Repeated fasting events sensitize enhancers, transcription factor activity and gene expression to support augmented ketogenesis

Abstract

Mammals withstand frequent and prolonged fasting periods due to hepatic production of glucose and ketone bodies. Because the fasting response is transcriptionally regulated, we asked whether enhancer dynamics impose a transcriptional program during recurrent fasting and whether this generates effects distinct from a single fasting bout. We found that mice undergoing alternate-day fasting (ADF) respond profoundly differently to a following fasting bout compared to mice first experiencing fasting. Hundreds of genes enabling ketogenesis are ‘sensitized’ (i.e. induced more strongly by fasting following ADF). Liver enhancers regulating these genes are also sensitized and harbor increased binding of PPARα, the main ketogenic transcription factor. ADF leads to augmented ketogenesis compared to a single fasting bout in wild-type, but not hepatocyte-specific PPARα-deficient mice. Thus, we found that past fasting events are ‘remembered’ in hepatocytes, sensitizing their enhancers to the next fasting bout and augment ketogenesis. Our findings shed light on transcriptional regulation mediating adaptation to repeated signals.

Graphical Abstract

Introduction

Episodes of fasting are an inherent aspect of physiology with most animals experiencing frequent and sometimes prolonged bouts of fasting (1). Mammals are exquisitely fitted to tolerate extended periods without food intake due to an adaptive response to fasting. The fasting response includes a diverse set of endocrine, metabolic and neural cues that together affect metabolism and behavior to maintain homeostasis in the face of energy shortage. A central aspect of the fasting response is the hepatic production of fuels in the form of glucose and ketone bodies that supply the energetic needs of extra-hepatic tissues. Glucose is produced by both glycogenolysis (the breaking down of glycogen) and gluconeogenesis, the de novo synthesis of glucose from precursors such as amino acids, lactate and glycerol. In ketogenesis, acetyl-CoA is used as a precursor to producing ketone bodies, mainly β-hydroxybutyrate (BHB). An abundant supply of acetyl-CoA is achieved by lipolysis of triglycerides (TG) in adipose tissue, the release of free fatty acids (FFA) to circulation, their uptake by hepatocytes and fatty acid oxidation (FAO) into acetyl-CoA. Glycogen breakdown is a quick and efficient way to supply fuel but hepatic glycogen depots quickly dwindle and are thus not a principal source of fuel during prolonged fasting bouts. In contrast, both gluconeogenesis and especially ketogenesis serve to provide fuel during prolonged fasting (2–5).

Hepatic fuel production is heavily regulated at the transcriptional level. Hundreds of genes related to gluconeogenesis, FAO and ketogenesis as well as to cellular processes enabling these pathways (uptake of extracellular precursors, inter-organelle transport, lipid catabolism etc.) are transcriptionally regulated during fasting. This widespread regulation of gene transcription is mediated by various transcription factors (TFs), some of which are activated by fasting-related extracellular signals such as hormones and metabolites. These TFs include cAMP response element-binding protein (CREB) which is activated by glucagon signaling, peroxisome proliferator-activated receptor alpha (PPARα) which is activated by fatty acids, glucocorticoid receptor (GR) which is activated by glucocorticoids and forkhead box O1 (FoxO1) which is activated by decreased insulin levels during fasting (6–8).

TFs bind cis-regulatory DNA elements termed enhancers. Upon activation and enhancer binding, TFs recruit histone-modifying enzymes, co-activators and chromatin-remodeling enzymes that together increase enhancer accessibility and promote gene transcription. Therefore, an increase in enhancer accessibility is often used as a proxy for increased enhancer activity (9,10). In addition to causing a bulk increase in enhancer accessibility, TFs also leave a ‘footprint’ on their binding site within the enhancer (measured by local protection from nuclease cleavage) (11). During fasting, thousands of enhancers are activated by fasting-related TFs, mediating the transcriptional response to fasting (12). These activated enhancers show abundant TF binding, TF footprints and increased accessibility, all resulting in fasting-dependent gene induction (6,12,13).

A multitude of studies in mammals, especially rodents and humans, examined the effects of nutritional regimens that involve recurring fasting events (i.e. intermittent fasting). The most extensively studied nutritional regimens are alternate-day fasting (ADF) and time-restricted feeding. In ADF, food is consumed ad libitumfor 24 h followed by a complete lack of food consumption (or severely restricted food consumption) for the next 24 h. In time-restricted feeding, within a 24 h window a fasting period of around 16 h is imposed following a shorter period (around 8 h) of ad libitumfeeding. In each of these regimens, far-reaching health benefits were reported in rodents and humans. These benefits include improvements in glucose tolerance, insulin sensitivity and lipid profiles as well as weight loss. Also, amelioration of obesity, diabetes, steatosis, hypertension, inflammation, certain cancers and neurodegenerative diseases were observed (14–17). A concern was raised that intermittent fasting regimens lead to an overall reduction in calorie intake, raising the possibility that the observed health benefits stem from the ensuing caloric restriction rather than from fasting per se. However, data has accumulated to show that the metabolic/hormonal/physiological state that fasting imposes leads to the striking health benefits rather than mere caloric restriction (18–20). Attempts were made to decipher the health-promoting underpinnings of intermittent fasting (21–27). However, the biological effects of intermittent fasting and their link to the resulting health benefits are still largely undetermined.

As detailed above, the response to fasting is dictated by considerable changes in enhancer activity and transcriptional regulation. Also, intermittent fasting produces long-term metabolic effects distinct from a single bout of fasting. Here, we aimed to examine if repeated fasting bouts are ‘remembered’ in gene expression programs, alter enhancer status and TF activity, thereby augmenting the response to future fasting events. Using a series of gene expression and enhancer accessibility profiling combined with gene knockout experiments, we found that repeated fasting bouts sensitize enhancers and gene expression programs to augment fuel production. We show that recurring fasting events are ‘remembered’ by transcriptional regulatory components, which prepare hepatocytes for the next fasting bout.

Materials and methods

Animals

Female, 6 weeks-old mice (C57BL/6JOlaHsd) were randomly assigned to either the unrestricted feeding (URF) or ADF groups (12 mice per group). The experiment started after 1 week of acclimation. The URF group had ad libitum access to food (Teklad TD2018) and water for 30 days. The ADF group had ad libitumaccess to food and water for 24 h followed by ad libitum access to only water for 24 h. The ADF group went through 15 cycles of fasting-refeeding (a total of 30 days). Food was removed at the beginning of the inactive phase (shortly after lights on, zeitgeber time 1). Food was put back in the cage 24 h later. On day 31 all mice in the experiment underwent 24 h of fasting. Half of the mice were euthanized at the end of the fasting period (six mice from the URF group and six mice from the ADF group). The remaining mice were refed and euthanized 24 h after the food was put back in the cage (six mice from the URF group and six mice from the ADF group). An additional group of mice that never experienced fasting (i.e. ad libitum fed continuously) was also included. At the end of the experiment mice were anesthetized and euthanized (ketamine:xylazine 30:6 mg/ml), the liver was excised and plasma was collected.

For the metabolic cages experiment, 16 mice (female, 6 weeks-old mice, C57BL/6JOlaHsd) were singly housed in metabolic phenotyping cages (Promethion Core, Sable Systems). The experiment started after 1 week of acclimation in the cages. The ADF experimental design described above was replicated with all mice undergoing ADF for 30 days. In metabolic cages, access to food was controlled remotely by programmed opening and closing of the food access control door.

Pparα hepatocyte-specific knockout (Pparα^hep−/−) mice were generated at INRAE’s rodent facility (Toulouse, France) by mating the floxed-Pparα mouse strain with C57BL/6J albumin-Cre transgenic mice, as described previously (28), to obtain albumin-Cre^+/−Pparα^flox/flox mice. Albumin-Cre^−/−Pparα^flox/flox (Pparα^hep+/+) littermates were used as controls. Genotyping was performed using an established protocol (28). A total of 14 18-week-old females Pparα^hep+/+ and 14 18-week-old females Pparα^hep−/− were randomly assigned to either the URF (six mice per genotype) or ADF groups (eight mice per genotype). The experiment started after 1 week of acclimation. The URF group had ad libitumaccess to food (Safe 04 U8220G10R) and water for 30 days. The ADF group had ad libitum access to food and water for 24 h followed by ad libitumaccess to only water for 24 h. The ADF group went through 15 cycles of fasting-refeeding (a total of 30 days). Food was removed at the beginning of the inactive phase (shortly after lights on, zeitgeber time 1). Food was put back in the cage 24 h later. On day 31 all mice in the experiment underwent 24 h of fasting and were euthanized at the end of the fasting period. Following sacrifice by cervical dislocation, plasma was collected, the liver was removed, weighed, snap frozen in liquid nitrogen and stored at −80°C until use.

All animal procedures are compatible with the standards for the care and use of laboratory animals. The research has been approved by the Hebrew University of Jerusalem Institutional Animal Care and Use Committee (IACUC). The Hebrew University of Jerusalem is accredited by the NIH and by AAALAC to perform experiments on laboratory animals (NIH approval number: OPRR-A01-5011). The Pparα^hep−/− experiments were approved by an independent ethics committee under the authorization number 45717–2023111017412475.

Glucose tolerance test

Basal Glycemia was measured after 24 h of fasting. Then, glucose (2g/kg) was injected intraperitoneally and blood glucose was evaluated every 20 min from the tip of the tail using a FreeStyle Optium Neo and glucose strips (Abbott cat# 7069470).

Insulin tolerance test

Basal glycemia was measured after 4 h of fasting. Then, insulin (0.5U/kg) was injected intraperitoneally and blood glucose was evaluated every 20 min from the tip of the tail using a FreeStyle Optium Neo and glucose strips (Abbott cat# 7069470).

RNA preparation, reverse transcription and quantitative polymerase chain reaction

Total RNA was isolated from liver pieces (30 mg) using a NucleoSpin kit (Macherey-Nagel cat# 740955) according to the manufacturer's protocol. For quantitative polymerase chain reaction (qPCR), 1 μg of total RNA was reverse transcribed to generate complementary DNA (Quantabio cat# 76047–074). qPCR was performed using C1000 Touch thermal cycler CFX96 and CFX Opus 384 instruments (Bio-Rad) using SYBR Green (Quantabio cat# 101414–276). Gene values were normalized with housekeeping genes (Gapdh). The sequences of primers used in this study are:

• Gapdh - Fwd: CAGGAGAGTGTTTCCTCGTCC, Rev: TTTGCCGTGAGTGGAGTCAT

• Hmgcl - Fwd: ACTACCCAGTCCTGACTCCAA, Rev: TAGAGCAGTTCGCGTTCTTCC

• Vnn1 - Fwd: CGCACCTGTGGTAGTTCAGT, Rev: GGTTAACACAGGTCCCGAGG

• Slc27a1 - Fwd: CAGAACTTCCCAGTCCAGACTTC, Rev: ACGTACACACAGAACGCCG

• Eci1 - Fwd: CGAGGTGTCATCCTCACGTC, Rev: GGGTTCCGGCCATACATCTC

• Got1 - Fwd: AATGATCTGGAGAATGCCCCC, Rev: GGCTGAGTCAAAGAAGGGGA

• Ppargc1a - Fwd: AAAAAGCTTGACTGGCGTCAT, Rev: TCAAGTTCAGGAAGATCTGGGC

• Fh1 - Fwd: CCAAAGAGTTTGCGCAGGTC, Rev: TCCTGGTGTTTAACCCCGTC

• Nr2c2 - Fwd: ATAATTTCCACCGACTCTGCG, Rev: GCTTGGCACTGGATGTTTCC

Western blot

Protein lysis buffer [50 mM Tris-HCl, 150 mM NaCl, 1% Triton, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulfate (SDS)] with protease inhibitors (Sigma, cat# P2714) was added to liver pieces (70 mg) followed by 1 min of homogenation (Bead Ruptor 12, Omni international) and centrifugation. Protein samples (50 μg) were loaded on 12% polyacrylamide SDS gels. Proteins were transferred (Trans-Blot Turbo, Bio-Rad; cat# 1704158) to a nitrocellulose membrane (Trans-Blot Turbo Transfer Pack, Bio-Rad; cat# 1704158), blocked for 1 h with 5% low-fat milk and incubated for 16 h with primary antibody (PPARα, Santa Cruz Biotechnology cat# sc-398394; histone H3, Cell Signaling Technologies cat# 14269) diluted 1:1000 in tris-buffered saline (0.5% Tween, 5% bovine serum albumin). Membranes were incubated with secondary peroxidase AffiniPure goat anti-mouse (1:10 000, Jackson Laboratory; cat# 115–035-146) for 1 h, followed by washes and a 1-min incubation with western blotting detection reagent (Cytiva Amersham ECL prime, cat# RPN2232). Imaging and quantification were performed with ChemiDoc (Bio-Rad).

Blood, plasma and liver measurements

Plasma samples were analyzed for TG, cholesterol, low-density lipoprotein (LDL) and high-density lipoprotein (HDL) using a Cobas c111 (Roche Diagnostics) automated clinical chemistry analyzer that was calibrated according to manufacturer guidelines. Glucose and BHB were measured directly on blood with FreeStyle Optium Neo and glucose strips (Abbott cat# 7069470) or beta ketone test strips (Abbott, cat# 7074270).

Alanine aminotransferase (ALT), aspartate aminotransferase (AST), TG and FFA were determined from plasma samples using a Pentra 400 biochemical analyzer (Anexplo facility, Toulouse, France). Plasma insulin concentration was measured using the rat/mouse insulin enzyme-linked immunosorbent assay kit (Merck).

Liver neutral lipid analysis

Hepatic lipids were extracted as previously described (29). Briefly, tissue samples were homogenized in Lysing Matrix D tubes with in 2:1 (v/v) methanol/ethylene glycol tetraacetic acid (EGTA) (5 mM). Lipids corresponding to an equivalent of 2 mg of tissue were extracted in chloroform/methanol/water (2.5:2.5:2, v/v/v), in the presence of the following internal standards: glyceryl trinonadecanoate, stigmasterol and cholesteryl heptadecanoate (Sigma-Aldrich, Saint-Quentin-Fallavier, France). Total lipids were suspended in 160 μl ethyl acetate, and the TG, free cholesterol and cholesterol esters were analyzed with gas-chromatography on a Focus Thermo Electron system using a Zebron-1 Phenomenex fused-silica capillary column (5 m, 0.32 mm i.d., 0.50 μm film thickness). The oven temperature was programmed from 200°C to 350°C at a rate of 5°C/min, and the carrier gas was hydrogen (0.5 bar). The injector and the detector were set to 315°C and 345°C, respectively.

Chromatin immunoprecipitation

Chromatin immunoprecipitation (ChIP) was performed as previously described (30) with modifications: Liver pieces (150 mg) were cross-linked with phosphate-buffered saline (PBS) containing 2 mM disuccinimidyl glutarate (Santa Cruz Biotechnology, cat# sc-285455). Livers were homogenized with a Dounce homogenizer and rotated for 30 min at room temperature. Samples were centrifuged and the pellet was resuspended with PBS containing 1% formaldehyde (Electron Microscopy Sciences, cat# 15714) for further crosslinking. After 10 min, samples were quenched with 0.125 M glycine for 5 min. Samples were then centrifuged, washed with PBS, resuspended in ChIP lysis buffer [0.5% SDS, 10 mM ethylenediaminetetraacetic acid (EDTA), 50 mM Tris-HCl (pH = 8)] and sonicated (Bioruptor Plus, Diagenode) to release 100–1000 bp fragments. Samples were diluted 1:5 with ChIP dilution buffer [170 mM NaCl, 17 mM Tris-HCl (pH = 8), 1.2 mM EDTA, 1.1% Triton X-100, 0.01% SDS]. Antibodies used: PPARα (Merck Millipore MAB3890) H3K4me1, H3K27ac and H3K27me3 (active motif, 39297, 39133, 39155, respectively). A total of 3 μg of antibody per sample was conjugated to magnetic beads (Sera-Mag, Merck, cat# GE17152104010150) for 2 h at 4°C. Chromatin was immunoprecipitated with antibody-bead conjugates for 16 h at 4°C. Immunocomplexes were washed sequentially with the following buffers: low salt buffer [0.01% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl (pH = 8), 150 mM NaCl], high salt buffer [0.01% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris-HCl (pH = 8), 500 mM NaCl], LiCl buffer [0.25 M LiCl, 1% IGEPAL CA630, 1% deoxycholic acid, 1 mM EDTA, 10 mM Tris (pH = 8.1)] and twice with Tris-EDTA (TE) buffer [10 mM Tris-HCl, 1 mM EDTA (pH = 8)]. Chromatin was eluted, de-crosslinked for 4 h at 65°C and deproteinized with proteinase K (Hy Labs, cat# EPR9016) for 1 h at 50°C. DNA was subsequently isolated using MinElute DNA purification kit (Qiagen cat# 20–28006).

The sequences of primers used in ChIP-PCR are:

• Non-enhancer region - Fwd: CAGGTCCCGGCAGACTTATC, Rev: CATCGCCTACGAGGATGAGG

• Vnn1 enhancer - Fwd: CTGCCCTCTGTGTCTTAAAAGC, Rev: GCTCGTAAACAGCCGCGA

• Hmgcl enhancer - Fwd: ATGGGGACACTGATGCGTTC, Rev: GGGAGACCTTCCCCTCGTAG

• Slc27a1 enhancer - Fwd: ACAAGCTGGATCAGGCAAGC, Rev: GGCAACCAAGCGGTGAAAAC

• Eci1 enhancer - Fwd: TCTCCAGTGTTGCCTAGCAG, Rev: TGGAACAAGGTGCATAGCAGA

RNA-sequencing (RNA-seq)

For quality control of RNA yield and library synthesis products, the RNA ScreenTape and D1000 ScreenTape kits (both from Agilent Technologies), Qubit RNA HS Assay kit and Qubit DNA HS Assay kit (both from Invitrogen) were used for each specific step. Messenger RNA (mRNA) libraries were prepared from 1 μg RNA using the KAPA Stranded mRNA-Seq Kit, with mRNA Capture Beads (KAPA Biosystems, cat# KK8421). The multiplex sample pool (1.6 pM including PhiX 1%) was loaded on NextSeq 500/550 High Output v2 kit (75 cycles) cartridge, and loaded onto the NextSeq 500 System (Illumina), with 75 cycles and single-read sequencing conditions.

ATAC-sequencing (ATAC-seq)

ATAC-seq was performed as described in our published protocol (31). Briefly, nuclei were isolated using a hypotonic buffer and Dounce homogenizer. Nuclei were tagmented using Tn5 transposase loaded with Illumina adapters. Tagmented DNA was PCR-amplified with sample-specific indices. The resulting library was size-selected to DNA fragments of 150–800 nt. The multiplex sample pool (1.6 pM including PhiX 1%) was loaded on NextSeq 500/550 High Output v2 kit (75 cycles) cartridge with paired-read sequencing conditions. Each sample was sequenced at a depth of at least 5 × 10⁷ reads.

Sequencing data analyses

Fastq files were mapped to the mm10 mouse genome assembly using Bowtie2 (32) with default parameters. Tag directories were made using the makeTagDirectory option in the HOMER suite (33). Selected gene loci were visualized by the integrated genome browser (34).

Differential gene expression

Differential gene expression was evaluated by DESeq2 (35) via the HOMER suite under default parameters. Genes were determined as differentially expressed between two conditions if they pass these cutoffs: fold change (FC) ≥ 1.5, adjusted P-value ≤ 0.05.

t-distributed stochastic neighbor embedding

Performed by the Rtsne package (R version 4.2.1). The analyzed values are log₂(Reads Per Kilobase per Million mapped reads (RPKM) + 1). Genes with RPKM < 0.5 were excluded.

k-means clustering

All genes regulated in at least one treatment (FC ≥ 1.5, adj. P-value ≤ 0.05) were included in the analysis. The normalized tag counts of each gene were used for the clustering analysis. Morpheus (https://software.broadinstitute.org/morpheus) was used to cluster genes under these parameters: k = 9; metric – one minus Pearson correlation; maximum iterations – 1000. Blue – minimum value of the gene. Red – maximum value of each gene (minimum and maximum values of each gene are set independently to other genes).

ATAC-seq analyses

Peak-calling was performed by MACS2 (narrowPeak option) (36). Differential enhancer activity was measured by DEseq2 (FC ≥ 1.5, adj. P-value ≤ 0.05). Normalized tag counts were used to visualize DESeq2 results in a scatter plot. Genomic annotations were performed by the HOMER suite (annotatePeaks option, parameter -annStats).

Bivariate genomic footprinting (BaGFoot)

BaGFoot was performed as described (13). Briefly, the three replicates from each condition (URF_f, ADF_f) were merged into a single BAM file. Accessible sites were called for each BAM file using MACS2. The footprint depth (FPD) and flanking accessibility (FA) were calculated for each known motif across all accessible sites. The difference (Δ) between ADF_f and URF_f was calculated and plotted on the bag plot.

De novo motif enrichment analysis

To unbiasedly detect enriched motifs, we performed a de novomotif enrichment analysis using the findMotifsGenome option in HOMER (parameter -size given). The entire enhancer landscape (all ATAC accessible sites across all conditions) was used as background to account for possible sequence bias. Using the entire enhancer landscape as background ensures that prevalent motifs appearing across liver enhancers will not be falsely detected as specifically enriched in the examined subset of enhancers. In motif enrichment analyses of total ATAC accessible sites and PPARα binding sites, the background was automatically selected by HOMER to account for GC bias and other sequence biases.

Aggregate plots and box plots

Tag density of ATAC or ChIP signal around ATAC site center or transcription start site were analyzed using the HOMER suite. In aggregate plots, the tag count (averaged across all sites) per site per base pair was calculated using the HOMER suite (annotatePeaks, option -size 8000 -hist 10). In box plots, tag count ±200 bp around the site center (averaged across all sites) was calculated using the HOMER suite (annotatePeaks, option -size 400 -noann). In box plots for H3K27ac, which is a broader signal, tag count ±500 bp around the site center (averaged across all sites) was calculated using the HOMER suite (annotatePeaks, option -size 1000 -noann). In both aggregate plots and box plots, the data is an average of all available replicates. In all box plots, the 10–90 percentiles are plotted.

Pathway enrichment analysis

Performed by GeneAnalytics (37).

Analysis of data from the literature

PPARα target genes in hepatocytes were found from analyzing previously published data (38,39). Differential gene expression was evaluated by DESeq2 via the HOMER suite under default parameters. Genes were determined as differentially expressed between two conditions if they pass these cutoffs: FC ≥ 1.5, adjusted P-value ≤ 0.05. ChIP-seq data for enhancer markers and liver pioneer factors were analyzed from published datasets: H3K27ac (12), p300 (40), HNF4α (41), C/EBPβ (12), FoxA1 and FoxA2 (42).

Results

To study the effects of intermittent fasting, we designed the following experiment: 8-weeks-old female mice were subjected to a 4-week ADF regimen in which animals had ad libitum access to food and water for 24 h followed by 24 h of access to only water; this group was termed ADF, for alternate-day fasting. In parallel, a control group of mice had unrestricted access to food and water throughout the 4 weeks; this group was termed URF, for unrestricted feeding (Figure 1A). Food intake of ADF mice during the 4-week period was reduced by 12% compared to URF mice, showing that during the 24-h feeding period, ADF mice almost entirely compensated for the lack of food intake during the fasting day (Supplementary Figure S1A and B). Therefore, ADF mice are only mildly calorie restricted, as shown elsewhere (21,23,26,43,44). In accordance, body mass was not reduced in ADF mice compared to URF mice (Figure 1B). To more extensively compare the effects of ADF on mouse metabolism, we housed a different group of mice in metabolic phenotyping cages, placed them on a 4-week ADF regimen and measured several parameters. Food intake and body mass were comparable to values measured in conventional cages (Supplementary Figure S1C and D). Mice ran significantly higher distances on the wheel during fasting days, a known phenomenon (45) presumably reflecting food-seeking behavior (Supplementary Figure S1E). Importantly, the respiratory exchange ratio (RER) during fasting days was 0.77 on average, showing reliance on fat oxidation. In contrast, during feeding days the RER was 0.97 on average, showing reliance on carbohydrate utilization (46) (Figure 1C and D). To check the effect of ADF on glucose homeostasis, we first measured fasting glycemia. Glycemia levels after 24 h of fasting were comparable between ADF and URF mice (Figure 1E). We then checked glucose tolerance by a glucose tolerance test (GTT) and insulin sensitivity by insulin tolerance test (ITT). ADF significantly improved both glucose tolerance and insulin sensitivity (Figure 1F and G). Taken together, these data show that young female mice undergoing a 4-week ADF regimen show slightly reduced food intake and unaltered body mass. To maintain homeostasis, mice readily switch between fuel sources, using lipids as the principal fuel during fasting days and carbohydrates during feeding days. ADF leads to improved glucose tolerance and insulin sensitivity.

Figure 1.

Robust fuel source switching and mildly reduced food intake maintain normal weight during ADF. (A) Scheme of experimental design. Mice (six per group) were put on either an URF or an ADF regimen for 30 days. Fasting periods lasted for 24 h. Livers and plasma were collected at the end of the last fasting or refeeding period. For full details, see Methods. (B) Body mass was measured weekly, showing unchanged body mass in ADF mice as compared to URF mice. (C) RER was measured continuously in metabolic phenotyping cages throughout 31 days, demonstrating a switch from carbon utilization in fed days to fat utilization in fasted days. The RER prior to fasting (day 1) and at the end of the ADF period (days 30 and 31) is statistically indistinguishable. (D) Average RER throughout the experiment duration shows preferential carbohydrate utilization on fed days and preferential lipid utilization on fast days. (E) Blood glucose levels were measured following 24 h of fasting in ADF and URF mice, showing no effect of ADF on glycemia. (F) Glucose tolerance was measured by a GTT, showing improved tolerance in ADF mice in most time points. Overall tolerance was measured by area under the curve (AUC), again showing improved tolerance after ADF. (G) Insulin sensitivity was measured by an ITT, showing improved sensitivity in ADF mice in most time points. Overall sensitivity was measured by AUC, again showing improved sensitivity after ADF. Data are presented as median (B, C) or as mean (E–G). *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001 by: unpaired two-tailed t-test (E), ordinary one-way analysis of variance (ANOVA) (B, C) or two-way ANOVA (F, G) followed by Holm–Sidak post hoc analysis. Data are presented as mean ± standard deviation (S.D.) (E-G). Biological replicates: 5–12 (B, C), 16 (D, E).

Fuel production during fasting takes place in hepatocytes where it is controlled by transcriptional and chromatin regulation (6,7). To explore whether hepatic gene expression is differentially regulated during ADF and how ADF may affect future fasting and refeeding responses, we exposed the URF and ADF groups to acute fasting and refeeding at the end of the 4-week period. Thus, URF mice served as the control whereby half of URF mice were euthanized at the end of a 24 h fasting period (this group was termed URF_f) while the other half fasted for 24 h followed by refeeding for 24 h at the end of which they were euthanized (this group was termed URF_re). Similarly, half of ADF mice were euthanized after 24 h of fasting (ADF_f) and the other half at the end of 24 h of refeeding (ADF_re; Figure 1A).

We collected livers from all groups and profiled their transcriptome by RNA-seq. To deduce fasting-dependent gene regulation, differential gene expression analyses were performed between the fasted and the refed states in both URF and ADF mice. As expected, we found that fasting elicits a major transcriptional program with thousands of genes altered in both the URF and ADF groups. A total of 2773 genes were regulated in at least one condition, i.e. induced or repressed by fasting compared to refeeding in the URF and/or ADF groups (Supplementary Table S1). Strikingly, fasting-dependent gene regulation was very different between the URF and ADF groups. Fasting led to the induction of 1109 genes in the URF group (URF_f compared to URF_re) while only 413 genes were induced in ADF_f compared to ADF_re (Figure 2A). Among these, 235 genes were induced in both URF_f and ADF_f. A similar trend was observed in fasting-repressed genes (Figure 2B). This suggests that the hepatic transcriptional response to fasting is markedly affected by previous fasting events. To examine this on a more comprehensive scale, we analyzed global gene expression patterns using t-distributed stochastic neighbor embedding (t-SNE). We found that while URF_re and ADF_re cluster closely, the ADF_f and URF_f conditions are far apart, showing they meaningfully differ in their gene expression patterns (Figure 2C). To further explore this, we compared the FC (fasting over refeeding) of each regulated gene in both the URF and ADF groups. We plotted the FC values of URF mice (x-axis) and ADF mice (y-axis). In a scenario where the fasting history of mice does not affect current fasting events, we would expect similar FC values in both ADF and URF groups, resulting in a 45° angle trendline. However, we found that most genes fell below the 45° line, resulting in a 22.3° angle trendline (Figure 2D). Thus, the general trend was that of dampened fasting-dependent gene induction in ADF, i.e. the induction level of many genes was lower in ADF compared to URF (Figure 2D, blue-shaded area). Conversely, a smaller, albeit not negligible group of genes showed higher fasting-dependent induction in ADF (Figure 2D, pink-shaded area). Given the obvious effect of ADF on fasting-dependent gene regulation, we directly compared gene expression between acute fasting in ‘first-time-fasters’ (URF_f) and ‘experienced fasters’ (ADF_f). We found that 1146 genes are differentially expressed between these two conditions (Supplementary Figure S2A and Supplementary Table S1), aligning well with data from Figure 2A–D showing marked transcriptional differences between URF_f and ADF_f. Taken together, these data show that the transcriptional response to fasting is dramatically different in mice that previously experienced recurrent fasting events compared to ‘first-time fasters’.

Figure 2.

Previous fasting events drastically affect the transcriptional response to acute fasting. (A) Normalized mRNA expression values of genes regulated by acute fasting in URF (left-hand side) and ADF (right-hand side). Inclusion criteria for induction and repression: FC ≥ 1.5, adj. P-value ≤ 0.05. (B) Overlapping the set of genes regulated by acute fasting in URF and ADF mice shows only partial overlap, indicating that the regulatory program of gene expression during fasting is affected by past fasting events. (C) All expressed genes in each condition were analyzed by t-SNE analysis, demonstrating that in the fed state URF and ADF have similar global gene expression patterns. Conversely, gene expression patterns of ADF_f are overtly different from URF_f. In all conditions, replicates cluster together, attesting to the technical quality of the data. (D) The FC values for all genes regulated by acute fasting are plotted; i.e. genes altered in URF_f compared to URF_re and/or genes altered in ADF_f compared to ADF_re. The x-axis value shows the URF_f over URF_re FC and the y-axis value shows the ADF_f over ADF_re FC. Comparing FC values between URF and ADF reveals two groups of genes: genes more strongly induced in ADF (pink-shaded area) and genes more weakly induced in ADF (blue-shaded area). Data are presented as mean (A, D). Biological replicates: 3 (A-D). Genes with RPKM < 0.5 were excluded (A, C).

Next, we aimed to get a bird's-eye view of gene expression patterns of all regulated genes across all four conditions. Therefore, we plotted the expression pattern of every gene regulated in at least one condition and clustered similar expression patterns (Figure 3A). This revealed several predominant gene expression patterns represented in clusters. In line with the t-SNE analysis, two clusters showed fasting-repressed genes which were largely unaffected by ADF (Clusters 3 and 7). Clusters 6 and 9 show fasting repression only after ADF. However, these clusters contain a relatively low number of genes and pathway enrichment analysis did not uncover liver-relevant enriched pathways involving more than a few genes (Supplementary Table S2). In contrast to fasting-repressed genes, we found two gene expression patterns showing a clear effect of ADF on genes induced in the fasted state as compared to the refed state [termed hereafter fasting-induced genes (FIGs)]. Some FIGs were less strongly induced following ADF compared to URF_f (Clusters 1, 4 and 5) while other FIGs were more strongly induced following ADF (Cluster 8). The most extreme example of the effect of ADF on FIGs is seen in Cluster 2 where only following ADF, fasting-dependent induction is observed. The expression pattern of an example gene from each cluster is depicted in Figure 3A.

Figure 3.

ADF sensitizes genes, augmenting ketogenic gene induction upon acute fasting. (A) Gene clustering of all fasting-regulated genes reveals several predominant gene expression patterns. Each row represents a gene and each column represents a mouse liver sample. Notably, a group of FIGs is induced more strongly after ADF while in a different group of genes, fasting-dependent induction is dampened by ADF. Blue: minimum expression value of the gene. Red: maximum expression value of each gene (minimum and maximum values of each gene are set independently to other genes). The RPKM values of a selected gene from each cluster are presented. (B, C) Genes from relevant clusters were filtered by the specified cutoffs to strictly define sensitized and de-sensitized genes. (D) The RPKM values of selected sensitized genes are presented, showing an increase in ADF_f compared to URF_f. (E) The RPKM values of selected de-sensitized genes are presented, showing a decrease in ADF_f compared to URF_f. Data are presented as mean ± S.D. (A, D, E). *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001 by ordinary one-way ANOVA followed by Holm–Sidak post hoc analysis (A, D, E). Only relevant pairwise comparisons are shown. Biological replicates: 3 (A–E). Ad-lib, ad libitum fed mice.

Collectively, the above data (Figure 2 and Supplementary Figures S2A and S3A) show that repeated fasting events sensitize the induction of certain transcriptional programs while de-sensitizing other transcriptional programs. To strictly define sensitization and de-sensitization of FIGs, we determined the following criteria. Sensitized FIGs (FIGs whose fasting-dependent induction is augmented in ADF) were defined as genes from Clusters 2 and 8 which pass at least one of the following statistical cutoffs: they are induced by fasting only in ADF but not in URF (Figure 2A) and/or they are induced in the ADF_f condition compared to the URF_f condition (Supplementary Figure S2A). De-sensitized FIGs (FIGs whose fasting-dependent induction is dampened in ADF) were defined as genes from Clusters 1, 4 and 5 which pass at least one of the following statistical cutoffs: they are induced by fasting only in URF but not in ADF (Figure 2A) and/or they are repressed in the ADF_f condition compared to the URF_f condition (Supplementary Figure S2A). This analysis revealed 369 sensitized FIGs and 560 de-sensitized FIGs (Figure 3B and C, and Supplementary Table S2). Taken together, these findings reveal that repeated fasting events profoundly affect gene regulation of future fasting bouts, with some genes sensitized and others de-sensitized for future fasting-dependent induction.

To get insights as to the biological functions governed by sensitized FIGs, we searched for statistically enriched pathways within this group of genes. Strikingly, 7 out of 14 highly enriched pathways were related to FAO, ketogenesis and the major transcriptional regulator of these processes – PPARα (Supplementary Table S2). To directly quantify how many sensitized FIGs are bona fide PPARα target genes, we analyzed two published datasets that determined the PPARα-dependent hepatic gene transcriptome. The two studies determined PPARα target genes by comparing the transcriptome of wild-type mice to that of liver-specific PPARα knockout mice following either fasting- or agonist-dependent stimulation of PPARα (38,39). We compiled a list of hepatic genes induced by PPARα (i.e. induced by fasting and/or agonist only in wild-type mice and not in PPARα knockout mice). We found that 41% (153) of sensitized FIGs are PPARα target genes (Supplementary Table S2). Many of these genes play a critical role in lipid catabolism and ketogenesis (e.g. Acad, Acat, Cpt2, Hadhb, Cyp4a14, Slc27a1, Slc25a20, Hmgcl, Fgf21, Vnn1, Eci2) (47). Selected examples of FAO- and ketogenesis-related, PPARα-regulated FIGs sensitized by ADF are shown in Figure 3D. We then partitioned sensitized genes based on their reliance on PPARα and found that sensitized genes not regulated by PPARα are not enriched with fasting and/or liver metabolism pathways. In contrast, pathways enriched in sensitized PPARα target genes are those related to lipid catabolism and ketogenesis (Supplementary Table S2). Therefore, a ketogenic gene program is sensitized in ADF. In this program, the induction of PPARα-regulated lipid catabolism and ketogenesis genes is augmented.

In contrast to sensitized FIGs, pathway enrichment analysis of de-sensitized FIGs did not result in an apparent pathway (selected examples of de-sensitized FIGs are shown in Figure 3E). Interestingly, a negative regulator of PPARα (TR4, encoded by Nr2c2) was also de-sensitized by ADF (Figure 3E), aligning well with the observed sensitization of PPARα target genes.

To check if sensitization also occurs in male mice, we repeated the experiment in male mice and measured gene expression. Surprisingly, genes sensitized by ADF in females were not sensitized in males, with one gene even showing an opposite pattern (Supplementary Figure S2B). In contrast, one de-sensitized gene showed a similar de-sensitization pattern also in male mice (Supplementary Figure S2C).

We wondered if the refed condition shows a gene expression pattern similar to the basal pre-fast condition or if the two states differ. In other words, we wanted to find if refeeding brings FIGs back to their basal levels. Examining the ad libitum fed condition (a group of mice collected without experiencing fasting) revealed that gene expression in the ad libitumfed condition is similar to the refed condition (Figure 3D and E). Collectively, we found that the comprehensive PPARα-controlled FAO/ketogenic program is augmented in female mice experiencing repeated fasting bouts.

The chromatin environment is a central factor in regulating the transcriptional response to fasting, with thousands of enhancers activated during fasting to regulate gene expression (12). We hypothesized that sensitized and de-sensitized responses to repeated fasting are driven by alterations in enhancer activity. Therefore, we mapped accessible regions in a genome-wide manner via ATAC-seq (48). We found a total of 182151 accessible hepatic sites across the genome in all conditions. Previous data suggests most accessible regions in the genome are cis-regulatory regions (in particular promoters or enhancers) (49). Only 6% of hepatic accessible sites we found were promoter-proximal regions (Supplementary Figure S3A), suggesting that the majority of accessible sites we found are enhancers. Accordingly, the motifs enriched in accessible hepatic sites are motifs associated with hepatic enhancers such as C/EBP, HNF4α and FoxA (Supplementary Figure S3B) (41).

To evaluate the dynamics in enhancer accessibility imposed by ADF, we measured differential accessibility between the ADF_f and URF_f groups. Remarkably, although both groups were collected after fasting, their enhancer landscape was considerably different with 23629 sites changing their accessibility due to ADF. The enhancers showing altered accessibility were roughly evenly divided between sensitized enhancers (showing increased accessibility in ADF_f compared to URF_f; 11571 sites) and de-sensitized enhancers (decreased in ADF_f; 12058 sites, Figure 4A and B, and Supplementary Table S3). This demonstrates that repeated fasting events lead to vast changes in chromatin organization with 13% of liver accessible sites showing altered accessibility following ADF.

Figure 4.

Enhancer accessibility near ketogenic genes is sensitized by ADF. (A) ATAC accessible sites with differential accessibility in ADF_f as compared to URF_f are shown. Inclusion criteria for induction and repression: FC ≥ 1.5, adj. P-value ≤ 0.05. (B) Quantification of chromatin accessibility at sensitized and de-sensitized enhancers. (C) Bivariate Genomic Footprinting (BaGFoot) analysis reveals TFs predicted to have increased (top right region) or decreased (bottom left region) activity in ADF_f compared to URF_f. TFs with a similar or identical DNA binding motif were marked with the same color. (D) Motif enrichment analysis shows TFs whose motifs are enriched in sensitized enhancers. (E) Motif enrichment analysis shows TFs whose motifs are enriched in de-sensitized enhancers. Data are presented as log₂ normalized mean tag counts (A). *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001 by unpaired two-tailed t-test (B). Biological replicates: 3 (A–C). Values with RPKM < 0.5 were omitted (A). Only motifs with −log₁₀(P-value) > 10 are shown (D, E).

A change in TF activity leading to enhancer activation is a plausible explanation for the observed widespread effect on enhancer accessibility, as has been previously shown (12,50–54). We sought to uncover the TF driving this effect on chromatin. Thus, we analyzed accessible sites in two independent unbiased approaches. First, we used BaGFoot, a tool that predicts TF activity from TF footprints as well as chromatin accessibility changes (13). BaGFoot detects all known TF motif occurrences across all accessible regions. Then, it quantifies both the accessibility flanking the motif (termed ‘flanking accessibility’) and the FPD within the motif. An increase in FPD and/or FA suggests the TF is more active in the tested condition on a genome-wide scale (13,55). Using BaGFoot, we compared the ADF_f and URF_f conditions for changes in FA and FPD across all accessible regions. We found multiple TFs with altered activity between conditions (Figure 4C), TFs with increased activity in ADF_f appear in the top right area and TFs with decreased activity in ADF_f appear in the bottom left area. The most prominent TFs with increased activity in ADF_f were PPARα, PAR bZIP TFs, Fox proteins, HNF4α and other nuclear receptors. The predominant TFs with decreased activity in ADF_f were CREB, AP-1 and ETS proteins. Thus, BaGFoot revealed several TFs associated with sensitized and de-sensitized enhancers. In addition to BaGFoot, we directly analyzed the groups of sensitized and de-sensitized enhancers (Figure 4A) to find significantly enriched TF motifs in each population. Motif enrichment analysis was highly concordant with BaGFoot results with PPARα and PAR bZIPs highly enriched in sensitized enhancers (Figure 4D). Motifs enriched in de-sensitized enhancers were also concordant with BaGFoot results with AP-1 and ETS motifs highly enriched and the CREB motif also among the top enriched motifs (Figure 4E). Fox motifs were enriched in both sensitized and de-sensitized enhancers, presumably due to the high abundance of these motifs in liver enhancers or due to the various TFs able to bind these motifs.

We also performed motif enrichment analysis in enhancers activated by fasting or refeeding in the URF or ADF conditions. As expected, we found motifs for fasting-activated TFs in URF_f-activated enhancers: PPARα, GR, CREB and C/EBPβ. The PPARα half-site was the top enriched motif in ADF_f-activated enhancers but GR, CREB and C/EBPβ were not enriched (fitting with Figure 4C and D). Among the top motifs enriched in refeeding-activated enhancers were the motif for thyroid hormone receptor (ThR), a nuclear receptor known to regulate lipogenesis (56). Of note, the motif of ThR is very similar to that of liver X receptor (LXR), a well-known factor involved in the fed response (57,58). Therefore, it is possible that the motif enrichment here reflects either ThR activity and/or LXR activity. Motifs for other feeding response TFs were also enriched: USF, E box (bound by both SREBP and ChREBP – two known TFs regulating the feeding response). Lastly, the motif for BCL6, a known repressor of PPARα (30) was also enriched (Supplementary Table S3).

The evidence from both gene expression (Figure 3 and Supplementary Table S2) and chromatin data (Figure 4C and D) clearly suggested a role for PPARα in enhancer sensitization. We examined the possibility that this is mediated via an ADF-dependent increase in PPARα protein levels. However, total PPARα protein levels were unchanged in ADF (Supplementary Figure S4A). Thus, we examined the possibility of preferred PPARα binding at sensitized enhancers. We performed chromatin immunoprecipitation sequencing (ChIP-seq) for PPARα in URF_f and ADF_f. Most PPARα binding sites were located in promoter-distal regions (89%, Supplementary Figure S4B) and the PPARα response element was the top enriched motif among PPARα binding sites, as expected (Supplementary Table S4). Strikingly, we found that PPARα occupancy was enriched in sensitized enhancers compared to de-sensitized enhancers (Figure 5A and B). Moreover, PPARα occupancy in sensitized enhancers was significantly higher in the ADF_f condition compared to URF_f (Figure 5C). PPARα occupancy was also increased near sensitized genes in the ADF_f condition as compared to the URF_f condition (Figure 5D). Therefore, even though we compared two groups in which mice were fasted for a period known to potently increase PPARα activity and binding, ADF mice show heightened PPARα activity. This aligns with the increased PPARα BaGFoot signal, PPARα motif enrichment and augmented PPARα-related gene expression in ‘experienced fasters’ (ADF_f). Indeed, the loci of ADF-sensitized PPARα-induced genes show enhancer sensitization as well as increased PPARα occupancy following ADF_f (compared to URF_f, Figure 5E). Taken together, these data show that a ketogenic gene program is increased in ADF, together with increased accessibility of enhancers and heightened binding of the major ketogenic TF – PPARα.

Figure 5.

PPARα preferentially binds sensitized enhancers following ADF. (A, B) Quantification of PPARα occupancy at sensitized and de-sensitized enhancers shows PPARα preferentially binds sensitized enhancers with minimal occupancy at de-sensitized enhancers. (C) Quantification of PPARα occupancy at sensitized enhancers in either the ADF_f or URF_f states shows higher PPARα occupancy following ADF_f. (D) Quantification of PPARα occupancy near sensitized genes in either the ADF_f or URF_f states shows higher PPARα occupancy following ADF_f. (E) Genome browser images of selected sensitized enhancers proximal to sensitized PPARα target genes show enhancer sensitization and increased PPARα binding following ADF. *P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001 by unpaired two-tailed t-test (B–D). Biological replicates: 2 (A–D). Representative replicate is presented (E).

While accessibility is a major determinant of enhancers, other factors also indicate enhancer activity. These include acetylation of lysine 27 in Histone H3 (H3K27ac) and occupancy of p300 (9). Additionally, in liver, the binding of the lineage-determining TFs FoxA1, FoxA2, HNF4α and C/EBPβ is also indicative of enhancer activity as these TFs facilitate the binding of other TFs (41). These factors are often termed pioneer TFs as they are the first to bind enhancers, leading to their increased accessibility and activity (59). To evaluate these enhancer markers in sensitized enhancers, we quantified ChIP-seq signal strength of each marker in sensitized and de-sensitized enhancers. We found that H3K27ac, p300 binding and liver lineage-determining TF binding are all substantially heightened in sensitized enhancers compared to de-sensitized enhancers (Supplementary Figure S4C). We then stratified sensitized enhancers into two classes: sensitized enhancers with or without PPARα binding sites. Sensitized enhancers harboring PPARα binding sites were dramatically more enriched with lineage-determining TF binding and enhancer marks (Supplementary Figure S4C). These data suggest that PPARα-bound sensitized enhancers are pre-bound with liver lineage-determining TFs and are enriched with enhancer markers, making them focal points with strong enhancer characteristics. Although the motifs of two other key fasting TFs, CREB and GR, were not enriched in sensitized enhancers, they serve a central role in fasting gene regulation. Thus, we wanted to compare a possible co-occurrence of liver lineage-determining TF binding with CREB and GR (in addition to PPARα). We partitioned sensitized sites to three groups: sites containing only a PPARα motif, sites containing only a CREB motif and sites containing only a GR motif (sites not containing any of these motifs and sites containing more than one motif were omitted). Comparing lineage-determining TF occupancy signal revealed that only HNF4α showed preferred binding at PPARα motif-containing sensitized sites (Supplementary Figure S4D). All other factors as well as H3K27ac levels were comparable among the different sub groups of sensitized sites.

To examine the role of histone modifications in sensitization, we performed ChIP for enhancer-related markers: H3K4me1 (a general enhancer marker), H3K27ac (a marker for activated enhancers) and H3K27me3 (a marker for de-activated enhancers) (60,61). As expected, sensitized enhancers were enriched with the H3K4me1 enhancer mark as compared to a non-enhancer region. Moreover, these enhancers were enriched with H3K27ac, an active enhancer mark but showed decreased levels of the de-activation mark, H3K27me3. No significant change in histone modifications was observed between nutritional condition, suggesting sensitization is not accompanied with changes in histone modification (Supplementary Figure S4E).

To evaluate the effects of ADF outside the liver, we measured several circulating parameters. Plasma TG and HDL levels were unchanged between groups (Supplementary Figure S5A and B). LDL levels were decreased in ADF_f (Supplementary Figure S5C), aligning with LDL lowering in humans undergoing intermittent fasting (17). The levels of BHB, the predominant ketone body, were elevated following fasting in both URF_f and ADF_f as compared to their refed counterparts. Notably, the fasting-dependent increase in ketogenesis was significantly more pronounced in ADF mice as BHB levels were higher in ADF_f compared to URF_f (Figure 6A). The augmented production of BHB in ADF mice is in complete agreement with the observed increases in FAO/ketogenic gene program, enhancer sensitization and increased PPARα binding at sensitized enhancers following ADF.

Figure 6.

The levels of BHB are augmented following ADF. (A) The plasma levels of BHB were measured after 4 weeks of ADF, showing an increase in BHB in ADF_f as compared to URF_f. (B) The plasma levels of BHB were measured after 4 weeks of ADF, showing an increase in BHB in ADF_f as compared to URF_f only in Pparα^hep+/+ mice and not in Pparα^hep−/− mice. (C) The plasma levels of FFA were measured after 4 weeks of ADF, showing a decrease in FFA in ADF_f as compared to URF_f only in Pparα^hep+/+ mice and not in Pparα^hep−/− mice. (D) The mRNA levels of selected sensitized genes were examined by qPCR after 4 weeks of ADF, showing sensitization only in Pparα^hep+/+ mice and not in Pparα^hep−/− mice. (E) The plasma levels of BHB were measured after 1 week of ADF, showing an increase in BHB in ADF_f as compared to URF_f. (F) The mRNA levels of selected sensitized genes were examined by qPCR after 1 week of ADF. Data are presented as mean ± S.D. P ≤ 0.05, **P ≤ 0.01, ***P ≤ 0.001, ****P ≤ 0.0001 by ordinary one-way ANOVA (A, E, F) or two-way ANOVA (B–D) followed by Holm–Sidak post hoc analysis. Biological replicates: 6–8. In panel (E), one outlier removed based on Grubbs’ test (alpha = 0.005).

The results in Figures 4 and 5 strongly suggest that sensitization is governed by PPARα. To causally link PPARα to sensitization and augmented ketogenesis in ADF, we employed hepatocyte-specific PPARα deficient mice (PPARα^hep-\-). We compared the effect of fasting in first-time fasters (URF) to experienced fasters (ADF). PPARα^hep-\- mice showed mildly reduced food intake following ADF (Supplementary Figure S5D). In both PPARα^hep+\+ and PPARα^hep-\- mice, ADF did not affect body mass, fasting glucose levels or plasma insulin levels (Supplementary Figure S5E–G). The plasma levels of ALT and AST were increased only in URF_f PPARα^hep-\- mice (Supplementary Figure S5H and I), as previously described (25). This indicates that lack of hepatic PPARα leads to liver damage upon acute fasting and this is resolved following repeated fasting events. In line with this, we found that liver weight is increased only in URF_f PPARα^hep-\- mice (Supplementary Figure S5J). This raises the possibility that in PPARα^hep-\- mice, acute fasting leads to liver damage accompanied by increased liver size. Indeed, high liver enzymes were associated with increased liver size (62). Hepatic free cholesterol was unaffected by PPARα or ADF, while esterified cholesterol increased in PPARα^hep-\- mice (Supplementary Figure S5K and L), a known phenomenon (28). Plasma TG were decreased in PPARα^hep-\- mice in URF but increased in PPARα^hep-\- mice following ADF (Supplementary Figure S5M). This interesting result warrants further investigation and is most likely a result of intertwined effects of fasting as well as PPARα on systemic lipoprotein metabolism. There was a trend toward an ADF-dependent increase in hepatic TG, as previously described (Supplementary Figure S5N) (25).

BHB levels were increased by ADF in PPARα^hep+/+ mice as compared to URF. In contrast, BHB levels were prominently lower in PPARα^hep−/− mice, showing no statistically significant increase in ADF (Figure 6B). Circulating FFA were reduced following ADF in control mice but not PPARα^hep−/− mice (Figure 6C). It is plausible that the high demand for ketogenic precursors in ADF mice leads to increased hepatic uptake of fatty acids, explaining the observed reduction in plasma FFA. Indeed, this reduction is absent in PPARα^hep−/− mice whose ketogenic capacity is severely impaired. Aligning with the observed elevation in ketogenesis, sensitization of genes functioning in lipid catabolism and ketogenesis was evident in control mice but abolished in PPARα^hep−/− mice (Figure 6D).

We were interested to evaluate whether the ADF-dependent changes we observed require 4 weeks of an ADF regimen or whether a shorter regimen will suffice. Similarly to BHB levels after 4 weeks of ADF, we found that BHB levels were also increased by ADF after only 1 week of ADF (Figure 6E), as was gene sensitization (Figure 6F). Collectively, these data show that ADF sensitizes the liver to produce more BHB through a PPARα-governed transcriptional program.

Taken together, in this study we show that PPARα-bound enhancers regulating lipid catabolism and ketogenesis are sensitized by repeated fasting events. Accordingly, the ketogenic transcriptional program is sensitized by ADF and as a result, the plasma levels of BHB are augmented.

Discussion

Organisms are often exposed to recurring environmental signals such as light/darkness, seasonal rhythms, cold/heat, scarcity/availability of food, etc. The response to many of these signals is brought about through regulating gene transcription. This raises two intriguing questions: (i) Do mammals adapt to frequently encountered challenges by a cellular memory mechanism? (ii) Do recurrent signals sensitize transcriptional programs to maximize future responses and increase survival?

Mammals evolved to maintain homeostasis during frequent and prolonged fasting periods. Dietary regimens that include repeated fasting bouts such as ADF are becoming increasingly popular as studies have shown their outstanding health benefits. Our study, as well as many others, show no change in body mass and little-to-no change in food intake during ADF (23,24,26,43,44). These findings preclude the possibility that the benefits of ADF are a result of reduced body mass or calorie restriction but are rather an effect of fasting per se. Therefore, while the benefits of ADF are well described, how recurrent fasting events promote them is unclear. To better understand this, we set out to decipher how ADF affects the liver’s response to future fasting events.

We found that ADF profoundly changes the transcriptional program and chromatin landscape of the liver to support a robust ketogenic program. Mice that have experienced several fasting bouts respond in a profoundly different manner to a following acute fast when compared to ‘first-time fasters’. The hepatic chromatin landscape of ‘experienced fasters’ is altered with many enhancers being sensitized for activation by previous fasting events. In turn, the induction of ketogenic genes is sensitized and the production of ketone bodies is augmented. Thus, by sensitizing enhancers and priming them for activation, ADF leads to a more robust fasting response in the following fasting bouts. The effects of repeated fasting events are only exerted during fasting periods and not in the intermittent feeding periods. Gene levels and BHB go back to basal levels during refeeding but are augmented in the next fasting bout, surpassing levels seen in mice fasting for the first time.

The transcriptional response to fasting was considerably affected by previous fasting events with some FIGs being de-sensitized (i.e. their induction was dampened following ADF) while others sensitized by ADF (i.e. their induction was augmented following ADF). What we term here as ‘gene sensitization’ is reminiscent of the coined term ‘transcriptional memory’ which was studied mostly in non-mammalian models or cultured cells. In the transcriptional memory model, a cell is able to mount a more robust transcriptional response to a signal it has previously encountered (63–65). While there are certain similarities between transcriptional memory and our observation of gene sensitization, there are important differences: First, we describe here a mammalian response to a recurring and prevalent nutritional state. Second, we describe a bifurcated effect whereby some FIGs are sensitized while others are de-sensitized. Third, transcriptional memory was not reported to be associated with enhancers but rather driven by other factors.

The group of sensitized genes was highly enriched with genes responsible for lipid catabolism and ketogenesis. Accordingly, BHB levels were increased in fasted mice following ADF. A previous study reported variable levels of ketone bodies in ADF_f that were either reduced or unchanged when compared to URF_f (25). The difference between our observations and those of Li et al. might stem from a different experimental setup. For example, we found sensitization in female mice while Li et al. used male mice, where we did not find evidence for gene sensitization. Nevertheless, in female mice we observed a reproducible increase in BHB following 1 and 4 weeks of ADF. Thus, our findings suggest that recurring fasting rewires the hepatic fasting response to augment lipid catabolism and ketogenesis. Indeed, a study in humans showed increased ketonemia after 4 weeks of ADF (66). While this observation aligns with our findings, it is important to note the difficulty in comparing mouse and human ADF as mice have a higher metabolic rate and reach fasting-induced ketosis quicker than humans. The health benefits of ADF include an improved lipid profile. It is therefore tempting to speculate that some of the health benefits of ADF are due to a shift toward lipid catabolism.

We found a major effect of ADF on the enhancer landscape with thousands of enhancers either sensitized or de-sensitized. Sensitized enhancers were associated with a FAO/ketogenic gene program. This shows that previous fasting events ‘prime’ ketogenic-related enhancers, leading to their increased activity in the next fasting bout. Several pieces of evidence link the TF PPARα to enhancer sensitization: (i) its motif is enriched within sensitized enhancers; (ii) the accessibility flanking the motif is increased following ADF; (iii) PPARα occupancy is increased in sensitized enhancers following ADF; (iv) Gene sensitization does not occur in hepatocyte-specific PPARα-deficient mice. We note that further studies are needed to fully characterize the effect of PPARα on the entire set of sensitized genes and enhancers. (v) Importantly, augmented ketonemia is significantly perturbed in hepatocyte-specific PPARα-deficient mice.

The reason for increased PPARα binding and activity during ADF is still unclear. A possible explanation may be increased production of endogenous PPARα ligands. PPARα is activated by a variety of lipophilic ligands, some of which are absorbed by the diet while others are endogenously produced by specific enzymatic pathways. Several genes were previously shown to encode proteins involved in PPARα ligand production (67–72). Additionally, FABP1 was reported to shuttle PPARα ligands into the nucleus and co-localize with PPARα (73,74). From this group of genes, six genes were induced by fasting in our dataset (Pla2g7, Pnpla2, Fabp1, Acot1, Acot2 and Acot4). Among them, three genes were also sensitized genes (Fabp1, Acot1, Acot2). This raises the possibility that repeated fasting events increase PPARα activity by increasing PPARα ligand levels. In addition, one of the genes de-sensitized by ADF is Nr2c2. This gene encodes TR4, a factor known to antagonize PPARα activity (75). Thus, the increased activity of PPARα in ADF may be supported by lowered TR4 expression.

The increase in PPARα binding and the associated increase in enhancer accessibility fits with PPARα’s known capability of recruiting co-activators, histone-modifying enzymes and chromatin remodelers (76). Indeed, PPARα as well as other TFs were shown to increase enhancer accessibility by recruitment of such factors (77).

In summary, we have shown here that repeated fasting bouts sensitize enhancers and gene induction to produce more robust ketogenesis. We showed that these effects are evident as early as 1 week after the commencement of the regimen, suggesting sensitization is a relatively quick response. We show that recurring fasting events are ‘remembered’ by transcriptional regulatory components, which prepare hepatocytes for the next fasting bout in a PPARα-dependent manner. Considering that, in addition to repeated fasting, organisms are routinely exposed to other recurring environmental signals, our findings may shed light on transcriptional regulation as a mediator of cellular adaptation to repeated signals and physiological habituation to the environment.

Data availability

All RNA-seq, ChIP-seq and ATAC-seq data have been deposited in the Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/) under accession number GSE212776.

Supplementary data

Supplementary Data are available at NAR Online.

Acknowledgements

We would like to thank Dr Walter Wahli and Dr Alexandra Montagner for providing key reagents and advice; Dr Asaf Marco and Dr Ofri Karmon for help in experiments; Dr Idit Shiff for support in high throughput sequencing and to Michael Fadi Saikali for help in R analyses. Graphical abstract created by HandMed (Dr Yitzchak Yadegari).

Author contributions: Contribution categories were adopted from CRediT (Contributor Roles Taxonomy). N.K.: Conceptualization, methodology, visualization, software, validation, formal analysis and investigation. M.C.-N.: Conceptualization, methodology, software, validation, formal analysis and investigation. J.B.: Investigation, validation and formal analysis. D.G.: Investigation, visualization and formal analysis. D.M.-A.: Investigation and validation. D.R.: Investigation. T.G.: Investigation. T.R.-F.: Investigation. A.P.: Investigation. A.N.: Investigation and formal analysis. O.G.: Investigation. M.B.-S.: Project administration. A.F.: Conceptualization, methodology, validation, investigation and formal analysis. H.G.: Conceptualization, methodology, validation, formal analysis, resources, supervision and funding acquisition. I.G.: Conceptualization, methodology, software, validation, formal analysis, investigation, resources, writing, visualization, supervision, project administration and funding acquisition

Funding

European Research Council [947907]; Israel Science Foundation [1469/19 and 3533/19]; Canadian Institutes of Health Research; International Development Research Centre; Azrieli Foundation; Agence Nationale de la Recherche (ANR-21-CE14-0079-01); Fondation pour la Recherche Medicale (Equipe FRM EQU202303016327). Funding for open access charge: European Research Council.

Conflict of interest statement. None declared.

Author notes

The first two authors should be regarded as Joint First Authors.