Integrative epigenomic and transcriptomic analyses reveal metabolic switching by intermittent fasting in brain

Abstract

Intermittent fasting (IF) remains the most effective intervention to achieve robust anti-aging effects and attenuation of age-related diseases in various species. Epigenetic modifications mediate the biological effects of several environmental factors on gene expression; however, no information is available on the effects of IF on the epigenome. Here, we first found that IF for 3 months caused modulation of H3K9 trimethylation (H3K9me₃) in the cerebellum, which in turn orchestrated a plethora of transcriptomic changes involved in robust metabolic switching processes commonly observed during IF. Second, a portion of both the epigenomic and transcriptomic modulations induced by IF was remarkably preserved for at least 3 months post-IF refeeding, indicating that memory of IF-induced epigenetic changes was maintained. Notably, though, we found that termination of IF resulted in a loss of H3K9me₃ regulation of the transcriptome. Collectively, our study characterizes the novel effects of IF on the epigenetic-transcriptomic axis, which controls myriad metabolic processes. The comprehensive analyses undertaken in this study reveal a molecular framework for understanding how IF impacts the metabolo-epigenetic axis of the brain and will serve as a valuable resource for future research.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11357-022-00537-z. 

Introduction

Intermittent fasting (IF) is a dietary regimen that restricts energy intake via alternating periods of fasting and ad libitum food consumption, without compromising nutritional composition. Many animal studies have established that IF ameliorates the development of age-related cardiovascular, neurodegenerative, and metabolic diseases and promotes longevity. Despite extensive experimental evidence supporting such protection by IF, our understanding of the underlying molecular mechanisms is still poor. Epigenetic modifications have been shown to be pivotal in mediating the influence of environmental factors on genomic status. Indeed, many age-related diseases are polygenic and confounded by environmental influences [1], suggesting that interactions between the environment and the genetic framework may underlie the complex pathophysiology of chronic diseases.

Calorie restriction (CR) is one form of environmental manipulation that can impact the epigenome through various epigenetic modifications and consequently affect a plethora of biological pathways that modulate age-related epigenetic events. For instance, CR can influence the epigenetic regulation of immuno-metabolic adaptation and attenuate age-dependent epigenetic drift, thus reducing the pathogenesis of age-related diseases [2–4]. IF is a lifestyle intervention that represents a potential environmental factor capable of influencing an individual’s epigenome. However, an understanding of IF-induced epigenetic changes is currently lacking, and potential epigenetic effects of refeeding after IF are unknown. Thus, it is of great interest to study whether IF serves as an environmental trigger that can influence the epigenomic profile.

H3K9me₃ is often associated with constitutive heterochromatin or inactive euchromatin, and has previously been reported to be modulated during CR [5, 6]. There is a paucity of data on whether this modulation of histone mark can be a result of IF. In studies using the epigenetic clock [7], the cerebellum, a major brain region harboring a large population of neuronal cells [8, 9], was recently reported to age more slowly than other parts of the human body, and to contain a circadian oscillator that responds robustly to energy restriction and food anticipation [10, 11]. It was therefore hypothesized that, as a large population of cells within the cerebellum would have a high energy demand, the cells should be sensitive to energy restriction and induce robust metabolic changes in response. Thus, we decided to investigate the effects of IF and refeeding on the modulation of H3K9me₃ in the cerebellum. We studied the epigenomic and transcriptomic profiles resulting from two common IF regimens—time-restricted fasting for 16 h (IF16) or 24 h on alternate days (i.e., “every other day”; EOD)—in mice for 3 months. In addition, we assessed changes in epigenomic and transcriptomic patterns in mice refed for 2 months (IF16.R or EOD.R), following the IF regimens. We found that both types of IF regimens were capable of distinctly affecting the modulation of H3K9me₃, which in turn differentially regulated different aspects of the transcriptome, especially in the metabolic axis. Moreover, we found that following refeeding, mice that had been subjected to either IF regimen demonstrated differential epigenomic and transcriptomic profiles. However, there was a loss of epigenetic memory in the transcriptome following the abolition of IF. Collectively, our data provide novel insights into the epigenetic-transcriptomic axis of IF, and refeeding mechanisms in the cerebellum.

Materials and methods

Ethical compliance statement

All animal procedures were approved by the National University of Singapore Animal Care and Use Committee and performed according to the guidelines set forth by the National Advisory Committee for Laboratory Animal Research (NACLAR), Singapore. All aspects of the study were performed in accordance with the ARRIVE (Animal Research: Reporting In Vivo Experiments) guidelines.

Intermittent fasting and refeeding regimen

C57/BL6NTac male mice (InVivos Pte Ltd., Singapore) were raised for 3 months with ad libitum access to food, which was a standard Teklad Global 18% protein rodent diet (Envigo, UK), and water. Mice were then randomly assigned to three study groups subjected to ad libitum (AL) feeding, IF for 16 h per day (IF16), or for 24 h on alternate days (i.e., EOD) for 3 months. Mice placed under the IF16 regimen were given daily access to food from 7 a.m. to 3 p.m., and then the food was removed for 16 h. For mice in the EOD regimen, food was provided from 7 a.m. to 7 a.m. the next day, after which it was removed for the next 24 h. All mice had ad libitum access to water, while the AL mice had ad libitum access to both food and water. In the refeeding regimen, all three groups of mice had ad libitum access to both food and water for 2 months.

During the entire experiment, the mice were housed in animal rooms at 20–22 °C, with 30–40% relative humidity under a 12-h light/dark cycle. A series of physiological tests were performed on the mice. Body weight was measured weekly. Nasoanal length was measured, using the Lee index (body weight/nasoanal length), on the day of animal euthanasia. Blood glucose and ketone levels were measured using a FreeStyle Optimum Meter and corresponding test strips (Abbott Laboratories, UK) at baseline and monthly via the tail bleed method. Both tests were performed at 7 a.m. Lastly, monthly food/energy consumption was recorded to measure calorie intake (weight of food consumed × kcal/g of food).

Cerebellum tissue collection

We have embarked on a preliminary selection of a specific brain part to be investigated in our studies. However, we discovered that the cerebellum is highly responsive to H3K9me₃ modulation following different IF regimens, whereas other brain parts show little to no changes to H3K9me₃ modulation. Therefore, following the fasting or refeeding regimen, animals were anesthetized and euthanized. All mice were euthanized between 7 a.m. and noon on a food deprivation day. The cerebellum was harvested, immediately flash-frozen, and stored at − 80 °C.

Chromatin immunoprecipitation

The frozen cerebellum was crushed into a fine powder with liquid nitrogen using a pestle and mortar. The tissue was then crosslinked with 1% formaldehyde (Merck, New Jersey, USA) for 10 min at room temperature, and the reaction stopped by adding glycine (Abcam, Cambridge, UK) to a final concentration of 0.125 M for 5 min at room temperature. Fixed cells were rinsed twice with phosphate-buffered saline, containing a protease (Thermo Scientific, Massachusetts, USA) inhibitor at 1:1000 ratio and resuspended in 10 mL of lysis buffer (10 mM Tris–HCl pH 8, 0.25% Triton X-100, 10 mM EDTA, 0.1 M NaCl). The lysates were subjected to further lysis using a Dounce tissue grinder (Sigma-Aldrich, Missouri, USA) for 14–16 strokes. The lysates were then suspended in 1% SDS lysis buffer (50 mM HEPES–KOH pH 7.5, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 1% SDS) and ultracentrifuged at 18000 rpm for 30 min. Subsequently, the chromatin jelly was resuspended in 0.1% SDS lysis buffer (50 mM HEPES–KOH pH 7.5, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 0.1% SDS) and subjected to another round of ultracentrifugation at 18,000 rpm for 30 min. Next, the lysate was resuspended in 0.1% SDS buffer, sonicated with 15 cycles of 30 s on and 30 s off in a sonicator (Diagenode, New Jersey, USA), and centrifuged at 14,000 rpm for 10 min. Approximately 500 ng of the sonicated chromatin was stored at − 80 °C as an input DNA control.

The remaining sonicated chromatin was incubated with 30 μL of Protein G Dynabeads (Invitrogen, California, USA) previously conjugated overnight to 4 μg of the chromatin immunoprecipitation (ChIP)-grade H3K9me₃ antibody (Abcam, Cambridge, UK). The beads were then washed for 5 min, once in low salt 0.1% SDS FA lysis buffer (50 mM HEPES–KOH pH 7.5, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 0.1% SDS), and then in high salt 0.1% SDS FA lysis buffer (50 mM HEPES–KOH pH 7.5, 350 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 0.1% SDS), followed by a LiCl buffer wash (10 mM Tris–HCl pH 8, 0.25 mM LiCl, 1 mM EDTA, 0.5% NP40, 0.5% sodium deoxycholate). Finally, the immunoprecipitated material was washed twice with TE buffer (10 mM Tris–HCl pH 8, 1 mM EDTA), and eluted with ChIP elution buffer (50 mM Tris–HCl pH 8, 10 mM EDTA, 1% SDS) incubation for 2 h at 68 °C for both input and immunoprecipitated DNA. Finally, chromatin was incubated for 1 h with RNase at 37 °C and digested with 10 μL of 5 mM NaCl and 5 μL of Proteinase K overnight at 68 °C. DNA was extracted using phenol–chloroform, and was eluted in nuclease-free water.

The Extracted DNA was quantified using a Quantifluor double-stranded DNA kit (Promega Corporation, Australia) as per the manufacturer’s instructions. 20 × TE buffer was first diluted 20 times, before diluting the Quantifluor dsDNA dye in a 1:400 ratio to prepare the working solution. For each sample, 1 μL of DNA was diluted in 200 μL of working solution, whereas for the blank and standard, 2 μL of 1 × TE buffer and 2 μL of provided DNA standard (100 ng/μL) were diluted in equal volume of working solution. Both blank and standard solutions were used for calibration, to plot a standard curve before the actual measurement of DNA concentration. For fragment size analysis, 100 ng of DNA was first purified using the QIAquick PCR purification kit (Qiagen, Japan) according to the manufacturer’s protocol. The resulting purified DNA was then mixed with a loading dye (Thermo Scientific, Massachusetts, USA) at a 1:50 ratio, run on a 1% agarose gel (Lonza, Switzerland) at 100 V for 30 min, and stained with SYBR Safe dye (Invitrogen, California, United States). Imaging was carried out using a ChemiDocXRS + imaging system (Bio-Rad Laboratories, California, USA). Fragments of the ideal size (100–500 base-pairs were selected for the analysis).

A library was prepared using New England Biolabs Ultra II DNA Kit for Illumina (New England Biolabs, USA) according to the manufacturer’s instructions. The immunoprecipitated material was sequenced using the 150-bp paired-end protocol provided by Illumina Genome Analyzer 1.9 (Novogene, Beijing). All data obtained from each sample were pooled for analysis.

ChIP sequencing bioinformatics analysis

Raw reads were checked for quality, using FastQC (Babraham Bioinformatics, UK). Reads with low quality (proportion of low-quality bases larger than 50%) or N ratio (unsure base) more than 15% were discarded. Reads with adaptors at the 5′-end or those without adaptors and inserted fragments at the 3′-end were also discarded. Next, the reads were trimmed at the 3′-end adaptor sequence and those that were less than 18 bp in length were discarded. The reference genome for Mus musculus (mm10) and gene model annotation files were downloaded from the National Center for Biotechnology Information (NCBI) genome database. Reference genome indexing and mapping of the quality checked paired-end reads to the reference genome were carried out using the Burrows-Wheeler Alignment Tool (v0.7.17) [12]. The “callpeak” and “bdgdiff” options in MACS2 software [13] were used for peak detection and comparisons, respectively. Strand cross-correlation was computed on the resulting peaks and plotted using the GraphPad Prism software (v5). Heatmaps, fingerprints, and principal component analysis (PCA) plots were generated using the DeepTools2 software [14]. A three-dimensional PCA plot was constructed using the Sigmaplot software (v1.3). Venn diagrams were prepared using the Bioinformatics and Evolutionary Genomics online software (Bioinformatics and Evolutionary Genomics, Belgium). The functional significance of peaks was analyzed using webserver g:Profiler [15] and DAVID [16]. GO terms with adjusted p-value < 0.05 were considered significantly enriched among the pool of differentially expressed peaks. Diagrams representing GO enrichment analysis results were plotted using the GraphPad Prism software (v5) and the ggplot2 R package (v3.1.1).

Total eukaryotic mRNA extraction

RNA from the cerebellum tissue was isolated using the EZ-10 DNAaway RNA Mini-Preps Kit (Bio Basic, Canada) according to the manufacturer’s protocol. Briefly, frozen cerebellum samples were homogenized and lysed in lysis buffer. Contamination by genomic DNA was prevented using the gDNA eliminator column. RNA purity was determined using Nanodrop ND-1000 (Thermo Fisher Scientific, USA), while RNA integrity was assessed through agarose gel electrophoresis and the Agilent 2100 Bioanalyzer (Agilent, USA). Enriched RNA was of high quality, demonstrating an OD₂₆₀/OD₂₈₀ ratio of 1.9–2.0 from Nanodrop readings, two distinct bands indicating 28S and 18S rRNA following agarose gel electrophoresis, and RNA integrity number ≥ 6.8 with a smooth base line using the Agilent 2100 Bioanalyzer.

Following the isolation of high-quality and pure total RNA from cerebellum tissues, a cDNA library was constructed using the NEBNext® Ultra™ RNA library preparation kit (New England BioLabs, USA) as per the manufacturer’s protocol. mRNA was first purified via the addition of poly T-oligo-attached magnetic beads, and then subjected to random fragmentation using a fragmentation buffer. The first strand of cDNA was synthesized using a random hexamer primer and RNase H- (M-MuLV reverse transcriptase). The second strand of cDNA was synthesized using DNA polymerase I and RNase H, and the resulting double-stranded cDNA was then purified using AMPure XP beads. The overhangs of these purified double-stranded cDNA were further processed using an exonuclease and a polymerase to create blunt ends, and the 3′-ends of these DNA fragments were adenylated and subsequently ligated with the NEBNext hairpin loop structure adaptor on both ends for hybridization. For optimal isolation of cDNA fragments of approximately 150–200 bp in length, the DNA fragments were purified using the AMPure XP system (Beckman Coulter, USA), followed by PCR amplification and purification with AMPure XP beads to obtain the DNA fragments representing the complete library. The resulting libraries were sequenced on the HiSeq™ 2500 Illumina platform, resulting in a minimum of 12 GB of raw data per sample (Illumina, USA).

mRNA sequencing bioinformatics analysis

The reference genome for Mus musculus (mm10) and gene model annotation files were downloaded from NCBI genome database. The reference genome was indexed and paired-end quality checked reads were mapped to the reference genome using the STAR aligner (v2.5) [17]. Reads mapping to each gene were qualified using HTSeq (v0.6.1) [18]. Fragments per kilobase of exon model per million mapped reads (FPKM) for each gene were computed based on the gene length and the number of reads mapped to the gene. The FPKM value was then used to estimate gene expression levels. A total of 35,275 unique RNA transcripts were quantified in the cerebellum datasets. Differential gene expression analysis was performed using the DESeq2 R package (v2_1.6.3) [19] with a negative binomial distribution model for the gene counts. The resultant p-values were then adjusted using the Benjamini and Hochberg’s test to control false discovery rate. Genes with adjusted p < 0.05 were assigned as differentially expressed.

Partial least squares-discriminant analysis (PLS-DA) plots were generated using the mixOmics R package (v6.6.2) [20]. Venn diagrams were prepared using an online software (Bioinformatics and Evolutionary Genomics, Belgium). Heatmaps were built using the pheatmap R package (v1.0.12) [20], while volcano plots were prepared using the ggplot2 R package (v3.0.0). GO enrichment analysis for differentially expressed genes was carried out using g:Profiler [15], DAVID [16], and Enrichr [21] webservers. GO terms with an adjusted p-value < 0.05 were considered to be significantly enriched among the pool of differentially expressed genes. Enrichment analysis diagrams were generated using the GraphPad Prism software (v5) and the ggplot2 R package (v3.1.1).

Integrative ChIP and mRNA sequencing analysis

To identify the number of differentially expressed genes that were modulated by H3K9me₃, a Venn analysis was performed (Bioinformatics and Evolutionary Genomics, Belgium) and illustrated, using the Inkscape software (v0.92), as the Euler diagrams. Next, we performed a series of GO interrogation and visualization of overlapped genes using the GOplot R package (v1.0.2) [22] and Inkscape software (v0.92), respectively. ChIP peaks and mRNA tracks were visualized using the Integrative Genome Browser (v2.5.3) [23–25]. Network analysis and visualization of overlapped genes were achieved using the STRING (v11.0) webserver [26] and the Cytoscape software (v3.7.1) [27]. The ClusterOne (v1.0) plugin in Cytoscape was used with default parameters to obtain significant gene clusters. Metabolic pathways were depicted as metabolic maps, using the iPath3 online software tool (v3) [28]; and metabolic cellular functions performed were depicted as metabolic maps, using the BioCyc Omics dashboard software (19.0) [29, 30].

Statistical analysis

The GraphPad Prism software (v5) was used for statistical analysis of experimental data. Two-way analysis of variance (ANOVA) was used, followed by Dunnett’s post hoc test, for weight, glucose, and ketone measurement. Numerical data were expressed as mean ± standard error of the mean. A p-value < 0.05 was considered statistically significant. Correlation was determined using the “cor.test” function in R, with options set to alternative = “greater” and method = “Spearman.” All experiments were performed using at least three biological replicates per condition, and Pearson’s correlation coefficients of at least 0.9 were considered to demonstrate high coverage and reproducibility (Supplementary Fig. 8).

Data availability

High-throughput sequencing data have been submitted to the NCBI Sequence Read Archive under accession number GSE135945.

Results

IF mice showed decreased body weight without a change in fuel preferences

The summarized study design includes the timing of interventions and blood and tissue collection (Fig. 1). Male C57BL/6 N mice were fed a normal chow diet (on a caloric basis: 58%, 24%, and 18% of carbohydrate, protein, and fat, respectively). Mice were randomly assigned to one of the three study groups, AL (ad libitum), daily IF16, or EOD schedules, beginning at 3 months of age. First, we monitored the physiological effects of IF and refeeding regimens on these mice. Both IF16 and EOD groups had a lower body weight than AL mice after 3 months of IF (Supplementary Fig. 1A), but were similar to each other. After a 2-month refeeding regimen, refed mice, that is, IF16.R, EOD.R, and AL.R groups, did not differ in body weight (Supplementary Fig. 1A). Aging in mice was associated with an increase in average body weight (Supplementary Fig. 1A). Overall, the energy intake of mice subjected to IF or refeeding did not differ from that of control mice (Supplementary Fig. 1B). Next, we investigated the composition of energy intake, to examine whether mice compensated for energy restriction through a preferential shift in the specific macromolecules from which energy was obtained. Notably, energy intake from each compositional source (carbohydrate, fat, and protein) did not differ across the three study groups (AL, IF, and EOD) (Supplementary Fig. 1C). However, mice in the corresponding refeeding groups (AL.R, IF.R, and EOD.R) consistently showed a higher energy intake for each compositional source (Supplementary Fig. 1C). Our findings, therefore, exclude the possibility that differences in body weight during IF were due to an altered composition of energy take in.

Fig. 1.

Experimental design for intermittent fasting and refeeding regimen. A Mice were raised to 3 months with ad libitum access to food before being randomly assigned and subjected to AL feeding, IF16 or EOD for 3 months. All mice had AL access to water, with AL mice also having free access to food. During the refeeding regimen, the three groups of mice were allowed AL access to both food and water for a further 2 months. A series of physiological tests was performed on the mice as depicted. B Cerebellar tissue was harvested, frozen and subjected to chromatin immunoprecipitation at the H3K9me₃ locus, and eukaryotic mRNA was extracted and sequenced on an Illumina platform. Five mice per groups for intermittent fasting and refeeding regimen are assigned for each experiment. AL, ad libitum; IF16, intermittent fasting 16 h; EOD, intermittent fasting 24 h on alternate day or “every other day”; AL.R, AL and refeed; IF16.R, IF16 and refeed; EOD.R, EOD and refeed; H3K9me₃, histone 3 lysine 9 trimethylation

Blood analyses showed that both IF16 and EOD mice had lower glucose and higher ketone levels than AL mice (Supplementary Fig. 2A and B). Interestingly, prolonged energy restriction resulted in a greater change in both glucose and ketone homeostasis, suggesting that the differential impact of energy restriction on fuel utilization is time dependent. However, no differences were observed in blood glucose levels across study groups following refeeding, although a significant change was observed in blood ketone levels in EOD.R mice. (Supplementary Fig. 2A and B). Concordantly, body composition analysis indicated a reduction in the Lee index or body fat mass in both IF16 and EOD mice compared with that in the AL mice (Supplementary Fig. 2C). No differences were observed in body fat mass following refeeding (Supplementary Fig. 2C). Thus, our data show that IF and refeeding have differential impacts on physiological homeostasis in mice.

IF induced epigenetic modifications in the cerebellum that were maintained following refeeding

To establish whether IF influences the epigenetic landscape, we carried out ChIP (chromatin immunoprecipitation) of the H3K9me₃ locus in the cerebellum and sequenced the resulting fragments, using next-generation sequencing techniques. ChIP-seq data from the three study groups, AL, IF16, and EOD, were compared on a three-dimensional principal component analysis (3D-PCA) score plot (Fig. 2A). The 3D-PCA plot showed differential occupancy sites for IF16 and EOD compared with AL, suggesting that the gene expression patterns modulated by H3K9me₃ may be distinct between the two IF regimens. ChIP-seq data profiles at the H3K9me₃ mark around transcriptional start sites (TSSs) of annotated genes in both IF16 and EOD were different from those of AL (Fig. 2B). Gene expression at the H3K9me₃ locus was enriched in IF16 relative to AL, but appeared to be repressed in EOD mice, supporting the data from 3D-PCA plot showing that the IF regimens have distinct effects on gene expression patterns. Comparision of peak (significantly enriched genomic intervals in the ChIP-seq dataset) profiles identified 939 differential peaks in IF16, and 241 differential peaks in EOD, compared with AL (Fig. 2C). Of these, 123 differential peaks were common between IF16 and EOD (Fig. 2C). Collectively, our observations demonstrate that IF modulates the epigentetic landscape in the cerebellum by inducing differential modulation of the H3K9me₃ mark. Interestingly, some of the genes that were modulated as a result of H3K9me₃-induced changes due to IF were observed to be influenced in both the IF regimens, whereas others were found to be distinct in either the IF16 or the EOD group.

Fig. 2.

Epigenomic analysis of IF and refeeding at the H3K9me₃ locus in the cerebellum. A Three-dimensional principal component analysis (3D-PCA) plot of H3K9me₃ ChIP dataset from expression profiles of IF groups. The three most significant principal components (PC1, PC2, and PC3) are displayed on the x-, y-, and z-axes, respectively. PCA discriminated AL, IF16, and EOD into three unique cluster regions relative to their input controls. B Summary and heatmap plots displaying H3K9me₃ ChIP-seq signal mapping to a 2-kb window around the TSS of genes revealed distinct expression patterns in the AL, IF16, and EOD groups. ChIP-seq signals are sorted according to mean score, and the scale bar illustrates log₂ ratio of ChIP signal vs control signal. C Venn diagram illustrates the number of differentially expressed peaks that are common and distinct in each IF regimen, compared with those in AL. D 3D-PCA plot of H3K9me₃ ChIP dataset obtained from expression profiles of refeeding groups. The three most significant principal components (PC1, PC2, and PC3) are displayed on the x-, y-, and z-axes, respectively. PCA discriminated AL.R, IF16.R, and EOD.R into three unique cluster regions relative to their input controls. E Summary and heat map plots displaying H3K9me₃ ChIP-seq signal mapping to a 2-kb window around the TSS of genes revealed distinct expression patterns in the AL.R, IF16.R and EOD.R groups. ChIP-seq signals over the body of genes are sorted according to mean score, and the scale bar illustrates log₂ ratio of ChIP signal vs control signal. F Venn diagram illustrates the number of differentially expressed peaks that are common and distinct in each refeeding regimen, compared with those in AL.R. G, H Top 20 differentially expressed peak-associated gene ontologies of IF and refeeding compared with control plotted against statistical significance (represented as (− log₁₀ p-value). The number of differentially expressed peaks belonging to a single gene ontology term is shown in brackets beside the term. TSS, transcriptional start sites; GO, gene ontologies; DEGs, differentially expressed genes

Histone modifications induced by environmental influences are stable throughout the somatic cell division [31–35]. Hence, we investigated whether the effects of IF on H3K9me₃ were maintained following refeeding. Interestingly, our findings revealed that mice subjected to refeeding (IF16.R and EOD.R) continued to display differential occupancy sites as compared to the control group (AL.R), suggesting that the modulation of gene expression by H3K9me₃ was maintained after the termination of IF (Fig. 2D). The ChIP-seq data profiles at the H3K9me₃ mark around the TSSs of annotated genes were strikingly different for IF16.R and EOD.R compared with those for AL.R (Fig. 2E). Overall, histone marks were downregulated in IF16.R, but upregulated in EOD.R (Fig. 2E). Following refeeding, 219 and 47 differential peaks were identified in IF16.R and EOD.R, respectively, compared with those in AL.R (Fig. 2F). Of these, seven differential peaks were common between IF16.R and EOD.R (Fig. 2F). Further bioinformatic analysis to decipher the functional significance of genes associated with the differential peaks in IF16 and EOD showed enrichment of gene ontology (GO) terms related to metabolic processes and cellular transportation (Fig. 2G, Supplementary Table 1), whereas terms related to redox processes, metabolic processes, DNA damage and repair response, and transcriptional-related events were enriched for genes associated with the differential peaks in IF16.R and EOD.R (Fig. 2H, Supplementary Table 1).

Transcriptomic analysis of IF and refeeding in the cerebellum

Following the changes detected in the epigenetic landscape, we next profiled the global cerebellum transcriptome through RNA sequencing, to ascertain the alterations in gene expression following IF and refeeding. Partial least square-discriminant analysis (PLS-DA) grouped the data into three clusters largely corresponded to the three study groups—AL, IF16, and EOD (Fig. 3A). The clustering patterns suggested that IF16 and EOD induced differential effects on global gene expression, compared with AL. Notably, unsupervised hierarchical clustering of the data also resulted in distinct segregation into AL, IF16, and EOD groups (Fig. 3B). Global gene expression profiles of IF16 and EOD were more similar to each other, and strikingly different from that of AL, although specific gene clusters showed stark differences between IF16 and EOD (Fig. 3B), suggesting that IF induces time-dependent changes in the transcriptome. This finding is consistent with the results of PLS-DA analysis, and further reinforces the findings from the blood analyses. Volcano plots of the results of differential expression analysis (Fig. 3C) identified 892 genes that were significantly differentially expressed in IF16 compared with AL. Of these, 453 genes were upregulated, while 439 genes were downregulated in IF16 (Fig. 3C). On the other hand, 1472 genes were significantly differentially expressed in EOD compared with AL, with 552 genes being upregulated and 920 genes being downregulated (Fig. 3C). We carried out GO enrichment analysis for the differentially expressed genes, to identify the biological processes affected by both the IF regimens. The top 20 enriched GO terms for IF16-induced differentially expressed genes were associated with circadian rhythm process, transcription-related events, and steroid lipid metabolic processes (Fig. 3D, Supplementary Table 2). In addition to transcription-related events and steroid lipid metabolic processes, the differentially expressed genes in EOD were associated with cell proliferation, and differentiation through modulation of signaling pathways, such as Ras protein, platelet-derived growth factor receptor (PDGFR), and epidermal growth factor receptor (EGFR) signaling pathways, suggesting a wider repertoire of transcriptional modulation with prolonged IF (Fig. 3D, Supplementary Table 2). In summary, IF modulates the transcriptome in a time-dependent manner, which is in agreement with the results of ChIP-seq analysis.

Fig. 3.

Transcriptomic analysis of IF and refeeding in the cerebellum. A Partial least square-discriminant analysis (PLS-DA) plot of IF transcriptomic expression profiles. Axis values are the explained variation of each variate. PLS-DA categorized the AL, IF16, and EOD transcriptomic datasets into three unique cluster regions represented by the respective ellipse and background color. B Heatmap of transcriptomic expression data showing differentially expressed genes in IF. Unsupervised hierarchical clustering segregated the AL, IF16, and EOD transcriptomes distinctly. Gene expression is shown in log₁₀(FPKM + 1) and differentially expressed genes were selected based on p-value < 0.05. C Volcano plot of statistical significance (− log₁₀ q-value) against enrichment (log₂ fold change) of differentially expressed genes in IF16 and EOD against AL. Total number of differentially expressed genes is shown in brackets. Upregulated genes are shown in orange and downregulated genes are shown in blue. Non-significant differentially expressed genes are shown in black. D Top 20 differentially expressed gene ontologies of IF compared with the control plotted against statistical significance (represented as (− log₁₀ p-value). Number of differentially expressed genes belonging to a single gene ontology term is shown in brackets beside the term. E PLS-DA plot of transcriptomic expression profiles in refeeding mice. Axis values show the explained difference between each variate. PLS-DA categorized the AL.R, IF16.R, and EOD.R transcriptomic datasets into three unique cluster regions represented by the respective ellipse and background color. F Heatmap of transcriptomic expression data showing differentially expressed genes in refeeding mice. Unsupervised hierarchical clustering distinctly segregated the AL.R, IF16.R, and EOD.R transcriptomes. Gene expression is shown in log₁₀(FPKM + 1), and differentially expressed genes were selected based on p-value < 0.05. G Volcano plot of statistical significance (− log₁₀ q-value) against enrichment (log₂ fold change) of differentially expressed genes in IF16.R and EOD.R against AL.R. Total number of differentially expressed genes is shown in brackets. Upregulated genes are presented in red and downregulated genes are presented in blue. Non-significant differentially expressed genes are shown in black. H Top 20 gene ontologies of differentially expressed genes in refeeding vs. control groups plotted against statistical significance (represented as (− log₁₀ p-value). Number of differentially expressed genes belonging to a single gene ontology term is shown in brackets beside the term. FPKM, fragments per kilobase of transcript per million mapped reads

Fasting and refeeding have differential effects on the transcriptome profile in different organisms and organs [36–38]. We examined the impact of refeeding on the transcriptome, using PLS-DA analysis, and found distinct clusters of AL.R, IF16.R, and EOD.R, with minimal overlap, indicating significant differences in the transcriptome profiles across the groups despite similar and abundant food intake (Fig. 3E). Unsupervised hierarchical clustering further showed distinct expression profiles across the three groups (Fig. 3F). Volcano plots identified 1326 genes that were significantly differentially expressed between IF16.R and AL.R. Of these, 457 genes were upregulated, while 867 genes were downregulated in IF16.R (Fig. 3G). On the other hand, 1180 genes were significantly differentially expressed between EOD.R and AL.R; 630 genes were upregulated, and 550 genes were downregulated in EOD.R (Fig. 3G). Analysis of the functional significance of genes that were differentially expressed in IF16.R, compared with AL.R, showed association with cellular transportation process; translation-related events; nervous system development, including dendrite morphogenesis and neuron migration and projection; neurotransmitter activities; learning; fatty acid biosynthesis; and oxidative phosphorylation (Fig. 3H, Supplementary Table 2). On the other hand, differentiation of genes propagated by EOD.R, compared with AL.R, was related to neurotransmitter activities, cellular transportation process, translational-related events, and sphingolipid and glycogen metabolic processes, as well as autophagy mechanisms (Fig. 3H, Supplementary Table 2). These results indicate that fasting and refeeding have heterogeneous effects on the transcriptome, and provide further evidence that these regimens induce differential adaptive molecular responses to energy restriction and abundance.

IF and refeeding regulated distinct as well as common biological processes

We performed an integrative analysis of the ChIP-seq datasets from IF and refed mice, to compare the epigenetic landscape at H3K9me₃ sites following the two regimens. The 3D-PCA plot showed a clear separation of the ChIP-seq dataset from each biological condition (Fig. 4A). Notably, IF16 and IF16.R were closer to each other, compared with EOD and EOD.R. Results obtained from the normalized strand cross-correlation curve showed strong fragment-length peaks, whereas fingerprint plots demonstrated patterns of broad repressive mark, typical of H3K9me₃, highlighting high-quality control of the ChIP-seq dataset (Fig. 4B, C). Moreover, the normalized strand cross-correlation curve demonstrated a higher focal enrichment signal for IF16 and a lower localized enrichment signal for EOD than for AL (Fig. 4B). Overall, the enrichment signal of the IF ChIP-seq dataset was apparently higher than that of the refeeding dataset, which was less distinct across the biological conditions (Fig. 4C). Normalized heatmap analysis across the groups showed that the overall histone mark patterns around the H3K9me₃ locus were enriched in IF16, but were downregulated in EOD, relative to AL (Fig. 4D). In contrast, the overall gene expression pattern for IF16.R was downregulated while that for EOD.R was upregulated, compared with that for AL.R (Fig. 4D). Further, global gene expression around the TSS at the H3K9me₃ locus was generally higher in the refeeding group than in the IF group. Broadly, these results further suggest that the regimens exert a differential impact on the H3K9me₃ landscape.

Fig. 4.

Integrative epigenomic analysis at the H3K9me₃ locus in the cerebellum, on IF and refeeding. A 3D-PCA plot of normalized H3K9me₃ ChIP dataset shown by expression profiles of IF and refeeding groups. The three most significant principal components (PC1, PC2, and PC3) are displayed on the x-, y-, and z-axes, respectively. PCA categorized all the biological groups based on unique cluster occupancy relative to their input control. B Normalized strand cross correlation (SCC) plot for ChIP-seq data of IF and refeeding mice. The ChIP-seq SCC curves show local maxima in the fragment sizes. C Fingerprint plots assessing the relative focal signal strength and specificity of ChIP signal vs input signal of both IF and refeeding dataset. Both input and biological group datasets demonstrated good coverage of reads across the genome. D Summary and heatmap plots displaying normalized H3K9me₃ ChIP-seq signal mapping to a 2-kb window around the TSS of genes revealed distinct expression pattern in the biological samples of IF and refeeding groups. Normalized ChIP-seq signals over the body of genes are sorted according to mean score, and the scale bar illustrates log₂ ratio of ChIP signal vs. control signal. E, F Venn diagram illustrates the number of differentially expressed peaks at the H3K9me₃ locus that are common and distinct between IF and refeeding regimens. G Dot plot for age-associated changes on differentially expressed peak ontologies. Gene ontology semantic is shown on the left axis, and gene ratio is shown on the x-axis. Dot size is proportional to the number of genes, and p-values are represented by a color range, from blue (low) to red (high)

We carried out a network analysis on the functional annotations of genes associated with differential peaks (Supplementary Fig. 3). Cluster profiling of the resulting network revealed that genes associated with differential peaks in IF16, compared with AL, orchestrate a myriad of functions (Supplementary Fig. 3A). These functions can be categorized into several metabolic processes, such as fatty acid, glycerolipid, carbohydrate (glucose and glycogen), and protein metabolism. The genes associated with the differential peaks are also related to a plethora of signaling pathways, including the G-protein coupled receptor (GPCR), Janus kinase (JAK), signal transducer and activator of transcription protein (STAT), bone morphogenetic protein (BMP), Wnt, and integrin-mediated signaling pathways. Furthermore, they also facilitate transcription-related events, transport, circadian rhythm, and chromatin organization. Cluster profiling of genes associated with the differential peaks in IF16.R, compared with AL.R, showed fewer functional annotations (Supplementary Fig. 3A). These functional annotations showed terms associated with metabolic processes, such as glycogen, peptide, quinolinate, collagen, and chondroitin sulfate proteoglycan metabolism. Other functions of the differential peak-associated genes included those related to the GPCR signaling pathway, cyclic adenosine monophosphate (cAMP) kinase activity, pyrimidine nucleotide transport into mitochondria, and small RNA interference. Despite the presence of fewer clusters in the profile output following refeeding, we observed a total of 118 differential peaks that were maintained following refeeding after the IF16 regimen (Fig. 4E). Cluster profiling of these differential peaks highlighted functions related to lipoprotein lipase activity, protein ubiquitination, quinolinate catabolic process, tetrahydrofolate metabolism, and the glycogen biosynthetic process (Supplementary Fig. 3A).

Compared to AL, EOD differential peaks were associated with functions related to lipid metabolism (e.g., fatty acid biosynthetic process, low-density lipoprotein particle receptor catabolic process, and sterol metabolic process and signaling), carbohydrate metabolism (e.g., malate and acetyl-coA metabolic processes), protein catabolism, tetrahydrofolate metabolic process, and mucopolysaccharide and heparan sulfate proteoglycan metabolic processes (Supplementary Fig. 3B). These differential peaks were also associated with neurotransmitter biosynthesis (e.g., serotonin) and endosomal transport, complement activation, histone modification and RNA processing, and regulation of the Wnt signaling pathway. Furthermore, cluster profiling of EOD.R vs. AL.R differential peaks revealed fewer functional annotations compared with EOD vs. AL (Supplementary Fig. 3B). These peaks were enriched for glycogen and proteoglycan biosynthetic processes, and quinolinate and tetrahydrofolate metabolic processes. The peaks were also related to the GPCR signaling pathway, DNA repair, anatomical structure development, and sensory perception of mechanical stimuli (Supplementary Fig. 3B). A total of 156 differential peaks were maintained between EOD and EOD.R, compared with AL and AL.R, respectively (Fig. 4F). These peaks mediate key metabolic processes, such as carbohydrate metabolism (e.g., carbohydrate utilization, and GDP L-fucose and glycogen metabolic processes), quinolinate, and exogeneous drug catabolism as well as catalyzing processes, such as equilibrioception, RNA interference, complement activation, and neurotransmitter secretion (Supplementary Fig. 3B). To determine whether age-associated changes have a confounding effect on the observed functional annotations, we investigated age-associated differential peaks between AL and AL.R (Fig. 4G, Supplementary Table 3). Our results revealed significant ontologies belonging to terms such as lipid and steroid metabolic processes, transcriptional-related events, organism development processes (e.g., palate, organ morphogenesis, motor neuron axon guidance, brain and cerebellum development), and redox processes. Many of these age-associated changes were evident in functional ontologies of both IF16.R and EOD.R, as well as in IF.

Our findings raise the possibility that age is an important factor influencing epigenetic reprogramming at the H3K9me₃ locus, and it may have a confounding effect on the changes brought about by refeeding. Thus, it is unlikely that the epigenetic maintenance reported in this study is due only to IF, with no age-associated effects.

Integrative transcriptomic dataset of IF and refeeding demonstrated distinct and common regulation of a plethora of metabolic processes

We conducted an integrative transcriptomic analysis between the IF and refeeding datasets. The PLS-DA plot showed that the AL, IF16, and EOD transcriptomic datasets were separated from each other, occupying three unique cluster regions (Fig. 5A). Heatmap analysis revealed stark variations in gene expression patterns among the different groups (Fig. 5B). Gene network analysis revealed that genes differentially expressed in IF16, compared with AL, belonged to various functional clusters spanning metabolic processes, such as fatty acid catabolism, sphingolipid metabolism, amino acid, and mitochondrial pyruvate transport, as well as protein ubiquitination and deubiquitination (Supplementary Fig. 4A). Genes differentially expressed in IF16.R, compared with those in AL.R, were related to, among other processes, protein ubiquitination and peptide and ATP metabolism (Supplementary Fig. 4A). A total of 83 differentially expressed genes were maintained between IF16 and IF16.R, compared with AL and AL.R, respectively (Fig. 5C). These genes are involved in a plethora of metabolic processes, including 3′-phosphoadenosine 5′-phosphosulfate and S-adenosylmethionine biosynthesis, quinolinate catabolism, and xylulose and estrogen metabolic processes (Supplementary Fig. 4A). In addition, genes involved in the GPCR and Hippo signaling pathways, transport (e.g., zinc ion import into synaptic vesicles, nuclear retention of unspliced RNA, and urea transmembrane transport), cellular differentiation (e.g., B cell and epithelial cell differentiation), replication fork maintenance, interferon-gamma secretion, cellular response to chromate, and collagen fibril organization were also maintained between the feeding conditions.

Fig. 5.

Integrative transcriptomic analysis of IF and refeeding in the cerebellum. A PLS-DA plot of normalized IF and refeeding transcriptomic expression profiles. Axis values are the explained differences in each variate. PLS-DA categorized the AL, IF16, and EOD datasets, distinctly represented by the respective ellipse and background color; however, the AL.R, IF16.R, and EOD.R transcriptomic datasets demonstrated little segregation. B Heatmap of normalized IF and refeeding transcriptomic expression data showing differentially expressed genes across biological groups. Gene expression is shown in log₁₀(FPKM + 1), and differentially expressed genes were selected based on p-value < 0.05. C, D Venn diagram illustrates the number of differentially expressed genes that are common and distinct between IF and refeeding regimens. E Dot plot for age-associated changes on differentially expressed gene ontologies. Gene ontology semantic is shown on the left axis, whereas gene ratio is shown on the x-axis. Dot size is proportional to the number of genes, and p-values are represented by a color range, from blue (low) to red (high)

Compared to AL, differentially expressed EOD-induced genes were involved in steroid biosynthesis, protein ubiquitination, and glycosylphosphatidylinositol metabolism (Supplementary Fig. 4B). These genes also mediate the GPCR and TOR signaling pathways, anatomical structure development, and ribosome biogenesis. Conversely, relative to AL.R, differentially expressed EOD.R-induced genes were involved in GPCR signalling pathways as well as metabolism-related activities, such as peptide biosynthesis and transport, protein ubiquitination, RNA biosynthesis, and generation of precursor metabolites and energy (Supplementary Fig. 4B). Despite a lower GO output compared to both IF16 vs. AL and IF16.R vs. AL.R, a greater number of differentially expressed genes was maintained (a total of 132) between EOD vs. AL and EOD.R vs. AL.R (Fig. 5D). Notably, these genes are involved in metabolism (e.g., citrate metabolic process, amylopectin biosynthesis, quinolinate catabolism, protein ubiquitination, dipeptide, and lipid transport), transcription-related events, and the GPCR and target of rapamycin complex 1 (TORC1) signaling pathways (Supplementary Fig. 4B). These genes also mediate negative regulation of other processes (e.g., dendritic spine maintenance, FasL, and CD4 biosynthesis), modifications (e.g., histone succinylation, tRNA aminoacylation for protein translation, peptidyl-aspartic acid hydroxylation, and peptidyl-glutamic acid modification), anatomical structure development, regulation of muscle system process, cerebellar granule cell precursor proliferation, cellular response to glial cell–derived neurotrophic factor, sodium transport, and establishment of cell polarity.

Since we observed age-associated changes in the epigenome, we investigated whether age was a confounder in the transcriptome dataset (Fig. 5E). Gene ontologies of differentially expressed, age-associated genes belonged to transcription and translation events, transportation, DNA damage and repair responses, mitochondrial electron transport, TOR signaling, protein folding and destabilization, and apoptosis. Many of these gene ontologies overlapped with some of the terms that were enriched during refeeding and maintained in differentially expressed genes suggesting a possible influence of age in addition to feeding effects.

Integrative epigenomic and transcriptomic data highlighted robust metabolic processes regulated by H3K9me₃ modulation, during IF, in a temporal-dependent manner

We investigated the profile of differentially expressed genes that were regulated by H3K9me₃ modulation in both regimens, using integrative multi-omics. A total of 208 differentially expressed genes governed by H3K9me₃ were common between IF16 and AL (Fig. 6A). As IF modulation of the H3K9me₃ mark resulted in robust metabolic process changes, we focused on gene ontologies related to the term “metabolism” and categorized them into three major arms: carbohydrate, lipid, and protein (Fig. 6B). Among the 208 differentially expressed genes, 89 were responsible for regulating metabolism. Between IF16 and AL, genes associated with carbohydrate and steroid metabolic processes were upregulated, whereas those involved in lipid and protein metabolic processes were downregulated (Fig. 6B, Supplementary Table 4). In addition, many of these genes were involved in more than one function, suggesting that the conglomerate of these 89 genes is necessary for reflecting the overall metabolic changes observed. We then broadly separated the upregulated and downregulated differentially expressed genes governed by the H3K9me₃ locus and analyzed their metabolism-related pathway changes (Fig. 6C). Notably, upregulated genes mediated the pathways related to circadian rhythm, and various signaling pathways (e.g., mitogen-activated protein kinases [MAPK], Ras-related protein 1 [Rap1], and Ras), as well as aspects of metabolism (e.g., sphingolipid, sulfur, and vitamin B6). Conversely, downregulated genes were involved in sphingolipid, phospholipase D, mTOR, Hippo, and adipocytokine pathways, as well as in the regulation of lipolysis in adipocytes and bile secretion (Supplementary Fig. 5A, Supplementary Table 5). Only 25 H3K9me₃-controlled differentially expressed genes were observed between IF16.R and AL.R (Fig. 6D). However, gene ontologies of these 25 genes revealed only two broad categories of positive regulation in the apoptotic signaling pathway and regulation of DNA-templated transcription (Fig. 6E). We analyzed the overlap between H3K9me₃-regulated genes that were differentially expressed between IF16 and IF16.R, relative to AL and AL.R, respectively, to investigate whether metabolic changes were maintained following refeeding, and found that no differentially expressed genes intersected the two regimens (Fig. 6F).

Fig. 6.

Integrative epigenomic and transcriptomic analyses of IF16 and IF16.R at the H3K9me₃ locus in the cerebellum. A The Euler’ diagram illustrates the number of differentially expressed genes governed by H3K9me₃ modulation during IF16, compared with that in the control. B Bar chart categorizing the H3K9me₃-governed differentially expressed genes during IF16 vs AL into three categories of metabolic gene ontologies; carbohydrate, lipid, and protein metabolic processes. Gene ontology subsets of each category are shown at the bottom of each bar chart. Statistical significance is plotted as − log₁₀(adj p-value) on the y-axis, whereas z-score is represented in the form of a color bar to illustrate whether a particular semantic term is more likely to increase or decrease. Red represents an increasing z-score, whereas blue represents a decreasing z-score. C Chord diagram showing the most enriched carbohydrate, lipid, and protein-related metabolic processes with H3K9me₃-governed differentially expressed genes during IF16 vs AL. In each chord, enriched gene ontologies are presented on the right, whereas differentially expressed genes contributing to this enrichment are presented on the left. Each differentially expressed gene is represented by a rectangle whose color is correlated with the expression level determined by log(fold-change), where red represents upregulation and blue represents downregulation. Chords connect these differentially expressed genes with gene ontology terms, with each term being represented by a single colored line. D The Euler’ diagram illustrates the number of differentially expressed genes governed by H3K9me₃ modulation during IF16.R, compared with that in the control. E 3D-pie chart to illustrate H3K9me₃-governed differentially expressed genes during IF16.R vs AL.R. Gene ontologies are presented at the bottom of the pie chart, whereas gene symbols belonging to each semantic gene ontology are shown at each category of the pie chart. F Venn diagram illustrates the number of differentially expressed genes that are governed by H3K9me₃ modulation and are maintained between IF16 and IF16.R, compared with that in the controls

Similar analysis revealed 423 differentially expressed genes governed by H3K9me₃ between EOD and AL (Fig. 7A). GO analysis revealed upregulation of carbohydrate, protein, fatty acid, and triglyceride metabolic processes, and downregulation of steroid metabolism (Fig. 7B, Supplementary Table 4). Chord diagram analysis revealed that 175 of the 423 differentially expressed genes were responsible for regulating metabolism (Fig. 7C). Two other notable details were observed in the case of EOD compared to IF16. EOD showed a large number of differentially expressed genes, involved in a number of metabolic processes; many of the upregulated genes controlled the protein metabolic axis (Fig. 7C). Analysis of the differentially expressed genes revealed a higher number of upregulated than downregulated terms. The upregulated, differentially expressed genes governed a plethora of signaling pathways (e.g., Rap1, Forkhead box O [foxO], sphingolipid, longevity regulating, cyclic adenosine monophosphate [cAMP], 5′-adenosine monophosphate kinase [AMPK], Ras, MAPK, phosphatidylinositol, Toll-like receptor, glucagon, mTOR, thyroid hormone, adipocytokine, and relaxin), as well as circadian rhythm, sphingolipid metabolism, folate and unsaturated fatty acid biosynthesis, insulin resistance, parathyroid hormone synthesis, secretion and action, and fatty acid elongation. Downregulated, differentially expressed genes mediated the pathways related to steroid and unsaturated fatty acid biosynthesis, metabolism (e.g., cysteine, methionine, pyrimidine, glutathione), and vitamin digestion and absorption, as well as circadian rhythm (Supplementary Fig. 5B, Supplementary Table 5). Analysis of differentially expressed H3K9me₃-controlled genes revealed 18 genes between EOD.R and AL.R (Fig. 7D), categorized into two broad groups: phospholipid metabolic process and myelination (Fig. 7E). An analysis of the overlap in H3K9me₃-regulated differentially expressed genes between EOD and EOD.R, relative to AL and AL.R, respectively, showed no differentially expressed genes that intersected the two regimens (Fig. 7F).

Fig. 7.

Integrative epigenomic and transcriptomic analyses of EOD and EOD.R at the H3K9me₃ locus in the cerebellum. A The Euler’ diagram illustrates the number of differentially expressed genes governed by H3K9me₃ modulation during EOD, compared with that in the control. B Bar chart categorizing the H3K9me₃-governed differentially expressed genes during EOD vs AL into three categories of metabolic gene ontologies; carbohydrate, lipid, and protein metabolic processes. Gene ontology subsets of each category are shown at the bottom of each bar chart. Statistical significance is plotted as − log₁₀(adj p-value) on the y-axis, whereas z-score is used in the form of a color bar to illustrate whether a particular semantic term is more likely to increase or decrease. Purple represents increasing z-score value, whereas cyan represents a decreasing z-score value. C Chord diagram showing the most enriched carbohydrate, lipid, and protein-related metabolic processes with H3K9me₃-governed differentially expressed genes during EOD vs AL. In each chord, enriched gene ontologies are presented on the right, whereas differentially expressed genes contributing to this enrichment are presented on the left. Each differentially expressed gene is represented by a rectangle whose color is correlated to the expression level determined by log(fold-change), wherein red represents upregulation and blue represents downregulation. Chords connect these differentially expressed genes with gene ontology terms, with each term represented by one colored line. D The Euler’ diagram illustrates the number of differentially expressed genes governed by H3K9me₃ modulation during EOD and EOD.R, compared with that in the controls. E 3D-pie chart illustrates the H3K9me₃-governed differentially expressed genes during EOD.R vs AL.R. Gene ontologies are presented at the bottom of the pie chart, whereas gene symbols belonging to each semantic gene ontology are shown at each category of the pie chart. F Venn diagram illustrates the number of differentially expressed genes that are governed by H3K9me₃ modulation and are maintained between EOD and EOD.R, compared with those in the control

As there was a temporal difference in the modulation of differentially expressed genes by H3K9me₃ in both IF16 and EOD, we looked at plausible interactions between H3K9me₃ peaks and the transcriptome track of selected genes, to reinforce our findings (Fig. 8). Evaluation of representative genes, distinctly modulated only in IF16 and EOD groups (e.g., Aldob and Gbe1, respectively), showed that a peak change at the H3K9me₃ locus resulted in upregulated expression of both genes (Fig. 8A, B). Next, we compared representative genes that were modulated in both IF16 and EOD (e.g., Elovl2 and Fads2) by H3K9me₃. Similar to previous observations, H3K9me₃ modulation resulted in upregulation of Elovl2 expression and downregulation of Fads2 expression (Fig. 8C, D), highlighting the distinct impact of varying the period of energy restriction, on relative gene expression patterns. Thus, IF can affect gene expression in a temporal manner, impacting metabolic changes differently. Indeed, when we mapped the differentially expressed genes that were governed by H3K9me₃ changes in both IF regimens onto a metabolic map, more nodes were observed during EOD than IF16, and the metabolic profile was very different (Supplementary Fig. 6A and B). During IF16, most differentially expressed genes were involved in carbohydrate metabolism, whereas there was a metabolic switch from carbohydrate metabolism to lipid and amino acid metabolism in EOD (Supplementary Fig. 6B). When we focused on the anabolic and catabolic changes in these metabolic pathways as a whole, we observed that the metabolic utilization of key macromolecules differed greatly between IF16 and EOD. For instance, biosynthesis of carbohydrates, amines, and polyamines was higher in EOD than in IF16, but the reverse was true for biosynthesis of amino acids, fatty acids, and lipids (Supplementary Fig. 7A). On the other hand, there was a lower degradation of carbohydrates, amines, and polyamines, but a higher degradation of amino acids, fatty acids, and lipids in EOD than IF16 (Supplementary Fig. 7B). Collectively, there was decreased maintenance of differentially expressed genes at the H3K9me₃ locus following refeeding. This reinforces the notion that neither IF16 nor EOD are robust enough to induce epigenetic reprogramming at this locus to maintain the transcriptomic changes, even with the abolition of IF. Instead, it appears that IF16 and EOD induce metabolic changes in response to energy restriction through modulation of the H3K9me₃ locus in the cerebellum. However, our findings indicate that EOD can induce more robust metabolic switching than IF16, providing a novel insight that temporal regulation of H3K9me₃ leads to distinct metabolic switching processes in the cerebellum.

Fig. 8.

Integrative Genome Visualization (IGV) tracks illustrating H3K9me₃ binding sites at selected gene loci following IF. A, B IGV track displaying H3K9me₃ binding sites at Aldob and Gbe1 gene loci following IF16 and EOD. C, D IGV track displaying H3K9me₃ binding sites at Elovl2 and Fads2 gene loci following IF16 and EOD. Blue background show H3K9me₃-dependent regulation of chromatin accessibility at the selected sites. Location of selected gene locus is shown at the top of each plot

Discussion

Accumulating evidence indicates that IF has numerous benefits for metabolic health, and is a plausible medical intervention. However, many studies have shown differing benefits of IF, because of confounding factors arising from differences in periods of time-restricted feeding as well as inter-individual differences in response to IF [39–42]. As such, the present study evaluated the temporal effects of IF by adopting different timepoints during IF and refeeding, and considered the epigenetic milieu of IF and refeeding to understand how this regulatory cascade can help to explain any response differences.

Considerable evidence has shown that CR and epigenetics are interlinked, thereby providing a clearer understanding of various aspects of regulation of gene expression patterns through this axis. H3K9me₃ is a key player in CR-induced modulation, which in turn affects metabolic adaptation and aging processes [5, 6]. This epigenetic locus has also been widely implicated in metabolic modulation in different organs and disease states, suggesting that H3K9me₃ has an important role in metabolic adaptation [43–45]. However, there is no available information on whether IF may also affect the epigenome. Here, we established that both IF and refeeding regimens can distinctly modulate the H3K9me₃ tag. Our data indicate a novel mechanism through which both IF and refeeding influence metabolo-epigenetics at the H3K9me₃ locus, similar to observations in CR studies. Moreover, H3K9me₃ can affect biological processes, such as redox and DNA damage and repair responses under excessive energy balance [46–48]. This was also observed in our study following refeeding. Our dataset is therefore consistent with other studies and indicates a role for H3K9me₃ modulation as a notable player in this dietary context.

The cerebellum affects circadian oscillation, which controls food anticipation behavior [10, 11], and functions as an epigenetic clock during aging [7]. Given the large population of neuronal cells residing within the cerebellum, and the corresponding energy demand, it is hypothesized that the cerebellum should be highly sensitive to energy restriction [49–51]. Indeed, the transcriptomic dataset for both IF and refeeding reveals robust changes in gene expression, compared with the control, indicating that both regimens act via a transcriptomic axis. It was also observed that the duration of fasting and refeeding distinctly affected the transcriptome. IF16-induced differentiation of genes is associated with the circadian rhythm process, which may explain the chronobiological changes in the cerebellum during food anticipation towards energy deprivation [10, 11]. Moreover, both IF regimens modulated steroid lipid metabolic changes, highlighting metabolic switching processes typical of fasting, which in turn may affect neurodevelopment and neurodegeneration in the cerebellum [36, 52]. In addition, 24-h fasting induces stem cell regeneration by metabolic switching [53]. Here, we also established that EOD differentially expressed genes modulate cellular proliferation and differentiation by modulating signalling pathways, such as Ras protein, PDGFR, and EGFR. Our data offer a transcriptomic lens for understanding the mechanistic aspects of how fasting impacts different signalling arms to regulate proliferation and differentiation in the cerebellum.

Integrative epigenetic and transcriptomic analyses of fasting and refeeding datasets revealed both to have a profound effect on H3K9me₃; however, IF appeared to have a greater effect on the transcriptome than refeeding. Network construction analysis of the epigenome showed that both IF16 and EOD induce H3K9me₃ modulation, which in turn regulates fatty acid (e.g., fatty acid oxidation and biosynthesis), carbohydrate (e.g., acetyl-CoA and tricarboxylic acid cycle), and protein (e.g., protein catabolism) metabolism. At the transcriptomic level, we also observed such changes in lipid (e.g., fatty acid catabolism and steroid biosynthetic process), carbohydrate (e.g., amino acid and mitochondrial pyruvate transport), and protein (e.g., protein de-ubiquitination and ubiquitination) metabolism, which are suggestive of coordinated control of such metabolic events. Interestingly, our findings also show that, despite only an 8-h difference in energy restriction, different facets of metabolic processes are distinctly modulated in IF16 and EOD. For instance, we observed glucose and glycogen metabolic changes during IF16, but malate and acetyl-CoA metabolic changes were only observed during EOD, as a result of H3K9me₃ modulation. Therefore, different arms of cellular metabolic pathways are distinctly modulated in different IF regimens to provide different sources of energy metabolites to meet energy demands. Besides metabolism, both IF16 and EOD could induce other differential biological changes at the epigenomic and transcriptomic levels, including various signalling pathways, anatomical structure development, and RNA splicing events.

Refeeding following IF also appears to affect both the epigenome and transcriptome, thereby governing several distinct biological processes in a temporal-dependent manner. In contrast to fasting, fewer genes and peaks were impacted upon refeeding, indicating reduced effects on the epigenome and transcriptome. Moreover, certain functional maintenance was apparent at both the H3K9me₃ locus and in gene expression, suggesting that epigenetic reprogramming of gene expression may occur as a result of IF. Our study highlighted age as a potential confounding factor in the precise mapping of the effects of refeeding following IF, as observed through differences in body weight and expression profiles of AL and AL.R mice. Indeed, many previous studies have also emphasized that aging has a profound effect on the level of H3K9me₃ marks and transcriptome patterns in the cerebellum [54–57]. Consistent with these reports, we found that many differences in genes and peaks observed to be mediating biological effects in refeeding overlapped with age-associated biological changes. Hence, it is possible that age affects epigenetic reprogramming at the H3K9me₃ locus, which makes it indeterminate that most of the epigenetic maintenance reported in this study between both types of regimens is induced solely by IF.

Collectively, our study shows, for the first time, that fasting can affect H3K9me₃, which regulates many biological changes relevant to IF, some of which have been reported in different organs. In addition, since both IF regimens affected the epigenome and transcriptome distinctly, our data provide a plausible explanation that a distinct modulation of both the epigenome and transcriptome axes may produce different outcomes, which in turn may help to explain the lack of standardization in reported IF effects, due to different IF regimens being employed. Given that several metabolic processes at both epigenomic and transcriptomic levels were modulated fasting, whether the transcriptomic changes governing these metabolic process changes are governed by H3K9me₃ modulation needs to be studied. Our multi-omics analysis revealed that differentially expressed genes from both the IF regimens governed by H3K9me₃ were involved in a variety of metabolic processes. Between IF16 and AL, carbohydrate and steroid metabolic processes appeared to be upregulated, whereas lipid and protein metabolic processes were downregulated. However, a comparison of EOD with AL revealed a metabolic shift, whereby carbohydrate, protein, fatty acid, and triglyceride metabolic processes were upregulated, with a corresponding downregulation of steroid metabolism. These metabolic changes in both the IF regimens appeared to be orchestrated by both unique and common genes. For instance, Aldob (responsible for glycolysis) [58, 59] and Gbe1 (responsible for glycogen storage) [60, 61] were uniquely upregulated in IF16 and EOD, respectively. However, Elovl2 (involved in de novo lipogenesis, lipid storage, and subsequent fat mass expansion) [62, 63] and Fads2 (involved in biosynthesis of polyunsaturated fatty acids) [64, 65] were upregulated and downregulated, respectively, in both IF16 and EOD, in a time-dependent manner. In addition, different signalling arms were distinctly modulated in IF16 and EOD. For example, common pathways, such as the adipocytokine, mTOR, and sphingolipid metabolic pathways, were downregulated in IF16 but upregulated in EOD. In contrast, unique signalling pathways, such as FoxO and AMPK, were distinctly modulated only in EOD. For certain signalling pathways, we observed that different genes were modulated in the two regimens. For instance, in the MAPK pathway, Angpt1 was involved only in IF16, whereas Egfr was involved only in EOD. Indeed, global metabolic mapping revealed different metabolic profiles in IF16 and EOD, with asymmetrical triggers of different milieu of carbohydrate, lipid, and protein cellular metabolic pathways, as well as varying degrees of anabolism and catabolism of these macromolecules. While most of the signalling pathways have previously been reported to be influenced by fasting [66–71], our findings suggest a temporal link between the signalling response and the period of energy restriction. This is achieved via differential manipulation of the epigenetic and transcriptomic cascade, resulting in varied metabolic arms being triggered to respond to energy demand.

In the case of refeeding, differentially expressed genes governed by H3K9me₃ were involved in the positive regulation of apoptotic signalling and transcription in IF16.R and phospholipid metabolism and myelination in EOD.R. Many studies have reported the role of H3K9me₃ in influencing these biological processes [72–76]. However, despite subjecting biological groups to excess energy balance following IF, it is observed that the differentiation of certain genes is distinctly modulated in a time-dependent manner at the H3K9me₃ site. While age may be a confounding factor in mediating these changes, we explored the possibility of epigenetic reprogramming and memory due to IF. Our findings show a lack of maintenance of differentially expressed genes at the H3K9me₃ locus. This may mean that IF16 and EOD are not sufficiently robust to induce epigenetic reprogramming at this locus following refeeding, or that H3K9me₃ may not be the sole epigenetic mechanism of IF. Therefore, the roles of other epigenetic regulatory mechanisms of IF should be considered for epigenetic maintenance.

While other epigenomic modifications may be modulated in response to IF regimens in a spatial-oriented manner, here we focused on H3K9me₃ epigenomic locus to answer a novel mechanism that shows how IF affects the metabolo-epigenetics axis in regulating a myriad of metabolic processes to bring about robust metabolic switching responses in the cerebellum. Decoding the precise crosstalk between epigenetics and transcriptional rewiring in the context of IF, and its impact on molecular memory and other mechanisms, will be useful in elucidating and gauging the overall benefits of IF.

Supplementary information

Supplementary data associated with this article can be found in the online version.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

G. Y. Q. N., R. S. Y. F., D. G. J., and T. V. A. conceived and designed the study. G. Y. Q. N., D. P. L. K. S., S. W. K., D. Y. W. F., J. K., and A. A. S. conducted animal experiments. G. Y. Q. N., D. P. L. K. S., S. W. K., H. B., J. P., J. L., E. K., S. P., J. W. H., and V. K. carried out the analyses and prepared the figures. G. Y. Q. N. and D. P. L. K. collected the sequencing data. E. O., T. D., M. P. H., R. V., K. M., L. H. K. L., and B. K. K. participated in the discussion of the project. G. R. D., C. G. S., J. G., and M. P. M. provided technical support to the project. G. Y. Q. N., R. S. Y. F., D. G. J., and T. V. A. drafted the manuscript. C. G. S., J. G., and M. P. M. edited the manuscript. R. S. Y. F., D. G. J., and T. V. A. supervised the analysis. The authors read and approved the final manuscript.

Funding

The Singapore National Medical Research Council Research Grants (Grant No. NMRC-CBRG-0102/2016 and NMRC-OFIRG-036/2017) supported this work. This study also supported by the National Research Foundation (NRF) funded by the Korean Government (Grant no. NRF-2019R1A2C3011422 and NRF-2019R1A5A2027340).

Availability of data and material

High-throughput sequencing data have been submitted to the NCBI Sequence Read Archive (SRA) under accession number GSE135945.

Code availability

Not applicable.

Declarations

Ethics approval

All animal procedures were approved by the National University of Singapore Animal Care and Use Committee and performed according to the guidelines set forth by the National Advisory Committee for Laboratory Animal Research (NACLAR), Singapore. All aspects of the study were performed in accordance with the ARRIVE (Animal Research: Reporting In Vivo Experiments) guidelines.

Consent to participate

Not applicable.

Consent for publication

Not applicable.

Conflict of interests

The authors declare no competing interests.