role in affecting HTO, or their suppressive role can be overridden by other HTO drivers during REM sleep. It is also possible that this 5-HT effect observed in prior studies is mediated by MnR Vglut3 neurons that express the 5-HT1a receptor. Inactivation of these Vglut3+ neurons via 5-HT could release its suppressive effect in the MS/DBB, which in turn promotes HTO.
On the other hand, our results suggest that Vglut3, GABA, and unidentified MnR neurons play important roles in modulating HTO, especially during REM sleep. Most GABA & others and a small subset of Vglut3 MnR neurons showed rhythmic activity phase locked to HTO and, respectively, increased and decreased firing rates during the REM stage. Regarding the unidentified MnR neurons, two recent studies raise the possibility that at least some of them are Vglut2 neurons, which can influence HTO via the MS/DBB (Szőnyi et al., 2019; Xu et al., 2021).
Converging evidence suggests that the MnR-to-hippocampus projection is predominantly glutamatergic; ∼58–80% of the neurons projecting from the MnR to the hippocampus express Vglut3 (Jackson et al., 2009; Varga et al., 2009; Domonkos et al., 2016; Senft et al., 2021). Although MnR 5-HT neurons project modestly to the hippocampus, MnR GABAergic and Vglut2 neurons do not directly project to the hippocampus but, instead, to the MS/DBB (Jackson et al., 2009; Szőnyi et al., 2019; Senft et al., 2021; Xu et al., 2021). Based on these anatomic connections, the MnR can potentially influence hippocampal activity directly via its Vglut3 and 5-HT neurons and indirectly via its 5-HT, GABAergic, Vglut3, and Vglut2 neurons projecting to the MS/DBB (Kinney et al., 1996; Leranth and Vertes, 1999; Crooks et al., 2012; Szőnyi et al., 2019). Taking this into consideration, we suggest that HTO could be modulated by the neural pathways originating from either MnR Vglut2, Vglut3, GABAergic neurons, or any combination of the three neuronal populations: 1) MnR Vglut2 neurons could modulate HTO via the MS/DBB; 2) MnR Vglut3 neurons could modulate HTO through the direct projection to the hippocampus and through the MS/DBB; 3) MnR GABAergic neurons could modulate HTO through the MS/DBB and through MnR Vglut3 neurons.
MnR P-wave modulates hippocampal activity
Slow oscillation known as the P-wave (∼1 Hz) has been associated with REM sleep (Hobson et al., 2000; Siegel, 2001; Vertes, 2004; Stickgold, 2005; Montgomery et al., 2008) and likely plays a role in regulating theta activity (Karashima et al., 2001, 2004; Ramirez-Villegas et al., 2021). We found that P-wave emerged in the MnR during REM sleep and that both MnR and CA1 theta oscillations increased their frequency at the trough of MnR P-wave. This is consistent with previous findings that reported theta oscillation acceleration but not amplitude changes synchronized with P-waves in rats (Karashima et al., 2005). Additionally, we found that MnR 5-HT neurons decreased their activity at the trough of P-wave, indicating a release of 5-HT-mediated suppression on P-wave activity. This corroborates with a previous finding that raising 5-HT level in the brain suppresses P-wave activity in cats (Brooks and Gershon, 1977). Finally, we found that many putative Vglut3, GABA, and unidentified MnR neurons, as well as CA1 interneurons, significantly increased their firing rates at the trough of the P-wave activity, which has not been reported previously. Therefore, P-wave may coordinate MnR and hippocampal activities at a slower scale and plays an important role in memory processes associated with REM sleep.
In conclusion, these results suggest that MnR nonserotonergic neurons play an important role in modulating HTO during REM sleep, whose functions include memory processes.
Footnotes
This work was supported by the National Institute of Mental Health–National Institutes of Health (Grant R01 MH119102), Drexel University College of Medicine, and the National Institute on Drug Abuse–Intramural Research Program (NIDA–IRP).We thank Dr. Thomas Kash for providing a breeding pair of Vglut3-Cre mice (originally generated at the Bradford Lowell Lab), Dr. Jen-Hui Tsou for help with the RNAscope assay (Vglut3-Cre mice), Dr. Hualou Liang and Dr. Lu Zhang for discussions and sharing MATLAB codes on Granger causality analysis, and Ashley Opalka for editing this article. We also thank the NIDA–IRP animal facility for breeding and caring for the Vglut3-Cre, ePet-Cre, and Vgat-Cre mice.
The authors declare no competing financial interest.
- Correspondence should be addressed to Dong V. Wang at dw657{at}drexel.edu
