Genetics-nutrition interactions control diurnal enhancer-promoter dynamics and liver lipid metabolism

SUMMARY

The circadian clock controls 24-hour rhythmic processes. However, how genetic variations outside clock genes impact peripheral diurnal rhythms remains largely unknown. Here, we find that genetic variation contributes to different diurnal patterns of hepatic gene expression in both humans and mice. Nutritional challenges alter the rhythmicity of gene expression in mouse liver in a strain-specific manner. Remarkably, genetics and nutrition interdependently control more than 80% of rhythmic gene and enhancer-promoter interactions (E-PIs), with a noncanonical clock regulator ESRRγ emerging as a top transcription factor during motif mining. Knockout of Esrrγ abolishes strain-specific metabolic processes in response to diet in mice, while SNPs associated with rhythmic gene expression are enriched in E-PIs in steatotic human livers and correlate with lipid metabolism traits. These findings reveal a previously underappreciated temporal aspect of genetics-environment interaction in regulating lipid metabolic traits, with implications for individual variations in obesity-associated disease susceptibility and personalized chronotherapy.

In brief

Noncanonical clock regulators bridge genetics and nutrition to mediate 3D enhancer-promoter interactions contributing to diurnal rhythm variation and metabolic disease risk.

Graphical Abstract

INTRODUCTION

Diurnal rhythms allow mammals to anticipate daily environmental changes and maintain physiological homeostasis ¹. Disruption of diurnal rhythm caused by shift work and chronic jet lag is a risk factor for many chronic metabolic diseases ^2–4. The genetic basis of diurnal rhythm was first discovered by Konopka and Benzer, who measured the rest and activity in flies ⁵. This discovery led to the identification of multiple core clock genes, including the first core clock gene Period (Per) from Nobel Prize awardees Hall, Young, and Rosbash, which provided insight into molecular mechanisms controlling the 24-hour rhythmic gene expression and physiological processes ⁶. Since the central circadian clock in the suprachiasmatic nucleus (SCN) acts as a master pacemaker for rhythmic behaviors ⁷, previous studies have advanced our understanding of how genetic variations affect the expression of core clock genes and contribute to the trait known as chronotype ⁸, which exhibits different sleeping and wake rhythms and related nervous system diseases, such as stress and depression ^9–11. In addition to central clocks in SCN, diurnal rhythms in peripheral tissues, while synchronized with the central SCN, also exhibit independent rhythms and are influenced by local cues to regulate tissue-specific functions ¹². For example, recent studies, including ours, show hepatic diurnal rhythms can be synchronized by nutritional signals and control rhythmic metabolism ^13,14. Moreover, although core clock genes are rhythmically expressed in all tissues, rhythmic transcriptomes in various tissues have indicated tissue-specific transcription regulation. Indeed, only less than 20 genes are concertedly rhythmically expressed across 10 tissues, indicating additional factors, such as lineage factors or environmental sensors, are critical to regulating tissue-specific rhythmic gene expression ^15,16. Yet, the impacts of genetic variations beyond core clock genes in regulating diurnal rhythms in peripheral tissues have barely been studied.

Utilizing newly developed mouse models to tissue-specific knockout canonical clock genes and analyses of unbiased genome-wide enhancer activities and transcriptomes, we and others found many genes retained rhythmic expression profiles upon core clock genes knockout ^13,17–19. For example, deleting the core clock genes Rev-erbα and Rev-erbβ in adult mouse hepatocytes disrupted the diurnal rhythms of other clock genes, including Bmal1 and Npas2; however, two-thirds of hepatocyte genes retained rhythmic expression ¹³. In revisiting previous circadian studies under various conditions, such as nutritional challenges, exercise, lung cancer, and aging ^18,20–24, we noted that canonical clock genes retained robust rhythmic expression in the liver, whereas thousands of circadian output genes and metabolites gained or lost rhythmic expression. Moreover, time-restricted feeding (TRF) altered lipid and amino acid metabolite rhythmicity without perturbing clock gene expression in humans ¹⁷, and TRF could restore the metabolic defects in core clock-deficient livers ²⁵. These results indicate that noncanonical clock regulators are crucial in regulating diurnal rhythms and led us to hypothesize that genetic variants could genome-widely affect the binding of trans-regulatory factors, including both core clock genes and noncanonical clock regulators, thereby regulating rhythmic gene expression and physiological processes.

In addition to genetic impacts, epigenetic regulation can link environmental cues with intrinsic clock machinery to regulate diurnal rhythms. Enhancer activity, quantified by the intensity of histone modifications and enhancer RNA abundance, is tightly correlated with rhythmic gene expression ^26,27, highlighting the essential role of enhancers in driving gene rhythmicity. The oscillating enhancer activities and gene transcription can cause genome-wide remodeling in response to nutritional challenges ¹⁸, allowing us to perform transcription factor (TF) motif analysis on these remodeled enhancer sites and identify key regulators responding to environmental stimuli. Moreover, core clock components, such as BMAL1 and REV-ERBα, dynamically modulate the oscillating enhancer-promoter interaction to control rhythmic gene transcription ^28–30. However, the role of noncanonical clock regulators in mediating enhancer-promoter interactions and diurnal rhythms is largely unknown, especially under pathological conditions. The impact and underlying mechanism of genetic variation on the rhythmic enhancer-promoter interactions have barely been studied. This could provide insights into how the interactions between genetics and environmental pacemakers regulate diurnal rhythms.

Genome-wide association studies (GWAS) have shown that genetic variants within cis-regulatory elements (CREs) are associated with variations in human gene regulation, complex traits, and diseases ^31,32. Quantitative trait loci (QTL) are genetic loci that influence the phenotypic variation of a complex trait, often through genetic interactions with trans-regulatory factors and the environment ³³. The effects of genetic variants on mRNA abundance, also known as expression quantitative trait loci (eQTL), explain a proportion of GWAS signals related to traits and diseases ³⁴. However, the specific interactions between genetic variations and altered biological rhythms are largely unknown in humans and mice. Except for sleep disorders and chronotypes ^9,35, whether these interactions contribute to complex traits and diseases in humans have not been well-studied. We recently identified a class of genetic variations that contribute to rhythmic gene expression across various tissues ³⁶, while the underlying regulatory mechanisms still need to be defined.

Here, we incorporated the genetic variants and mRNA expression with computed time information and reported that single-nucleotide polymorphisms (SNPs) in CREs are highly associated with the variation in hepatic 24-hour oscillating gene expression. We further measured the oscillating gene expression, metabolites, and 3D enhancer-promoter interactions in mouse livers, which were collected every 4 hours throughout the day from mice with two different genetic backgrounds. Unexpectedly, we found that a majority of rhythmic genes and enhancer-promoter interactions (E-PIs) are interdependently regulated by genetics and nutritional challenges. Noncanonical clock regulators, such as Estrogen Related Receptor Gamma (ESRRγ), were enriched in these rhythmic E-P interacting sites. Liver-specific Esrrγ knockout disrupted rhythmic E-PIs and gene expression, involved in lipid droplet formation, lipid metabolism, and other metabolic processes, which are influenced by genetics and nutritional challenges. Furthermore, we mapped individual-specific E-PIs in steatosis livers (also known as metabolic dysfunction-associated fatty liver disease; MAFLD) from obese patients. We found that SNPs associated with rhythmic gene expression were enriched in these E-PI sites and show enrichment in ESRRγ binding motifs. These SNPs contribute to multiple metabolic traits related to liver diseases. Together, these results indicate the interdependent impact of genetics and nutrition on diurnal rhythms and provide an additional regulatory layer for diurnal rhythms responding to environmental stress. The impacts of genetic variation on peripheral diurnal rhythms also provide a novel explanation for SNP-trait association identified through GWAS, particularly for those related to metabolic traits and complex diseases. More broadly, the current study could pave the path for future investigations of time-dependent diagnostics and therapeutics to advance precision chronobiology.

RESULTS

Human genetic variations contribute to individual-specific rhythmic gene expression

To determine the regulatory effect of genetic variation on rhythmic gene expression, we recently established a bioinformatic method to assess the interaction between SNPs and the rhythmic expression of their nearby genes ³⁶. By integrating genetic variation information and RNA-seq data in human livers from the GTEx dataset, we identified 3962 rhythmic genes in individuals with specific genotypes, defined by genetic variations (including SNPs and small insertions/deletions) located near each rhythmic gene (Figure 1A). These rhythmic genes are nearly three times as many as rhythmic genes identified in the whole population, and only 97 rhythmic genes are not associated with genetic variations. The explanation for identifying more rhythmic genes by considering genetic variation information could be that the rhythmicity of the same gene in one subpopulation with a specific genotype was masked by non-rhythmic patterns in other subpopulations with different genotypes, and shows no rhythmic expression in the whole population. For example, PLIN2, which is required to form and maintain lipid droplets, is only rhythmic in the subpopulation with AA genotype at SNP rs578602, which involves the A to G variation, but not in other subpopulations (Figure 1B). The overall expression level of PLIN2 has no difference across different genotypes (Figure 1C), suggesting the regulatory mechanism of this SNP on PLIN2 differs from that of expression quantitative trait loci (eQTL), which mediates overall gene expression levels. To determine physiological processes that were affected by these SNP-associated rhythmic genes, we performed pathway enrichment analysis and found that the top enriched pathways are related to lipid metabolism (Figure 1D). Moreover, based on ChIP-seq data using anti-H3K27ac, H3K4me3, H3K4me1, H3K36me3, and H3K9me3 antibodies in the human liver from the EN-TEx project ³⁷ to define enhancer and promoter regions, we found that these genetic variants are enriched in active enhancer and promoter regions compared to random control SNPs (Figure S1A). As previous studies have shown tissue-specific rhythmic gene expression across various tissues ^15,16, we further examined whether the identified genetic variants contributing to an individual’s hepatic rhythmic gene expression are tissue-specific. We found that these liver rhythmic gene–associated genetic variants primarily influence rhythmic gene expression in the liver, but not in other tissues (Figure S1B). In line with this result, liver rhythmic gene–associated genetic variants show the strongest enrichment in GWAS loci associated with liver-related traits (Figure S1C). Together, these results demonstrate that SNPs in enhancer and promoter regions tissue-specifically contribute to the variation of rhythmic gene expression in human liver.

Figure 1. Human genetic variations contribute to individual-specific rhythmic gene expression.

(A) Overlap between rhythmic genes identified in genotype-specific subpopulations and those detected in the overall population. (B) SNP rs10116792 as a representative genetic variation that determines the rhythmic expression of PLIN2 in the liver of individuals with the AA genotype but not in those with other genotypes. Rhythmicity is determined by harmonic regression. (C) The average mRNA abundance across 24 hours of PLIN2 in subpopulations with specific genotypes. (D) Pathway enrichment analysis on rhythmic genes associated with genetic variation in the liver. See also Figure S1.

Distinct rhythmic transcriptomes in mice with different genetic backgrounds

Because direct genetic manipulation and repeating tissue collection over 24 hours are impractical in humans, we used C57BL/6J (B6) and 129S1/SvImJ (129Sv) mice to dissect the mechanism underlying the variation of gene rhythmicity under different genetic backgrounds. The genomes of these two strains differ by 5.2 million SNPs, a number similar to that observed between any two unrelated individuals in humans ³⁸ . Consistent with previous studies, B6 mice were more active and showed stronger rhythmic locomotor activities than 129Sv mice (Figure S2A). From liver samples collected every 4 hours around the day in both mouse strains, we observed more rhythmic transcripts in 129Sv mice than in B6 mice (Figures 2A, 2B and Table S1), although the peak expression of rhythmic transcripts is enriched in similar phases (Figure S2B). Only 450 out of 5316 rhythmic transcripts are differentially expressed between the two strains in their average expression level across the 24-hour period (Figures 2C, 2D, and Table S1), suggesting that most strain-specific rhythmic transcripts differ from the differentially expressed transcripts. Consistent with genetic variation-mediated rhythmic genes in humans, we also found that lipid metabolism-related pathways are enriched in differential rhythmic genes (Figure 2E), while interferon signaling pathways are enriched in differentially expressed genes across strains (Figure S2C). To determine whether the individual-specific rhythmic genes are aligned with mouse strain-specific rhythmic genes, we overlapped these rhythmic genes in two species and found that around one-third of them are rhythmically conserved in both species. These conserved rhythmic genes are enriched in similar biological processes, such as metabolism, protein processing in the endoplasmic reticulum (ER), and fatty acid metabolism (Figure 2F), which is similar to the mouse strain-specific rhythmic genes. Although humans do not live under exact 12-hour light: 12-hour dark (LD) conditions due to artificial lighting and modern lifestyles, we still observed human-mouse orthologous rhythmic gene pairs in metabolism pathway exhibit opposite phases between two species (binomial test, p = 9.0 × 10–7) (Figure 2G and Table S1). Thus, these data provide experimental evidence for distinct diurnal rhythmic transcriptomes in peripheral liver tissue from mice with different genetic backgrounds, without changes in average expression levels. The rhythmic genes also exhibit conserved pathway enrichment across species.

Figure 2. Distinct rhythmic transcriptomes in mice with different genetic backgrounds.

(A and C) Heatmaps showing Z-score–normalized expression of rhythmic transcripts specific to B6 and 129Sv strains (A) and differentially expressed genes between the two strains (C). Rhythmic transcripts are defined using the criteria: Meta2d p-value ≤ 0.05, 20 ≤ period (t) ≤ 24 hours, and oscillation amplitude > 1.5. Differentially expressed transcripts were defined by a p-value ≤ 0.01 and an absolute log₂ fold change (|log₂FC|) > 1. (B and D) Venn diagram showing the overlap of rhythmic genes (B) and differentially expressed transcripts(D) between B6 and 129Sv strains. (E-F) Pathways enriched in mouse strain-specific rhythmic genes (E) and conserved rhythmic genes associated with genetic variants in both humans and mice (F). (G) Phase distribution of rhythmic genes enriched in metabolic pathways. Genes are grouped based on their phase in humans, with daytime on the left (6:00 – 18:00) and nighttime on the right (18:00 – 6:00). Each dot represents a human (red)–mouse (blue) orthologous gene pair (n = 82) with rhythmic expression and annotated in the metabolism terms shown in panel F. A binomial test revealed significant enrichment of gene pairs phased in opposite time windows between species (p = 9.0 × 10⁻⁷). In mice, daytime and nighttime phases were defined based on light-on and light-off periods, respectively. See also Figure S2 and Table S1.

Genetics-nutrition interactions control diurnal rhythmic genes and metabolic processes

In addition to being shaped by genotype, environmental factors are crucial to regulating diurnal rhythms, so-called “genotype × environment” (G × E) interactions. Recent major shifts in diet and lifestyle have led to dysregulation of regulatory programs of genotypes that evolved under past environmental conditions. To determine whether mice with different genetic backgrounds respond differentially to diet-induced obesity (DIO), we compared the rhythmic transcriptome in B6 and 129Sv mice fed with normal chow (NC) and a 12-week obesogenic diet. Thousands of transcripts show rhythmic expression in a strain-specific manner (Figures 3A, 3B, and Table S2). This strain-specific rhythmic remodeling in response to DIO is independent of locomotor activities since they are similar in DIO condition (Figure S3A). We observed B6 gained more body weight than 129Sv, which provides endocrine signals and may contribute to strain-specific diurnal rhythms in liver (Figure S3B). We also observed strain-specific differential expression levels of transcripts in response to DIO (Figure S3C and Table S2), which largely differ from differential rhythmic genes (Figure S3D and Table S2). Consistent with previous findings in livers of mice with nutritional challenges ^18,24, core clock genes, such as Bmal1, Rev-erbα, Rev-erbβ, Cry1, and Clock (Figures 3C and S3E) retain rhythmic expression across all conditions, and we defined these genes as common rhythmic transcripts. We also identified genes that are rhythmic in only one strain and remain unchanged under nutritional challenges, such as Cyp8b1 (Figure 3D), a key enzyme involved in bile acid synthesis ³⁹, such as glycocholic acid (Figure 3E). These are defined as strain-dominant rhythmic transcripts. The transcripts whose rhythmicity is affected in DIO regardless of the strains are defined as DIO-dominant rhythmic transcripts, as exemplified by Nnmt related to nicotinamide metabolism ⁴⁰ (Figure 3F). Indeed, metabolite nicotinamide is regulated by DIO regardless of strain background (Figure 3G).

Figure 3. Genetics-nutrition interactions control diurnal rhythmic genes and metabolic processes.

(A–B) Heatmaps showing rhythmic gene expression remodeling in response to DIO in B6 (A) and 129Sv (B) mice. (C) Diurnal expression pattern of Bmal1 as a representative rhythmic gene commonly affected by DIO in both strains. (D-E) Example of a strain-dominant rhythmic pathway, bile acid biosynthesis, featuring the representative gene Cyp8b1 (D) and its associated metabolite glycocholic acid (E). (F-G) Example of a DIO-dominant rhythmic pathway, nicotinate and nicotinamide metabolism, featuring the representative gene Nnmt (F) and the associated metabolite nicotinamide (G). (H-I) Example of a strain × DIO interaction-dependent pathway, urea cycle and amino acid metabolism, highlighting the representative gene Acy1 (H) and the metabolite N-acetyl-glutamine (I). All metabolite data are presented as mean ± SEM, n = 3 per group. Statistical significance: p < 0.01 (**), p < 0.05 (*). (J) Classification of rhythmic transcripts based on changes in rhythmicity, defined by Meta2d p-value ≤ 0.05, 20 ≤ period (t) ≤ 24 hours, and oscillation amplitude > 1.5. Representative expression patterns are shown for each category. (K) Proportions of rhythmic transcripts in each category. (L) Pathway enrichment analysis of rhythmic genes grouped by category. See also Figure S3 and Table S2.

Interestingly, we found that Acy1, which catalyzes the hydrolysis of N-acetylated amino acids ⁴¹, gained rhythmicity expression in DIO with 129 genetic background, while the expression of Acy1 showed arrhythmicity in both NC and DIO mice with B6 genetic background (Figure 3H). Notably, N-acetyl-glutamine, a crucial metabolite in the urea cycle and regulated by ACY1 ^42,43, also exhibits independent regulation by genetics and DIO (Figure 3I). Based on transcriptome-wide analysis, only 3.8% of rhythmic transcripts were rhythmic across all conditions, including most of the core clock genes (Figures 3J and 3K). Strain-dominant transcripts represented 8.8% of rhythmic transcripts that were enriched in ER protein processing and bile acid biosynthesis. DIO dominantly regulated 3.5% of rhythmic transcripts, in which nicotinate and nicotinamide metabolism are enriched (Figures 3K and 3L). Remarkably, 83.9% of rhythmic transcripts were interdependently regulated by strain and DIO and involved multiple patho/physiological processes, including triglyceride biosynthesis, urea cycle, and Keap1-Nrf2 pathways (Figures 3K, 3L, and S3F). In sum, the integrative transcriptomic and metabolomics analyses demonstrate that the interdependence of the genetic background and DIO regulates the majority of rhythmic transcripts and biological processes.

Enhancer-promoter dynamics underlie genetics-nutrition interactions in diurnal gene expression

To further explore the molecular mechanisms underlying the observed strain and DIO interdependent rhythmic transcripts, we performed in situ chromatin interaction analysis with paired-end-tag sequencing (ChIA-PET) using anti-H3K27ac antibodies to determine enhancer-promoter interactions in liver collected at ZT10 and ZT22 from NC and DIO mice with B6 and 129Sv genetic backgrounds (Figure 4A) and E-PIs were identified based on ChIA-PET data. Compared with Hi-C ⁴⁴, in situ ChIA-PET has: (1) a higher resolution for detecting E-PI; (2) the ability to detect associations with a protein of interest on the chromatin looping for functional studies; and (3) sharp narrow peaks that improve the prediction of TF-binding sites ⁴⁵.

Figure 4. Enhancer-promoter dynamics mediate genetics-nutrition interactions in diurnal gene expression.

(A) Overview of in-situ ChIA-PET experiment. Detailed descriptions can be found in Extended Experimental Procedures. (B) Proportion of time-dependent ChIA-PET sites affected by genetic background and DIO. (C-D) In-situ ChIA-PET identified time-dependent E-PI sites. Heatmaps show strain × DIO interaction-dependent E–PI sites in DIO-B6 (C) and DIO-129Sv (D) mice. (E) Percentage of E-P sites containing one or more SNPs between B6 and 129Sv strains. (F) Enrichment of TF binding motifs at anchor regions of time-dependent E–PI sites. The enriched motifs are identified using IMAGE. (G) Diurnal expression pattern of ESRRs in control and DIO B6 mice (mean ± SEM, n = 3 per group). (H) Diurnal expression pattern of Esrrγ in DIO mice with B6 and 129Sv genetic backgrounds (mean ± SEM). See also Figure S4 and Table S3.

We identified a total of 35, 311 rhythmic enhancer-promoter looping sites in livers from NC and DIO mice with B6 and 129Sv backgrounds and observed common, DIO-dominant, strain-dominant, strain × DIO-interdependent enhancer-promoter looping sites (Figure 4B). Consistent with rhythmic gene expression, strain and DIO-dependent enhancer-promoter looping sites represented 80.8% of rhythmic E-P looping sites (Figures 4B - 4D, S4A, S4B, and Table S3). For example, there are greater fold changes of E-PIs between ZT22 and ZT10 associated with Plin2 gene in 129Sv mice than in B6 mice (Figure S4C). As expected, SNPs are more enriched in strain-dominant sites compared to other groups, indicating genetic effects in strain-dominant peaks, highlighting the genetic basis of strain-specific regulatory differences (Figure 4E).

To determine the functional role of rhythmic E-PIs and identify the TFs that bind these enhancers and promoters, we performed TF motif analysis using differential rhythmic enhancers in three mouse nutrition groups: 1) NC-B6 vs. NC-129Sv, 2) DIO-B6 vs. NC-B6, and 3) DIO-129Sv vs. NC-129Sv (Figure 4F). We identified that the noncanonical clock regulator PPAR was enriched in DIO-B6, which we also identified in our previous circadian enhancer mapping experiments using global run-on sequencing in MAFLD livers ¹⁸. The common motif that was enriched in the above three groups was Estrogen-Related Response Element (ESRRE) motif (Figure 4F), suggesting a critical role of ESRR in regulating the rhythmic expression in strain-specific and DIO-specific contexts. ESRRs belong to a small subfamily of nuclear receptors consisting of three members: ESRRα, β and γ. ESRRα is highly expressed in the liver and regulates mitochondrial activity, biogenesis, and turnover ⁴⁶. The expression of ESRR β and γ is very low in liver ⁴⁷. It has been reported that there is functional redundancy of ESRRα and γ in heart and skeletal muscle ^48,49. To determine which ESRR member binds to the ESRRE motif, we quantified the diurnal gene expression of Esrrα, β, and γ in healthy and MAFLD mouse livers. Only the rhythmic expression of Esrrγ was enhanced in MAFLD livers (Figure 4G), and its expression level was higher in livers of DIO-129Sv mice than DIO-B6 mice (Figure 4H). Together, these results indicate that ESRRγ is a candidate regulator of time-dependent E-PI interactions and rhythmic gene expression.

ESRRγ controls the strain-specific rhythmic gene expression in response to DIO

To further unbiasedly validate whether ESRRγ mediates these rhythmic enhancer-promoter looping, we performed transcription factor binding similarity screening based on all published liver cistrome using the CistromeDB ⁵⁰ and our generated ESRRγ ChIP-seq data in this study. In addition to core clock genes, including ARNTL (also known as BMAL1), PER2, CRY1, and CRY2, and noncanonical clock regulators, including HNF4α, USF1, PPARα, STAT5β ^18,51–54, ESRRγ is the top transcription factor bound on the enhancer-promoter looping sites (Figure 5A). Although HNF4α and PPARα were also identified as one of the key TFs, ESRRγ has distinct chromatin binding sites from HNF4α and PPARα, as well as ESRRα, another nuclear receptor in ESRR family that is expressed in liver (Figure S5A). Even in the fatty livers from DIO-B6 mice, with an increased number of ESRRγ chromatin binding sites, only one-third of these sites overlap with PPARα binding sites (Figure S5B), further indicating distinct roles of hepatic ESRRγ from other hepatic nuclear receptors.

Figure 5. ESRRγ mediates the interplay between genetic background and nutrition in circadian transcriptional regulation.

(A) GIGGLE score–based similarity analysis between ESRRγ chromatin binding sites and H3K27ac-associated rhythmic E-P loops. (B) Proportion of H3K27ac-associated E–P loops that overlap with ESRRγ binding–associated loops (left), and percentage of these loops disrupted following ESRRγ knockout (right). (C) Proportion of ESRRγ ChIP-seq binding sites that contain one or more SNPs between B6 and 129Sv mouse strains. (D) Representative ChIA-PET example showing ESRRγ-associated chromatin looping at the Plin2 locus. (E) Diurnal expression of Plin2 in wild-type and ESRRγ knockout mice under 129Sv and B6 genetic backgrounds. (F) Comparison of hepatic TG levels between control and Esrrγ knockout mice under NC and DIO conditions. Data are shown as mean ± SEM. * p < 0.05; ** p < 0.01. (G and L) Percentage of ESRRγ-dependent rhythmic transcripts (left) and the proportion of these transcripts that are regulated by ESRRγ through different mechanisms (right) in B6 (G) and 129Sv (L) mice. Among the rhythmic transcripts disrupted in DIO-B6, ESRRγ exerts regulatory control via three distinct mechanisms: direct binding, chromatin looping with co-regulators (e.g., SREBF1), and downstream transcriptional effectors (e.g., USF1). (H-I and M-N) SREBF1 and NFE2L2 are identified as co-regulatory factors of ESRRγ that modulate rhythmic transcription in B6 (H-I) and 129Sv (M-N) mice. Diurnal expression patterns of SREBF1 (H) and NFE2L2 (M), along with their putative target genes (I and N), are shown. (J-K and O-P) USF1 and STAT1 were identified as downstream effectors of ESRRγ that contribute to rhythmic transcription in B6 (J-K) and 129Sv (O-P) mice. Diurnal expression patterns of USF1 (J) and STAT1 (O), along with their respective putative target genes (K and P), are shown. See also Figure S5.

To determine the role of ESRRγ in controlling diurnal enhancer-promoter interactions and transcriptomes to induce strain-specific responses to over-nutrition, we performed ChIA-PET using anti-ESRRγ antibodies in livers from DIO-B6 mice. We found that 48% of H3K27ac-associated E-PI loops overlapped with ESRRγ-associated loops (Figure 5B). Moreover, hepatocyte-specific knockout of ESRRγ disrupts 68% of the above E-PI loops, indicating the direct role of ESRRγ in forming or maintaining these E-PI loops (Figure 5B). The strain-specific ESRRγ binding sites in DIO mice were highly enriched for the occurrence of SNPs between 129Sv and B6 mice (Figures 5C and S5C). For example, Plin2 is rhythmic in livers of 129Sv mice but not B6 mice, and the enhancer-promoter interaction is also higher in DIO-129Sv than in DIO-B6. SNP rs33020361 within the ESRRE DNA binding motif is located in ESRRγ-associated looping anchors (Figure 5D), suggesting that this genetic variation could be a regulatory variant causing strain-specific ESRRγ-associated looping. Indeed, hepatocyte-specific knockout of Esrrγ disrupts the rhythmic expression of Plin2, a lipid droplet protein that plays a crucial role in lipid metabolism and storage, in livers from DIO-129sv (Figure 5E). In sum, these results indicate that transcription factor ESRRγ is a key factor in mediating enhancer-promoter interactions and regulating strain-specific rhythmic gene transcription in response to DIO.

Although the physiological functions of ESRRγ in the liver are still not well-studied due to a lack of liver-specific knockout mouse models, ESRRγ has been studied in the brain, brown adipocytes, and kidney using tissue-specific knockout mouse models and is generally related to metabolism and mitochondrial regulation ^55–57. Liver-specific Esrrγ knockout did not lead to hepatic triglyceride accumulation in normal chow-fed mice but caused 1.88- and 2.61-fold increases in hepatic triglyceride levels in DIO-B6 and DIO-129Sv mice, respectively (Figure 5F). However, the mitochondrial oxygen consumption rate in liver tissue and isolated mitochondria did not show a significant difference between control and Liver-specific Esrrγ knockout with both strains (Figure S5D), indicating that the function of ESRRγ is different from its role in other tissues. To dissect the role of hepatic ESRRγ, we first performed RNA-seq in livers collected every 4 hours from liver-specific ESRRγ knockout DIO mice with B6 or 129Sv genetic background and then determined ESRRγ-dependent rhythmic genes and differentially expressed genes. We found that 58.6% and 72.3% rhythmic genes, identified respectively in DIO-B6 and DIO-129Sv, are ESRRγ dependent (Figures 5G, 5L, and S5E), while less than 12% of genes are differentially expressed either in all time points pooled samples or in individual time points (Figure S5E). Moreover, liver-specific ESRRγ overexpression in DIO-B6 mice is not sufficient to restore ESRRγ-dependent genes in DIO-129Sv mice. These results confirmed that ESRRγ is a key regulator for strain-specific rhythmic gene expression, but does not affect global hepatic gene expression.

To understand how ESRRγ regulates its target rhythmic gene expression, we integrated the ChIP-seq analysis with ESRRγ antibody to identify the direct regulation of ESRRγ-dependent rhythmic genes. We incorporated ChIA-PET analysis with ESRRγ antibody and transcription factor motif analysis using integrated analysis of motif activity and gene expression (IMAGE) ⁵⁸ to identify co-regulators that could recruit ESRRγ via chromatin looping for regulating ESRRγ-dependent rhythmic genes. For livers from DIO-B6 mice, we identified promoters of 98 ESRRγ-dependent rhythmic transcripts that were bound by ESRRγ, suggesting a direct regulation (Figure 5G). Promoters of 232 ESRRγ-dependent rhythmic transcripts did not contain ESRRγ binding sites, but ESRRγ-associated long-distance enhancers interacted with their promoters based on ChIA-PET analysis (Figure 5G). Further transcription factor motif analysis identified the motif of SREBP1, a key regulator for lipid synthesis ⁵⁹, was enriched in these E-PI sites, suggesting that SREBP1, acting as a co-regulator, could recruit ESRRγ to regulate the rhythmic expression of these genes. Indeed, the average expression of SREBF1 target genes and itself expression were rhythmic (Figures 5H and 5I). For ESRRγ-dependent rhythmic genes, whose promoters did not contain ESRRγ-binding sites and were not looped to ESRRγ-associated enhancers, we performed motif enrichment analyses on their promoter and associated enhancer region to identify putative downstream effectors of ESRRγ. For example, USF1, an important modulator of core clock gene BMAL1 transcription activity ⁵⁴, was identified as one of the putative TFs responsible for mediating the expression of ESRRγ-dependent rhythmic genes (Figures 5G and 5J). The expression of Usf1 is also ESRRγ-dependent (Figure 5K) in DIO-B6 mice, further indicating its role as a downstream effector of ESRRγ. Similarly, the rhythmic expression of 434 transcripts was directly regulated by ESRRγ (Figure 5L). NFE2L2 (also known as NRF2), a crucial transcription factor in cellular stress responses ⁶⁰ was identified as a co-regulator that mediate the enhancer-promoter interaction for the rhythmic expression of 598 ESRRγdependent transcripts (Figures 5L, 5M, and 5N). Moreover, STAT1, playing key roles in immune response, was identified as one of the transcription factors for the indirect regulation of ESRRγ-dependent rhythmic transcripts in DIO-129Sv mice (Figure 5L). In line with this finding, the putative targets of STAT1 show rhythmic expression in livers from DIO-129Sv mice, and the rhythmic expression of STAT1 is ESRRγ-dependent (Figures 5O and 5P). These results show that the lipid metabolism regulators, such as SREBF1 and USF1, coordinate with ESRRγ to regulate DIO-B6-specific rhythmic genes, whereas stress and inflammation regulators, including NRF2 and STAT1, collaborate with ESRRγ to regulate DIO-129Sv-specific rhythmic genes. These findings provide potential mechanisms for how ESRRγ regulates specific rhythmic genes under different genetic backgrounds in response to diet-induced obesity.

Hepatic ESRRγ regulates strain-specific rhythmic metabolic processes in response to DIO

To define the pathophysiological function of ESRRγ and given that ESRRγ regulated the rhythmic expression of lipid metabolism-related genes from the above transcriptomic analyses, we first performed Oil Red O staining. Interestingly, we found that the size of liver lipid droplets shows rhythmicity in DIO-129Sv mice but not in DIO-B6 mice (Figures 6A, 6B, S5G, and S5H). In line with the finding that the rhythmic expression of the lipid droplet formation gene, Plin2, is ESRRγ-dependent (Figure 5E), the rhythmic size of lipid droplets in livers from DIO-129Sv mice is also ESRRγ-dependent (Figures 6A and 6B). To further characterize the impacts of hepatic ESRRγ in lipid metabolism, we quantified 1487 lipid species in control and hepatocyte-specific Esrrγ knockout livers collected at ZT10 and ZT22. We identified 105 and 243 rhythmic lipids, respectively, in DIO-B6 and DIO-129Sv mice (Figure 6C). Among these lipids, ceramide exhibited rhythmicity only in DIO-B6 mice, while diglycerides (DG), which are related to insulin resistance ⁶¹, phosphatidylethanolamine (PE) and triglycerides, which are major components of lipid droplets, showed more rhythmicity in DIO-129Sv mice (Figures 6C, 6D, and Table S4). Although TG was accumulated in Esrrγ deficient livers of DIO-129Sv mice at ZT10, a larger size of lipid droplets was observed in control livers at ZT10, indicating the large lipid droplets are not dependent on TG accumulation. Furthermore, metabolomic analysis in control and hepatocyte-specific Esrrγ knockout livers, collected every 4 hours over a 24-hour period from DIO-B6 and DIO-129Sv mice, revealed that ESRRγ is essential for multiple rhythmic metabolic processes, such as rhythmic aspartate metabolism in DIO-B6 mice and urea cycle and branch chain amino acid metabolism in DIO-129Sv mice (Figures 6E - 6H and Table S4). These results indicate that Hepatic ESRRγ regulates strain-specific rhythmic metabolic processes in response to DIO.

Figure 6. Esrrγ-dependent rhythmic and strain-specific regulation of hepatic lipid metabolism in DIO mice.

(A) Representative Oil Red O staining of liver sections collected at ZT10 and ZT22 from Control (Ctrl) and liver-specific Esrrγ knockout (Esrrγ LKO) 129Sv mice under DIO conditions. Scale bar, 20 μm. Enlarged views of selected areas from the images above are shown below. (B) Quantification of lipid droplet size in Ctrl and Esrrγ LKO 129Sv mice at ZT10 and ZT22 under DIO conditions by using Image J (data are expressed as mean with SEM, **p<0.01). (C) Heatmaps showing ESRRγ-dependent rhythmic lipid species in DIO-B6 and DIO-129Sv mice. (D) Heatmap showing time-dependent changes in triglyceride (TG) levels at ZT10 and ZT22 in DIO-B6 and DIO-129Sv mice upon liver-specific Esrrγ knockout. (E) Heatmaps showing ESRRγ-dependent rhythmic metabolites in DIO-B6 and DIO-129Sv mice. Rhythmic metabolites are defined as those with a Meta2d p-value < 0.05, 21≤ period≤ 24 hrs, and a peak-to-trough amplitude > 1.5. (F) Enriched metabolic pathways in ESRRγ-dependent rhythmic metabolites specific to DIO-B6 (upper) and DIO-129Sv (down) mice. (G–H) Temporal abundance of representative ESRRγ-dependent metabolites specific to DIO-129Sv mice, urea (G), valine (H). See also Figure S5 and Table S4.

Human rhythmicity-related genetic variants are associated with E-P interactions and lipid metabolism traits

Although the conservation of non-coding sequences is relatively poor between humans and mice, emerging studies indicate that conserved regulatory regions across species are enriched for SNPs that regulate metabolic processes ^31,62,63. To determine whether genetic variants that contribute to variations in human rhythmic gene expression are also enriched in enhancer-promoter looping sites, we mapped the E-PIs using Micro-C ⁶⁴ in human MAFLD liver specimens (Figures 7A, S6A-S6C, and Table S5). The overall topologically associating domain boundaries were conserved among individuals as expected (Figure S6D), and we also observed a mean of 5,202 individual-specific E-PIs (Figure 7B and Table S4). We recently genome-widely identified genetic variations that contribute to the rhythmic gene expression in various human tissues ³⁶. To determine whether these genetic variations were enriched in enhancer-promoter looping sites, we compared the occurrence of rhythmic gene-associated genetic variants in these sites with random genetic variations as a control and found that these rhythmic gene-associated genetic variants are enriched in these sites (Figures 7C and 7D). To identify trans-regulatory factors that convert genetic information to rhythmic gene expression, we performed transcription-factor motif analysis in the 500 bp regions surrounding genetic variants associated with rhythmic genes and located in E-P looping sites (rlGVs). Lipid metabolism-related transcription factors, including ESRRγ, which was enriched in 49.7% of rlGV-flanking regions, as well as HNF4α, PPARα, and CEBP, were identified (Figure 7E). Indeed, HNF4α, PPARα, and CEBP have been reported to mediate rhythmic lipid metabolism in response to DIO ^18,51,53,65. We found some of the rlGVs that are associated with the rhythmic expression of genes, such as MYC, JUND, CYP4F3, are also located in the ESRRγ DNA binding motif (Figure S7). These chromatin loci harboring the above SNPs suggest potential cis-regulatory effects of these SNPs on rhythmic gene expression by affecting ESRRγ DNA binding. However, more SNPs associated with rhythmic gene expression in subpopulations with specific genotypes are not located in specific TF DNA binding motifs, which indicates a trans-regulatory mechanism for such regulation.

Figure 7. Human genetic variations associated with rhythmic gene expression are enriched in E-P interaction sites and contribute to lipid metabolism traits.

(A-B) Capture of enhancer-promoter interaction loops in human fatty liver using Micro-C. (A) Summary of donor information. (B) Overlap of identified E-PI loops across three donors. (C) Mapping of genetic variants associated with rhythmic genes that are located in E–PI sites, referred to as rhythmic loop-associated genetic variants (rlGVs). (D) Enrichment of rhythmic gene-associated genetic variants in chromatin loop anchors. (E) TF motif enrichment analysis on rlGV-flanking regions. ±500 bp regions surrounding each rlGV were extracted, and transcription factor binding motifs enriched in these regions were identified using HOMER. (F) Diseases associated with rlGV-related genes. (G) Enrichment of rlGVs in 20 metabolic-related traits estimated by stratified LDSC. The p-value indicates enrichment of rlGVs for the heritability of a particular trait. (H-I) Partitioned heritability (H) and per-SNP heritability (I) mediated by rlGVs across 20 metabolic-related traits. The per-SNP heritability is estimated using parameter τ from LDSC regression. See also Figures S6, S7 and Table S5.

Pathway analysis indicates lipid metabolism is enriched in rlGV-related genes (Figure S6E). Further disease association analysis showed that rlGV-related genes are associated with multiple liver diseases, including fatty liver disease, cholestasis, and hyperlipidemia (Figure 7F). To determine whether these rlGVs contribute to lipid-related traits, we performed a stratified linkage disequilibrium score (LDSC) regression analysis to quantify the enrichment of rlGVs for the heritability of 20 lipid-related traits (Table S6), including apolipoprotein, cholesterol, and triglyceride levels. These rlGVs were significantly enriched for heritability of these traits (Figure 7G), and the proportions of heritability mediated by rlGVs are around 0.06 (Figure 7H), which were comparable to that of eQTLs and splicing QTL (sQTLs) ⁶⁶. Moreover, we found that the per-SNP heritability mediated by rlGVs, as estimated by LDSC regression, was significantly higher than control SNPs (Figure 7I). Overall, the analyses suggest that rlGVs play a distinct role in lipid metabolism-related traits by regulating rhythmic gene expression.

Discussion

In this study, we found that genetic variants located in enhancers and promoters contribute to the variation of diurnal rhythms. Remarkably, noncanonical clock regulators bridge the connection between genetics and nutritional challenges to regulate more than 80% of rhythmic transcripts and enhancer-promoter interactions. Moreover, human SNPs associated with rhythmic gene expression are enriched in enhancer-promoter looping sites. These SNPs contribute to liver-related metabolic traits and diseases, including lipid and cholesterol metabolism and VLDL-mediated lipid secretion.

Because it is not practical to sample tissues (except blood) from humans in a circadian fashion, our understanding of the role of genetic variation in diurnal rhythm is largely from GWAS of chronotypes studies ^9,67,68. These studies explored the contributions of genetic variations to dysregulation of sleep and related illnesses, and discovered common genetic variants associated with complex sleep-related phenotypes in humans. However, how genetic variants contribute to the variation of diurnal rhythms in peripheral tissues is largely unknown. Moreover, although these GWAS studies identified some novel genes, such as AK5, APH1A, and ERC2 ³⁵, the functions of these noncanonical clock regulators still need to be further studied. The underlying mechanism of how noncanonical clock regulators contribute to the variation of diurnal rhythms remains unknown. Here, we demonstrate that human genetic variants contribute to variations of thousands of rhythmic genes in the liver by applying our newly established algorithm. By integrating with chromatin coordinates of genetic variants and genome-wide enhancer-promoter interactions, we found these variants were located in the cis-regulatory elements of their associated rhythmic genes. The following TF motif analysis identified that noncanonical clock regulators, including ESRRγ, were enriched in these loci, and liver-specific knockout of ESRRγ could change the rhythmicity of target genes. Therefore, these findings provide a new mechanistic perspective on how genetic variants interact with noncanonical clock TFs to regulate rhythmic gene expression, contributing to the variations of diurnal rhythms in peripheral tissues. Future investigations of the function of the identified genetic variants from humans in the proper contexts will likely yield important insights into biomarkers for disease development and chronotherapy, such as optimizing the time for meal and drug administration.

Transcription factors play crucial roles in mediating E-PI and shaping the 3D chromatin conformation, influencing gene regulation via several mechanisms, including 1) bridging enhancer and promoter by oligomerization by themselves or recruiting cofactor; 2) recruiting chromatin modifier and histone modifiers to alter local chromatin state; and 3) defining topological associating domains and boundaries, such as CTCF ⁶⁹. Based on our integrative analyses of RNA-seq, ChIP-seq, ChIA-PET analysis, and motif mining, ESRRγ can bind to promoter regions and recruit other strain-specific cofactors, such as SREBF1 and NRF2, which could fit the mechanisms of bridging enhancer and promoter by oligomerization. The SNPs could disrupt or enhance the chromatin bindings of ESRRγ and its cofactors. Regarding recruitment of chromatin modifiers and histone modifiers, ESRRγ has been reported in the complex of PPAR-gamma coactivator (PGC)-1 and chromatin-modifying enzymes like the p300 and SRC-1 histone acetyl transferases in other tissues ⁷⁰. In the liver, ESRRγ may also affect chromatin state and disrupt H3K27ac-associated E-PIs, as indicated by our ChIA-PET results. Although we did not observe changes in CTCF expression following ESRRγ knockout, our RNA-seq analysis showed that the expression of some mediator complex components was altered in DIO-129Sv mice. In summary, ESRRγ may influence rhythmic 3D genome remodeling through multiple mechanisms, and different ESRRγ-dependent rhythmic genes may be regulated in distinct ways.

In addition to the identification of an unexpectedly large number of rhythmic genes in humans and mice with different genetic backgrounds, we also unexpectedly found that a majority (above 80%) of rhythmic genes are interdependent on the regulation of strain and DIO. These results suggest a key role in the interactions between environmental cues and genetic backgrounds in the regulation of gene rhythmicity. Emerging evidence indicates that environmental stimulations and internal patho/physiological cues, including diet, mealtime, exercise, aging, cancer, and microbiota, can reprogram the rhythmic transcriptome in liver ^{21–23,71,72}. These studies were well-conducted by sampling tissues every 3 or 4 hours around the clock in a stable environment in mice with the same genetic background. The present study provides one more dimension about the genetic background in these contexts. We expect there will be different responses to the above environmental stimulations and internal patho/phyological cues under different genetic backgrounds. The genome-wide enhancer-promoter interaction mapping allows us to determine the trans-regulatory factors and cis-regulatory elements that directly regulate the variations of diurnal rhythms in DIO. In addition to SNPs located in well-defined TF binding motifs, we also found that the E-P interacting sites for some DIO and strain-independent rhythmic genes do not contain genetic variants, which may be due to indirect trans-regulatory TF effects and cannot be directly identified in cis-regulatory variants. These results provide experimental evidence for the consideration of the contribution of gene-environment interactions to phenotypic variations of complex traits in future GWAS analysis, especially for 24-hour rhythm-related traits, which can be reprogrammed by environmental cues and genetic variants.

Accumulating evidence suggests that the disruption of diurnal rhythm, including jet-lag, shift work, and other attributes of unhealthy lifestyles, accelerates the development of liver diseases such as MAFLD, hepatitis, cirrhosis, and HCC ^73–75. Of note, lipid, cholestasis, and HDL levels, which are critical for the development of above MAFLD-related diseases ^76,77, are enriched in the genetic variants associated with rhythmic genes in current studies, indicating the various diseases risks in subpopulations with specific genotypes. Chronotherapy includes chrono-nutrition and chrono-pharmacology, which refers to the principles of providing meals and administering drugs at an optimized time of the day to maximize the beneficial metabolic or drug effect while reducing detrimental metabolic outcomes or drug toxicity ^1,78. Some studies and clinical trials indicate that time-restricted eating within an 8-hour period could lead to body weight loss and improve metabolic syndromes ^79–81. However, several other studies reported that time-restricted eating was not effective for weight loss and did not improve metabolic syndrome ^82,83. Similar inconsistent chrono-studies and clinical trials were also found for optimizing the use of statins in the evening to maximize the effects of cholesterol synthesis ⁸⁴, but several studies have reported that dosing time does not affect the drug efficiency ^85,86. In light of our observation of the interplay between genetic variation and nutrition challenges in regulating diurnal rhythmic gene expression, individual genetic backgrounds in the examined population should be taken into consideration to address the inconsistency of these studies. Future efforts, along with the current study, extend beyond liver rhythmic physiology and are expected to pave the path for personalized chronotherapy regarding optimized meal and drug administration timing based on individual genetic background.

Our studies were majorly focused on lipid metabolism, while metabolomic data also indicated rhythmicity of other pathways, such as bile acids, urea cycle, and BCAA metabolism, which were also interdependently regulated by genetic backgrounds and nutrition and are ESRRγ-dependent. As these metabolic processes are crucial for maintaining physiological homeostasis and are extensively studied as therapeutic targets, our findings on their rhythmicity under different genetic backgrounds pave the way for advancing precision chronotherapy by incorporating time-dependent effects and patients’ genetic backgrounds.

Limitations of the study

Our current study highlights the impacts of genetics-nutrition interactions in understanding obesity-associated disease susceptibility. These interactions could have broader impacts on other metabolic disorders and liver diseases. For example, lipid metabolism dysregulation is not limited to obesity but is also central to alcoholic liver diseases. Recent studies indicated that ESRRγ affects alcoholic liver injury ⁸⁷. It would be interesting to investigate the independent effects of genetic background and alcohol on the progression of alcoholic liver disease. In addition to genetic variations, factors in other organs, such as the microbiota ^88,89, may also contribute to the observed differences since we observed that rhythmic bile acids, which influence microbiota composition through antimicrobial activity and activation of host signaling pathways that maintain gut homeostasis, are regulated by ESRRγ in 129-DIO mice ⁹⁰. These potential inter-organ crosstalks merit further investigation in future studies.

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dongyin Guan (dongyin.guan@bcm.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• RNA-seq, H3K27ac ChIA-PET, ESRRγ ChIA-PET data, ESRRγ ChIP-seq data, ATAC-seq data and processed files are available at GSE278920, GSE278921, GSE278922, and GSE297115.

• Raw lipidomic data, metabolomic data and processed files are available in Table S2 and Table S4.

• Source data for all figures are available in Data S1.

• Any additional information required to reanalyze the data reported in this paper is available from the Lead Contact upon request.

STAR★METHODS

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice model

All animal experiments were performed following the procedure approved by the Baylor College of Medicine Institutional Animal Care and Use Committee. Wild-type C57BL/6J (B6) and 129S1/SvImJ (129Sv) male mice were obtained from Jackson Lab. Esrrγ^f/f mice were originally established in C57BL/6J genetic background by the Ronald Evans Laboratory. Esrrγ^f/f mice with 129S1/SvImJ genetic background were generated by crossing 10 generations of Esrrγ^f/f mice with B6 genetic background with wild-type 129Sv mice. The 129S1/SvImJ genetic background of Esrrγ^f/f mice was verified by genotyping in the Jackson Laboratory. All mice were housed in Animal Core Facilities under a 12-hour light and 12-hour dark cycle (lights on at 7 a.m. [zeitgeber time 0, ZT0] and lights off at 7 p.m. [ZT12]). For the nutritional challenge, diet-induced obesity (DIO) mice were fed for 12 weeks with a 60% high-fat rodent diet starting at 8 weeks of age. Age-matched control mice were fed with normal chow. Baylor College of Medicine (BCM) vector core generated the AAV vectors (AAV8-TBG-GFP for control and AAV8-TBG-Cre for liver-specific Esrrγ knockout). AAVs were intravenously injected with 2 × 10¹¹ genome copies (GC) per mouse at 8 weeks of age prior to high-fat diet feeding. Livers were collected every 4 hours over a 24-hour period and immediately flash-frozen in liquid nitrogen for subsequent processing.

METHODS DETAILS

Mice locomotor activity monitoring

Mice locomotor activity was measured using the comprehensive laboratory animal monitoring system (CLAMS) in the Mouse Metabolism & Phenotyping core at BCM. NC or DIO mice were habituated for one week to single housing with bedding material and fed with normal chow or high-fat diet. Water was provided ad libitum via standard water bottles. The mouse locomotor activities were monitored continuously for 4 days. Data are expressed as the mean ± SEM (n = 5 per group).

Mapping of genetic variations associated with rhythmic gene expression

The method for mapping genetic variants associated with rhythmic gene expression followed our previously established approach for identifying rhyQTLs across different tissue types ³⁶. In brief, genotype data from the GTEx project were downloaded from dbGaP under accession number phs000424.v8.p2. Genetic variants, including single-nucleotide polymorphisms (SNPs) and small insertions/deletions (indels), within ±1 Mb of each protein-coding gene’s transcription start site (TSS) were extracted, excluding those with fewer than 50 samples in two or more genotype-defined subpopulations. To evaluate the interaction between genetic variation and rhythmic gene expression, expression data were normalized and covariate-adjusted using linear regression to reduce confounding and improve detection accuracy. Covariates included five genotype principal components (PCs), sequencing platform, library preparation protocol, donor sex, ischemic time, age, and type of death. For each genetic variation, we assessed the rhythmicity of its nearby gene in each genotype group using harmonic regression ¹⁰⁸. If the nearby gene’s rhythmicity met the criteria of a p-value ≤ 5 ×10⁴ and a peak-to-trough fold change ≥ 1.5 in at least one genotype group, we further evaluated the differential rhythmic expression across genotypes.

During differential rhythmicity analysis, a downsampling strategy was applied to mitigate potential biases from unequal sample sizes. First, the two genotype groups with the largest sample sizes were selected. The larger group was then downsampled to match the sample size of the smaller group, ensuring balanced representation. This process minimized data loss while preserving comparability between groups. To reduce bias from single-time sampling, downsampling was repeated 20 times. Differential rhythmicity was assessed using the R package dryR ⁹⁴, and rhythmic pattern consistency across iterations was evaluated using a G-test to determine whether the frequency of fitted models deviated significantly from random expectation. In parallel, harmonic ANOVA (HANOVA) ¹⁰⁹ was used to quantify rhythmic differences between different genotype groups, and p-values were adjusted using the Benjamini–Hochberg (BH) procedure. Genetic variants were considered associated with rhythmic gene expression if nearby genes showed genotype-dependent rhythmicity, with significance thresholds of p < 0.05 (G-test) and q (BH) < 0.05 (HANOVA). To determine biological processes involved in the interested genes, pathway enrichment analysis was performed using Enrichr ¹¹⁰, and disease association analysis was performed using Metascape ¹⁰⁵.

Enrichment of genetic variations associated with rhythmic gene expression in CREs

To determine whether genetic variations associated with rhythmic gene expression are enriched in CREs, a permutation test was used to calculate the enrichment. In brief, we obtained normalized histone modification signals in candidate CREs from ChIP-seq data using anti-H3K27ac, H3K4me3, H3K4me1, H3K36me3, and H3K9me3 antibodies in human liver samples from the EN-TEx project (http://entex.encodeproject.org/main.html) ³⁷. Genetic variation loci-associated with rhythmic gene expression in the liver were intersected with the union of candidate CREs regions. These preprocesses established the link between the co-doccurrence of CREs and genetic variants. Enrichment was calculated by randomly selecting a number-matched set of genetic variants 1,000 times to generate control sets. The average signal for each histone modification was calculated for both the detected genetic variants and the control sets during the permutation process. Enrichment was assessed by comparing the histone modification signals of the detected genetic variant set to the distribution of signals from the permutation sets (Figure S1). The p-values represent the frequency at which the mean signal of the control sets exceeded the mean signal of the detected variant set, indicating how often the control sets showed higher signal values compared to the detected genetic variant set.

RNA extraction and quantitative PCR

Total RNA was extracted from frozen liver tissue using Trizol reagent and purified by RNeasy Mini Kit. Fresh 1 μg total RNA was reverse-transcribed into 20 μL cDNA by High-Capacity cDNA Reverse Transcription Kit according to its protocol. Quantitative PCRs were performed with gene primers (Table S7) and SYBR Green PCR Master Mix under a QuantStudio 6 Flex instrument. Data analysis was performed using the standard curve method, and gene expression was normalized to the mRNA level of the housekeeping gene Arbp.

RNA-seq library preparation and RNA-seq data processing

1 μg total RNA extracted from each biological replicate liver (n = 3 per group) was used for library construction: briefly, ribosomal RNA was first depleted using the RiboZero Magnetic rRNA removal kit, then the remaining RNA was converted to a DNA library for sequencing according to the TruSeq Stranded Total RNA Library Prep kit protocol.

The pair-end fastq reads sequenced from RNA-seq libraries were mapped to the mouse reference genome (version GRCm39) by STAR aligner. Low mapping quality reads (MAPQ < 30) were removed by SAMtools. Filtered BAM files were converted to bigwig files for genome browser visualization. Tag directories were created by Homer from filtered BAM files. The total number of reads for each transcript was quantified. Differentially expressed transcripts were identified using the R package DEseq2. Cutoffs of |log₂FC| >1 and p-value < 0.01 were used to identify differentially expressed transcripts. The transcript rhythmicity was determined by the R package Meta-cycle, meta2d_p_value < 0.05, and oscillation amplitude (peak/trough) > 1.5 were considered as rhythmic transcripts.

Metabolic profiling

Weigh approximately 3–15 mg of tissue (n = 3–4 per group) in Eppendorf tubes (keep tubes on dry ice). Extract in 1 mL ice-cold 80% methanol (prechilled at −80 °C; methanol:water 4:1, v/v) using a tissue homogenizer. After homogenization, keep the mix on ice for 10 min, then centrifuge at 20,000 × g for 10 min at 4°C. Transfer the supernatant (~400 μL) into new pre-cooled 1.5 mL tubes and immediately dry the samples in a vacuum concentrator. Adjust the volume of the supernatant taken for drying based on tissue weight to ensure that all samples contain metabolites extracted from the same tissue mass. Samples were reconstituted into ~30–80 μL sample solvent based on the tissue weight/tube on the day of LS-MS run in (water:methanol:acetonitrile, 2:1:1, v/v/v). Final injection volume is 3–5 μL, equivalent to a metabolite extract from 160 μg tissue. Differential metabolites analyses were performed by R package limma, and the rhythmicity of metabolites was determined by the R package Meta-cycle, meta2d_p_value < 0.05, and oscillation amplitude (peak/trough) > 1.5 were considered as rhythmic metabolites.

Lipidomic profiling

For lipidomic profiling, liver tissues were collected at ZT10 and ZT22 from B6 and 129Sv mice with or without Esrrγ knockout, under diet-induced obesity (DIO) conditions. Liver tissues were snap-frozen in liquid nitrogen and kept in −80°C refrigerator. 20 mg of samples were homogenized (30 HZ) for 20 s with a steel ball and centrifuged (3000 rpm, 4°C) for 30 s. Then add 1 mL of the extraction solvent (MTBE: MeOH =3:1, v/v) containing internal standard mixture. After whirling the mixture for 15 min, 200 μL of ultrapure water was added. Vortex for 1 min and centrifuge at 12,000 rpm for 10 min. 200 μL of the upper organic layer was collected and evaporated using a vacuum concentrator. The dry extract was dissolved in 200 μL reconstituted solution (ACN: IPA=1:1, v/v) for LC-MS/MS analysis. Differential lipids expression was analyzed using the R package limma. Lipids exhibiting rhythmicity were identified based on comparisons between ZT10 and ZT22, with thresholds set at p-value < 0.05 and fold change > 1.5 (n = 3–5 per group).

Tissue in situ ChIA-PET

ChIA-PET experiments were performed as described previously ¹¹¹ with minor changes. In brief, mouse livers (n = 3 per group) were harvested and cross-linked in 1% formaldehyde for 20 min, followed by quenching with 1/20 volume of 2.5M glycine solution for 5 min, and two washes with 1×PBS. Nuclear extracts were prepared by Dounce homogenization. 30 ml of AluI enzyme was used to digest the genome of dual-crosslinked liver nuclei at 37°C O/N. Exposed ends of genomic DNA were dA-tailed and ligated with dT-overhang biotinylated bridge linker (IDT, sequence is listed in Table S7) at 16°C O/N. The ligated chromatin was sheared by sonication and pulled down by anti-H3K27ac antibody. Immunoprecipitated DNA was subjected to Tn5 tagmentation, M280 Dynabeads streptavidin pull-down, and PCR amplification. PCR products were subjected to size selection and sequenced.

In situ ChIA-PET data processing and visualization

ChIA-PET data were processed using ChIA-PET Tool V3 ⁹⁸, a comprehensive pipeline for analyzing chromatin interaction data. Briefly, linker sequences were first identified and removed to retain valid paired-end tags (PETs). The filtered reads were then aligned to the mouse reference genome (GRCm39) using BWA, followed by read purification to eliminate PCR duplicates and low-quality mappings. Purified PETs were categorized into self-ligation and inter-ligation types based on their genomic orientations and distances. Self-ligation PETs were used to identify protein-binding peaks, while inter-ligation PETs were used to detect statistically significant chromatin interactions. All steps were performed using default parameters. Only contacts supported by ≥3 PETs and spanning >8 kb were retained for downstream analysis. Enhancer and promoter peaks were identified by generating tag directories from uniquely mapped reads using HOMER. Chromatin interaction maps were visualized as 2D contact matrices using Juicertools (Durand et al., 2016), and genome browser tracks were viewed using IGV ¹¹².

To assess the proportion of H3K27ac-associated E–PI harboring SNPs between B6 and 129Sv mice, only high-confidence interactions, defined by FDR < 0.05 and PET count ≥ 5, were included. Rhythmic E–P loops between ZT22 and ZT10 under each condition were identified using the diffloop R package (https://github.com/aryeelab/diffloop), with a merging window of 1000 bp and a fold change ≥ 2. Strain-specific SNPs between B6 and 129Sv mice were obtained from the Mouse Genomes Project (ftp://ftp-mouse.sanger.ac.uk/REL-1505-SNPs_Indels/strain_specific_vcfs/129S1_SvImJ.mgp.v5.snps.dbSNP142.bed.gz). Genomic overlap between rhythmic E–PI sites and strain-specific SNPs was assessed using the intersect function of BEDTools, and the percentage of rhythmic interaction anchors harboring SNPs was calculated accordingly.

To compare E–PI loops between the two mouse strains, PET loops from each strain were first merged using BEDTools, allowing a maximum gap of 1 kb between anchors. Only loops with a genomic span greater than 10 kb and less than 250 kb were retained for further analysis. PET counts from each strain were then summed within the merged loop regions to quantify interaction strength. Using the merged BED file, loops were classified as strain-specific or common based on a fold change ≥ 2 in PET counts between the two strains, with only loops supported by at least 5 PETs included in the analysis.

ChIP-seq library preparation and data processing

H3K27ac and ESRRγ ChIP experiments were performed on liver samples from NC or DIO mice (n = 3 per group) harvested at ZT10 or ZT22. DNA was amplified according to the ChIP-seq sample preparation guide provided by Illumina using adaptor oligo and primers from Illumina, enzymes from New England Biolabs, and a PCR purification kit and MinElute kit from QIAGEN.

Sequenced reads were aligned to the mouse reference genome (GRCm39) using Bowtie2 with default parameters. Only reads that uniquely mapped to the genome were retained for downstream analysis. Peak calling was performed using HOMER. To assess SNP enrichment within ESRRγ binding regions, peaks from all samples were merged and resized to 200 bp to represent core binding sites. The percentage of ESRRγ-bound regions overlapping with strain-specific SNPs between B6 and 129Sv mice was calculated.

In parallel, the percentage of SNP overlap was evaluated in two control sets: (1) common ESRRγ binding sites, defined as peaks detected in both B6 and 129Sv strains, and (2) randomized ESRRγ binding regions, generated by randomly permuting the genomic coordinates of the combined ESRRγ peaks across the GRCm39 genome. To account for sampling variability, the randomization procedure was repeated 100 times, and the median percentage of SNP-overlapping regions was used as the baseline.

GIGGLE score-based similarity analysis

To assess transcription factor (TF) binding similarity, the top 1,000 ESRRγ ChIP-seq peaks from each condition were used to generate a GIGGLE index representing ESRRγ cistrome. Genomic coordinates of merged peaks in all samples were resized into 200 bp to cover TF binding regions. For H3K27ac histone binding analysis, uniquely mapped paired-end reads from H3K27ac ChIA-PET data were used for peak calling with HOMER, and the merged peaks from all samples were resized to 1,000 bp to encompass the H3K27ac-enrichedregions. GIGGLE was then utilized to assess the similarity between H3K27ac-associated genomic coordinates and the ESRRγ cistromes in liver tissue. CistromeDB was then applied to determine the similarity between these genomic coordinates and published cistromes, and only those derived from mouse liver tissue were selected for downstream analysis.

TF enrichment analysis using IMAGE

To identify transcription factors enriched at time-dependent E-PI sites detected by ChIA-PET, we applied the IMAGE (Integrative Motif Activity and Gene Expression) ⁵⁸ analysis framework across three biological comparisons: NC-B6 vs. NC-129Sv, NC-B6 vs. DIO-B6, and NC-129Sv vs. DIO-129Sv. For each comparison, dynamic E-PI anchor regions were defined from ChIA-PET data, and enhancer activity was calculated as the ZT22/ZT10 signal ratio at enhancer anchors. This was integrated with rhythmic gene expression data from the corresponding comparisons, using the ZT22/ZT10 expression ratio of rhythmic transcripts. Motif enrichment and activity scores were computed using JASPAR transcription factor binding motifs. TFs showing strong motif enrichment and consistent activity across comparisons were defined as high-confidence candidates likely involved in circadian, diet-induced, or strain-specific transcriptional regulation.

Oil Red O staining

Mouse liver tissues (n = 3–5 per group) were fixed in 10% formalin solution at 4°C for 24hrs, and dehydrated in 30% sucrose overnight, then embedded dehydrated tissue with 30% sucrose: O.C.T compound (1:1) and slided into 10μm sections. Oil Red O staining was performed using the Oil Red O Stain Kit according to the manufacturer’s instructions.

Measurement of hepatic triglycerides

The Triglyceride Liquid Reagent for Diagnostic Set (SB2100430) was used for TG assay according to the manufacturer’s instructions. Briefly, mouse liver tissue pieces (50–100 mg) were weighed and resuspended in 500 μL of tissue lysis buffer (50 mM Tris-HCl pH 7.4, 140 mM NaCl, 0.1% TX-100). Samples were homogenized using a tissuelyser at 20Hz for 3 minutes, repeated 3 times. Homogenates were diluted 50-fold. Next, 20 μL of standards and samples were added to a 96-well microplate, followed by 20 μL of 1% deoxycholate to each well, and incubated at 37°C for 10 minutes. Subsequently, 200 μL of activated TG reagent was added to each well, followed by an incubation at 37°C for 20 minutes. Absorbance was read at 500 nm.

Oxygen consumption rate

Oxidative respiration in liver tissue (n = 4 per group) was assessed using the Seahorse XFe24 Extracellular Flux Analyzer (Agilent) following the manufacturer’s protocol. In brief, sensor cartridges were hydrated overnight in XF calibrant at 37°C under non-CO₂ conditions. Mice were perfused with 1 × PBS prior to liver tissue harvest. Liver samples were chopped into small pieces and then equilibrated for 1 hour at 37°C in XF Assay Medium (modified DMEM) supplemented with 25 mM glucose, 1 mM pyruvate, and 1 mM L-glutamine. The XFe24 assay included a 20-minute equilibration phase followed by measurements of basal respiration and responses to sequential injections of mitochondrial stress compounds: oligomycin (4 μM), FCCP (8 μM), and antimycin A/rotenone (10 μM/3 μM). Each compound was delivered over multiple cycles consisting of mixing, waiting, and measuring periods. Oxygen consumption rate (OCR) was recorded and normalized to tissue weight (mg).

Mitochondrial respiration was assessed using the Seahorse XF96 Extracellular Flux Analyzer. XF96 sensor cartridges were hydrated overnight at 37°C in a non-CO₂ incubator with XF calibrant. Mitochondria were isolated from mouse liver tissue (n = 4 per group) using a Dounce homogenizer in mitochondrial isolation buffer (MIB, d-mannitol 210mM, sucrose 70mM, HEPES pH 7.2 5mM, EGTA 1mM, BSA 0.1%), followed by differential centrifugation. Protein concentration was determined using the Bradford assay to normalize mitochondrial input per well. Isolated mitochondria were diluted in mitochondrial assay solution (MAS, d-mannitol 220mM, sucrose 70mM, KH₂PO₄ 10mM, MgCl₂ 5mM, HEPES pH 7.2 2mM, EGTA 1mM ) containing 0.02% BSA, loaded onto Seahorse XF96 cell plates, and centrifuged at 2,000 × g for 20 min at 4°C. Substrate buffers for complex I (5 mM pyruvate, 0.5 mM malate) were added, and treatment compounds (ADP 4mM, Oligomycin 5μg/mL, FCCP 1μM, Antimycin A/retenone 1μM/1μM) were prepared in MAS without BSA and preloaded into assay cartridge ports. Oxygen consumption rates (OCR) were measured following compound injections according to the pre-programmed assay protocol. All compounds and reagents were obtained from Sigma-Aldrich unless otherwise specified.

Micro-C library preparation and data analysis

Liver samples from three donors with MAFLD were collected at Baylor St. Luke’s Medical Center in accordance with approved IRB protocols and informed consent. Micro-C libraries were prepared using the Dovetail^™ Micro-C Kit, following the manufacturer’s protocol. Briefly, ~5 mg of frozen liver tissue was ground to a fine powder using a mortar and pestle in liquid nitrogen, then resuspended in 1× PBS. Tissue was fixed with 3 mM DSG (Thermo Fisher Scientific, A35392) for 30 minutes, followed by crosslinking with 1% formaldehyde for an additional 30 minutes. After two washes with wash buffer, the tissue pellet was resuspended in reconstituted cell isolation enzyme mix and incubated at 37°C without shaking for 30 minutes. MNase digestion was performed at 22°C for 15 minutes. Digested chromatin was lysed with 20% SDS and captured using chromatin capture beads. Subsequent steps, including end polishing, bridge ligation, and intra-aggregate ligation, were carried out using the corresponding enzyme mixes. Crosslinks were reversed with crosslink reversal buffer and proteinase K, and DNA was purified using SPRIselect beads. Sequencing libraries were then prepared using the NEBNext Ultra II DNA Library Prep Kit with NEBNext Unique Dual Index Primer Pairs and sequenced (2 × 150 bp) on the Illumina NextSeq platform.

Raw sequencing reads (fastq files) were first aligned to the human reference assembly genome (GRCh38) using BWA MEM. Alignments were parsed, paired, sorted and filtered for PCR duplicates to generate invalid pairs using pairtools. The resulting pairs files were used to generate matrix files at different resolutions using Juicer tools, and these files at 10 kb resolution were used for downstream analysis. Chromatin loops were identified using the Mustache loop-calling algorithm at 10 kb resolution, based on the .hic files derived from unique valid read pairs. A summary of this workflow and the quality control of the data can be found in Figures S6B and S6C. Micro-C data was visualized using scalable vector graphic (svg), as shown in Figure S7.

Enrichment of genetic variations associated with rhythmic gene expression in E-PI sites

To determine whether genetic variations associated with rhythmic gene expression are enriched in E-PI sites, we compared the number of these variations located in E-PI sites to the number of genetic variations randomly selected from the genome. We randomly selected SNPs with the same minor allele frequency (MAF) distribution as the observed genetic variations from all SNPs detected in the GTEx population, ensuring the number of selected SNPs matched the number of observed variations. We then counted how many of these randomly selected SNPs were located within E-PI sites to establish a baseline. This random sampling process was repeated 1,000 times, and the enrichment of observed genetic variations was defined as the ratio of the observed number to the median of the expected numbers obtained from these 1,000 iterations. The p-value was calculated using a permutation test.

To identify high-confidence promoter and enhancer regions, we intersected the E-PI sites with the union of open chromatin regions in human liver, as detected by ATAC-seq from 44 liver samples representing the full histological spectrum of nonalcoholic fatty liver disease, including healthy controls, steatosis, and fibrotic nonalcoholic steatohepatitis livers ⁹¹. The enrichment of genetic variants associated with rhythmic gene expression within these high-confidence E-PI sites was also calculated using the same method as for all E-PI sites.

Motif enrichment analysis on rlGVs

Genetic variations associated with rhythmic gene expression, located in open chromatin regions at E-PI sites, were defined as rlGVs. These rlGVs were extended by 500 bp upstream and downstream, and TF DNA binding motifs were scanned within these regions. Enriched motifs were identified using the findMotifsGenome.pl function from Homer.

Enrichment of rlGVs for lipid-related traits heritability

To evaluate the contribution of rlGVs within E-PI sites to the heritability of lipid-related traits, we partitioned the heritability of 20 lipid-related traits using stratified linkage disequilibrium score (LDSC) regression. This analysis included rlGVs alongside 53 other functional categories within the ‘baseline-LD model’, as described previously ¹¹³. In brief, GWAS summary statistics of 20 lipid-related traits were downloaded from GWAS Catalog ¹¹⁴ (Table S6). The binary annotation file for rlGVs was generated by assigning a value of one to rlGVs located within enhancer-promoter interaction regions and a value of zero to the remaining SNPs in baseline regions. As a control, we created a randomly selected set of SNPs that matched the number of rlGVs within enhancer-promoter interaction regions, with the same minor allele frequency (MAF) and distance to transcription start site (TSS) distribution. LD scores were computed based on the annotation files using SNP genotype data from individuals of European ancestry in the 1000 Genomes Project Phase 3, with a window size of 1 centimorgan (cM). The partitioned heritability for each of the 20 lipid-related traits was estimated as the ratio of heritability attributed to rlGVs or control SNPs relative to the overall SNP-based heritability. The enrichment p-values, which quantify whether each trait is significantly enriched in rlGVs or control SNPs, were also calculated using stratified LDSC. We further quantified per-SNP heritability using the parameter τ, a valuable metric for assessing the contribution of an annotation category to total SNP-based heritability, following the approach described in a previous study ⁶⁶.

QUANTIFICATION AND STATISTICAL ANALYSIS

The transcript or metabolite rhythmicity is determined by the R package Meta-cycle, meta2d_p_value < 0.05, 20 ≤ period (t) ≤ 24 hours, and oscillation amplitude (peak/trough) > 1.5 were considered as rhythmic. Continuous variables are reported as the mean value ± standard error of the mean (SEM). Data with a normal distribution are compared with a two-tailed Student’s t test, and the Mann-Whitney U test was used for data without a normal distribution. p value < 0.05 is defined as a significant difference, and the significance level is indicated in figure legends. The statistical parameter “n” represents the sample size in each experiment, and it is reported in the figure or in the figure legends. Schematics were created with BioRender.com.

Supplementary Material

1

Data S1. Source data.

3

Table S1. Expression data of B6 and 129Sv mice under NC condition and circadian phase of metabolism-related rhythmic genes in human and mouse, related to Figure 2. A.) Heatmap of B6 specific circadian transcripts. B.) Heatmap of 129Sv specific circadian transcripts. C.) DEG analysis of B6 and 129Sv. D.) Phase of metabolism-related rhythmic genes in human and mouse.

4

Table S2. Expression and metabolomic data of B6 and 129Sv mice under NC and DIO conditions, related to Figure 3 and Figure S3. A.) Heatmap of NC-B6 specific circadian transcripts compared with DIO-B6. B.) Heatmap of DIO-B6 specific circadian transcripts compared with NC-B6. C.) Heatmap of NC-129Sv specific circadian transcripts compared with DIO-129Sv. D.) Heatmap of DIO-129Sv specific circadian transcripts compared with NC-129Sv. E.) DEG analysis of NC-B6 and DIO-B6. F.) DEG analysis of NC-129Sv and DIO-129Sv. G.) Differential metabolite analysis of B6 and 129Sv mice under NC and DIO conditions classified into three categories.

5

Table S3. Time-dependent ChIA-PET sites of B6 and 129Sv mice under NC and DIO conditions, related to Figure 4 and Figure S4. A.) DIO-B6 specific time-dependent ChIA-PET sites. B.) DIO-129Sv specific time-dependent ChIA-PET sites. C.) NC-B6 specific time-dependent ChIA-PET sites. D.) NC-129Sv specific time-dependent ChIA-PET sites.

6

Table S4. ESRRγ dependent rhythmic lipidomic and metabolomic data of B6 and 129Sv mice mice under DIO condition, related to Figure 6. A.) ESRRγ dependent rhythmic lipids. B.) Fold change of TGs upon Esrrγ LKO. C.) ESRRγ dependent rhythmic lipids.

7

Table S5. Chromatin loops detected from Micro-C data in human livers, related to Figures 7A-7D.

8

Table S6. Summary of lipid-related traits used for heritability enrichment analysis, related to Figures 7G-7I

9

Table S7. qPCR primer and ChIA-PET bridge linker sequence, related to the STAR Methods.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
H3K27ac	Active Motif	39133
ESRRγ	Liming Pei’s Lab	Pei, L. et al.56
Chemicals and Reagents
TRIzol Reagent	Ambion	5457
SYBR Green PCR Master Mix	ThermoFisher Scientific	4334973
SPRIselect	Beckman coulter	B23318
AluI	NEB	R0137L
Dynabeads™ M-280 Streptavidin	Thermo Fisher	11206D
IGEPAL CA-630	Sigma	I8896
DynaBeads Protein G	Thermo Fisher	10004D
Sucrose	Sigma-Aldrich	S0389
VAHTS DNA Clean Beads	Vazyme	N411–01
37% Formaldehyde Solution	Sigma-Aldrich	F8775
DSG (Disuccinimidyl Glutarate)	ThermoFisher Scientific	A35392
DMSO	ThermoFisher Scientific	276855
Formalin solution, neutral buffered, 10%	Sigma-Aldrich	HT501320
Tissue-Tek® O.C.T. Compound	Sakura	4583
Critical commercial assays
RNeasy Mini Kit	QIAGEN	74106
High Capacity cDNA Reverse transcription Kit	ThermoFisher Scientific	4368814
Illunima Ribo-Zero Plus rRNA Deleption Kit	Illumina	20040526
TruSeq Standed Total RNA Library Prep Kit	Illumina	20020599
Triglyceride Liquid Reagent for Diagnostic Set	Stanbio™	SB-2100–430
Dovetail® Micro-C Kit	Cantata Bio	21006
Oil red O Stain Kit	Abcam	ab150678
Deposited data
RNA-seq	This study	GSE278920
H3K27ac ChIA-PET	This study	GSE278922
ESRRγ ChIA-PET	This study	GSE278922
ESRRγ ChIP-seq	This study	GSE278921
ATAC-seq	This study	GSE297115
Lipidomics	This study	Table S4
Metabolomics	This study	Table S2 and S4
Mirco-C in human livers	This study	Table S5
ATAC-seq in human livers	Kang, B. et al. 91	Table S1 in Kang, B. et al.
Experimental Models: Organisms/Strains
C57BL/6J	The Jackson Laboratory	664
129S1/SvImJ	The Jackson Laboratory	2448
C57BL/6J Esrrγf/f	Ronald Evans’ Lab	Liming et al., 2015
129S1/SvImJ Esrrγf/f	C57BL/6J Esrrγf/f cross with 129S1/SvImJ	N/A
Oligonucleotides
See Table S7 for qPCR primer list	N/A	N/A
Software and Algorithms
SAMtools 1.17	Li, H. et al.92	https://samtools.sourceforge.net
Bedtools 2.31.1	Quinlan, A.R. et al.93	https://bedtools.readthedocs.io/en/latest/
dryR	Weger, B.D. et al.94	https://github.com/naef-lab/dryR
Meta Cycle	Wu, G. et al.95	https://github.com/gangwug/MetaCycle
DESeq2	Love, M.I. et al.96	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
limma	Ritchie, M.E. et al.97	https://www.bioconductor.org/packages/release/bioc/html/limma.html
pheatmap		https://github.com/raivokolde/pheatmap
ChIA-PET v3	Li, G. et al98	https://github.com/GuoliangLi-HZAU/ChIAPET_Tool_V3
Homer 4.11.1	Heinz, S. et al.99	http://homer.ucsd.edu/homer/
IMAGE	Madsen, J.G.S. et al.58	https://github.com/JesperGrud/IMAGE
EnrichR	Chen, E.Y. et al.100	https://maayanlab.cloud/Enrichr/
R4.4.1		https://www.r-project.org
GraphPad Prism 10.0	N/A	https://www.graphpad.com/features
Adobe Illustrator 2024	N/A	http://Adobe.com
bwa v0.7.17-r1188	Li, H. et al.101	https://github.com/lh3/bwa
Pairtools 1.0.2	Abdennur, N. et al.102	https://github.com/open2c/pairtools
Juicertools 1.22.01	Durand, N.C. et al.103	https://github.com/aidenlab/JuicerTools
Mustache 1.0.1	Roayaei Ardakany, A. et al.104	https://github.com/ay-lab/mustache
Metascape	Zhou, Y. et al.105	https://metascape.org/gp/index.html#/main/step1
STAR 2.7.1a	Dobin, A. et al.106	https://github.com/alexdobin/STAR
Bowtie2 2.5.1	Langdon, W.B. et al107	https://github.com/BenLangmead/bowtie2
ImageJ	N/A	https://imagej.net/ij/
Other
Rodent normal chow	LabDiet	5010
Rodent 60% High fat deit	Research Diets	D12492i
AAV8-TBG-GFP	BCM Vector core	N/A
AAV8-TBG-Cre	BCM Vector core	N/A

Highlights.

• Genetic variants contribute to variation in peripheral diurnal rhythms

• More than 80% of rhythmic genes and E-PIs are interdependent on genetics and nutrition

• Noncanonical clock regulators govern rhythmic expression, lipid droplets and metabolism

• Human SNPs linked to rhythmic gene expression are tightly associated with metabolic traits