The circadian clock in the choroid plexus drives rhythms in multiple cellular processes under the control of the suprachiasmatic nucleus

Animal experimentation was approved by the Animal Care and Use Committee of the Institute of Physiology (33/2019) and by Taipei Medical University (LAC-2020-0576). It was performed in accordance with the Animal Protection Law of the Czech Republic as well as the European Community Council directives 2010/63/EU, and the Animal Protection Act of Taiwan. All efforts were made to reduce the suffering of the animals. Adult male C57BL/6J mice (Charles River, Germany) (4- to 6-month-old) and knock-in mPer2^Luc male mice [33] on C57BL/6J background (strain B6.129S6-Per2^tm1Jt/J, JAX, USA; a colony maintained at the Institute of Physiology, the Czech Academy of Sciences) (5- to 8-month-old) were housed individually in a temperature-controlled facility at 23±2 °C with free access to food and water. The light/dark cycle was maintained with 12 h of light and 12 h of darkness (LD12:12); lights were turned on and off at 06:00 and 18:00, respectively. Prior to sampling, C57BL/6J and mPer2^Luc mice were kept in constant darkness. Tissue-specific knockout FoxJ1-ERT2-Cre^+/− (JAX stock #027012) [34] and Bmal1^fl/fl mice (JAX stock #007668) [35], both on C57BL/6J background, were crossbred at the Taipei Medical University Laboratory Animal Center to produce FoxJ1-ERT2-Cre^+/−; Bmal1^fl/fl mice. Tamoxifen was provided ad libitum through the diet (tamoxifen citrate 0.4 g/kg; Envigo TD.130,860) to 8-month-old male FoxJ1-ERT2-Cre^+/−; Bmal1^fl/fl mice maintained under 12:12 LD cycle (07:00 light-on and 19:00 light-off) and controlled temperature (21.4 ± 0.2ºC) and humidity (59 ± 5%) conditions. The tamoxifen-induced flox recombination was tested in liver samples through qPCR (data not shown).

C57BL/6J (for RNAseq) and mPer2^Luc (for circadian bioluminescence microscopy) mice were divided into two groups: the control group (Ctrl) was subjected to sham operation, and the other group was subjected to bilateral electrolytic SCN lesions (SCNx, Fig. 1A). The surgical ablation of the SCN was performed as previously described [36]. The mice were anesthetized with isoflurane inhalation (Isoflurin, Vetpharma Animal Health, Spain) and mounted in a stereotactic instrument (David Kopf Instruments, USA). The eyes were treated with the ointment Ophthalmo-Septonex (Zentiva, Czechia). The electrode tips (0.2 mm diameter) were placed bilaterally into the SCN through a drilled hole in the skull (coordinates: 0.1 mm rostral to bregma; ± 0.35 mm to midline; 5.55 mm below brain surface, and 0.1 mm caudal to bregma; ± 0.35 mm to midline; 5.65 mm below brain surface), and 0.6 mA current pulse was delivered for 6 s (53,500 Lesion Making Device, Ugo Basile, Italy). The parameters were determined empirically to achieve complete SCN lesion and avoid damage of surrounding tissue. Mice recovered in their home cages and received analgesic (NUROFEN, UK) immediately after the surgery (applied orally) and then in drinking water (2.5 ml per 250 ml) during the next day. The success of the SCN lesions (SCNx) was carefully checked (1) behaviorally by monitoring the locomotor activity in constant darkness for at least one week (Fig. 1B) and through periodogram analysis (Fig. 1C and Fig. S1), and (2) histologically postmortem for the absence of the SCN throughout the rostrocaudal brain sections of the anterior hypothalamus (Fig. 1D). The mice with detectable rhythm in locomotor activity in constant darkness and/or those in which histological examination of the hypothalamic sections revealed the presence of the SCN or its remaining parts were excluded from the study. The Ctrl group underwent sham surgery using the same procedure, including anesthesia and insertion of electrodes, but without the application of electric current. None of the sham-operated mice exhibited a disruption in circadian locomotor rhythmicity under constant darkness.

The locomotor activity of mice was monitored throughout the experiment individually in cages equipped with infrared movement detectors positioned centrally above the top of each cage, and the activity was detected using a circadian activity monitoring system (Dr. H.M. Cooper, INSERM, France) as required. The activity was recorded every minute, and double-plotted actograms and Chi-square periodograms [37] were generated to visualize the data with Chi-square statistics set at P < 0.0001; significant periodogram peaks within the circadian range of SCNx mice were below 10% of an average Ctrl peak (Fig. S1). The parameters of circadian rhythmicity were analyzed using the ClockLab toolbox (Actimetrics, USA) and Fiji ImageJ plugin ActogramJ [38].

After exposure to constant darkness to verify absence of locomotor activity rhythms as a marker of SCN lesion success, the C57BL/6J mice of Ctrl and SCNx groups (n = 12 per group) were returned to the original LD12:12 regime for one week to synchronize with the LD cycle. Subsequently, the lights were not switched on in the morning (assigned as circadian time, CT0), and mice remained in constant darkness for the following two days during which they were sacrificed every 4 h (from CT0 to CT44, 12 time points). ChP from the 4th ventricle was removed by pincers and immediately immersed into RNAlater. FoxJ1-ERT2-Cre^+/-; Bmal1^fl/fl (KO) mice maintained in LD12:12 were sacrificed at two time points; 7 h after the lights on for the day time point, i.e., Zeitgeber time (ZT)7 and 6 h after the lights off for the night time point, i.e., ZT18, at Taipei Medical University, Taipei, Taiwan. Whole brains were removed, frozen on dry ice, embedded in OCT compound (Sakura Finetek), and then cold-shipped to Prague. Upon arrival, they were sectioned on cryocut and ChP from the 4th ventricle was sampled by laser-capture microdissection.

Total RNA from individual samples of ChP from mice of Ctrl, SCNx and KO groups was purified by RNeasy Micro Kit (Qiagen, Germany) following the manufacturer’s instructions, treated with On-Column DNAse I digestion set (Sigma) for 15 min at 37 °C, and then quantified by spectrophotometry (NanoDrop, ThermoFisher, USA) and fluorimetry (Qubit, ThermoFisher, concentration range 14–49 ng/µl). RNA integrity was tested by capillary electrophoresis (Agilent, USA) with average RIN = 9.1 (7.3 for frozen KO samples). Samples of purified total RNA (n = 24 for Ctrl, SCNx; n = 2 for KO separately) were processed at the Functional Genomics and Bioinformatics Service Laboratory of the Institute of Molecular Genetics (Prague, Czechia) as follows: rRNA depletion, polyA RNA enrichment, and library preparation using KAPA mRNA HyperPrep Kit (Roche, Switzerland) or Smarter Stranded totRNA-seq low input kit (Takara, Japan; for KO samples); sequencing using Illumina (USA) NextSeq 2000 P3 (50 cycles, SE). Bioinformatics was performed using the nf-core/rnaseq (v3.10.1, 10.5281/zenodo.1400710) workflow pipeline. Reads were mapped to the latest Genome Reference Consortium Mouse Build 39 (Mm_39_104_nf310); all samples passed quality control (MultiQC v1.13); read counts and transcripts per million (TPM) were calculated by Salmon and Stringtie. Only genes with average TPM ≥ 1 / sample (i.e., the sum of 12 TPM values in 12 samples in each group TPM_sum ≥ 12; referred to as filtered background) were analyzed further. Differential expression between all Ctrl and SCNx samples was analyzed by DESeq2 using RNAlysis 3.8 GUI [39].

To identify circadian cycling genes with medium-to-high amplitude, three different algorithms were used – eJTK [40–42], ARSER [43] and BIO_CYCLE [44, 45]. After examining the dependence of number of cycling genes on Benjamini-Hochberg (BH) false discovery rate (FDR)-adjusted score (FDR corrected empirical P value in case of eJTK, FDR adjusted Q value in case of ARSER and BIO_CYCLE) of all three algorithms, the rhythmicity threshold was set at below 0.4. To detect phase of the cycling genes, BIO_CYCLE variable LAG was used; to predict relative amplitude and avoid its masking by highly expressed medium amplitude genes, minimum TPM values of each gene were set to 1 before using BIO_CYCLE to calculate AMPLITUDE variable from the normalized TPM values. Rhythmic genes in Ctrl group were divided to six clusters according to their profiles by K-means clustering in RNAlysis 3.8.

Genes were divided into subsets using either DESeq2, rhythmicity threshold or K-means clustering and submitted to over-representation analysis (ORA, implemented in WebGestalt toolkit [46]) against filtered background, showing top ten enriched terms with BH FDR < 0.1 (0.25 in case of differential expression between pooled Ctrl and SCNx samples). Cytoscape v.3.10 was used to detect and visualize protein-protein interactions (STRING database) significantly enriched (FDR < 0.05) against filtered background. Bar plots were prepared in Prism (Graphpad, USA), other plots in seaborn + matplotlib, statistical comparisons (Wilcoxon signed rank test, Kruskal-Wallis and Dunn’s test) were performed in Prism, scipy or statsmodels Python packages.

mPer2^Luc mice with successfully ablated SCN (checked as described above; n = 6) were arrhythmic in constant darkness for 14 days. At 12:00 (corresponded to ZT5 on the previous LD cycle) on day 1, they were subjected to intraperitoneal injection of dexamethasone (SCNx + Dex; 0.15 mg/ml Dex in PBS solution, 0.2 ml per mice, approx. 1 mg/kg, Merck) in darkness using night vision goggles with minimal handling. The injections were repeated at the same time on day 2 and day 3. On day 3, the animals were sacrificed 1 h after injection and organotypic ChP explants were prepared as detailed below. Since the injection is a stressful procedure that increases endogenous glucocorticoid levels, the effect of Dex on the ChP clock was compared with the clock parameters in naïve SCNx mice.

mPer2^Luc mice from Ctrl (n = 3), SCNx (n = 3) and SCNx + Dex (n = 6) groups were sacrificed by rapid cervical dislocation between 12:00 and 13:00, corresponding to CT5-6 of the Ctrl group with intact SCN. ChP explants were dissected from the 4th ventricle as described previously [21]. The explants were immediately placed onto Millicell Culture Inserts (Merck) inside a 6-well plate containing 1 ml of air-buffered recording medium (DMEM supplemented with 100 U/ml penicillin, 100 µg/ml streptomycin, 1% GlutaMAX, 2% B27 supplement (ThermoFisher) and 0.1 mM D-Luciferin (Biosynth, Switzerland)). The bioluminescence traces were recorded in a motorized Luminoview LV200 luminescence microscope (Olympus, Japan) fitted with an autofocusing LUCPLFLN20X objective (Olympus), a stage top incubator set to 37 °C/100% humidity (Tokai HIT, Japan) and a water cooled ImageEM X2 EMCCD camera (Hamamatsu, Japan), with an exposure time of 9.5 min and electromagnetic gain 250. Six explants in a plate were imaged every 1 h for 120 h by CellSens Dimensions (Olympus) and exported as 16-bit TIFFs. Cosmic rays were removed by pixelwise subtraction of consecutive images. Whole ChP explants were manually traced in Fiji ImageJ and algorithmically subdivided into 4684–6434 approximately cell-sized regions of interest (ROIs). Data were exported as mean intensity signal and XY coordinates and analyzed after detrending and denoising by a custom modified Python script based on per2py package [47].

ChP samples collected from Ctrl and SCNx C57BL/6J mice were sequenced on the Illumina NextSeq 2000. We obtained 41.6–48 million reads uniquely mapped to the mouse genome per each wild type ChP sample. After exploring TPM cross-sample correlation and distribution (Fig. S2A, B), we further analyzed all genes with average TPM ≥ 1 / sample that were assigned a unique gene ID (n = 13,077).

Successful isolation of ChP samples without contamination from surrounding cerebral tissue was confirmed by the result that among the chromosomal genes with the highest expression were those coding for well-known ChP-enriched secreted proteins (supplemental Table S1), such as Ttr (transthyretin, which transports thyroxine from the bloodstream to the brain), Enpp2 (autotaxin, responsible for the production of lysophosphatidic acid in extracellular fluids), Clu (clusterin, extracellular chaperone), Igf2 (insulin-like growth factor 2) and Igfbp2 (insulin-like growth factor-binding protein 2) [30, 31], as well as ChP markers such as Folr1 (folate receptor alpha [48]).

We used DESeq2 to explore differential expression between the Ctrl and SCNx groups. Only 11 genes were significantly differentially expressed between the groups (FDR < 0.05, max. log fold change ± 0.3) when all time points were analyzed together; 174 genes with FDR < 0.25 (Fig. 1E) were submitted to over-representation analysis (ORA), which revealed several significantly enriched Gene Ontology (GO) terms related to Endoplasmic reticulum (ER), ER membrane, Golgi complex and cation transport (Fig. 1F). Among the downregulated genes was Slc22a8 (organic anion transporter 3, Oat3), a previously described organic transporter responsible for organic anion distribution in CSF [49]. As the overall differences between the whole Ctrl and SCNx groups were minor, we performed time-resolved transcriptomic analysis.

To identify circadian cycling transcripts, we took advantage of two-day sequential sampling in constant darkness. We employed three previously described algorithms, each based on different rhythmicity detection method – eJTK, ARSER and BIO_CYCLE (see methods for details). We empirically set alpha value for FDR-adjusted Q-values of all three methods to 40% based on exploring how they influence the number of identified cycling genes in both groups (Fig. 1G). This allowed us to detect considerable number of rhythmic transcripts in the Ctrl group (n = 732, Fig. 1H, I left) while simultaneously minimizing the number of low amplitude and/or false positives in the SCNx group (n = 27, Fig. 1H, I right). This resulted in a large number of false positives with clearly non-circadian variation in transcript levels. Our data underscore the need for careful use of rhythmicity prediction algorithms, ideally together with arrhythmic or randomized control.

When ranked by relative amplitude, the four top scoring genes in Ctrl group were clock genes Nr1d1/Rev-Erbα, Ciart/Chrono, Arntl/Bmal1 and Dbp. Additionally, all other known clock genes were identified as rhythmic, with the exception of Cry1 and Rora (which were below the rhythmicity threshold, see Fig. 4A and discussion), and Rorb and Bmal2 (which were not expressed in ChP). Most rhythmic genes peaked around the beginning of the subjective night at CT10-14 or at CT18-20 (Fig. 1J, left), while very few rhythmic genes peaked in the first half of subjective day between CT0 and CT9.

After surgical destruction of the SCN, ChP’s global transcriptome displayed almost no detectable high or medium amplitude circadian oscillations (Fig. 1K). Although we cannot rule out the possibility that widespread shallow oscillations persisted, a comparison between the same genes in Ctrl and SCNx groups showed a dramatic and highly significant (Wilcoxon signed-rank test, P = 8 × 10^{− 114}) decrease in amplitude after SCN lesion (Fig. 1L, left). Of the genes identified as rhythmic in the SCNx group only 27 had amplitude marginally higher than that of their Ctrl counterparts (P = 0.0039, Fig. 1L, right).

To assess involvement of the local ChP clock in the circadian rhythmicity, we analyzed the transcriptome of additional ChP samples from mice with genetic knockout of essential clock gene, Bmal1, in epithelial cells with motile cilia, including those in ChP (referred to as KO, see Materials and Methods). These samples were collected at two time points: during the day (ZT7) and at night (ZT18). A comparison of the absolute values of normalized day-night differences among the top cycling genes across the three groups suggested (Fig. 1M) that the effects of the targeted genetic disruption of the local clock in ChP were at least comparable to the effects of the SCN lesion on ChP rhythms.

We employed ORA, analyzing all rhythmic genes against filtered background, to characterize the circadian transcriptome of ChP in Ctrl mice. First, we visualized the gene network using the STRING protein-protein interactions database in Cytoscape (Fig. 2A). The majority (77%) of the identified rhythmic genes formed a single large network, with three major subnetworks enriched in Circadian rhythm (clock genes), ER Protein processing (mainly ER chaperones, see details below), and C-terminus of histone H2A (mainly histones and associated proteins). Moreover, related GO (Fig. 2B) and KEGG (Fig. 2C) terms were significantly enriched. Chaperone Protein class, ER and histone-related Reactome terms were similarly enriched (Fig. 2C). Finally, we identified several enriched transcription factor (TF) target sequences (Fig. 2C) – among them, known circadian regulators, such as E4BP4/NFIL3, CREBP1/ATF2 and HLF, as well as the heat shock factor (HSF).

To identify the time of day when specific rhythmic ChP functions are activated in Ctrl mice, we used K-means clustering to categorize all rhythmic genes (Fig. 3A) into six distinct phase/amplitude clusters (Fig. 3B and C, right), each containing between 79 and 181 genes.

The largest cluster peaked around CT20 with medium amplitude (CT20/medium) (Fig. 3C, blue). This, together with similarly phased cluster peaking at CT20 but with a high amplitude (CT20/high) (Fig. 3C, violet), was enriched for HSF transcription targets, the Protein processing in ER KEGG term, the ER GO cellular component, and the Protein folding GO biological process terms. The cluster CT20/medium was enriched for the Unfolded protein binding GO molecular function term, while the cluster CT20/high for the Heat shock protein binding GO molecular function term. A STRING protein-protein interactions analysis showed that the cluster CT20/medium was enriched for Protein processing in ER, Oligosaccharyltransferase complex and Chaperone complex. In contrast, the cluster CT20/high was enriched for the Unfolded protein response, Chaperone complex, Nucleosome and other STRING cluster terms (Fig. S2C).

The cluster that peaked around CT14 with medium amplitude (Fig. 3C, red) was enriched for E4BP4 (NFIL3) transcription targets, among them immune response genes involved in NF-κB signaling (Fig. S2C).

The cluster that peaked at CT8 with high amplitude (Fig. 3C, green) was enriched for HIF1 (hypoxia inducible factor) transcription targets and the LRRC8/pannexins STRING cluster term (volume-regulated anion channel subunits, Fig. S2C).

The cluster that peaked at CT12 with high amplitude (Fig. 3C, orange) contained clock and clock-controlled genes regulated via E-box. It was enriched for the Circadian rhythm KEGG and STRING terms, Rhythmic process and related GO biological process terms, the Transcription regulatory region DNA binding GO molecular function term and several TF target terms, including C/EBP, CREBP1, E4BP4 and ARNT. It also contained genes involved in WNT, glucocorticoid, TGF and PDGF signaling (Fig. S2C).

The cluster that peaked around CT23 with high amplitude (Fig. 3C, brown) was enriched for RORA TF targets, with the Positive regulation of developmental process GO Biological process and the Plasma membrane receptor complex GO cellular component terms.

In ChP of Ctrl mice, the six most highly expressed rhythmic genes (Hspa8, Hsp90aa1, Calr, Hsp90b1/GRP94, Kcne2, and Hspa5/GRP78/BiP) and more than half of the remaining top 50 most highly expressed rhythmic genes (including transcription factor Xbp1 and its target Sdf2l1 [50]) play a role in endoplasmic reticulum stress (ERS) response or unfolded protein response (representative genes from this group are shown in Fig. 4B). We visualized the corresponding KEGG pathway (Protein processing in Endoplasmic reticulum, Fig. S3), showing that Chaperone-mediated protein recognition, Folding and ER-associated degradation (ERAD) processes, as well as the main regulatory complexes XBP and CHOP contain genes under circadian regulation. All of them belong to clusters peaking at CT20 with medium or high amplitude (Fig. 3C, S2C), with the exception of Fus (fused in sarcoma) which peaked at CT12. The gene codes an RNA-binding component of hnRNP complex involved in splicing that actually induces ERS only in its mutated form commonly expressed in several types of cancer and amyotrophic lateral sclerosis [51, 52]. Interestingly, several other splicing related genes (e.g., Lsm5, Cwc27; Fig. S5) followed similar expression pattern as Fus, suggesting that both RNA and protein processing is rhythmic but with distinct timing.

We identified circadian cycling transcripts of genes previously described to play specific roles in ChP functions in Ctrl mice (Fig. 4C). Here, we list only a few examples.

Hydin (hydrocephalus-inducing protein homolog) is a gene required for ciliary motility. Its mutation leads to the accumulation of CSF within the ventricles [53]). Ccdc88c (coiled-coil domain containing 88 C, or Daple) is a gene whose mutation causes congenital hydrocephalus [54]. Its product interacts with and negatively regulates the cytoplasmic WNT signaling transducer disheveled, whose transcript Dvl1 was also significantly rhythmic in ChP. Moreover, the gene for the Frizzled ligand (Wnt5b, see Fig. 4H) was rhythmic with high amplitude, as were other genes involved in both canonical (Porcn, Sfrp5, Bambi, Nkd2) and non-canonical WNT pathways (Nkd2, Nfatc2, see Fig. S4).

Another group of rhythmically expressed genes is involved in the modulation of CSF composition. The product of Slc22a8, OAT3, controls the organic anion distribution in CSF [49], including prostaglandins (reviewed in [55]). Kcne2 (Potassium Voltage-Gated Channel Subfamily E Regulatory Subunit 2) regulates the CSF ion composition [56]. All of these genes lost circadian expression in the SCNx mice; in the case of Slc22a8, this was accompanied by significant downregulation (DESeq2, FDR < 0.05; Fig. 1E). Additionally, other genes potentially involved in CSF composition, such as genes coding for ion channels and secreted signaling proteins, are listed in the following subchapter.

On the other hand, some genes critical for CSF production, such as Ttr (Fig. 4D), Folr1, Enpp2, Clu (Table S1), were either constitutively expressed or their mRNA variation was below the rhythmicity threshold. We also found non-circadian expression of several genes previously hypothesized [19] to be responsible for diurnal changes in CSF volume and composition (Fig. 4D). These include the ion cotransporter Slc4a10, Glut1/Slc2a1, ATPase subunit genes, tight junction genes such as Cldn2, and aquaporins Aqp1/4/11, suggesting that their potential circadian regulation might need to occur at the level of protein abundance or activity, as was recently demonstrated for Ttr [32].

With the exception of known clock genes, we identified many previously uncharacterized, highly rhythmic genes in ChP of Ctrl mice (Fig. 4E). Among the highest amplitude non-clock transcripts (Fig. 4E) were several lncRNAs and pseudogenes, as well as genes coding for components of the ERAD pathway (Derl3), various enzymes (Entpd1, Neu2, Dnase1l2, Fmo2, Lpl), and histones (H2bc24, H3c4).

We identified at least 9 genes for ion channels or their regulatory subunits (including K⁺, Ca²⁺, and Cl⁻ channels that may affect the CSF composition, Fig. 4F) and more than 20 genes for various membrane receptors and their interacting partners. These include genes for NOTCH4 receptor and its ligand JAGGED1 (Jag1), ciliary neurotrophic factor receptor (Cntfr), interleukin 1 receptor (Il1r1), platelet-derived growth factor receptor A (Pdgfra), nerve growth factor receptor (Ngfr), growth hormone receptor (Ghr), and tumor growth factor receptor β subunit (Eng; Fig. 4G), all exhibiting medium or high amplitude rhythms.

We detected high-amplitude expression of genes coding for secreted signaling proteins and known mitogens (Fig. 4H), such as above-mentioned Wnt5b, two angiopoietin genes, and cellular communication network factor 2 (Ccn2). We also found evidence of rhythmic activation of the glucocorticoid receptor (GR) signaling pathway, with its markers such as Gilz (Tsc22d3) tracking the peak in blood glucocorticoids during first half of the night (Fig. 4I).

Many genes involved in immune response (including interleukin 1 and lipopolysaccharide receptors, Fig. 4G; NF-κB signaling components Nfkbia and Nkiras1, suppressor of cytokine signaling Cish and others, Fig. S5), cell cycle regulation (including G2 checkpoint kinase Wee1 and p21^CIP1/WAF1 G1 checkpoint inhibitor gene Cdkn1a, Fig. S5), splicing (Fig. S5), and vesicle trafficking (Fig. S5) were rhythmically expressed with high amplitude. We also found rhythmic transcripts for at least 15 histones and histone-associated proteins (Fig. S5), 4 cadherin genes coding transmembrane adhesion proteins, including recently reported circadian cadherin Cdh3 (Fig. S5; [32]), at least 20 transporters potentially involved in circadian regulation of CSF composition (including Abcb6 and Magt1), 8 G-proteins and G protein-coupled receptors (Fig. S6).

Unlike ERS genes, which all clustered around CT20 (Fig. S2C, 4B), the phase of these functional genes was much more varied. For example, WNT, NOTCH and GR signaling genes clustered around CT12, indicative of E-box driven expression, while interleukin 1 receptor gene was expressed in antiphase and E4BP4/NFIL3 targets governing immune response clustered around CT4. Similarly, genes for K⁺ and Ca²⁺ channels were expressed in different phases.

Together, our data demonstrate broad circadian rhythmicity of ChP transcriptome involved in diverse time-specific functions, with a very high emphasis on nighttime expression, which is, nevertheless, highly dependent on systemic rhythms driven by the central circadian pacemaker in the SCN.

The massive dampening of ChP circadian transcriptome oscillations in response to lesion of the SCN (Fig. 1) could be attributed to a loss of rhythms at the single-cell level, a loss of synchrony among cellular oscillators, or, more likely, a combination of both. To address this, we explanted ChP from mPer2^Luc mice subjected to sham surgery (Ctrl; n = 3) or SCN lesion (SCNx; n = 3; SCNx was verified as detailed above) to monitor PER2-driven bioluminescence in individual cell-sized regions of interest (ROIs) of the whole ChP tissue using a luminescence microscope (LV200, Fig. 5A). The recording started immediately after harvesting the explants in the presence of luciferin and lasted for 5 days. While explants from both Ctrl and SCNx mice were rhythmic, the overall amplitude and PER2 protein luminescence levels were noticeably reduced in the SCNx explants (Fig. 5B-E, upper vs. middle plot). Analysis of individual ROIs confirmed that the overall rhythm in SCNx explants was compromised. Setting the rhythmicity threshold of cosine curve goodness of fit at R² = 0.97 revealed dramatically fewer high-amplitude oscillating ROIs in the SCNx compared to the Ctrl explants (Fig. 5B, C). The phases of the ROI rhythms in SCNx explants were more dispersed (Fig. 5D), and their amplitudes were lower on average and more uniform (Fig. 5E) across the whole explant. A statistical comparison (Fig. 5F, blue vs. red) between all recorded ROIs across Ctrl and SCNx explants showed that the amplitude, mesor (average PER2 level), and goodness of fit were significantly reduced in the SCNx explants (Mann-Whitney, P < 0.0001). In addition, the period was shorter (P < 0.0001), and the phase was more spread out (P < 0.0001) in the SCNx than in the Ctrl explants. Thus, the presence or absence of SCN in vivo has lasting effects on ChP circadian cycling in static culture ex vivo, despite the fact that local clocks still drive the rhythms in ChP cells.

In a last step, we attempted to uncover a possible mechanism of how the SCN controls the ChP clock. Based on our previous data [21], we hypothesized that glucocorticoids and subsequent GR activation play a role as the major SCN-driven signal that controls the ChP clock. Indeed, our transcriptomic analysis showed that GR activation was highly rhythmic in the Ctrl ChP, but the rhythmicity was lost due to SCNx (Fig. 4I). SCN ablated arrhythmic mPer2^Luc mice (n = 6) kept in constant darkness received injection of corticosterone analog dexamethasone at 12.00 h (corresponding to ZT5 of the previous LD cycle) on three consecutive days (SCNx + Dex), and we analyzed the rhythms in ChP explants harvested 1 h after the last injection (Fig. 5A). We hypothesized that if the GR signaling pathway is involved in ChP clock entrainment in vivo, Dex injections should be able to at least partially rescue the number of rhythmic ROIs ex vivo. Indeed, analysis of ex vivo rhythms in single cell-sized ROIs, as described above, showed that both the number of rhythmic ROIs (Fig. 5B, C, lower plot), their amplitude, mesor (average level of PER2 luminescence over the course of the analysis), goodness of fit (R²) and period (Fig. 5E, lower plot and Fig. 5F, orange) increased dramatically in SCNx ChP after Dex injections (Kruskal-Wallis with Dunn’s test, P < 0.0001), even above and beyond the rhythm in Ctrl group. Moreover, the ChP clock was phase advanced by 5–9 h when compared with Control or SCNx group (Fig. 5D, F), consistent with the advanced timing of Dex injections relative to the endogenous corticosterone surge. Overall, the data provide evidence that the SCN-regulated glucocorticoid signaling is one of the possible drivers of circadian rhythms in ChP.

The authors declare no competing interests.

Raw files and technical details about the RNA-Seq data have been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO Series accession no. GSE243858↗.