The global and promoter-centric 3D genome organization temporally resolved during a circadian cycle

To study global genome architecture during a circadian cycle, we performed in-nucleus Hi-C (see “”) on mouse adult liver at four different timepoints of the circadian cycle (ZT0, 6, 12 and 18), with ZT0 and ZT12 being the start of the light and dark phase, respectively, in three biological replicates. Samples of individual livers were processed in parallel for RNA-seq. We produced high-quality Hi-C data sets with a high percentage of valid pairs (~ 80%), low PCR duplicates (less than 2%), and high cis:trans interaction ratios obtained (~ 80:20%) (Table S). In total, we obtained ~ 2 billion valid Hi-C read pairs from mouse adult liver across a circadian cycle (Table S). Methods 1 1

To relate chromatin compartments with transcription, we performed stranded total RNA-seq. We obtained ~ 500,000,000 of 150-bp reads per timepoint (4 biological replicates each) for a total of ~ 2,000,000,000 reads (Table S2). Spearman correlation analysis showed good correlation between the 4 biological replicates (Additional file 2: Figure S2A, ZT0 and 12 shown). Genes with differential expression between at least one pair of timepoints were identified (q-value < 0.01) and classified as circadian. In total, we detected 1257 circadian gene transcripts (Additional file 2: Figure S2B, Table S3). Inspection of individual gene expression profiles from our RNA-seq showed the expected oscillatory expression pattern for examples of both core-clock and output circadian genes in the liver (Additional file 2: Figure S2C,F). Gene Ontology and KEGG pathway analysis identified circadian rhythm and metabolism as significantly enriched categories in our identified circadian gene set (Additional file 2: Figure S2D,E) confirming efficient detection of circadian oscillating genes. RT-qPCR of primary and mature mRNAs confirmed oscillation of circadian gene expression (Additional file 2: Figure S2G). Analysis of the RNA content in A and B compartments at all timepoints revealed that A compartments are RNA-rich and B are RNA-poor (Fig. 1c, p < 0.0001, one-way ANOVA test). The RNA content in OCCs falling into A or B at different times of the day was also significantly different (Fig. 1d, p < 0.0001, Mann-Whitney test). In addition to RNA abundance, we analyzed time resolved enrichment of histone modifications characteristic of open chromatin H3K4me3 and H3K4me1 [13] in OCCs around the clock. Both histone marks were significantly enriched in regions at times of A compartment assignment compared to B compartment across all timepoints (Fig. 1e, AABA [ZT0 = A, ZT6 = A, ZT12 = B, ZT18 = A] and BABB [ZT0 = B, ZT6 = A, ZT12 = B, ZT18 = B], Additional file 1: Figure S1F, AABB, ABBB, BABA. All p values < 0.001, one-way ANOVA test, Tukey post hoc test). Histone deacetylase 3 (HDAC3) binding to chromatin is enriched in deacetylated “closed” chromatin and it has been measured in mouse liver at ZT22 and ZT10 [14]. We quantified HDAC3 occupancy at OCCs at ZT0 and 12 (our closest timepoints to ZT22 and 10). HDAC3 binding was higher at OCCs that fell into the B compartment at ZT12 compared to the A compartment at ZT0 (Fig. 1f, Wilcoxon p < 0.001). Examples of OCCs overlapping the circadian Arntl gene TAD and the circadian Npas2 gene TAD are shown in Fig. 1g and h respectively. For all circadian genes presented as examples throughout the paper, the RNA-seq signal tracks for the four timepoints examined can be found in Additional file 3: Figure S3. Together, these results show that both transcription and chromatin state fluctuate in accordance with compartment switching during a circadian cycle.

CTCF functions as an architectural protein that establishes chromatin domains together with cohesin in mammalian chromatin [16–18]. We performed ChIP-seq against CTCF at ZT0 and ZT12 on the same liver samples used for Hi-C. In total, 33,262 CTCF binding sites (75.3%), out of a total of 44163 sites, were shared between ZT0 and 12 and CTCF showed similar enrichment between the two timepoints (Additional file 4: Figure S4D). When plotting the Hi-C contacts in an aggregate peak analysis centered on the CTCF binding sites, CTCF-bound regions exhibited robust contact insulation properties, independent of the timepoint examined (Fig. 2b right panel and Additional file 4: Figure S4E top panels). Additionally, we assessed preservation of CTCF insulation between mouse ES cells and liver cells by overlaying regions occupied by CTCF in mESCs [19], onto our liver Hi-C data sets. Robust insulation was observed when using either ZT0 or ZT12 Hi-C data suggesting large agreement of CTCF chromatin occupancy and insulation properties between mESCs and adult liver tissue (Additional file 4: Figure S4 E bottom panels). The conservation of CTCF binding between ESCs and adult liver tissue shows that the insulation of circadian genes by CTCF precedes the onset of their rhythmic transcription, as ESCs do not harbor functional circadian oscillators [20].

Next, we assigned genes to TADs. We found that on average TADs harbored 7.2 genes. TADs containing one or more circadian genes (cTADs) were larger than non cTADs and on average contained more genes (14.2) (Additional file 4: Figure S4F, all p values < 2.2e−16, Wilcoxon rank sum test). We observed that 70% of cTADs contained only one circadian gene (Fig. 2c). The remaining cTADs contained more than one circadian gene and remarkably, the circadian genes sharing TADs exhibited peak transcriptional expression at shared times during the day (40% for TADs with 2 circadian genes compared to the expected 28%, 18% of TADs with 3 circadian genes compared to the expected 8.8%, p < 0.0001, chi square test) (Fig. 2d and e, only data from cTADs with 2 circadian genes shown). We next analyzed whether TADs encoding circadian genes switched chromatin compartment over the circadian cycle. Indeed, most cTADs (73%, 623) overlapped with OCCs more than expected when compared to a random set of the same number of non-cTADs (Wilcoxon test, p < 0.0001) (Fig. 2f). Examples of cTADs containing one circadian gene (cTADs with Mical2, MicalcI, Arntl and Copb1, respectively), or more than one circadian gene (cTAD with Wee1 and Swap70 genes, both with acrophase at ZT12) are shown in Fig. 2g (left and right panels respectively). Examples of cTADs overlapping OCCs are shown for Arntl (Fig. 1g) and Npas2 (Fig. 1h) as previously described. These results show that cTADs often set transcriptional phase coherence between multiple circadian genes in the same TAD. While most TADs maintain their structural boundaries over time, they overlap with compartments that switch between active and silent states throughout a circadian cycle.

First, we focused on gene promoter-promoter contacts. We built promoter-promoter networks at all timepoints and found large disconnected networks as expected for promoters scattered across the different chromosomes (Additional file 5: Figure S5C, ZT0 shown). We then evaluated the larger clusters of the promoter-promoter network containing hundreds of connected gene promoters (Additional file 5: Figure S5D, the 4 larger clusters shown including inter- and intrachromosomal promoter-promoter interactions) and performed gene enrichment analysis on these using gProfiler [25] (Additional file 5: Figure S5E). We found the Major Histocompatibility Complex forming one cluster with genes encoded on chromosomes 7 and 17. MHC genes are lowly expressed in the healthy liver and constitute dense, highly connected chromatin [26] (Additional file 5: Figure S5D and S5E, cluster 1). Similarly, a transcriptionally repressed cluster is formed between olfactory receptor genes on chromosome 7 (Additional file 5: Figure S5D and S5E, cluster 3). In addition to repressed genes clustering in spatial proximity, we identified actively transcribed genes involved in glutathione synthesis and amino acid metabolism essential for liver detoxification function arranged in a promoter network (Additional file 5: Figure S5D and S5E, cluster 2). Another cluster encompassed constitutive histone genes located in chromosomes 3, 13, 11, 18, 12, and 7, among others (Additional file 5: Figure S5D and S5E, cluster 4). Thus, prominent constitutive and liver-specific promoter-promoter networks both transcriptionally active and inactive were identified in the adult liver.

Next we examined the circadian component in promoter-promoter networks. Circadian promoters are dispersed across chromosomes (Additional file 6: Figure S6A, blue dots represent circadian gene promoters). Nevertheless, circadian promoters establish significantly more contacts among them compared to non-circadian gene promoters (Additional file 6: Figure S6B. Above, comparison of the number of edges formed between circadian promoters compared to a random set of non-circadian gene promoters. Below, Z-scores compared to the random sampling of non-circadian promoters). Next we looked at the reads supporting significant interactions between circadian gene promoters in contrasts with interactions between non circadian gene promoters. The result shows that circadian promoter-promoter contacts are more robust compared to non-circadian promoter-promoter contacts (Additional file 6: Figure S6C, p values < 0.001, Mann-Whitney test). Finally, we compared the time of maximal mRNA abundance of circadian genes whose promoters were contacting each other. To do this, we separated the circadian gene promoters into diurnal and nocturnal based on their transcriptional acrophase and then analyzed the time of maximal RNA expression of their contacted circadian gene promoters. We found that diurnal promoters preferentially contacted other diurnal genes. Likewise, nocturnal circadian genes preferentially contacted other genes with nocturnal expression maxima (Additional file 6: Figure S6D, our intronic circadian gene set, p values < 0.0001 Wilcoxon signed rank test). Similar results were obtained for circadian genes detected by GROseq [27] reflecting primary transcription (Additional file 6: Figure S6E, p values < 0.0001 Wilcoxon signed rank test). This is exemplified by the interaction between the Tef and the Aco circadian gene promoters with shared pre-messenger expression peak at ZT12 (Additional file 6: Figure S6F) as well as the interaction between Rorc and Cgn circadian genes with shared acrophase at ZT18 (Additional file 7: Figure S7A). Thus, in summary the results show that circadian promoter-promoter contacts are more robust than non-circadian promoter-promoter contacts. Furthermore, pairs of interacting circadian gene promoters tend to share their transcriptional acrophase.

In order to characterize the full range of promoter interactions, we next examined contacts between promoters and non-promoter genomic regions. Genomic regions that interact with gene promoters including circadian gene promoters in mouse liver, were enriched for histone modifications characteristic of regulatory elements such as H3K27ac, H3K4me1, H3K4me3, and H3K27me3, as well as DNase I hypersensitive sites and the structural protein CTCF [13, 28, 29], compared to distance matched non-interacting regions (Fig. 3b left panel, all p values < 8e−166, t test, CTCF, p value <8e−18; t test). A set of enhancers from which enhancer RNAs (eRNAs) are produced have been described in the liver [27]. We found that these enhancers were enriched at the contacted regions (Fig. 3b, left panel p value < 9e−165; t test). Overall, the chromatin features at promoter-contacting regions suggest that our P-CHi-C captured potential structural and/or regulatory elements.

Next we identified transcription factor binding motifs at genomic regions involved in constant or rhythmic interactions with circadian promoters in an unbiased manner using the MEME suite (“Methods”) [30]. We found TF binding motifs both in regions forming constant and dynamic contacts with circadian promoters. Other DNA binding motifs were found enriched in regions engaged in either dynamic or constant contacts with circadian gene promoters (Fig. 3f, Table S5). The binding motif of the nuclear receptor Nr5a2/LRH-1 was highly enriched at interacting regions of rhythmic gene promoters. Nr5a2 is a key metabolic sensor that modulates bile acid synthesis and cholesterol homeostasis by regulating the expression of Cyp7a1 and Cyp8a1 circadian genes among others, and controls triglyceride synthesis and lipid composition and metabolism [31–34]. Interestingly, a pro-inflammatory function of Nr5a2 deficiency is linked to non-alcoholic fatty liver disease [35]. In addition, loss of Nr5a2 in the adult liver leads to disruption of hepatic lipid homeostasis and composition [36]. To confirm Nr5a2 occupancy at circadian gene promoters and interacting regions, we performed Chromatin IP (ChIP-qPCR) against Nr5a2 at the four timepoints during the day for three loci in which the DNA binding motif was found enriched. We found Nr5a2 dynamically occupying the Arntl circadian gene promoter with peak occupancy at ZT6 when Arntl gene expression starts to decrease (Fig. 3g, Additional file 3: Figure S3). A similar binding pattern over time was observed on the Arntl interacting enhancer (Fig. 3g). We then analyzed Nr5a2 binding to a chromatin region stably contacting the circadian gene promoter of Ppp1r3c. Interestingly, we found that Nr5a2 binding was dynamic with maximum occupancy of Nr5a2 at this region is also dynamic and peaks at ZT6 when Ppp1r3c gene expression starts to decrease (Fig. 3g, Additional file 3: Figure S3). In addition, we assessed Nr5a2 occupancy in a chromatin region dynamically interacting with the circadian gene promoter of Gsk3a but did not find the receptor bound to this region (Fig. 3g). Thus, our results suggest that Nr5a2 is dynamically occupying regulatory elements of circadian genes irrespective of whether the contacts to circadian gene promoters are constant or rhythmic over time.

Other binding sites for nuclear receptors enriched in regions contacting circadian promoters included VDR (vitamin D receptor) and ESR2 (estrogen receptor 2).

The Tcfap2c/AP-2 gamma binding motif was found to be highly enriched at dynamic interacting regions of circadian gene promoters. In the liver, Tcfap2c has been associated with repression of fatty acid synthesis pathways [37] and was identified as a key TF involved in lipid droplets biogenesis [38]. Fos:Jun/AP1 binding motifs were found in genomic regions forming stable contacts with circadian promoters. AP1 factors are a well-characterized immediate-early transcription factors induced in response to signals in the serum and that regulate the expression of circadian genes in liver and cultured cells [39] as well as the suprachiasmatic nucleus [40–42]. Recently, AP-1 was shown to bring together key genes and enhancers through stable and dynamic loops during macrophage development bringing together key macrophage genes and enhancers [43].

In summary, a set of DNA binding motifs for distinct liver nuclear receptors and immediate early genes are enriched in regions contacting circadian promoters and could function in the wiring of the circadian promoter 3D interactome in the liver.

A set of enhancers are transcribed in a circadian fashion in the mouse liver [27]. We found that these enhancers preferentially contact circadian gene promoters, suggesting that rhythmically transcribed genomic regions, protein-coding and non-coding, interact with each other in the nuclear space (Fig. 4a). This is also true for our subset of circadian genes oscillating at the intronic level (see “Methods”) and for circadian genes detected through GRO-seq [27] reflecting primary transcriptional oscillation (Fig. 4b,c).

We then compared the transcriptional phases between promoters of circadian genes and their corresponding contacted enhancer elements with rhythmic transcription. To do so, we separated the circadian gene promoters into diurnal and nocturnal depending on their transcriptional acrophase and then analyzed the time of maximal RNA expression of their contacted enhancer elements. We found a significant contact preference between diurnal promoters and diurnal enhancers as well as nocturnal promoters and nocturnal enhancers (Fig. 4d, left). These preferences were more pronounced when analyzing our circadian intronic gene set reflecting primary transcription (see “Methods”) (Fig. 4d, center) or detected by GRO-seq (Fig. 4d, right, all p values < 0.0001 Wilcoxon signed rank test). For example, the Rnf125 circadian gene promoter with peak transcription at ZT0 contacts 12 rhythmically expressed enhancers with acrophases between 19 and 1 h during the circadian cycle. Furthermore, as it can be observed, the number of contacts with the enhancers, increase during the acrophase (Fig. 4e).

Finally, we focused on the genomic interactions formed by circadian core clock gene promoters including Npas2, Clock, Arntl, Cry1, Cry2, Per1, Per2, Rorc, Nr1d1, and Nr1d2 as defined by [44]. Notably, all core clock genes displayed fewer overall contacts compared to a random set of the same number of other circadian genes in the liver (12 vs 19.6 mean number of contacts for core-clock vs other circadian genes, Fig. 4f, p < 0.0001, t test). However, the contacts formed by the core clock gene promoters were more dynamic than a random set of the same number of contacts for other circadian genes in the liver (42.3% vs 26.8% mean proportion of dynamic contacts for core clock vs other circadian genes, Fig. 4g, p < 0.0001, t test). For instance, the Arntl circadian gene promoter engages in contacts with two enhancer elements at ZT18, the time when Arntl expression increases, in three contacts at ZT0, at time of maximal transcriptional output, and does not engage in contacts at either ZT6 or ZT12, when Arntl1 transcription decreases (Fig. 4h, see Additional file 3: Figure S3 for the expression profile). The Nr1d1 gene promoter engages in more contacts at the gene’s maximal time of expression, around ZT6 (Fig. 4i, see Additional file 3: Figure S3 for the expression profile). In contrast to core clock contact profiles (additional examples are shown for Rorc, Nr1d2, Npas2, and Per2 (Additional file 7: Figure S7A-D, see Additional file 3: Figure S3 for expression profiles), promoters of circadian output genes engage in numerous contacts that are constant during the day as exemplified by Dhr3 and Ppp1r3c gene promoters (Additional file 7: Figure S7E and F, see Additional file 3: Figure S3 for expression profiles). In conclusion, core-clock gene promoters engage in more dynamic genomic contacts compared to other circadian genes in the liver.

C57BL/6 male mice were maintained in the Babraham Institute Animal Facility and all applied procedures were approved considering the animal welfare practices according to the Home Office in the UK. Then, 8-week-old male mice were maintained under a 12-h light:12-h dark cycle for 2 weeks and fed ad libitum. Livers were dissected from three biological replicate pools (Pool A, B and C) composed of 2 livers each at four timepoints ZT0, 6, 12, and 18, with ZT0 = lights on and ZT12 = lights off. The liver tissue was chopped into 5–6-mm³ pieces and directly fixed in 2% formaldehyde for 10 min.

In nucleus, Hi-C library generation was performed as previously described [21]. Briefly, fixed adult liver tissue from three biological replicates at ZT0 0, 6, 12, and 18 was sieved through a 70-μM cell strainer and dounce homogenized in 10 ml of ice-cold lysis buffer with a tight pestle for a total of 30 strokes on ice. Nuclei were washed and permeabilized with 0.3% SDS for 45 min at 37 °C and then incubated overnight with HindIII at 37 °C, DNA ends were labeled with biotin-14-dATP (Life Technologies) in a Klenow end-filling reaction and ligated in nuclei overnight. DNA was purified by phenol-chloroform, and the concentration was measured using Quant-iT PicoGreen (Life Technologies). A total of 10 μg of DNA was sheared to an average size of 400 bp using a Covaris machine and following the manufacturer’s instructions. The sheared DNA was end-repaired, adenine tailed, and subject to a double size selection using AMPure XP beads to isolate DNA ranging from 250 to 550 bp in size. Ligated fragments marked by biotin-14-dATP were pulled-down using MyOne Streptavidin C1 DynaBeads (Invitrogen) and ligated to paired-end adaptors (Illumina). The in nucleus Hi-C libraries were amplified using the PE PCR 1.0 and PE PCR 2.0 primers (Illumina) using 6–9 PCR amplification cycles as required.

Promoter Capture was performed as previously described [21–23]. Briefly, Biotinylated 120-mer RNA baits were designed to target both ends of HindIII restriction fragments overlapping the Ensembl promoters of protein-coding and noncoding transcripts and UCEs as described in detail in [22]. Promoter Capture was carried out using in nucleus Hi-C libraries derived from three biological replicates at ZT0 0, 6, 12, and 18 with the SureSelect target enrichment system and the biotinylated RNA bait library according to the manufacturer’s instructions (Agilent Technologies). After library enrichment, a post-capture PCR amplification step was carried out using the PE PCR 1.0 and PE PCR 2.0 primers (Illumina) with 4–6 PCR amplification cycles as required. In nucleus Hi-C and CHi-C libraries were sequenced on the Illumina HiSeq 2000 platform.

For ChIP-seq, liver tissue for two biological replicates at ZT0 and ZT12 was dissected as processed as for Hi-C and then fixed in 1% formaldehyde for 5 min. Chromatin immunoprecipitation was performed as described [21] using 10 μg of α-CTCF (Millipore, 07-729). DNA was purified using Zymo Research DNA purification columns. Sequencing libraries were prepared with the NEBNext ChIP-seq library prep kit (NEB) according to the manufacturer’s instructions. DNA was purified using AMPure beads (Agencourt). For quantitative chromatin immunoprecipitation experiments (ChIP-qPCR), liver tissue, collected at ZT0, ZT6, ZT12, and ZT18 per duplicate, was fixed with a 1% formaldehyde for 10 min. Cell lysis and chromatin immunoprecipitation were performed essentially the same way as for the ChIP-seq experiments, using 5 μg of a monoclonal antibody against the NR5A2 receptor (Santa Cruz Biotechnology, sc-393369) per assay. ChIP DNA templates were purified using the Zymo Research ChIP DNA Clean & Concentrator kit (Zymo Research, D5201). Quantitative protein occupancy at specific regions was performed by qPCR (Jena Bioscience, PCR-369), according to the manufacturer’s instructions. Mean values for all the analyzed regions across ZTs were expressed as fold enrichment compared to a mock control. Statistically significant differences between ZTs and mock were analyzed using one-way ANOVA followed by the Tukey post hoc test.

For RNA-seq libraries, total RNA was purified from the same livers processed for in nucleus Hi-C from four biological replicates at ZT0, 6, 12, and 18. Sequencing libraries were prepared with TruSeq Stranded Total RNA Ribo Zero Gold Library Prep Kit v2 (Illumina).

Additional file 1: Figure S1. OCCs present fluctuations in histone modifications. A. Heatmap of PC1 values for stable compartments from the three independent Hi-C biological replicates and B, the merged Hi-C biological replicates. C. PC1 eigenvector correlations between every pair of timepoints (Kruskal-Wallis rank sum test p-value < 2e-16). D. Proportion of OCCs and constant compartments in the mouse genome. E. Barplot of the frequency of the different OCCs categories with compartments switching at different times of the 24 hours. For example AABA refers to OCCs with compartment assignment A at ZT0, A at ZT6, B at ZT12 and A at ZT18. F. Chromatin features (H3Kme4 and H3Kme1) at OCCs across time points (AABB, ABBB and ABAB OCCs are shown) (*** all p-values < 0.001 except for the H3K4me1 AABB, one way ANOVA and Tukey post hoc test).Additional file 2: Figure S2. Total RNA-seq analysis at four timepoints during the circadian cycle. A. Spearman correlation analysis of the four biological replicates from ZT0 and ZT12 (r ≥ 0.85). B. Heat map of the relative transcription (Z scores) of 1,257 oscillating genes sorted by oscillation phase. C. Individual expression profiles and genome browser tracks showing examples of the RNA-seq signal for Arntl, Per1, Nr1d1, Cry1, Ppp1r3c and Gsk3a oscillating genes (Fragments Per Kilobase of transcript per Million mapped reads, FPKM) (n=4, q-value <0.01). D. GO analysis for the detected oscillating genes. Significantly enriched categories are shown. E. KEGG analysis for detected oscillating genes. Significantly enriched categories are shown. F. Additional individual transcriptional profiles from the RNA-seq for Rorc, Clock, Npas2, Cry2, Per2, Per3 circadian genes. G. RT-qPCR validation for candidate circadian genes both at mRNA and pre-mRNA level (p-values < 0.0001, one-way ANOVA test).Additional file 3: Figure S3. Expression profiles of the circadian genes displayed along the paper. RNA-seq signal tracks for circadian genes shown across the paper for the four timepoints during the circadian cycle. Arntl, Npas2, Cgn, Nr1d1, Dhrs3, Nr1d2, Per2, Ppp1r3c, Tef, Rnf125, Rorc and Aco2 circadian gene expression profiles are presented.Additional file 4: Figure S4. TAD structure and CTCF binding is preserved during the 24 hours. A. TADs overlap between time points. B. TADs size distribution at all time points. C. 50kb resolution Median Observed/Expected Hi-C signal around 1000 randomly selected TADs from ZT6, 12 and 18 plotted on the indicated time points. TADs were scaled to fit the five central bins. D. CTCF ChIP-seq motif analysis and peak overlap between ZT0 and 12. Genomic tracks of CTCF ChIP-seq signal at example regions harbouring circadian genes. E. Above, the same metaplots as in C but for 1000 randomly CTCF peaks found at ZT12 plotted using ZT12 and ZT0 Hi-C contacts. Below, metaplot using CTCF peaks from mESC and plotted using liver ZT0 and 12 Hi-C contacts. CTCF peaks are at the central bin of the metaplot. F. TAD and cTAD size comparison (p-value < 2.2e-16, Wilcoxon rank sum test).Additional file 5: Figure S5. Promoter-Promoter networks in the mouse liver over a circadian cycle. A. Partial view of a virtual 4C from the Arntl gene promoter using the Hi-C and P-CHi-C raw valid pairs. Histograms of read counts per restriction fragment around the bait region corresponding to the captured promoter are shown. B. Number of total read counts comparing Hi-C and P-CHi-C recovered using the Arntl1 gene promoter as bait on a virtual 4C (restriction fragments with at least 5 reads in the P-CHi-C experiment where used) (p-value < 0.0001, Wilcoxon ranked test). C. Promoter-promoter contact network at ZT0. Each color represents a chromosome. Nodes with degree=1 are not included for simplicity. D. The four largest promoter-promoter interaction clusters of the network at the different time points during the day. E. Significantly enriched GO categories for the genes on the most prominent promoter-promoter clusters shown in D.Additional file 6: Figure S6. Circadian gene promoter-promoter interactions. A. Promoter-promoter contact network at ZT0 with the circadian genes marked in blue. B. Above, number of edges (interactions) between circadian gene promoters at ZT0, 6, 12 and 18 hours of the day (blue line) compared to a random set of non circadian promoters. Below, z-score for circadian gene promoter contacts compared to the random sampling. C. Quantification of the read counts supporting all circadian gene promoter-promoter interactions compared to a random set of non circadian promoter-promoter interactions (*** p-value < 0.001, Mann Whitney test) D. Transcriptional phase distributions of circadian promoters in contact with diurnal or nocturnal circadian promoters (all p-values < 0.0001, Wilcoxon signed rank test) for our circadian intronic gene set. E. The same as in D for the circadian gene set detected through GRO-seq (Fang et al., 2014) (all p-values < 0.0001, Wilcoxon signed rank test) F. Virtual 4C landscape for the Tef circadian gene promoter from P-CHi-C data at all time points during the day. The acrophase of Tef is written next to the gene name. Expression profiles for both genes can be found in Figure S3. Genomic tracks show significant contacts as arcs and chromatin features including liver H3K4me3, H4K4me1, H3K27ac, DNAseI, eRNAs and TADs. Tef gene promoter contacts Aco2 gene promoter both with acrophase at ZT12.Additional file 7: Figure S7. Core clock gene promoter vs output circadian genes interaction landscapes.A-D. Virtual 4C for Rorc, Npas2, Nr1d2 and Per2, core clock circadian gene promoters from P-CHi-C data at all time points during the day. The Rorc circadian gene promoter contacts Cgn circadian gene promoter and both peak in transcription at ZT18. E-F Virtual 4C for Dhrs3 and Ppp1r3c output circadian genes in the liver. Acrophases are written next to the gene names. Genomic tracks show significant contacts as arcs and chromatin features including liver H3K4me3, H3K4me1, H3K27ac, DNAseI, eRNAs and TADs. Expression profiles for all genes can be found in Figure S3. The interaction profiles of core clock genes display less contacts that dynamically change over time. The two output gene contact profiles show more saturated contacts that are mostly constant during the 24 hours.Additional file 8. Custom code. This file contains a summary of the custom code used in this work.Additional file 9. Review history.Additional file 10: Table S1. Hi-C statistics. HiCUP summary results for the independent Hi-C replicates.Additional file 11: Table S2. Total stranded RNAseq number of unique read pairs.Additional file 12: Table S3. Circadian genes detected from total RNAseq.Additional file 13: Table S4. Promoter Capture Hi-C statistics. HiCUP summary results for independent P- CHi-C replicates and significant interactions detected with CHiCAGO.Additional file 14: Table S5. Transcription factor DNA binding motif enrichment analysis. This table contains the information from all time points for both dynamic and stable interactions separately.