Methods and Protocols

The immortalized mouse hepatocyte cell‐line AML12 was obtained from ATCC (CRL‐2254) and cultured as previously described (Ploton et al, 2018). Mouse primary hepatocytes (MPH) were prepared from livers of 10‐week‐old male C57BL/6J mice (Charles River) as described in Bantubungi et al (2014) and grown on collagen‐coated plates in serum‐free William's medium (Ploton et al, 2018). Non‐parenchymal cells (NPC) from the same livers were obtained by differential centrifugation. Briefly, liver homogenates obtained after perfusion were pressed through a 70‐μm cell strainer and centrifugated for 5 min at 27 g. Pellets from this first centrifugation were washed and centrifuged twice again for 5 min at 27 g to obtain the MPH fraction. Supernatants from the first centrifugation were collected and centrifuged for 5 min at 400 g to obtain the NPC fraction. Separation of MPH and NPC was confirmed by monitoring expression of selected marker genes (Appendix Fig S2D). Acute endoplasmic reticulum stress (ERS) treatment in MPH is defined as 4‐h treatment with 1 μM thapsigargin. Vehicle (0.04% DMSO) was used as control. In all figures from this study, ERS in MPH is defined as 4‐h treatment with 1 μM thapsigargin unless indicated otherwise (shorter or longer treatment times with different concentrations were also used in some experiments as specifically indicated in the figures and their legends). Experiments involving MZ1 or JQ1 were performed by pre‐treating MPH for 3 h with 0.01, 0.1, or 1 μM MZ1 or 1 h with 500 nM JQ1 before addition of 1 μM thapsigargin for 4 h. Experiments involving C646 were performed by co‐treating MPH with 5, 10, or 20 μM C646 and 1 μM thapsigargin for 4 h. Experiments involving trichostatin A were performed by co‐treating MPH with 1 μM trichostatin A and 1 μM thapsigargin for 4 h. Experiments involving cycloheximide were performed by co‐treating MPH with 0, 1, or 10 μg/ml cycloheximide and 1 μM thapsigargin for 4 h. Experiments involving PBA, ISRIB, MG132, or PP2 were performed by pre‐treating MPH for 30 min with 5 mM PBA, 1 μM ISRIB, 10 μM MG132, or 10 μM PP2 before addition of 1 μM thapsigargin for 1 or 4 h. Experiments involving inhibitors of the three arms of the UPR were performed by pre‐exposing AML12 cells to 30 μM STF083010, 200 μM ISRIB, or 100 μM AEBSF for 2 h and subsequently treating them for 4 h with 1 μM thapsigargin or 2 μg/ml tunicamycin.

Fluorescence‐activated cell sorting (FACS)‐isolated hepatocytes were obtained by directly running the MPH fraction into an Influx sorter (Becton Dickinson) equipped with a 200 μm nozzle and tuned at a pressure of 3.6 psi and a frequency of 6.3 kHz. Sample fluid pressure was adjusted to reach an event rate of 2,000 events/s. Hepatocytes were identified as FSChi SSChi events and sorted on a “pure” mode with 80% sorting efficiency.

HEK293 cells were grown as in Ploton et al (2018) and transfected using jetPEI (Polyplus Transfection) according to the manufacturer's instructions.

All chemicals used in this study are provided in the Reagents and Tools table.

Male C57BL/6J wild‐type (WT) mice were purchased from Charles River at 8 weeks of age and housed in standard cages in a temperature‐controlled room (22–24°C) with a 12‐h dark–light cycle. They had ad libitum access to tap water and standard chow and were allowed to acclimate for 2 weeks prior to initiation of the experimental protocol. ERS was induced by intraperitoneal injection of tunicamycin using 1 μg/g mouse body weight (Sigma‐Aldrich, #T7765) or vehicle (150 mM dextrose), and liver was collected 8 h later (five mice per group). The Nfil3 ^−/− (NFIL3 KO) mice used in this study (C57BL/6J background) were previously described (van der Kallen et al, 2015). WT littermates were used as controls. Mice of 10 weeks of age were treated with tunicamycin or vehicle as described above, and liver was collected 8 h after injection (eight mice per group). To induce ERS in muscle, 30 μg tunicamycin was injected intramuscularly into the gastrocnemius muscle. The contralateral leg was injected with a saline solution and used as control. Muscles were collected 24 h after injection (nine mice per group).

Two different models of sepsis were used. For the bacterial injection model (BIM) of sepsis (sepsis^BIM), mice were injected intraperitoneally with 8 ×  10⁸ CFU of live E. coli (DH5α) bacteria or PBS (controls) and liver was collected 16 h later (six mice per group). In a separate experiment, mice were pre‐treated for 4 consecutive days with tauroursodeoxycholic acid (TUDCA; intraperitoneal injection of 500 mpk/day) or vehicle (PBS) followed by bacterial injection 2 h after the last TUDCA administration on the fourth day (10 mice per group), and sacrificed 6 h after bacterial injection which is sufficient to induce LIVER‐ID TF loss (Appendix Fig S22C). For the cecal ligation and puncture (CLP) model of sepsis (sepsis^CLP), male C57BL/6J wild‐type mice of 24 weeks of age were randomly allocated to sepsis^CLP or healthy pair‐fed control and sacrificed after 10, 30 h, or 3 days (15 mice per group per timepoint). Mice in the sepsis^CLP groups were subjected to single‐puncture CLP followed by intravenous fluid resuscitation as previously described (Derde et al, 2017). Briefly, mice were anesthetized, a catheter was inserted in the central jugular vein, and the surgical CLP procedure was performed (50% ligation of the cecum at half the distance between the distal pole and the base of the cecum and a single‐puncture through‐and‐through) followed by intravenous fluid resuscitation. They received pain medication and antibiotics 6 h after CLP and from then on every 12 h for the remainder of the experiment and mice of the “day 3” group (prolonged phase) received parenteral nutrition from the morning after surgery to mimic the human clinical situation. The data reported for the sepsis^CLP design correspond to the 10‐h timepoint (acute phase) unless indicated otherwise. Healthy pair‐fed mice were used as control.

All animal studies were performed in compliance with EU specifications regarding the use of laboratory animals and approved by the Nord‐Pas de Calais Ethical Committee (for ERS treatments and the sepsis^BIM design) or the KU Leuven Ethical Committee (P093/2014) (for the sepsis^CLP design).

Plasma aspartate aminotransferase (AST) and alanine aminotransferase (ALT) activities were determined by colorimetric assays (Thermo Fischer Scientific) using serum obtained following retro‐orbital blood collection.

RNA extraction, reverse transcription, and real‐time‐quantitative PCR (RT–qPCR) were performed as previously described (Dubois‐Chevalier et al, 2017). The primer sequences are listed in Table EV5. All primers were designed to hybridize to different exons, and generation of single correct amplicons was checked by melting curve dissociation. Murine gene expression levels were normalized using hypoxanthine‐guanine phosphoribosyltransferase (Hprt) (sepsis^CLP experiments) or cyclophilin A (PPia) (all other experiments) housekeeping gene expression levels as internal control. Human gene expression levels were normalized using 18S ribosomal RNA (RNA18S5). For gene expression analyses, in vitro experiments (AML12 and MPH) were repeated at least three times (independent experiments), each experiment being performed in technical triplicates. For in vivo mouse studies, we used at least five animals per experimental condition (genotype or treatment). The number of biological replicates is indicated in the figure legends.

RNA was extracted from MPH treated for 4 h with 1 μM thapsigargin (three independent experiments), livers of NFIL3 KO and WT littermates treated for 8 h with 1 μg/g tunicamycin (five mice per genotype per treatment), or gastrocnemius muscles of WT mice treated for 24 h with 30 μg tunicamycin (treated and contralateral control muscles from nine mice) and was checked for quantity and quality using the Agilent 2100 Bioanalyzer (Agilent Biotechnologies) before being processed for analysis using MoGene‐2_0‐st Affymetrix arrays according to the manufacturer's instructions. Data were analyzed as described hereafter and have been submitted to GEO under accession number {"type":"entrez-geo","attrs":{"text":"GSE122508","term_id":"122508"}}GSE122508.

The liver‐specificity index was calculated as the difference in normalized expression in liver and mean of normalized expression in control tissues using data from BioGPS (Table EV6).

Raw transcriptomic data from Affymetrix microarrays were normalized with Partek Genomics Suite 6.6 using background correction by Robust Multi‐array Average (RMA), quantile normalization, and summarization via median polish. Principal component analyses (PCA) were used for quality control of the data. RMA values were also used to display expression changes for selected gene sets in different figures. Differential expression analyses were performed at probeset level with Partek Genomics Suite. Dysregulated genes were defined taking into account any potential factor interaction in the original experimental design and using a Benjamini–Hochberg corrected P‐value cut‐off (FDR) set at 0.05.

Raw counts from single‐cell transcriptomic data (447 cells from E10.5 to E17.5; Yang et al, 2017) were normalized by estimation of library size factor with DESeq 1.26.0 (Anders & Huber, 2010) according to Brennecke et al (2013). PCA was performed on normalized data using FactoMineR 1.41 (Lê et al, 2008). Then, the average expression of ERS DOWN or ERS UP genes was projected for each cell on 2D PCA plot.

Genes with different temporal expression profiles were identified using the Short Time‐series Expression Miner (STEM v1.3.11; Ernst et al, 2005), which fits dynamic patterns of gene expression to model profiles. Normalized gene expressions (rpkm) were obtained from Rib et al (2018), and average expression from replicates was used. Parameters were set at “log normalize data”, 4 for “max unit change in model profiles between timepoints”, −0.05 for “minimum absolute expression change”, and FDR for “Correction method”.

To make fold changes comparable with those obtained using RNA‐seq, microarray data were normalized using the Affymetrix Power Tool (Thermo Fisher Scientific) run through the GIANT tools suite (Vandel et al, 2018) on a local instance of Galaxy (Afgan et al, 2018). Normalization was set to “scale intensity + rma” and normalization level to “probeset”. Normalized expression values retrieved for the studies which used RNA‐seq (Table EV6) were log₂‐transformed. For each dataset, a single expression value per gene was defined using gene symbols as identifiers and by averaging values obtained from replicates. Fold changes (log₂) were next computed on scaled data, which were obtained using the scale function of the «graphics» R package (R Core Team, 2015) on each dataset separately. This was performed using global mean (mean of all expression values under all conditions of interest in a given study) for “center” parameter and global standard deviation for the “scale” parameter. Only genes common to all analyzed datasets were considered for subsequent analyses, and for each dataset, the bottom 20% genes with lowest expression in the liver were discarded. Bagplots were drawn using the “bagplot” function of the “aplpack” (v1.3.2) R package using default parameters (R Core Team, 2015). Bagplots are bivariate boxplots showing the spread of the data using a “bag” containing 50% of the data points with the largest depth (around the median) and its extension by a “loop” whose limit excludes outliers (Rousseeuw et al, 1999).

Transcriptomic data of liver injuries were pooled, and batch effects were corrected with the “ComBat” function of the “sva” R pakage (Leek et al, 2019) using the mouse liver differentiation study as the batch of reference (see Table EV6 for details regarding used datasets). Parameters were set to “mean.only = T” and “par.prior = T”. Each study was defined as a different batch where the control condition (i.e., non‐injured livers) was matched to the adult liver stage of the reference dataset. Next, a PCA was computed only on the mouse liver differentiation study with the “PCA” function of FactomineR (Lê et al, 2008; using “scale.unit = F”). Liver injury studies were considered as supplemental individuals. Finally, the first principal component (representing 63.55% of the variability of the mouse liver differentiation study) was plotted and used to project the liver injury studies. Data corresponding to prenatal mouse livers were used in the analyses but were discarded for data visualization.

Functional enrichment analyses were performed using the ToppGene Suite (Chen et al, 2009). KEGG Pathways with Bonferroni‐corrected P < 10⁻³ and Gene Ontology (GO) Biological Processes with Bonferroni‐corrected P < 10⁻⁶ were considered, and similar terms were merged.

Gene set enrichment analyses (GSEA) were performed using the GSEA software (v3.0) developed at the Broad Institute (Subramanian et al, 2005). We used 1,000 gene‐set permutations and the following settings: “weighted” as the enrichment statistic and “difference of classes” as the metric for ranking genes. Ranking was performed by the GSEA software using the average expression value per gene when multiple probesets were present in the microarray. In addition to enrichment plots, figures also provide NES and FDR, which are the normalized enrichment score and the false discovery rate provided by the GSEA software, respectively. In experiments with multiple conditions, the BubbleGUM tool (GSEA Unlimited Map v1.3.19; Spinelli et al, 2015) was used to integrate and compare numerous GSEA results with multiple testing correction. Non‐oriented GO term enrichment analyses were performed using the “MousePath_GO_gmt.gmt” set of genes from the Gene Set Knowledgebase (GSKB; preprint: Lai et al, 2016).

MPH (3 × 10⁶ cells) were fixed for 30 min at room temperature with disuccinimidyl glutarate followed by a 10‐min incubation with 1% formaldehyde and a 5‐min incubation with 125 mM glycine. After two washes with ice‐cold PBS, cells were scraped in PBS, pelleted by centrifugation at 400 g for 5 min, resuspended in Lysis Buffer (50 mM Tris–HCl pH 8.0, 10 mM EDTA, 1% SDS, and 1× PIC from Roche), and sonicated for 4 min (four cycles 30 s ON/30 s OFF using Bioruptor Pico from Diagenode). Mouse liver (200 mg of tissue) was cut in small pieces in ice‐cold PBS, pressed through a 70‐μm cell strainer followed by a few passages through a 18G needle. Fixation, lysis, and sonication were performed as described for MPH. Chromatin (50 μg for H3K27ac ChIP and 200 μg for BRD4 ChIP) was diluted 10‐fold in Dilution Buffer (20 mM Tris–HCl pH 8.0, 1% Triton X‐100, 2 mM EDTA, 150 mM NaCl) and incubated overnight with 2 μg of H3K27ac antibody (Active Motif, #39685) or 3 μg of BRD4 antibody (Bethyl Labs, #A301‐985A100) at 4°C. The next day, A/G sepharose bead mix (GE Healthcare) was added during 4 h at 4°C in the presence of 70 μg/ml yeast tRNA (Sigma‐Aldrich). Beads were washed three times with RIPA buffer (50 mM HEPES pH 7.5, 1 mM EDTA, 0.7% Na deoxycholate, 1% NP‐40, and 500 mM LiCl) containing 10 μg/ml yeast tRNA and once with TE buffer (10 mM Tris–HCl pH 8.0, 1 mM EDTA). DNA was then eluted in 100 mM NaHCO₃ containing 1% SDS and incubated overnight at 65°C for reverse‐crosslinking. DNA purification was performed using the MinElute PCR purification kit (Qiagen, #2800), and samples were subjected to qPCR analyses. The primer sequences are listed in Table EV5.

H3K27ac ChIP and input samples from MPH treated for 4 h with 1 μM thapsigargin or vehicle (0.04% DMSO) from three independent experiments were additionally sent for sequencing on an Illumina Hi‐seq 4000 as single‐end 50‐base reads according to the manufacturer's instructions. Data were analyzed as described hereafter and have been submitted to GEO under accession number {"type":"entrez-geo","attrs":{"text":"GSE122508","term_id":"122508"}}GSE122508.

ChIP‐seq data quality control and uniform reprocessing including mapping to the mm10 version of the mouse genome and signal normalization have been described in Dubois‐Chevalier et al (2017) except Bowtie 2 (sensitive mode; Langmead et al, 2009) was used for the BRD4 ChIP‐seq analyses. ChIP‐seq data were visualized using the Integrated Genome Browser (IGB 9.0.1; Freese et al, 2016).

H3K4me3 ENCODE ChIP‐seq data from several mouse tissues (Shen et al, 2012; Table EV6) were used to call broad H3K4me3‐enriched regions using MACS2 as described in Chen et al (2015). Broad H3K4me3 domains were defined as those spanning more than three times the median size of all H3K4me3‐enriched regions in a given tissue. Broad H3K4me3 domains from mouse liver were separated into liver‐identity (LIVER‐ID) domains, which were defined as broad H3K4me3 domains specific to liver (i.e., detected in < 25% of other analysed tissues), and in ubiquitous (UBQ) domains. LIVER‐ID and UBQ domains were then assigned to genes according to overlapping TSS from the GENCODE (M9) database (Frankish et al, 2019), resulting in 621 LIVER‐ID genes and 657 UBQ genes which are listed in Table EV1. TFs within these gene lists were subsequently obtained using comparison with mouse TFs listed in the Animal TFDB 2.0 (Zhang et al, 2015). Muscle‐identity (MUSCLE‐ID) and UBQ genes, listed in Table EV4, were defined in a similar way using H3K4me3 ChIP‐seq data from the ROADMAP consortium processed by Chen et al (2015). Human to mouse gene name conversion was performed using the dbOrtho tool from bioDBnet (Mudunuri et al, 2009).

To define BRD4 super‐enhancers (SE), we first used MACS2 to identify enriched peaks (effective genome size = 2150570000, bandwidth = 300, mfold = 5–50, FDR (q‐value) = 0.05, max duplicate tags at the same location = 1) using mapped reads previously filtered to remove duplicates and reads mapping to false positives regions we had identified in Dubois‐Chevalier et al (2017). SE were identified by applying rank ordering of super‐enhancers (ROSE; Loven et al, 2013; Whyte et al, 2013) on the BRD4 peak‐calling results using mouse liver ChIP‐seq inputs ({"type":"entrez-geo","attrs":{"text":"GSE26345","term_id":"26345"}}GSE26345) as control (setting: ‐s 12500, ‐t 0).

Regions with significant changes in H3K27ac ChIP‐seq signals induced by ERS were identified using csaw 1.6.1 (Lun & Smyth, 2014, 2016). Mapped reads were previously filtered to remove duplicates and reads mapping to ENCODE blacklisted regions (Encode_Project_Consortium, 2012) or mouse ChIP‐seq false‐positive regions we had identified in Dubois‐Chevalier et al (2017). The command lines and full list of used parameters are provided in Computer Code EV1. Briefly, the genome was binned and reads counted, bins with background level signal as defined using input samples were discarded before normalization using a loess regression. Finally, after dispersion estimation with the function estimateDisp, a paired‐differential analysis was performed on this filtered and normalized data using glmQLFit. Bins overlapping H3K27ac peaks (broad regions called with MACS2 using a pool of all H3K27ac ChIP‐seq datasets and inputs as control—parameters: q‐val narrow = 0.001 and q‐val broad = 0.01) were identified using findOverlaps from GenomicRanges 1.24.3 (Lawrence et al, 2013; parameters : minoverlap = 75, maxgap = 0). Bins overlapping a single H3K27ac peak were combined using combineOverlaps, and only merged bins with FDR ≤ 0.05 were considered (merged bins with FDR > 0.05 were defined as unchanged H3K27ac regions). The ratio of UP to DOWN bins in the merged regions was next calculated, and H3K27ac UP or DOWN regions were defined as those having a ratio ≥ 2 or ≤ 0.5, respectively. Coordinates for H3K27ac UP, DOWN, and unchanged regions are provided in Dataset EV1. The bigwig signals were computed using the loess normalized signal on each dataset and/or averaging the loess normalized signal between replicates. Genes were assigned to H3K27ac regions as follows: First, genes whose TSS from the GENCODE (M9) database (Frankish et al, 2019) directly overlaps H3K27ac regions were retrieved. In addition, distal H3K27ac was linked to potentially regulated genes using CisMapper (O'Connor et al, 2017) as previously described in (Dubois‐Chevalier et al, 2017).

In order to identify TFs whose binding is enriched in H3K27ac UP and H3K27ac DOWN regions, we used Locus Overlap Analysis (LOLA 1.4.0; Sheffield & Bock, 2016) to compare TF binding within UP, DOWN, and ALL (i.e., also including H3K27ac unchanged) regions. Mouse TF‐binding sites were retrieved from the Gene Transcription Regulation Database (GTRD) (Metaclusters of GTRD release 16.07; Yevshin et al, 2017). Inputs were discarded, and ChIP‐seq datasets were ascribed to TFs using nomenclature information provided by the authors. A heatmap of log‐odds ratio was generated using the heatmap.2 function of the R package “gplots” (v3.0.1; Warnes et al, 2016) and hierarchical clustering using the hclust function of the R package “Stats” (using Euclidean distance and ward.D2 agglomeration method; R Core Team, 2015). A list of the TFs of each cluster is shown in Table EV2.

H3K27ac UP, DOWN, or unchanged regions were overlapped with cis‐regulatory modules (CRMs) defined in Dubois‐Chevalier et al (2017) based on co‐binding of 47 transcriptional regulators in mouse liver. Combinatorial co‐binding of transcriptional regulators at H3K27ac UP, DOWN, or unchanged was analyzed using multidimensional scaling (MDS) analyses as described in Dubois‐Chevalier et al (2017). Plots were performed with the smoothScatter function of the «graphics» R package (R Core Team, 2015) using a conserved color scale.

Public data used in this study were downloaded from Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/; Edgar et al, 2002), ENCODE (Yue et al, 2014), UCSC Genome Browser (Dreszer et al, 2012), or BioGPS (Mouse MOE430 Gene Atlas; Wu et al, 2016) and are listed in Table EV6.

Postmortem liver biopsies from patients admitted to the intensive care unit (ICU) of Leuven University Hospital with sepsis (n = 64), who died after a median ICU stay of 10 days (IQR 6–20 days), were compared with matched patients undergoing elective restorative rectal surgery (n = 18). Written informed consent was obtained from the patients or their closest family member and from the volunteers. The study protocols and consent forms were approved by the KU Leuven Institutional Review Board (ML1094, ML1820, and ML2707). Bilirubin was quantified with the use of a standard routine automated assay in the University Hospital Clinical Laboratory.

MPH and AML12 cells were scraped in ice‐cold PBS, pelleted by centrifugation at 400 g for 5 min, lysed in Laemmli buffer 6× (175 mM Tris–HCl pH 6.8, 15% glycerol, 5% SDS, 300 mM DTT, and 0.01% Bromophenol Blue) and sonicated for 10 min. Mouse liver was cut in small pieces in ice‐cold PBS and pressed through a 70‐μm cell strainer. The pellet obtained after centrifugation at 400 g for 5 min was lysed and sonicated as described for MPH. Western blottings shown in this study were obtained using total cellular extracts unless indicated otherwise.

MPH were scraped in ice‐cold PBS, and mouse liver was cut in small pieces in ice‐cold PBS and pressed through a 70‐μm cell strainer. Pellets were obtained by centrifugation at 400 g for 5 min, lysed in Hypotonic Buffer (20 mM Tris–HCl pH 8.0, 10 mM NaCl, 3 mM MgCl₂, 0.2% NP‐40, and 1× PIC from Roche) and incubated for 5 min at 4°C. Samples were centrifuged at 600 g for 5 min at 4°C and supernatants constituted the cytoplasmic fraction. Nuclear pellets were lysed in Nucleus Lysis Buffer (25 mM Tris–HCl pH 8.0, 500 mM NaCl, 1 mM EDTA, 0.5% NP‐40, and 1× PIC from Roche). Hypotonic Buffer and Nucleus Lysis Buffer were supplemented with Phosphatase Inhibitor Cocktail (#P044 from Sigma‐Aldrich) as well as with 5 mM Sodium Butyrate and 5 μM Trichostatin A for deacetylase inhibition. After incubation for 30 min at 4°C, samples were sonicated for 10 min and centrifuged at 16,000 g for 5 min at 4°C. Laemmli 6× was added to the supernatants which were used for Western immunoblotting.

MPH were scraped in ice‐cold PBS, pelleted by centrifugation at 400 g for 5 min, lysed in Buffer A (50 mM HEPES pH 7.5, 10 mM KCl, 1.5 mM MgCl₂, 340 mM sucrose, 10% glycerol, 1 mM DTT, and 1× PIC from Roche), and incubated for 10 min at 4°C. Samples were centrifuged at 1,300 g for 5 min at 4°C, and supernatants were discarded. Nuclear pellets were washed with Buffer A and subsequently lysed in solution B (3 mM EDTA, 0.2 mM EGTA, 1 mM DTT, and 1× PIC from Roche). After incubation for 30 min at 4°C, samples were centrifuged at 1,700 g for 5 min at 4°C and supernatants were discarded. Chromatin pellets were washed with solution B, resuspended in Buffer C (50 mM Tris–HCl pH 8.0, 1 mM MgCl₂, and 83 U/μl benzonase), and incubated for 20 min at 4°C. Laemmli buffer 6× was added before loading for Western immunoblotting.

The pcDNA3.1‐mNFIL3 (Addgene 34572) and pSGG5‐hHNF4A constructs were used for in vitro transcription and translation (in vitro TNT) using the TnT^® Quick Coupled Transcription/Translation System (Promega).

One hundred μg of proteins was separated by 10% SDS‐PAGE and immunodetected by Western immunoblotting using the primary antibodies listed in the Reagents and Tools table. Primary antibodies were detected using HRP‐conjugated secondary antibodies (Sigma‐Aldrich). Images were acquired using a G‐box (Syngene, Cambridge, UK) or using the iBright™ CL1500 Imaging System (Thermo Fisher Scientific). Quantifications were performed using Image Studio Lite v5.2 (LI‐COR Biosciences, Lincoln, USA), and band intensities were defined using the signal value (sum of the pixel intensity in a shape minus the background value).

Simple Western size‐based assays were run on a WES system as recommended by the manufacturer (ProteinSimple, San Jose, USA). Protein concentrations ranged from 0.25 to 0.8 μg/μl depending on the target protein. Primary antibodies are listed in the Reagents and Tools table. Secondary antibodies were provided by the manufacturer (PS‐MK14 and PS‐MK15, ProteinSimple). Data were analyzed using the Compass software (ProteinSimple). Quantifications were obtained using the area under the peak of the protein of interest.

MPH cells were resuspended into Hypotonic Buffer (20 mM Tris–HCl, pH 7.5, 10 mM NaCl, 3 mM MgCl₂, 0.2% NP‐40, and protease inhibitors), and the pellet was lysed for 30 min. After 10‐min sonication (30‐s on/off cycles with a Bioruptor (Diagenode) and centrifugation, 500 μg nuclear proteins from the soluble fraction were diluted with two volumes of a buffer containing 25 mM Tris–HCl pH7.5, 1 mM EDTA, 1.5 mM MgCl₂, and incubated overnight with 2 μg of p300 antibody (Active motif, #61401) or control mouse IgG (sc‐2025, Santa Cruz). Samples were then incubated for 4 h with magnetic beads (Life technologies) previously blocked with 5 mg/ml of serum albumin bovine and washed 4 times using ice‐cold washing buffer containing 25 mM Tris–HCl pH7.5, 150 mM NaCl, 1 mM EDTA, 0.2% NP‐40, and protease inhibitors. Beads were finally eluted in Laemmli buffer 6×.

Statistical analyses were performed using the Prism software (GraphPad, San Diego, CA) and R (R Core Team, 2015). The specific tests and corrections for multiple testing which were used as well as the number of samples per condition are indicated in the figure legends. In all instances, statistical significance was considered to be reached when P‐values were below 0.05, which was indicated by * or ^#. All bar graphs show means ± SD (standard deviations). Box plots are composed of a box from the 25^th to the 75^th percentile with the median as a line and min to max as whiskers.