Animal welfare

Animal handling was in compliance with the guidelines of the local animal welfare committee (permission number: 38/2013). Details on mouse treatment were described in [15]. Mice were housed in a 12/12 h light/dark cycle under constant conditions (temperature: 22 °C ± 2 °C; relative humidity: 55% ± 10%) with food and water ad libitum. Female control C57Bl6/JxDBA/2 were randomly divided into the experimental groups at the age of 3 weeks. The control group (Co) received normal chow (#1320, Altromin, Lage, Germany). The other group was fed the Lieber-DeCarli (LDC, #F1258SP, BioServ, Flemington, NJ, USA) diet as the only food source. The composition of the diets with regard to metabolic energy was 24% kcal from protein, 12% kcal from fat, and 64% from carbohydrates for the Co diet and 17.2% kcal from protein, 40.9% kcal from fat, 15.4% from carbohydrates, and 26.5% from ethanol, which equals 4% ethanol, for the LDC diet. Mice were sacrificed at the age of 9 weeks.

The diet was prepared as recommended by the manufacturer’s instructions and animals were fed as published [15, 16] with a magnetic stirrer and a magnetic stir bar. To one-third of the dry mix one-third of warm water was added, and mixed until the product dispersed. This step was repeated, ethanol was added and the product was dispensed into liquid diet feeding tubes (#13260, BioServ, Flemington, NJ, USA). Mice received the LDC diet for 1 week without ethanol, followed by 1 week of increasing ethanol concentrations: 2 days 1%, 2 days 2%, 3 days 4%. The lipid composition induced by LDC feeding was analyzed in mice treated in parallel to the animals used for hepatocyte isolation within this project and confirmed massive lipid deposition [15]. For analysis of DNA methylation related to aging, hepatocytes were isolated from female control C57Bl6/J mice at the age of 10 weeks (young) and 40 weeks (mid-aged); for DNA methylation analyses in liver tissues, livers from female control mice aged 10 (young) and 84–85 (aged) weeks were used [17].

Hepatocyte isolation

Isolation of primary hepatocytes from mouse livers was performed as described previously [18]. In brief, the liver was perfused through the vena cava with an EGTA-containing buffer, followed by a perfusion with collagenase buffer. After digestion, the liver was excised and the liver capsule was opened under sterile conditions, and the cells were released into a suspension buffer. The cell suspension was filtered through a 100 µm gauze to remove tissue debris and centrifuged for 5 min at 4 °C and 50×g. The hepatocyte pellet was washed and the centrifugation step was repeated. Aliquots of hepatocytes were cryopreserved and stored at − 80 °C until further analysis.

Colorimetric Sulfo-Phospho-Vanillin assay

The colorimetric Sulfo-Phospho-Vanillin assay was used to quantify total lipids in isolated hepatocyte samples. Freeze-dried samples were dispersed with 18 volumes of hexane/2-propanol (3:2 (v/v)) for 10 min and centrifuged for 10 min at 4 °C and 10,000×g. The supernatant was transferred into a new glass vial (#60500-1109, DURATEC Analysentechnik GmbH, Hockenheim, Germany), dried under nitrogen stream, re-dissolved in 200 µl chloroform–methanol (2:1 (v/v)), and stored at − 20 °C. As a standard solution olive oil was diluted in chloroform–methanol (2:1 (v/v)). 100 µg, 75 µg, 50 µg, 25 µg, 12.5 µg and 6.25 µg olive oil were used as a standard and handled like the samples. 5 µl of the lipid extracts were transferred into a 1.5 ml glass vial and the solvent was evaporated by incubation for 2 to 5 min at 90 °C in a drying closet. Samples were cooled to room temperature, 100 µl of sulfuric acid (95–97%, #100731.1000, Merck, Darmstadt, Germany) was added and incubated for 20 min at 90 °C. After cooling the vials down to room temperature, 50 µl vanillin-phosphoric acid (0.2 mg vanillin per ml 17% orthophosphoric acid (85%, #20624, VWR, Darmstadt, Germany) was added, followed by 10 min incubation at room temperature. 100 µl of the colored solution was transferred to a 96 well plate and the absorption was measured at 550 mm using the Sunrise™ absorbance microplate reader (Tecan Austria GmbH, Grödig, Austria).

RNA sequencing

RNA was isolated using the TRIzol method. Briefly, 1 ml TRIzol was added to 5,000,000 cells followed by vortexing, a 5-min incubation at room temperature, and addition of 200 μl chloroform. After mixing, further incubation at room temperature for 2–3 min and centrifugation (12,000g) at 4 °C for 5 min, the clear supernatant was mixed with 500 μl isopropanol and incubated at room temperature for 10 min. After further centrifugation (12,000g) at 4 °C for 10 min, the supernatant was discarded and the pellet washed with 1 ml cold 75% ethanol followed by vortexing and centrifugation (7500g, 4 °C, 5 min). The pellet was dried and dissolved in RNase-free water. The quality (QC) of total RNA as determined by TapeStation or Bioanalyzer (Agilent) was above RIN 9.5. For the preparation of the libraries, rRNA was removed with the Ribo-Zero™ Human/Mouse/Rat rRNA Removal Kit from Biozym Scientific GmbH.

The long RNA library was prepared using the TruSeq Stranded totalRNA Sample Prep Kit from Illumina (San Diego, CA) according to the manufacturer’s instructions. All samples were sequenced using an Illumina HiSeq 2500 sequencer v4 (Illumina, San Diego,CA) (1 sample/lane plus 2 × 125 bp for long RNA) at IKMB NGS core facilities.

Capturing DNA methylationBisulfite treatment and PCR

500 ng genomic DNA was subjected to bisulfite treatment using the EZ DNA Methylation-Gold Kit (Zymo Res.) according to the manufacturer's protocol. Two microliters of bisulfite treated DNA was used as template in a 30-μL reaction in the presence of 3 mM of Tris–HCl (pH 8.8), 0.7 mM of (NH4)2SO4, 50 mM of KCl, 2.5 mM of MgCl2, 0.06 mM of each dNTP, 3 U HotFire DNA polymerase (Solis BioDyne), and 167 nM of primers (Supp. Table 1). PCRs were performed at 95 °C for 15 min followed by 42 cycles at 95 °C/60 s, 54 °C (58 °C for Tns1 and Art3 reactions)/30 s, 72 °C/30 s, and a final extension 72 °C/5 min.

SNuPE/HPLC analysis

Primer extension was performed as previously described [19]. Five μl of PCR products were treated with 1U of ExoCIAP (mixture of Exonuclease I [Jena Bioscience] and Calf Intestine Alkaline Phosphatase [Calbiochem]) for 30 min at 37 °C. To inactivate the ExoCIAP enzymes, the reaction was incubated for 15 min at 80 °C. Afterwards, 14 μl of primer extension mastermix (50 mM of Tris–HCl, pH 9.5, 2.5 mM of MgCl2, 0.05 mM of ddNTPs, 1.6 μM of each SNuPE primer (Additional file 1: Table S1), and 2.5 U of Termipol DNA polymerase [Solis BioDyne]) were added. Primer extension reactions were performed at 96 °C for 2 min, followed by 50 cycles at 96 °C/30 s, 50 °C/30 s, and 60 °C/20 s. Separation of products was conducted on an XBridge BEH C18 2.5 µm 4.6 mm × 50 mm column (Waters) at 0.9 ml/ min at 50 °C by continuously mixing buffer B (0.1 M TEAA, 25% acetonitril) with buffer A (0.1 M TEAA) (Additional file 1: Table S1). Methylation indices were determined by measuring the height (h) of the methylated (M) and unmethylated (UM) peak using the equation h(M)/(h(M) + h(UM)).

Reduced representation bisulfite sequencing (RRBS)

Library preparation was conducted as described previously in a one-tube reaction [20]. Briefly, 500 ng genomic DNA was digested with MspI, repaired and A-tailed using Klenow fragment enzyme (NEB), subsequently Illumina TruSeq universal adaptors were ligated (T4 ligase, NEB). Bisulfite treatment was performed using the EZ DNA Methylation-Gold Kit (Zymo Res.) according to the manufacturer's protocol. Library preparation was accomplished by amplification with indexed TruSeq adaptor sequences (12 cycles) to add sample-specific 6 bp identifiers. Sequencing was conducted on the HiSeq2500 (Illumina) on a 100 bp single-read flow cell aiming at 25–30 Mio reads per sample.

DNaseI sequencing

DNase I sequencing (DNase-seq) was performed as previously described [21, 22]. Briefly, nuclei were isolated from 1 × 10^7 cells by using buffer A (60 mM KCl, 15 mM Tris–HCl (pH 8.0), 15 mM NaCl, 1 mM EDTA (pH 8.0), 0.5 mM EGTA (pH 8.0), 0.5 mM spermidine free base) supplemented with IGEPAL (0.1% final concentration) and incubation on ice for 15 min. Nuclei were treated with different DNase I concentrations (2,25 × 10^6 nuclei each, with 0–80U/ml) for 3 min at 37 °C and the reaction was stopped at 55 °C for 1 h with stop buffer (50 mM Tris–Cl (pH 8.0), 100 mM NaCl, 0.1% SDS, 100 mM EDTA (pH 8.0), 1 mM spermidine and 0.3 mM spermine) supplemented with proteinase K (50 µg/ml). DNA was then purified using phenol chloroform extraction and double-hit fragments of 100–500 bp were selected by sequential purifications with Agencourt AMPure XP Beads (Beckman Coulter, Brea, USA). Sequencing libraries were prepared from 8 ng of purified DNA (from 80 U/ml-digest) using the TruSeq ChIP Library Preparation kit (Illumina, San Diego, USA) according to the manufacturer’s protocol and sequenced on an Illumina HiSeq2500 (v3 paired-end flow cell). Raw reads were processed with the DEEP pipelines GALv1 and DHSv3 [23] (https://github.molgen.mpg.de/DEEP/comp-metadata).

Chromatin Immunoprecipitation (ChIP sequencing)

Cells were crosslinked in 1% formaldehyde for 5 min at room temperature under rotation, followed by quenching in 0.125 M glycine for 10 min. The crosslinked cells were then pelleted by centrifugation for 5 min at 4 °C. ChIP seq was performed according to the Nexson protocol [24]. In brief 625,000 cells were used per ChIP. Cell were first lysed in 500 μl of Farmen Buffer (5 mM PIPES pH8, 85 mM KCl, 0.5% Igpal, 1 × complete protease inhibitor) and then briefly sheared on a Diagenode Bioruptor Plus for 6 cycles (15 s ON/30 s OFF), to free nuclei. Nuclei were isolated by brief centrifugation at 2000 rpm for 5 min at 4 °C. Nuclei were then lysed and resuspended by homoginization with a 27 guage syringe in 440 μl of 0.33% SDS shearing buffer (100 mM NaCl, 50 mM Tris–HCl pH8.1, 0.2% NaN3, 0.33%SDS, 3% Triton X-100) and divided into 4 tubes for shearing on a Bioruptor Pico for 45 cycles (30 s ON/ 30 s OFF) on high frequency. The chromatin was then diluted to 0.11% SDS prior to peforming the ChIP on the IPstar with (50 mM Tris–HCl pH8.6, 100 mM NaCl, 5 mM EDTA pH 8.0, 0.2% NaN3). ChIPs were performed on the Diagenode IPstar automated machine using the following set up parameters: Indirect method, the Auto Histone ChIP Kit (Diagenode)-200 μl, 1 μg antibody—Diagenode H3K27ac (pAb-196–050), H3K27me3 (pAb-195–050), H3K36me3 (pAb-192–0500), H3K4me1 (pAb-194–050), H3K4me3 (pAb-003–050), H3K9me3 (pAb-193–050) and 10 h antibody incubation, 5 h bead incubation, 5 min washes. The ChIPs were then de-crosslinked on the IPstar for 4 h at 65∙C. ChIP samples were then removed and treated with 2 µl RNase A (10 mg/ml) 30 min at 37 °C followed by 3ul Proteinase K treatment for 3 h at 55 °C. DNA was purified with Zymo concentrator ChIP DNA clean up columns. ChIP DNA was then quantified by Quibit. ChIP libraries were generated using NEBNext® Ultra DNA Library Prep Kit for Illumina® (E7370S/L) according to the manufacturer’s instructions. The PCR cycles were as follows: 5 Steps: 1 Cycle: 98 °C for 30 s; 10 Cycles: 98 °C for 2 min; 1 Cycle: 98 °C for 10 s and 65 °C for 75 s; 1 Cycle: 65 °C for 5 min; Hold at 4 °C. The libraries were paired-end sequenced on an Illumina HiSeq 2500 platform.

Bioinformatic analysesDifferential gene expression analysis

RNA reads were trimmed for adapter and low-quality tails (Q < 20) with TrimGalore (http://www.bioinformatics.babraham.ac.uk/projects/trim_galore/) and subsequently aligned to the mm10 reference genome with gene models from GENCODE (version M2) 11 by the IHEC supported pipeline grape-nf (https://github.com/guigolab/grapenf/tree/35e44730f5da02a41e2aef7d97a722e20c5773f). tGrape-nf wraps STAR (version 2.4.0j)12 and RSEM (version 1.2.21)13.

DESeq2 (Version 1.18.1) was used to detect differentially expressed genes (DEGs) with maximal adjusted p-value of 0.05. Genes for which less than four samples had an FPM (fragments per million mapped reads) above 0.1 were discarded.

Differentially expressed genes (DEGs) with adjusted p < 0.05 were used for GO functional (MF, BP and CC) and biological pathway enrichment (KEGG, Reactome and WikiPathways) using g:Profiler [25].

Moreover, these DEGs were used to build the PPI network using STRING v11 [26]. The active interaction sources were taken from literature, experiments, databases, co-expression and co-occurrences of the nodes (or proteins). A confidence score (> 0.7) was chosen to get the interacting nodes. Following this, k-means clustering of the genes was performed. The optimal number of clusters (k = 4) was chosen based on the significant association and enrichment of GO terms and KEGG pathways for the genes (in each cluster).

Differential DNase peak analysis

TEPIC v2 [27] was used to compute the TF gene scores based on the differential DNase peaks regions for LDC over Co (pseudocount set to 1 e-6). A base mean cutoff of 10 was used to filter for expressed TFs from the results of differential gene expression analysis. This resulted in 265 motifs, out of 380 total motifs from JASPAR [28], HOCOMOCO [29] and KELLIS ENCODE Motif database [30] (included in TEPIC v2). The Differential DNase I peaks were obtained using edgeR [31]. The positive and negative peaks for the steatosis (LDC) vs control (Co) were identified based on positive log fold change i.e. logFC > 0 (and vice versa). GENCODE (vM21) was used for reference genome sequence annotations.

Prediction of transcription factors involved in gene regulation

DYNAMITE (from TEPIC v2 [27] package) was then used (parameters: –Ofolds = 10 –Ifolds = 6 –alpha 0.01) to identify the transcriptional regulators that regulate the differentially expressed genes (LDC over Co). The TFs were ranked based on absolute regression coefficient, and the top 15 TFs (absolute coefficient >  = 0.125) were used for further analysis. The TF effect score (or TF enrichment) is calculated (as in Eqs. 1 and 2) by the difference of the median log2 quotient Qt,i (for each TF, obtained from the TF gene score) for all the DE genes in each cluster and mean of the median Log2 quotients (of others) (Fig. 2b).

$$} = Median\left\_2}\frac^}}^}};\,\,\forall \,\,g \in cluste} \right\}$$

(1)

$$} = } - \frac\sum\limits_^ }}$$

(2)

where, At,g is the TF affinity score for TF t regulating gene g. The affinity score of a gene is the sum of TF affinity values for all open-chromatin regions (peaks) in LDC or Co in a window around gene g. Et,i is the TF effect score for transcription factor t in cluster i.

Differential chromatin domain annotation

Chromatin state segmentation tracks were computed using ChromHMM (v1.15) [32] with default parameters using the 18-state model published by the Roadmap Epigenomics Consortium [33] for the ChIP-seq data of the six histone modifications H3K36me3, H3K27ac, H3K27me3, H3K9me3, H3K4me3, H3K4me1 plus the Input control. The input short-read alignment (BAM) files were filtered with sambamba (v0.6.8) [34] to retain only properly paired reads that were not marked as duplicate or supplementary alignments, and that were aligned with a mapping quality (MAPQ) of at least 5 as reported by the read aligner.

Next, we used SCIDDO (development version #39de43a) [23] to identify larger domains of chromatin differences among the two conditions control (Co) and steatosis (Ldc). We executed SCIDDO with default parameters, using the emissions of the 18-state ChromHMM model (see above) to set the scoring scheme. The identified differential chromatin domains (DCDs) were then reduced to domains unique to the respective condition of interest (using bedtools v2.27.1 [35], command “intersect -v”), and the unique domains further intersected with gene bodies of protein-coding genes (GENCODE vM21, [36]) to obtain a list of genes putatively affected by differential chromatin marking.

Analysis of RRBS data

Raw DNA methylation data for hepatocytes and whole liver has been processed by the pipeline implemented within the DEEP project (https://github.molgen.mpg.de/DEEP/comp-metadata) to generate BED files. These files were used as input to RnBeads [37] for further analysis. We filtered for CpGs that are covered by at least 5 sequencing reads and conducted differential analysis between the aged and young samples using the limma method [38] as implemented in RnBeads [37]. The resulting p-values were corrected for multiple testing using the Benjamini–Hochberg method and the following criteria were used to select differentially methylated CpGs: mean methylation difference of at least 0.05, methylation variance in both of the groups less than 0.05 and a FDR-adjusted p-value less than 0.05.