5-HT levels exhibit ultraslow oscillations during NREM and WAKE.

A. Histology and experimental protocol. Left: expression of GRAB5-HT3.0 sensor (in green) in dorsal CA1 with optic fiber track above. Right: methodology for dual implantation surgeries. AAV9-hSyn-5HT3.0 was first injected into the right dorsal CA1. In the same surgery, an optic fiber was implanted above the injection site. After three weeks of viral expression, a silicon probe was implanted above the left dorsal CA1. Simultaneous recording of the GRAB5-HT3.0 sensor activity (fiber photometry) and electrophysiology was performed. B-E. Example dual fiber photometry-electrophysiology recording with times shown in E. B. Labeled sleep states resulting from automated sleep-scoring above an intracranial EMG trace. C. Spectrogram (Stockwell transform) showing normalized power of a hippocampal LFP channel during awake and sleep states. D. Z-scored 5-HT trace. E. Spectrogram (Stockwell transform) of the 5-HT trace shown in D. F. Left: Mean 5-HT level by state, across all experiments (total n=6 mice, 12 recording sessions of 1.5-3 hours). Right: p-values from a multiple comparisons test applied after fitting a Bayesian GLMM to the data. G. Pie chart showing proportion of time spent in different behavioral states, averaged across all experiments. H. Top: examples of ultraslow 5-HT oscillations in NREM and WAKE. Bottom: Power spectrum of 5-HT signals in WAKE vs. NREM sleep. Frequencies were separated into two groups (> 0.04 Hz. and <0.04 Hz.), and a GLMM was fitted to the data. P-values shown were derived from a post-hoc multiple comparisons test on the fitted model. Results indicated higher power in the lower half of frequencies (<0.04 Hz.) for both states (p < 0.0001), as well as a significant difference in the lower frequencies between NREM and WAKE (p<0.0001), with a NREM/WAKE ratio of 1.21 +/- 0.028. I. Control fluoxetine and saline injection experiments. A significant difference between the post-injection period of saline-injected and fluoxetine-injected animals (shaded in red) was observed (Wilcoxon ranked-sum test, p<0.001, n= 3 mice).

Ripples occur time-locked to ultraslow 5-HT oscillations.

A. Schematic showing the convolutional neural network used for ripple detection. 8-channel x 400 ms-LFP chunks were used as input. The bottom four channels (cyan) were taken from the dorsal CA1 and contained ripples, and the top four channels (magenta) were chosen from a non-adjacent part of the neocortex above the dorsal CA1. The model consisted of four convolution blocks (‘Conv2d’), each block comprising a 2D convolutional layer, a ReLU activation function, and batch normalization. Two dense layers with dropout and batch normalization (‘Dense’) followed and produced the final output, a 400 ms vector with values between 0-1, indicating the probability of a ripple occurring during the course of the input chunk. B. Example model output given the four LFP chunk inputs shown. First row: true positives. Second row: fast oscillations and movement artifacts not detected as ripples by the model. C. Spectrogram from a ripple detected by the model, 0-1 normalized. D. Characteristics of detected ripples. Ripples from all experiments were included, and probability distributions are shown. Top left: distribution of duration. Top right: distribution of z-score normalized ripple power. Bottom: distribution of ripple frequency. Ripple duration and normalized ripple power follow a log-normal distribution (duration: X2 (df = 7, N = 49,458) = 1.398e+03, p < .0001, normalized ripple power: X2 (df = 7, N = 49,458) = 422.1862, p < .0001). E. Example 5-HT trace and computed power in the ripple band (120-250 Hz). F. Same example 5-HT trace and individual detected ripples. G. Example of ripple cluster extraction. Ripple clusters were defined as having a minimum of 10 ripple events and an inter-ripple interval of less than 3 seconds. Note the few ripples occurring during the rising phase of 5-HT ultraslow oscillations in F are excluded from extracted ripple clusters in G. From these ripple clusters, the first (orange) and last (black) ripples in a cluster were extracted. H1-2. Ripple-triggered 5-HT in NREM (H1) and WAKE (H2) states. The first rows of H1 and H2 show all 50 s 5-HT segments centered around the ripple peak for different combinations of ripples (columns). In the first column, all ripples in the given state were included. The second and third columns used only the first or last ripple in extracted ripple clusters, respectively. The second rows of H1 and H2. show the mean ripple-triggered 5-HT traces (blue) and randomly shifted traces (orange) for each group of ripples. The orange traces were obtained by randomly shifting the ripple times for each condition and averaging the resulting 5-HT 50 s segments centered around those shifted times.

Ripple occurrence and power vary by the phase of ultraslow 5-HT oscillations.

A. Schematic showing one period of a slow 5-HT oscillation. The rising phase of the oscillation occurs from −180° to 0°, and the falling phase occurs from 0° to 180°. B. Mean z-scored inter-ripple interval (IRI) by 5-HT phase angle during NREM (left) and WAKE (right). C. Mean rising phase IRI - mean falling phase IRI, plotted by session and mouse level in WAKE (left) and NREM (right). Red point with error bar indicates predicted difference and confidence interval after fitting a GLMM to the data. P-values shown were derived from a post-hoc multiple comparisons test on the fitted model. (n=6 mice, 12 sessions). D. Box plots showing IRI in rising vs. falling phase in NREM (left) and WAKE (right). Significance levels shown reflect results of statistical analysis in C. E. Example 5-HT trace (top) and corresponding 5-HT phase angles and ripples (bottom) for NREM (left) and WAKE (right). The peak of the ultraslow oscillation (0°) is indicated by the dashed purple line. F1. Schematic polar plot showing one period for a slow 5-HT oscillation. The falling phase of the oscillation occurs from 0° to 180°, and the rising phase occurs from −180° to 0°. F2. Phase of all NREM ripples relative to the ultraslow 5-HT oscillation. A significant (p<0.0001) bias toward the falling phase was found after fitting a GLMM to the data with a binomial link function. F3. Mean phase vector of NREM and WAKE ripples. A significant (p<0.0001) bias toward the falling phase was found after fitting a GLMM to the data with a binomial link function. F4. Phase of all WAKE ripples relative to the ultraslow 5-HT oscillation. G. Z-scored ripple power by 5-HT phase angle during NREM (left) and WAKE (right). Red vertical dashed lines delineate analyzed phase segments: ‘center’ (−90° to 90°) vs. ‘side’(−180° to −90° and 90° to 180°). Representative ripples from each phase grouping are shown above. H. Mean center phase ripple power - mean side phase ripple power, plotted by session and mouse level in WAKE (left) and NREM (right). Red point with error bar indicates predicted difference and confidence interval after fitting a GLMM to the data. P-values shown were derived from a post-hoc multiple comparisons test on the fitted model.

EMG and MAs vary by the phase of ultraslow 5-HT oscillations.

A.-D2. Relationship between microarousal (MA) occurrence and the phase of slow 5-HT oscillations. A. Example trace showing 5-HT, EMG, and MAs during a NREM bout. B. Example trace showing extracted 5-HT phase angle and MAs. C. MA occurrence according to 5-HT phase angle. A significant (p<0.0001) bias toward the falling phase was found after fitting a GLMM to the data with a binomial link function. D1. MA-triggered 5-HT across all MA events. D2. Mean MA-triggered 5-HT trace (blue) plotted with mean of randomly shifted 5-HT trace (orange). The orange trace was derived by randomly shifting all MA times and averaging the resulting 5-HT segments around those shifted times. E.-G. Relationship between the EMG signal and phase of slow 5-HT oscillations. E. Example traces showing extracted 5-HT phase angle and the EMG signal during NREM (left, blue) and WAKE (right, black) states. F. Mean z-scored EMG signal by 5-HT phase angle during NREM and WAKE states. G. Mean rising phase EMG - mean falling phase EMG, plotted by session and mouse level. Red point with error bar indicates predicted difference and confidence interval after fitting a GLMM to the data. P-values shown were derived from a post-hoc multiple comparisons test on the fitted model.

Coherence varies by the phase of ultraslow 5-HT oscillations.

A. Schematic showing representative hippocampal and cortical traces used for coherence calculations. B. Mean z-scored hippocampal-cortical coherence by frequency for NREM (left) and WAKE (right). C. Mean coherence by 5-HT phase angle for delta (1-5 Hz.), theta (6-10 Hz.), slow gamma (30-60 Hz.), fast gamma (60-100Hz.) and high frequency oscillation (HFO, 100-150 Hz.) bands in NREM (left column) and WAKE (right column). D. Mean rising phase coherence - mean falling phase coherence, plotted by session and mouse level for different frequency bands (rows) and states (columns). Red point with error bar indicates predicted difference and confidence interval after fitting a GLMM to the data. P-values shown were derived from a post-hoc multiple comparisons test on the fitted model.

Relationship between ripple incidence and 5-HT levels depends on time-scale.

A. Ripple incidence by behavioral state shows an inverted-U dose response relationship, with a peak at intermediate 5-HT levels (see Figure 1F). B. Within states, ripple incidence depends on the phase of the ultraslow 5-HT oscillation. At the same absolute 5-HT level (e.g. green dots), therefore, different ripple incidences are observed.

Serotonin modulates ripple incidence in vitro.

A. Spectrogram of an example in vitro ripple, 0-1 normalized. B. Ripple detection in vitro. Given the absence of non-ripple fast oscillations and movement noise in vitro, ripples were detected in the standard way; namely, the raw LFP signal was bandpass filtered in the ripple range (120-300 Hz.), the amplitude of activity in this frequency band was extracted with the Hilbert transform, and finally ripples were detected as peaks greater than a threshold of two standard deviations and a minimum duration of 20 ms. C. Example traces from the baseline (blue) and low concentration 5-HT wash-in period (red) of an example recording. D. Example traces from the baseline (blue) and high concentration 5-HT wash-in period (red) of an example recording. E. Left: Ripple incidence over time in an example wash-in experiment. 5 µM 5-HT was added to the circulating ACSF after a 10-minute baseline. Right: Mean 5 µM 5-HT wash-in ripple incidence - mean baseline ripple incidence, plotted by mouse. Red point with error bar indicates predicted difference and confidence interval after fitting a GLMM to the data. P-value shown was derived from a post-hoc multiple comparisons test on the fitted model. F. Left: Ripple incidence over time in an example wash-in experiment. 50 µM 5-HT was added to the circulating ACSF after a 10-minute baseline. Right: Mean 50 µM 5-HT wash-in ripple incidence - mean baseline ripple incidence, plotted by mouse (n= 20 slices from 5 mice). Red point with error bar indicates predicted difference and confidence interval after fitting a GLMM to the data. P-value shown was derived from a post-hoc multiple comparisons test on the fitted model.

Waveform of ultraslow 5-HT oscillation is asymmetric.

A. Mean 5-HT waveforms for one period of the ultraslow 5-HT oscillation during WAKE (black) and NREM (blue). B. Mean rising phase 5-HT slope - mean falling phase 5-HT slope, plotted by session and mouse for WAKE (left) and NREM (right) data. Red point with error bar indicates predicted difference and confidence interval after fitting a GLMM to the data. P-values shown were derived from a post- hoc multiple comparisons test on the fitted model. C. Example 5-HT trace showing one period of the ultraslow oscillation with corresponding phase angles below to illustrate the calculation of the length ratio (rising phase length / falling phase length). D. Length ratio per session in NREM (top) and WAKE (bottom). Line at 1 indicates equal rising and falling length. Significant difference from equal rising and falling lengths (ratio = 1) was tested after fitting a GLMM to the log-transformed data (NREM: p <0.0001, WAKE: p = 0.003).

Ripples couple most strongly to ultraslow (0.01 - 0.06 Hz.) oscillations of serotonin.

Individual lines represent session-level mean phase vectors of ripples relative to different slow oscillations of 5-HT (red: 0.01-0.06 Hz., black: 0.001-0.01 Hz., blue: 0.5-1 Hz.) for WAKE (left) and NREM (right). Thicker lines with arrows represent the grand mean vectors (mean of session-level means) for each frequency band.

Ripple frequency, but not duration, is shaped by ultraslow 5-HT oscillations.

A. Mean ripple frequency by phase of ultraslow 5-HT oscillation in NREM (top) and WAKE (bottom). B. Mean falling phase ripple frequency - mean rising phase ripple frequency, plotted by session and mouse. Red point with error bar indicates predicted difference and confidence interval after fitting a GLMM to the data. P-values shown were derived from a post-hoc multiple comparisons test on the fitted model. C. Mean ripple duration by phase of ultraslow 5-HT oscillation in NREM (top) and WAKE (bottom). D. Mean falling phase ripple duration - mean rising phase ripple duration, plotted by session and mouse. Red point with error bar indicates predicted difference and confidence interval after fitting a GLMM to the data. P-values shown were derived from a post-hoc multiple comparisons test on the fitted model.

Power in different frequency bands is timed by ultraslow 5-HT oscillations.

A-D. Left columns: Mean power by phase of ultraslow 5-HT oscillation in NREM (top) and WAKE (bottom) for A. low gamma (30 - 55 Hz.), B. high gamma (60-90 Hz.), C. theta (8-12 Hz.), and D. cortical high frequency oscillation (HFO, 100-150 Hz.) A-D. Right columns: Mean falling phase low gamma power - mean rising phase low gamma power, plotted by session and mouse during NREM (top) and WAKE (bottom). Red point with error bar indicates predicted difference and confidence interval after fitting a GLMM to the data. P-values shown were derived from a post-hoc multiple comparisons test on the fitted model.

Units show preference for different phases of ultraslow serotonin oscillations.

A. Example spiking activity (green) of a neuron relative to the ultraslow serotonin oscillation with a preference for the rising phase. B. Example spiking activity (red) of a neuron relative to the ultraslow serotonin oscillation with a preference for the falling phase. C. Mean z-scored firing rates of pyramidal cells and interneurons by ultraslow serotonin oscillation phase in NREM. D. Same data as in C in the WAKE state. E. Percentage of units showing falling phase preference (red), rising phase preference (green), and no preference (blue) as per results of a circular V test (p < 0.05) in NREM. F. Same data as E in the WAKE state. (n=3 mice, 1 session per mouse, with 64, 59 and 30 extracted units, respectively)

Ripples missed by model do not affect ripple distribution along ultraslow 5-ht oscillations.

A. Example trace showing a correctly detected ripple (green) and a ripple missed by the model (red). B. Ripple power of correctly detected ripples (true positives, green) and missed ripples (false negatives, red). Model performance was evaluated manually in three mice with a total of 772 true positives, and 52 false negatives. C. Top: Recall (true positives / true positives + false negatives) by phase of ultraslow 5-HT oscillation. Bottom: Histogram showing true positives by 5-HT ultraslow phase angle (green) and true ripples (true positives + false negatives, blue). No statistical difference was found between the two distributions (Mardia-Watson-Wheeler test, p = 0.9375).