Male 8–12-week old, vGAT-iRES-Cre (Jackson Laboratory, JAX:016962) and vGAT-IRES-Cre/Gt(ROSA)26Sor_CAG/ZsGreen1 mice were used in this study. Gt(ROSA)26Sor_CAG/ZsGreen1 mouse strain originates from Jackson Laboratory, JAX:007906. Mice, two to three animals per cage, were housed on a reversed 12-h light/dark cycle. Food and water were available ad libitum. All experiments were approved by the Animal Care and Use Committee of the Institute of Experimental Medicine and the Committee for Scientific Ethics of Animal Research of the National Food Chain Safety Office under the project number PE/EA/200-2/2020 and were performed according to the 2010/63/EU Directive of the EC Council. All efforts were made to minimize pain and suffering and to reduce the number of animals used.

Virus injections were performed under general anesthesia by intraperitoneal injection of ketamine–xylazine mixture (dose by body weight, 100 and 10 mg/kg, respectively), and analgesic (Buprenorphine 0.1 μg/g, s.c.) was applied at the beginning of surgery. MRR region was targeted via a burr hole (AP, –4.1 mm, ML, 0.0 mm) perpendicular to the skull surface using a standard stereotaxic frame (David Kopf Instruments, Tujunga, CA, United States). AAV-EF1a-DIO-hChR2(H134R)-eYFP (UNC Vector Core, Chapel Hill, NC, United States) or AAV-CAG-FLEX-ArchT-GFP (UNC Vector Core, Chapel Hill, NC, United States) was injected at a depth of 4.6–4.5 mm from the skull surface through a pulled glass micropipette using Nanoject II precision microinjector pump (Drummond, Broomall, PA, United States). After injection, mice were allowed to recover for at least 2 weeks to ensure sufficient expression of the virus.

Surgeries for whole cell patch-clamp recordings were done under isoflurane anesthesia (0.5–1.5%), and analgesic (Buprenorphine 0.1 μg/g, s.c.) was applied at the beginning of surgery. A small lightweight headplate was attached to the skull using Optibond adhesive (Kerr, Brea, CA, United States) and Paladur dental acrylic (Kulzer, Hanau, Germany). During whole cell recordings, mice were head-restrained with a downward tilted head position (pitch angle: 20°). For optogenetic experiments, an optic fiber (100 μm core diameter, Thorlabs GmbH, Newton, NJ, United States) was implanted above MRR (AP, –5.4 mm, ML, 0.9 mm at 18° ML angle) and fixed with dental acrylic. On the day of recording, at least 2 days after headplate surgeries, craniotomies were drilled above MRR (AP, –5.6 mm, ML, –0.7 mm, patch electrode ML at 10° angle) and dorsal hippocampus (AP, –3.2 mm, ML, –2.2 mm, LFP electrode AP, 20° ML at 10° angle). The craniotomies were covered with fast sealant (Body Double, Smooth-On, Easton, PA, United States), and recording was performed at least 1 h after surgery to allow recovering from anesthesia.

Headplate surgeries for multichannel recordings were done under isoflurane anesthesia (0.5–1.5%), and analgesic (Buprenorphine 0.1 μg/g, s.c.) was applied at the beginning of surgery. A small lightweight headplate was attached to the skull using Optibond adhesive (Kerr, Brea, CA, United States) and Paladur dental acrylic (Kulzer, Hanau, Germany). During multichannel recordings, mice were head-restrained with a downward tilted head position (pitch angle: 20°). Two cranial windows (1.5 × 1.5 mm) were drilled above the left hippocampus (AP, –2.5 mm; ML, 2.5 mm) and MRR (AP, –6.1 mm; ML, 0.0 mm, MRR probe ML at 4° angle) under stereotaxic guidance. For the optic fiber and ground electrode a hole was drilled above MRR (AP, –6.1 mm; ML, 0.0 mm) and cerebellum (AP, –5.6 mm; ML, 2.0 mm) respectively. The craniotomies and drill holes were covered with fast sealant (Body Double, Smooth-On, Easton, PA, United States). After surgery, the mice were continuously monitored until recovered, and then they were returned to their home cages for at least 48 h before starting habituation to the head restraint.

Mice were habituated to the head restraint and experimental setup for 1–2 days before each recording session. Head restrained mice were free to run, walk, or sit on the treadmill equipped with a 2-m-long belt. Current-clamp recordings were performed from MRR, 4,200–4,800 μm from the pial surface using a Multiclamp 700B amplifier (Molecular Devices, CA, United States). The data were digitized using CED Micro 1401 laboratory interface (Cambridge Electronic Design Limited, Cambridge, United Kingdom) and recorded using Spike 2 acquisition software. No bias current was applied during recordings and series resistances ranged between 20 and 60 MΩ. Once we got a stabilized recording, optical stimulations (light source: 447 nm diode or 593 nm DPSS laser, Roithner Lasertechnik GmbH, Austria, Ikecool, Corporation, United States, respectively, the latter no longer operational) were applied following 1–2 min of the control period. Both whole cell patch-clamp and cell-attached recordings were stabile during quiet wakefulness but we abruptly lost the cells at the onset of movement. For the analysis, we selected periods with stabile membrane potential; thus, the usual recording length was 150–250 s (median 181 s, interquartile range 99.5–301 s) with the shortest period of 30 s (Supplementary Table 1). For anesthetized experiments urethane (0.007 mL/g of 20%) was ip. administered at the beginning of the recording. In vivo external solution contained (in mM) 150 NaCl, 2.5 KCl, 10 HEPES, 1.5 CaCl₂, and 1 MgCl₂ (pH 7.3). Patch pipettes (5–8 MΩ) were filled with (in mM) 125 K-gluconate, 7 KCl, 10 HEPES, 10 sodium phosphocreatine, 0.1 EGTA, 2 MgATP, 2 Na₂ATP, 0.5 Na₂GTP (pH 7.2, 280–295 mOsm), and 2 mg/mL biocytin was added before recording. At the end of the recording, mice were transcardially perfused with 4% paraformaldehyde (PFA) and the brain was removed for post-hoc immunohistochemistry.

Mice were habituated to head restraint and experimental setup for 1–2 days. Head-restrained mice were free to run, walk, or sit on air supported free-floating 20 cm diameter polystyrene ball. On the day of recording Buzsaki64 or Poly5 silicon probes (Neuronexus, Ann-Arbor, MI, United States) were lowered through the cranial window to the left dorsal hippocampus and MRR under isoflurane anesthesia (0.75–1.5%). The optical fiber was inserted in the MRR at 4.1 mm depth from the skull surface. Probes and optical fiber were coated with a lipophilic fluorescent dye, DiI (Thermo Fisher Scientific, Waltham, MA, United States), for later histological verification of the location. A ground electrode was placed above the cerebellum. Mice were allowed to recover from anesthesia for ∼1 h before recording.

The probe in the hippocampus was advanced using a micromanipulator (David Kopf Instruments, Tujunga, CA, United States) until the pyramidal layer was detected by increased unit activity and the occurrence of ripple events. In the MRR, unit activity was monitored from 4.2 to 5.2 mm depth, and the final position was determined based on the appearance of units ceasing their firing upon 2 s long laser stimulation (593 nm, Ikecool Corporation, United States). Once we found optically silenced units, the recording was commenced after an approximately 1 h waiting period for letting the tissue settle around the probe. Electrophysiological recordings were performed by a signal multiplexing head-stage (RHD 128, Intan Technologies, LA, United States) and an OpenEphys data acquisition board¹. Signals were acquired at 20 k sample/s (Open Ephys 0.4.4.1). Mouse locomotor activity was monitored with an optical computer mouse positioned close to the polystyrene ball at the equator. Speed was calculated by a custom written macro in Igor Pro. At the end of the recording, mice were transcardially perfused with 4% PFA and the brain was removed for post-hoc immunohistochemistry.

Neuronal spikes were detected and automatically sorted by a template-matching algorithm using the Spyking Circus software (Yger et al., 2018), followed by manual curation of the clusters using the Phy software (Rossant et al., 2016) to obtain well-isolated single units. Spike sorting quality was assessed with a refractory period violation, and visual inspection of auto- and crosscorrelations; poor quality clusters were discarded.

The location of the recorded cell was confirmed by biocytin labeling. We analyzed only those recordings in which the recording electrode, visible by the non-specifically labeled biocytin track, was inside the MRR region. Out of 108 recorded cells, we obtained 27 whole cell recordings. Of these, 13 were unequivocally identified as GABAergic or non-GABAergic. If we had been not able to find the biocytin labeled cell or there were more, equally labeled cells, we classified the registered neurons as non-identified (NI). The neurotransmitter phenotype of the recovered biocytin-labeled cells was determined by the colocalization of ZsGreen1 (selectively expressed by GABAergic neurons in the ZsGreen reporter mice, see above) or alternatively, by post hoc immunohistochemistry.

Optotagging was also used for cell identification. In AAV-EF1a-DIO-hChR2(H134R)-eYFP-injected vGAT-ires-Cre mice, if light stimulation (447 nm) had evoked depolarization and spiking, or hyperpolarization and spike suppression, the cell was identified as GABAergic or non-GABAergic, respectively. If the delivery of 593 nm light in AAV-CAG-FLEX-ArchT-GFP-injected vGAT-ires-Cre mice silenced the cell or evoked marked depolarization, the neuron was classified as GABAergic or non-GABAergic, respectively.

For the reconstruction of neuronal morphology and probe locations, coronal or parasagittal sections (60 μm) were cut using a vibratome (Leica Microsystems, Wetzlar, Germany). Sections were first incubated in blocking solution for 2 h (10% normal goat serum (NGS), 0.5% Triton X-100 in 0.1 M phosphate-buffered saline (PBS), and then incubated overnight in streptavidin AlexaFluor-488 or Cy3 (1:1,000, Jackson Immuno Research) to reveal the identity of the biocytin-filled neuron. To amplify the signal of the virally delivered fluorescent reporter protein, coexpressed with ChR2 or ArchT, sections from transduced brains were incubated in chicken anti-Green Fluorescent Protein (GFP, 1:2,000, Life Technologies) primary and anti-chicken AlexaFluor-488 secondary antibodies (1.1,000, Thermo Fisher Scientific). 5-HT and vGlut3 content were investigated by incubating the sections in guinea pig anti-vGlut3 (1:1,000, Frontier Institute, Nittobo Medical Co.) and rabbit anti-5HT (Immunostar) primary and anti-guinea pig AlexaFluor-647 or AlexaFluor-405 and anti-rabbit AlexaFluor-647 or 405 (1:1,000, Thermo Fisher Scientific or Jackson Immuno Research). All antibodies were diluted in carrier solution (0.1 M PBS, 1% NGS, 0.2% Triton-X100) and incubated overnight at room temperature. Fluorescence signals (DiI, streptavidin Alexa, or fluorescent reporter protein of ChR2, ArchT, and vGAT) were inspected using an Axioplan 2 microscope (Carl Zeiss, Oberkochen, Germany). To investigate the neurochemical content (vGlut3, 5-HT, vGAT) of biocytin-labeled cells, z stack images were acquired using a Nikon C2 confocal microscope (x40 objective, Nikon, Europe).

All in vivo data were analyzed in Igor Pro 8 (Wavemetrics, Lake Oswego, OR, United States). Action potentials from whole cell recording voltage traces were detected automatically using an amplitude threshold algorithm (TaroTools, custom written macro in Igor Pro)² and were manually verified. The membrane potential spike threshold was computed as the second derivative of the voltage trace. Average membrane potential was calculated after clipping spikes above the threshold voltage. To calculate theta phase coupling, recording channels with maximal ripple amplitude, corresponding to the pyramidal layer, were selected. LFPs were filtered in the theta range (4–12 Hz). The phase of the theta oscillation was calculated by computing the Hilbert transform. Each single unit spike had been assigned to its coincident phase value and then the phase distribution was computed. Theta-coupling was assessed by performing Rayleigh’s test for circular uniformity (p > 0.05). For computing theta rhythmicity index, autocorrelograms of every single unit were calculated, and theta rhythmicity index was defined as the ratio of the local minimum and maximum in a time window between 80 and 200 ms. To investigate the correlation between firing frequency and speed, the latter was normalized by its maximum in every animal and then divided into 25 bins. The firing rate of the neuron in each speed bin was then plotted and the resulting point cloud was fitted by linear regression. To analyze ripple coupling, hippocampal recording channels with maximal ripple amplitude were selected and then downsampled to 1 kHz. The LFPs were band-pass filtered in the 120–250 Hz range and the Hilbert transform was computed. Ripples were detected as events exceeding 3 SD of the signal with a minimal duration of 20 ms. The z-scored ripple peak triggered perievent histograms (PETH) were created for every single unit. Fast ripple-coupled units were defined if the z-scored PETH exceeded 2 SD in either the positive or negative direction for at least 20 ms in a 300 ms window around ripple peak. Slow reduction of ripple-coupled firing suppression was defined as a > 2 SD decrease of the z-scored PETH values for at least 500 ms in a 5-s-long window preceding the ripple peak. The onset and offset of ripple-coupled activity were defined as the 1 SD crossings. For the stimulus-triggered wavelet, the LFP was subjected to continuous wavelet transformation, utilizing a Morlet mother wavelet.

In whole cell experiments in each session shuffled PETHs were generated by selecting random timepoints matching the number of ripples in that session repeated 100 times. The ripple-triggered PETHs were plotted with the corresponding 95% confidence bounds of the shuffled data. If bin values exceeded the 95% confidence bounds for at least 20 ms, it was defined as a significant deviation.

All statistical analyses were performed with standard Igor Pro 8 functions. For comparisons, Student’s t-test or Mann–Whitney and Wilcoxon test were used. The value p < 0.05 was considered significant. All values indicate mean ± SD unless stated otherwise.

To uncover the physiological characteristics and state-dependent activity of GABAergic and non-GABAergic neurons in the median raphe nucleus and surrounding paramedian raphe [together termed median raphe region (MRR)], we performed whole cell patch-clamp or cell-attached recordings from MRR with the simultaneous registration of local field potentials (LFP) in the hippocampal CA1 region in head-fixed behaving mice (Figure 1A). LFP was recorded from the pyramidal layer of dorsal CA1, identified based on the occurrence of ripple waves during immobility. The location of the recorded cell was confirmed by biocytin labeling. We analyzed only those sessions in which the recording electrode (visible by the biocytin-labeled track) was inside the MRR (Figure 1B).

We obtained 27 whole cell recordings and 81 cell attached recordings (N = 62 mice) from the MRR in single cell recording experiments. There was no difference in the firing frequency (median 4.72 Hz, interquartile range 2.59–8.71 Hz vs. median 7.45 Hz, interquartile range 3.21–14.09 Hz, Wilcoxon signed-rank test) nor in the coefficient of variation of inter-spike intervals (CV_ISI, median 1.06, interquartile range 0.79–1.35 vs. median 1.13, interquartile range 0.80–1.66, Wilcoxon signed-rank test) between whole cell and cell attached recordings.

For the identification of GABAergic cells, we have used a double transgenic mouse line, generated by crossing ZsGreen1 fluorescent reporter mice with vesicular GABA transporter (vGAT)-Cre mice, in which GABAergic cells expressed the ZsGreen fluorescent protein. Non-GABAergic neurons were labeled post hoc for vesicular glutamate transporter 3 (vGlut3) and serotonin (5HT). The unequivocal neurochemical classification was possible in 25 out of 108 recorded MRR cells: 12 cells were GABAergic whereas 13 cells were proved to be non-GABAergic (NG). In the remaining 83 cells, the neurotransmitter phenotype could not be determined (non-identified cells, NI). Firing frequency varied widely in each group, from 0.33 to 13.9 Hz in GABAergic cells (n = 12, median 5.78 Hz), 0.35–34.18 Hz in non-GABAergic neurons (n = 13, median 5.54 Hz), and between 0.18 and 38.16 Hz in case of NI neurons (n = 83, median 7.41 Hz, Figure 1D but see Supplementary Table 1). Accordingly, both GABAergic and non-GABAergic cells displayed a broad range of ISIs without significant difference in their distribution (median 145.95 ms, interquartile range 118.12–457.87 ms vs. median 179.70 ms, interquartile range 58.33–315.30 ms, Wilcoxon signed-rank test). The firing pattern regularity of GABAergic neurons formed a continuum from regular to irregular reflected in the broad distribution of CV_ISI (median 1.18, range 0.51–1.81. Figures 1C,D). The majority of these cells typically displayed irregular firing patterns with intermittent bursting (Figure 1C), sharply distinct from the characteristic clock-like, tonic, serotonergic firing pattern (Aghajanian et al., 1978; Prisco et al., 2002; Kocsis et al., 2006). The resting membrane potential of GABAergic cells varied between –44.02 and –55.46 mV (n = 7, –49.50 ± 4.03 mV) and they displayed unimodal subthreshold Vm distribution, centered around –50 mV (Figure 1E). The membrane potential distribution of GABAergic and non-GABAergic cells (n = 6, –48.70 ± 4.03 mV, two-tailed t-test) was not statistically different. Action potential threshold of GABAergic cells (–33.77 ± 3.42) was also in the same range as that of non-GABAergic neurons (–35.6 ± 6.88, Figure 1F). However, large subthreshold Vm fluctuations characterized the former population thus, the standard deviation (SD) of Vm of GABAergic and non-GABAergic cells was significantly different (3.56 ± 0.22 vs. 3.04 ± 0.52, two-tailed t-test, *p < 0.05, Figure 1F).

To directly investigate the effect of local GABAergic cells on the spike output of non-GABAergic putative projection neurons (Jacobs and Azmitia, 1992; Hioki et al., 2009; Jackson et al., 2009; Varga et al., 2009; Szőnyi et al., 2014, 2019), we combined cell type-selective optogenetic manipulation of the former with whole-cell patch clamp recordings from MRR neurons. In several cases, we abruptly lost the cells at the middle of the stimulation because of movement, therefore some of the experiments (n = 6/6 at ChR2 and n = 3/11 at ArchT) were carried out under urethane anesthesia. Channelrhodopsin 2 (ChR2) was targeted to MRR cells using the vGAT-Cre transgenic mouse line and viral gene transfer (AAV vector). One millisecond-long optogenetic single pulse stimulation did not evoke a detectable effect in the membrane potential but lengthening the pulse to 3 or 10 ms resulted in marked depolarization or spiking in GABAergic cells (n = 3) and phasic hyperpolarization (–3.36 and –5.39 mV, respectively) in non-GABAergic cells (n = 3) (Supplementary Figure 1).

ArchaerhodopsinT [ArchT, Han et al. (2011)] was utilized for suppressing neuronal activity (Figure 2A). To assess the efficiency of ArchT, we recorded from GABAergic neurons and found that illumination (5 s pulse, 593 nm) caused marked hyperpolarization (delta Vm: –10.9 ± 12.32 mV, n = 3) and silencing of three out of four tested cells (n = 4, Figure 2B).

Inactivation of the GABAergic circuit in the MRR revealed a marked tonic inhibition of non-GABAergic projection neurons. ArchT mediated suppression (1–2 s pulses) of GABA-mediated inhibition caused depolarization and a rightward shift in Vm distributions in the neighboring, non-GABAergic cells (Figure 2D). The membrane potential depolarization (from –47.87 ± 2.07 mV to –45.08 ± 1.72 mV, n = 4, Figure 2F) was only observed during light stimulation and returned to baseline after stimulus cessation (onset and offset latencies were 30.72 ± 3.17 ms and 38.64 ± 14.98 ms, respectively). Depolarization during the suppression of local inhibition was paralleled by a marked increase in non-GABAergic cells’ firing rates (n = 7, from 0.88 to 3.66 Hz, median, *p < 0.05, Wilcoxon signed-rank test, Figure 2C). After the termination of light delivery, a small rebound hyperpolarization (–3.36 ± 2.39 mV) was observed in the majority of recorded non-GABAergic units (n = 3/4).

Among the intracellularly recorded neurons (n = 4), we found one 5HT positive/vGlut3 negative cell, one vGlut3 positive neuron (Figure 2E), and a putative vGlut2 positive cell, which was negative for both 5HT and vGlut3. These results confirm that each major type of non-GABAergic projection neuron in the MRR is under a tonic inhibitory control by local GABAergic cells.

State-dependence is a hallmark of neuronal activity in the median raphe, possibly shaped by the MRR GABAergic circuit. Hence, we compared the firing of MRR GABAergic cells during non-theta and theta states. Due to microwobbling amplified by the proximity of the basilar artery and the central aquaeduct to the MRR, whole cell recording was possible only during immobility (n = 27 of 108 neurons). Nevertheless, short theta segments could be isolated in 15 cells (n = 15/27), and the membrane potential of GABAergic cells was found to be unchanged while transient theta epochs emerged compared to the non-theta state (Supplementary Figure 2A). To study state-dependence in a larger population of GABAergic neurons, we deployed multichannel recording combined with inhibitory optical tagging (Figure 3A). ArchT was targeted to MRR GABAergic cells by viral gene transfer in vGAT-Cre mice. MRR and hippocampal activity were simultaneously sampled by silicone probes in head-fixed, awake animals running or sitting on air-supported free-floating (20 cm diameter) polystyrene balls. Delivery of 2-s long laser pulses enabled the tagging of even low activity cells (Supplementary Figure 2B). We could identify 39 GABAergic units from five animals. Their firing frequency varied in a wide range from 0.06 to 24.29 Hz (median 5.08 Hz) and did not differ significantly from the frequency of GABAergic neurons recorded by the whole cell patch-clamp (Supplementary Figure 2C). The appearance of theta was accompanied by the change of firing activity in the majority of GABAergic neurons (Figure 3B). During locomotion, the firing frequency increased, decreased, or remained unaffected in n = 23/39, 7/39, and 9/39 cells, respectively (Figures 3D,E). The direction of firing rate modulation did not depend on the magnitude of firing rate during quiet wakefulness (r = –0.114, p = 0.487), but GABAergic cells varied in their sensitivity to movement. That is, 21/39 GABAergic cells displayed a firing rate increase that was positively correlated with speed (Figure 3C), 16/39 cells showed no correlation, whereas 2/39 cells were inversely correlated. Besides the level of activity, we also explored how the firing pattern regularity of these neurons is affected by state transitions. Various patterns ranging from regular to irregular were observed spreading along a continuum reflected in the broad distribution of CV_ISI (Supplementary Figure 2C). Non-theta to theta state transition at the onset of movement altered the firing pattern, resulting in augmented rhythmicity and the occasional clustering of spiking (Figures 3B,F). In addition, there was no correlation between the rhythmicity during quiet wakefulness and the magnitude of the rhythmicity change (r = –0.223, p = 0.237), albeit the more rhythmic the cells were, the larger increase of frequency was detected during locomotion (r = 0.505, p = 0.004). Interestingly, there was an abrupt increase of theta power in the hippocampus with accompanying suppression of subtheta bands parallel with the light stimulation when the majority of MRR GABAergic cells were presumably silenced (Supplementary Figure 3). Importantly, the elevation of theta frequency preceded the onset of movement.

Next, we aimed to unravel the temporal relationship of the activity of MRR GABAergic cells to hippocampal theta. The majority of GABAergic cells showed theta phase-coupled firing (n = 31/39, Rayleigh test for uniformity, p < 0.05, mean resultant length 0.27 ± 0.02) covering the whole theta cycle as a population (Figure 4A). A subgroup of theta preferring MRR GABAergic neurons (n = 16/31, Figures 4A,B) exhibited ripple correlated-activity (absolute z-score values higher than 2 for at least 20 ms in a 300-ms window around ripple peak). Ripple activated cells covered mostly the ascending phase of theta, whereas cells with fast transient suppression of activity during ripples concentrated near the trough (Figure 4D). Transient activation (n = 9/31) lasted about 52 ms on average (51.67 ± 17.5 ms) centered around ripple peaks. On average, rapid suppressions (n = 7/31, 46.43 ± 22.3 ms) tended to start later compared with activation relative to ripple peaks (23.89 ± 11.11 vs. 10.71 ± 24.57 ms before ripple peaks (Figure 4C), but the difference did not reach significance. Ripple-associated transient increase of activity was also observed in two cells (n = 2/8) with no theta preference (n = 8/39) (Supplementary Figure 4A). Ripple-coupled fast transient membrane potential changes could not be observed in the identified whole cell recorded GABAergic cells (n = 0/7).

We aimed to explore if alteration of local inhibition during ripples is mirrored by a complementary change in the ripple-coupled firing of the MRR’s non-GABAergic output neurons. To this end, we investigated the ripple-correlated firing of these neurons recorded in whole cell mode.

One glutamatergic cell (5HT-, vGlut3+) showed fast transient suppression of spiking (Figure 4E) after ripple peak paralleled by a marked membrane potential hyperpolarization, reaching its maximum postpeak (Figure 4F). In the background of the rapid hyperpolarizing transient, we could also detect a slow depolarization on multiple second-long timescales, resulting in a significant elevation of activity around ripples (Figure 4G). In a serotonergic/glutamatergic cell (5HT+, vGlut3+), we observed slow gradual hyperpolarization and simultaneous decrease of activity concurrently with ripples (Figures 4H–J), but without any detectable ripple-locked fast transient. The diverging timescales and opposite direction of activity change suggest that different mechanisms govern the ripple-associated activity of MRR neurons.

The observed second-long ripple-correlated activity change prompted us to investigate the ripple-associated activity at a longer timescale in the identified GABAergic group. One whole cell recorded GABAergic cell showed slow hyperpolarization, and 9 out of 39 GABAergic neurons suppressed their activity for 1.50 ± 1.19 s around ripples (Supplementary Figure 4B). Strikingly, the majority of these neurons (n = 6/9) exhibited fast transient ripple-coincident activation embedded in longer-lasting suppression thus, exhibiting contrasting activity change simultaneously on different timescales. We did not detect multiple second-long elevations of activity among GABAergic cells around ripples.