Multiple internal clocks in the mouse hindbrain area that controls daily rhythms

Both manual visual inspection (Fig. 1a–d) and automated image analysis (Fig. 1e, f) revealed that the AP, NTS and the 4^thVep function as three independent circadian oscillators. A spatiotemporal sequence in PER2::LUC expression was evident whereby the initial peak of the AP bioluminescence preceded that of the NTS by ~2 h (p < 0.0001), and the 4^thVep by ~10 h (p < 0.001). Consequently, the rhythm in PER2::LUC in the 4^thVep was in near antiphase to that of the AP (Fig. 1a–c, e, f). The circadian period of these bioluminescence rhythms was similar in the AP and NTS (~24.7 h and ~24.3 h respectively), but significantly shorter in the 4^thVep (~22.9 h; compared with AP, p < 0.05) (Fig. 1d). Unbiased, grid-based whole-image analysis of bioluminescence also identified independent oscillatory clusters corresponding to these anatomical subregions in the DVC (Fig. 1e) and confirmed the inverted relationship between initial phase and period length, with slower oscillators being relatively advanced in phase (Fig. 1f). This unexpected relation suggests that in vivo, coordinated network coupling acts to stabilise phase alignment among these oscillators, whereas ex vivo, this coupling is absent and consequently, phase alignment among oscillators gradually collapses as each oscillator free-runs at its own intrinsic period, a process called detuning²⁷. Simulations reveal that the degree of synchronisation can decay in a quadratic time course when there is no coupling²⁸. However, we found that the AP and NTS ex vivo detuned differently from predicted time courses for complete decoupling. The AP sustained synchrony longer than the NTS (mean R, p < 0.01), and in both structures, the detuning time course revealed that some of the heterogenous coupling remained ex vivo (Supplementary Fig. 3). This suggests that in vivo, the coupling structure among oscillators is complex, with a subset of DVC oscillators preferentially connected.

Damping of the whole area PER2::LUC rhythms in the DVC can arise through reduced clock gene expression in single cells, progressive desynchronisation amongst these single cells (detuning) or a combination of the two. With manual visual inspection, putative single cell oscillators could be reliably discriminated in the AP and NTS (but not in the 4^thVep), allowing individual assessment of their PER2::LUC expression (Fig. 1g). Single cell oscillators desynchronised visibly throughout the recording coincident with the damping of their respective networks (Fig. 1g,i and Supplementary Fig. 4A). Initially, the average phase peak of the oscillating AP cells was CT9.4 ± 0.2 h, while that of NTS cells occurred ~1.5 h later at CT10.91 ± 0.6 h (day 2, p < 0.05). Cells desynchronised at similar rates in both structures, but throughout the recordings, single cells in the AP were more synchronised than those in the NTS (Fig. 1i). Consistent with this, the distribution of phase differences varied between these structures (2-way ANOVA interaction: p < 0.0001) (Fig. 1j). The average period of single oscillating AP cells was slightly longer than that of NTS cells (p < 0.05), but less variable (p < 0.05), which potentially accounts for the reduced synchrony in the latter structure (Supplementary Fig. 4B, C).

Following 7 days in culture, whole area PER2::LUC rhythms in the DVC were significantly dampened and subsequent application of 10 µM forskolin (n = 3) re-initiated whole area and single cell oscillations for up to a further 14 days. This treatment transiently re-synchronised phasing in the AP and NTS (Supplementary Fig. 5A, C, D). However, by the 5th day following forskolin treatment, the initial spatiotemporal sequence seen during baseline recordings of the AP leading the NTS and the 4^thVep had resumed (Supplementary Fig. 5B). This is further evidence that these oscillators can be phase synchronised, but that altered dynamics in coupling seen here ex vivo changes these phase relations. Detailed analysis of the PER2::LUC expression is reported in Supplementary Table 1.

Rhythmic clock gene expression in the DVC is not widely reported and we next assessed the in vivo Per2 and Bmal1 transcript expression in the AP and NTS separately at 6 h intervals over the circadian cycle, beginning at the onset of the circadian day (CT0). Both Per2 and Bmal1 expression in the AP and NTS varied significantly over these time points (AP: p < 0.01 and p < 0.0001; NTS: p < 0.01 and p < 0.05, respectively) (Fig. 1k and Supplementary Tab. 2), with a delayed peak in Per2 in the NTS consistent with the phasing observed in PER2::LUC rhythms. Thus, in the mouse NTS and AP, molecular clock components also vary in expression over 24 h in vivo.

The disconnection of the NTS from the AP also compromises the ability of the single cell NTS oscillators to maintain synchrony. Single cell rhythms in the NTSd desynchronised more quickly than single cells in the NTSc (p < 0.05, two-way ANOVA interaction; Supplementary Fig. 6B, C). Thus, anatomical and functional connections with the AP slow the oscillators in the NTS and promote single cell synchrony in this structure (see Supplementary Table 4 for statistical analysis).

Since neuronal signals from the AP regulate key rhythmic parameters of the NTS oscillator, we next examined the contribution of action potential-dependent communication in the maintenance of rhythmic PER2::LUC expression in the AP and NTS. To this end, we applied tetrodotoxin (TTX, 0.5 µM; a voltage-gated sodium channel blocker) to the culture media after 48 h of baseline recording (n = 5 brainstem slices) and manually assessed PER2::LUC bioluminescence. This visual inspection revealed that at both whole tissue and single cell level, TTX application had little effect in the AP (n = 5 slices; Fig. 4e, f; Supplementary Fig. 7A Supplementary Table 5) with neither damping of oscillations (p = 0.804; Fig. 4g) nor single cell synchrony altered (p = 0.238; Fig. 4h). By contrast, in the NTS, TTX rapidly accelerated the damping of whole tissue rhythms (p < 0.01; Fig. 4i, Supplementary Fig. 7B), and desynchronised single cell oscillations (p < 0.001; Fig. 4j, k, l; Supplementary Fig. 7B, C, Supplementary Table 5). Unbiased grid-based analysis confirmed these assessments (Fig. 4m, n) with the NTS cluster prominently desynchronised in response to TTX, whereas the synchrony index of the AP less obviously reduced (Fig. 4o). The cluster corresponding to the 4^thVep was unaffected by TTX (Fig. 4p). This result was expected, since this region contains non-neuronal ependymal cells that do not use action potential-dependent synaptic communication. Surprisingly, this automated analysis also identified a fourth correlation cluster that approximated the border between the AP and NTS that was not affected by the TTX (Fig. 4n, p). This area contains non-neuronal cells forming the specialised glial barrier between the AP and NTS³⁰ and thus synchrony here was also unaltered by blocking action potential- dependent communication.

These experiments reveal that both AP input as well as action potential-dependent communication shape the parameters of the NTS circadian oscillation. In addition, these findings allude to a TTX-resistant oscillator localised to the region of the glial barrier between the AP and NTS.

NTS neurons are also responsive to chemical cues that promote arousal and ingestive behaviour, including the hypothalamic neuropeptide orexin A (OXA) and the gastrointestinal hormone ghrelin^38–40. Ghrelin application elicited activations both day and night in the NTS MUA, with the amplitude of the response increased at night (p < 0.05, Fig. 6c). Suppression of firing to ghrelin was infrequently observed with no day-night differences in these response characteristics. OXA enhanced MUA from ~70% of NTS recording sites, both day and night, with an increased amplitude in the nocturnal response (p < 0.001, Fig. 6d). Ghrelin signals via the ghrelin receptor (Ghsr) while orexin acts via the HCRTR1 and HCRTR2 receptors, and to investigate whether NTS expression of these receptors varied across the circadian cycle, we examined Ghsr and Hcrtr1-2 expression in the NTS of mice culled at 4 timepoints in constant dark. No significant temporal changes were observed in Ghsr expression in the NTS (p = 0.132, Fig. 6f). However, expression of transcripts for orexin receptors (Hcrtr1 and Hcrtr2) was significantly elevated during the night, with the highest level at CT18 (p < 0.05, Fig. 6g). Statistical analysis of qPCR measurements and drug treatments are presented in Supplementary Tables 2 and 7, respectively. Therefore, NTS neurons exhibit elevated nocturnal responsiveness to metabolic and arousal cues as well as gene expression of some of their cognate receptors. This suggests that one function of circadian processes in the NTS is to enhance its night-time sensitivity and processing of signals associated with metabolic and arousal state.

All mice were kept under standard lighting conditions of 12:12 h light - dark (LD) cycles, unless stated otherwise, with ad libitum availability of food and water. C57BL6J mice were provided by Charles River, Kent UK, all other genotypes (details below) were bred in house by the University of Manchester Biological Services Facility. All procedures and experiments were carried out with approval by the Research Ethics committee of the University of Manchester and in keeping with the UK Animal (Scientific Procedures) Act 1986. Experimental procedures were designed to minimize the number of animals used and their suffering. Night vision goggles (Cobra, UK) were used to prevent animal exposure to light during procedures carried out during the dark phase or constant dark conditions.

Tissue preparation: A total of 21 adult male mPer2Luc knock-in (PER2::LUC) mice¹⁰ between the ages of 10–20 weeks old were used for all bioluminescent imaging experiments. 250 µm thick coronal brain slices −7.20 to −7.26 mm from bregma⁶³ were cut in ice cold Hank’s Balanced Salt Solution (HBSS; Sigma, Germany) supplemented with 1 mg/ml penicillin-streptomycin (Gibco Invitrogen Ltd, UK) and 0.01 M HEPES (Sigma) using a vibroslicer (Camden Instruments, UK). The DVC was dissected using a scalpel and prepared for culture. For the disconnection study, the unilateral NTS was cut out from the complex and further cultured in the same dish. The explant tissue was then transferred to 30 mm Millicell cell culture inserts (Merck, Germany) in glass-coverslip sealed Fluorodish culture dishes (World Precision Instruments Ltd., USA) with sterile culture medium (Dulbecco’s Modified Eagle’s Medium; DMEM, Sigma) supplemented with 0.1 mM luciferin (Promega, USA), B27 (Gibco Invitrogen Ltd, USA), 1 mg/ml penicillin-streptomycin (Gibco Invitrogen Ltd), 10 mM HEPES (Sigma) and 3.5 g/L d-glucose (Sigma).

Data acquisition: Images were taken using the Olympus Luminoview LV200 (Olympus, Japan) fitted with a cooled Hamamatsu ImageEM C900-13 EM-CCD camera and a 20 × 0.4 NA Plan Apo objective (Olympus, Japan) on a heated stage kept at 37 °C. Exposure time was 60 min and gain was kept constant throughout. 10 µM forskolin (Sigma) treatments were performed as a fresh media change. 0.5 µM tetrodotoxin (TTX; Tocris, UK) treatments were performed on day 3 of the recordings at CT2-3, administered as a microdrop to the existing media. Wash out of the TTX was performed as a complete media change at CT2-3 on day 7 Control treatments were made analogously, with a microdrop of water.

Image analysis—manual tracking: Images were primarily analysed in ImageJ, using a region of interest tool to select putative single cells or whole brain areas for assessing relative bioluminescence over time: AP, NTS, and 4^thVep for baseline recordings and TTX treatments and separately NTSc and NTSd (connected and disconnected from the AP, respectively) for the mechanical disconnection protocol. The first 12 h of all recordings were excluded and raw data were subject to a 3 h running average smooth before further analysis. Peaks and troughs of individual bioluminescence traces of whole structures and putative single cells were determined manually. An average of three peak to peak measurements were used to measure period. Damping rate was determined as relative to the amplitude of the peak on day 2. To determine the spatiotemporal amplitude heatmap for the whole DVC, the OriginPro 9.1 (OriginLab, USA) software was used. Rayleigh plots of peak phase and subsequent R values were created using El Temps (University of Barcelona, Spain). For these plots, triangles represent peak bioluminescent phases for individual cells, the arrow indicates the phase vector and its length represents significance with respected to the inner dotted circle of the p = 0.05 significance threshold. Variance is represented by the black box around arrow heads. Cross-correlograms indicating the differences in phasing were calculated in Statistica (StatSoft, USA).

Grid analysis: Time-lapse images were first treated with noise removal using Remove Outliers function in ImageJ (NIH, USA). The images were then stabilised using a set of ImageJ Plug-ins (Image Stabilizer and StackReg). From stabilised time-lapse images, spatiotemporal dynamics can be characterised by treating PER2::LUC oscillations in single-cell sized grids (5 × 5 pixels), analogous to voxels in brain imaging²⁸. To maximize the chance that a grid represents a single-cell at each time point, cropped 410 × 210-pixel images were convoluted with a 5 × 5 homogeneous kernel. The images were subsequently reduced to matrices of 82 × 42 grids. Actively bioluminescent grids were selected in the order of high time average values (n = 700–900 grids).

Oscillator phase and correlation analysis: Each grid had oscillating temporal profile with varying amplitudes, and its phase was extracted by delay embedding (Takens’ theorem). The period of its grid was estimated by the inverse of dominant frequency from Fast Fourier Transform (FFT) for the first 2 days in culture. This method tended to weigh higher on oscillations prior to damping. Amplitude-normalised oscillations were obtained by taking sine of extracted phases. Pair-wise correlations of the normalised oscillations were calculated against time-series of all grids, to make a correlation matrix and the partial correlation matrix was used to identify clusters²⁸. This revealed spatial heterogeneity of oscillatory activities that matched with known anatomical areas such as AP and NTS.

Detuning analysis: The circadian oscillators in AP, NTS, and adjacent regions are initially well-synchronised and start to detune (desynchronise) over time in culture. When the variance of the initial frequency distribution is known, the detuning dynamics can be mathematically predicted for the uncoupled case (decoupling; coupling constant K = 0 in the Kuramoto model). It can be shown to follow a quadratic time course proportional to frequency variance²⁸.\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Detuning }}\sim{\it{e}}^{{\mathrm{ - }}{\it{Dt}}^{\mathrm{2}}/2}$$\end{document}Detuning~e-Dt2/2

where D indicates frequency variance, which was estimated from initial 2-day’s FFT, as described in the previous subsection.

Statistics: Statistical analyses were performed using Prism 7 (GraphPad Software, USA). To compare periods, standard deviation of periods (SD period) and time of peaks, paired t test and repeated measures (RM) one-way ANOVA followed by Tukey’s multiple comparison were used. To assess the differences in values measured in the 3–4 subsequent days to asses damping and synchronisation (times of peaks, SD of single cell lags, Rayleigh R), RM two-way ANOVA followed by Sidak’s multiple comparison was used, with the interaction value used to evaluate the difference in rate of the parameter change. The interaction in RM two-way ANOVA was also used to compare the shape of frequency histograms of single cell lags to the whole structure peak. Linear regression model was used to measure the desynchronisation rates. For detuning analysis, we used Welch’s t test and paired t test using Mathematica (Wolfram Research, IL). All data was presented as mean ± standard error of mean (SEM). Both manual analysis and grid analysis yielded similar findings.

Tissue preparation: A total of 19 male C57BL6J mice, aged 10 weeks old, were placed into constant darkness for 36–48 h before being culled at four time points across the circadian cycle: CT0, CT6, CT12, CT18 (n = 4–5). Brains were excised and flash frozen in dry ice before being stored at −80 °C for up to 4 weeks. Sections of 20 µm were cut using a cryostat (Leica CM3050 S), onto PEN-membrane slides (Leica Biosystems, Germany) and then stored at -80 °C for up to 2 weeks. For laser-capture microdissection, slides were first defrosted and then stained with 1% cresyl violet (Sigma) in 70% ethanol before regions of interest were extracted on laser-capture microscope (LCM) system (Leica DM6000 B) and stored in lysis buffer (Promega) prior to RNA extraction (see below). LCM performed on the cresyl violet-stained tissue enabled precise extraction of the desired areas (i.e. those which demonstrated the strongest bioluminescent signal measured in the PER2::LUC mice): (1) the AP with the surrounding glial border and (2) the bilateral medial NTS adjacent to the AP, excluding the DMV (Supplementary Fig. 9).

RNA extraction for RT-qPCR: The dissected tissue was then immediately processed for RNA extraction using the ReliaPrep RNA Tissue Miniprep System (Promega, USA) and subsequently stored at −80°C. Reverse-transcription was performed using the High-Capacity RNA-to-cDNA Kit (Applied Biosystems, USA) after which quantitative RT-PCR was performed using Power SYBR Green Master Mix and measured using a StepOnePlus Real Time PCR system (Life Technologies, USA). Transcripts were amplified by QuantiTect primer assay (Qiagen, Germany) for all genes (Cckbr, Cldn5, Ghsr, Hcrtr1, Hcrtr2, Tbp < housekeeping gene> and Tjp1) except the following primers for Per2 (5′–3′: forward GCCTTCAGACTCATGATGACAGA, reverse TTTGTGTGCCTCAGCTTTGG) and Bmal1 (5′–3′: forward CCAAGAAAGTATGGACACAGACAAA, reverse GCATTCTTGATCCTTCCTTGGT).

Data analysis and statistics: The ΔΔCT method was used for data analysis with Tbp as the house keeping gene and CT0 values for relative target gene expression. One-way ANOVA and Kruskal-Wallis tests were performed using Prism 7 (GraphPad Software) for statistical analysis. Data was presented as mean ± SEM.

Tissue preparation: In total, 39 male adult C57BL6J mice were culled at ZT0 (for long-term and acute early day recordings) or ZT11 (for acute early night recordings). The brain was carefully excised and immersed in the pre-oxygenated with carbogen (95% oxygen, 5% CO₂), ice-cold incubation artificial cerebrospinal fluid (ACSF) composed of (in mM): NaCl 95, KCl 1.8, KH₂PO₄ 1.2, CaCl₂ 0.5, MgSO₄ 7, NaHCO₃ 26, glucose 15, sucrose 50 and Phenol Red 0.005 mg/l. The brainstem was then separated and sliced using a vibroslicer (Camden Instruments). 250 µm (for acute set up) or 300 µm thick (for long-term recordings) coronal brain slices (at the corresponding stereotactic coordinates to those prepared from PER2::LUC mice), were cut and incubated in the cold incubation ACSF constantly bubbled with carbogen. After 30 min, slices were transferred to the oxygenated pre-incubation chamber filled with recording ACSF warmed up to 32°C, composed of (in mM): NaCl 127, KCl 1.8, KH₂PO₄ 1.2, CaCl₂ 2.4, MgSO₄ 1.3, NaHCO₃ 26, glucose 5, sucrose 10 and Phenol Red 0.005 mg/l.

Recording and data acquisition: Slices containing the DVC were then transferred to the recording wells (at projected ZT1 for the early day and long-term recordings, or ZT13 for early night recordings) of the MEA2100-System (Multichannel Systems GmbH, Germany) and placed onto the 6 × 10 perforated multi-electrode arrays (MEAs; 60pMEA100/30iR-Ti, Multichannel Systems). Slices were perfused with fresh recording ACSF, preheated to 32 °C with a perfusion cannula (Multichannel Systems), throughout the whole experiment. The tissue was allowed to settle for 1 h following set up before starting recording. For long term recordings that lasted more than 24 h, suction was set to minimum and ACSF was enriched with 1% Penicillin-Streptomycin (Sigma) to maximize the survival of the tissue. In order to verify the tissue condition after over 24 hours of recording, 100 µM NMDA (Tocris Bioscience, UK) was applied to provide evidence of tissue responsiveness. After short-term recordings, 1 µM TTX (Tocris) was applied to verify that the registered signal comprised of TTX-sensitive action potentials and to assess the signal-to-noise ratio.

The raw signal was sampled at 25 kHz and acquired using MC Rack 4.6.2 Software (Multichannel Systems). For spike detection, MC DataTool 2.6.15 (MultiChannel Systems) and MC Rack were used. First, data was high pass filtered in forward and reverse directions (300 Hz). Then, spikes were classified as the signal exceeding the threshold of -17.5 µV (twice as high as determined to be a noise level after TTX application). Spontaneous multi-unit activity (sMUA) was further analysed only from recording spots that were classified to be localised in the AP or NTS.

Drugs: All drugs: orexin A (OXA, 200 nM; Tocris), ghrelin (200 nM, Tocris) and cholecystokinin (CCK; 1 µM, Tocris) were stored as 100x concentrated and were freshly diluted in recording ACSF prior to the application. To expose studied structures to elevated glucose, 10 mM of sucrose in the recording ACSF was exchanged to equimolar concentration of glucose (final concentration at 15 mM). All drugs were delivered by bath perfusion.

Stimulation: Stimulations were performed by switching the mode of four chosen electrodes from recording to stimulation mode. Stimulation electrodes were localised in the middle of the AP, which was assessed visually and electrophysiologically as recording locations in the AP are characterised by distinctly higher MUA than those in the NTS. Thus the stimulation area was of a triangular shape, with three electrodes in one row and one in the row below (Fig. 3a). The stimulation protocol was built from trains of ten negative voltage pulses (amplitude: 200 mV, duration: 5 ms, inter-stimulus interval: 5 s). Each train was repeated three times every 2 min (a 75 s stimulation break; Fig. 3b).

Data analysis and statistics: For short-term recordings, baseline sMUA was averaged from 1800 s intervals in NeuroExplorer 5 (Nex Technologies, USA). Recording locations were classified as “active” if the sMUA exceeded 0.005 Hz. The proportion of “silent” sites in each structure between early day and early night was compared using Fisher’s exact test and the sMUA frequency - with Mann-Whitney tests. For the long-term studies, daily sMUA profiles were calculated in 600 s bins. The time of peak activity was manually assessed for sMUAs for each recording electrode localised in the AP and NTS and further analysed in El Temps for phase clustering before comparison with unpaired t test. Activity heatmaps were plotted in OriginPro 9.1 (OriginLab) from averaged baseline sMUA values.

Drug responses were analysed on 1 s binned MUA histograms in Spike2 8.11 (Cambridge Electronic Design Ltd., UK) and further in MatLab R2013a (MathWorks, USA) with custom-written scripts. The change in MUA was classified as a response if it exceeded 3 standard deviations (SDs) from the baseline. The maximal amplitudes of the responses were compared between day and night using unpaired t tests or Mann-Whitney tests, and the relative proportion of responsive spots with Fisher’s exact tests. Responses were relicated across brain slices prepared from different animals.

Responses to electrical stimulation were prepared for analysis in Spike2 8.11 (Cambridge Electronic Design Ltd., UK). First, waveform data was transformed with the use of kernel function to generate Gaussian probabilities of spike generation over time (spike density function). Next, modified peri-stimulus time histograms (PSTHs) were generated, based on the sum of spike density functions around all 30 stimuli. Further analyses were performed with custom-made MatLab script (MatLab R2013a, MathWorks, USA). Each recording location in the NTS was classified as responsive to the AP stimulation, if the change in spike density function at the PSTH exceeded one standard deviation (SD) from the baseline mean. The amplitudes of responses were compared using Mann-Whitney test and the excitation-to-inhibition ratio between day and night recordings were analysed using Fisher’s exact test. All responses localised immediately adjacent and subjacent to the stimulation electrodes were excluded from further analysis, due to the possibility of a passive response. Responses were relicated across brain slices prepared from different animals.

Statistical analyses were performed in Prisms 7 (GraphPad) and all data were presented as mean ± SEM.

Tissue preparation: A total of 20 male C57BL6J mice, aged 10 weeks old, 11 of which were placed in constant darkness (DD) for 36–48 h prior to cull. 4 groups of mice were subjected to transcardial perfusions: n = 5 at ZT1, n = 4 at ZT13, n = 5 at CT1 and n = 6 at CT13. After terminal injection of pentobarbital (80 mg/kg), transcardial perfusions were performed with ice-cold phosphate-buffered saline (PBS) (flow rate: 7 ml/min, volume: 35 ml) followed by a 1% Evans Blue dye (EB, Sigma) and 4% PFA in PBS cocktail (flow rate: 7 ml/min, volume: 50 ml). Brains were subsequently excised and immersed in 4% PFA in PBS to post-fix overnight at 4 °C, after which they were immersed in 30% w/v sucrose solution for 2-3 days. Fixed hindbrains were separated before mounting on a freezing sledge microtome (Bright Instruments, UK) and cut into 35 µm thick to collect sensory circumventricular organs: the vascular organ of the lamina terminalis (OVLT), the subfornical organ (SFO) and the DVC.

Immunohistochemical staining: Slices containing the ROIs were rinsed off in PBS and permeabilised with 0.1% Triton X-100 (Sigma) for 20 min at room temperature (RT). Following washing in PBS, sections were blocked in 5% normal donkey serum (NDS; Sigma) with 0.05% Triton X100 for the next 30 min at RT. Then, slices were transferred to a PBS solution containing 0.5% NDS, 0.05% Triton X100 and the primary antisera: chicken anti-vimentin (1:4000, Abcam, Cambridge, UK), and mouse anti-GFAP (1:400, sigma) for 48 h at 4°C. Sections were subsequently washed in PBS and placed in the PBS solution of anti-chicken Alexa488 (1:800, Jackson ImmunoResearch, West Groove, USA) and anti-mouse Cy3 (1:800, Jackson ImmunoResearch) conjugated secondary antibodies raised in donkey. After 24 h at 4 °C, immunostained slices were rinsed off in fresh PBS and mounted on gelatine-coated glass slides with VectaShield medium (Vector Laboratories, USA). Details on antibodies used in this protocol are presented in Supplementary Table 9.

Data collection, analysis and statistics: Three brainstem sections (caudal AP, intermediate AP, rostral AP), one mid-level SFO and one mid-level OVLT section per mouse were imaged on a confocal microscope (Leica SP5 upright) under the 20x objective. Florescence emitted by EB was detected at 650 nm. Z-stacks were collected with 3 µm steps and analysed as maximal intensity projections. Further analysis was performed in ImageJ. As no EB signal was observed outside SFO or OVLT in any of the experimental groups (Supplementary Fig. 10), detailed image analysis was performed for brainstem sections only. Maximal intensity projections were created for each stack to analyse the intensity of EB-derived signal in the NTS. Regions of interest (ROI) were anatomically distinguished using vimentin-immunoreactivity (ir) to delineate the border between the AP and NTS and GFAP-ir was used to establish ventral border of the DMV and the lateral border of the NTS. Subsequently, the area of NTS (whole) was divided into two ROIs: medial and lateral, by halving the ventral and dorsal border of the NTS with a linear vector (Fig. 5a, top panel). The intensity of each ROI (mean grey value) was reduced by the background value, collected with the 40 × 40-pixel oval at the area ventral to DVC (which lacked EB staining). The intensity values for 3 ROIs were compared between corresponding slices in ZT/CT1 vs ZT/CT13, using unpaired t tests or Mann-Whitney tests in Prism 7 (GraphPad). Data were presented as mean ± SEM. Results obtained under LD were replicated under DD conditions.

Further information on research design is available in thelinked to this article. Nature Research Reporting Summary

Reporting Summary Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Peer Review File