Functional genomics uncovers the transcription factor BNC2 as required for myofibroblastic activation in fibrosis

Resources and reagents

All resources and reagents used in this study are listed in Supplementary Data File 10.

Animal studies

Mice were housed in standard cages in a temperature-controlled room (22-24 °C) with a 12-h dark–light cycle. Wild-type (WT) C57BL/6J mice were purchased from Charles River and were allowed to acclimate for at least 1 week prior to any experiment. The Ayu21-18 mouse line (B6;CBA-Bnc2Gt(pU21)18Imeg/Orl) was purchased from Infrafrontier (EM-05043). In this model, insertion of the gene trap vector pU21 in the intron separating exons 2a and 3 leads to disruption and inactivation of the Bnc2 gene⁵⁵. For genotyping, DNA was extracted from the tail with REDExtract-N-Amp Tissue PCR Kit (Sigma-Aldrich) and PCR was performed as described in ref. 55 using the primers listed in Supplementary Data File 10. Primers pairs allowing to detect the WT allele or the mutated allele (directed against the pU21 vector) were mixed in the same PCR (35 repetitions of the following amplification cycle were used: 95° for 15 s, 55° for 15 s, and 72° for 30 s). Amplicons were visualized after migration in ethidium bromide-stained agarose gels as shown in Supplementary Fig. 21. Experiments were performed using 13–19 weeks old heterozygous male mice (Bnc2^+/−) and their wild-type littermates (Bnc2^+/+) as control.

To induce liver fibrosis, WT mice (females, 10–14 weeks old) were intraperitoneally injected with CCl₄ (Sigma-Aldrich) three times a week for 8 weeks. Increasing doses of CCl₄ were used according to our previously established procedure⁶⁸ as follows: 0.1 mL/kg (week 1), 0.2 mL/kg (week 2), 0.25 mL/kg (week 3), 0.3 mL/kg (week 4) and 0.4 mL/kg (weeks 5–8). The HFSC diet and its associated NASH-like liver phenotype have been described previously³². The CDAA-HFSC diet consisted of ad libitum feeding with a choline-deficient, l-amino acid-defined (CDAA) diet supplemented with 35% sucrose, 21% fat, and 2% cholesterol. Control mice were fed a chow diet containing 12% sucrose and 7% fat (Ssniff, custom diets). Mice body weight was measured weekly throughout the course of the experiments.

All animal studies were performed in compliance with EU specifications regarding the use of laboratory animals and have been approved by the Nord-Pas de Calais Ethical Committee (APAFIS#15539-2018053011323354).

Human liver samples

Fibrotic liver samples were obtained from NASH patients from the ABOS prospective cohort (“Atlas Biologique de l’Obésité Sévère”; ClinGov NCT01129297)⁶⁹ and from patients of the TargetOH cohort (“Comparison of Inflammatory Profiles and Regenerative Potential in Alcoholic Liver Disease”; ClinGov NCT03773887; and DC-2008-642)^70,71. Patients with obesity enrolled in the ABOS cohort were deemed eligible for weight loss surgery⁶⁹. For transcriptomic analyses of biopsies from the ABOS cohorts³⁶, 53 NASH patients with a Kleiner score F3-F4 were matched for clinical parameters with 53 control patients with obesity and F0 Kleiner score (age, sex, BMI, alcohol, diabetes, statin, fenofibrate; Supplementary Data File 5) as in ref. 72. Follow-up liver biopsies from patients who underwent bariatric surgery were obtained 1 or 5 years after surgery⁶. Liver samples with alcohol-related cirrhosis were obtained from 42 patients with decompensated alcoholic-related cirrhosis who underwent liver transplantation at Huriez Hospital’s Liver Unit (Lille, France). Fragments of healthy liver samples were obtained from 20 patients who underwent liver resection for hepatic tumors (control patients; Supplementary Data File 4). The liver samples were immediately fixed for histology or frozen for RNA and protein extraction. Study is authorized by the Lille ethical committee (Lille University Hospital), and informed consent was obtained from all subjects.

Primary HSC isolation, cell culture, transfection, and treatments

Primary mouse HSCs were purified from C57BL/6J mice (male, 15–18 weeks old) according to the protocol described in⁷³. Briefly, livers were digested in situ with 14 mg pronase (Sigma-Aldrich) and 3.7 U collagenase D (Roche) followed by in vitro digestion with 0.5 mg/mL pronase, 0.088 U/mL collagenase D and 0.02 mg/mL DNase I (Roche). HSCs were then separated by a Nycodenz gradient and sorted on a FACS Aria II SORP (BD Biosciences) based on retinoid autofluorescence using the 355 nm laser for excitation and 450/50 nm band-pass filter for detection. Cytometry data were analyzed using FlowJo v10.5.3 (FlowJo, LLC). The purity of HSCs was assessed by measuring the percentage of ultraviolet (UV; retinol autofluorescence) and Desmin-positive cells (Supplementary Fig. 11A).

Primary HSCs and HSC cell lines [human LX2 (Merck, scc064) and murine EMS404 (Kerafast, #EMS404)] were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 5% fetal bovine serum (FCS, Deutscher). Human primary HSC (Supplementary Data file 10) were grown in DMEM supplemented with 15% FCS. LL-29 cell line was grown into F-12 media supplemented with 15% FCS (Deutscher). AML12 and IHH cells were cultured as described in^74,75. All cells were grown in the presence of 100 U/mL penicillin-streptomycin (Gibco) at 37 °C in a humidified atmosphere with 5% CO₂. As the EMS404 cell line⁷⁶ had not been thoroughly used in the literature, we verified the proper expression of MF markers and their ability to respond to TGFβ (Supplementary Fig. 14). For TGFβ stimulation, cells were serum-starved for 24 h in DMEM supplemented with 0.5% FCS and 0.2% BSA before incubation with TGFβ (R&D systems) at 1 ng/mL for 24 h in serum-free media. Primary mouse MF-HSCs were subjected to this protocol 4 days after isolation. Additional treatments with 1 μM verterporfin (Sigma-Aldricht), 0.5 μM MZ1 (Tocris) or 1 μM pomalidomide, thalidomide, CC-885 or Iberdomide (MedChemExpress) were directly added to the growth medium. The TGFβ signaling inhibitors SD208 (Tocris) and SB431542 (Stemgent) were used at 1 and 10 µM, respectively. Spheroids were obtained by growing mouse primary HSCs in U-bottom cell repellent plates (Greiner) for 9 days as described in ref. 47. Non-embedded spheroids were used to ensure inactivation of YAP1^47,77. Reagents used for cell culture treatments are listed in Supplementary Data File 10.

Transfection of siRNA (used at 20 nM) was performed using lipofectamine 2000 (Invitrogen, Carlsbad, CA, USA) for EMS404 cells, JetPrime (Polyplus transfection) for LX2 cells and HiPerfect (Qiagen) for primary HSCs, according to the manufacturer’s instructions. Transfected cells were cultured for 48 h prior to terminal assays. For Yap1 silencing in mouse primary HSCs, cells were transfected twice (at day 3 and day 6 after isolation) with a siRNA targeting Yap1 (siYAP1) or non-targeting control siRNA (siCTRL), and cells were harvested 48 h after the last transfection. Control and target siRNAs directed against human or mouse genes were purchased from Thermo Fisher Scientific (Supplementary Data File 10). Plasmid transfection in LX2 cells was performed using JetPEI (Polyplus transfection) according to the manufacturer’s instructions. Transfected cells were cultured for 48 h prior to terminal assays. Plasmids used in these experiments are listed in Supplementary Data File 10.

Isolation of primary mouse liver cell types

Simultaneous purifications of murine primary cell populations from the same livers were performed using a modified version of the method described in ref. 74. Briefly, livers were digested in situ with 100 U/ml of collagenase type IV (Sigma). Hepatocytes (Hep.) were enriched through a 50×g centrifugation prior to size-based FACS sorting, whereas the non-parenchymal fraction was further processed for the isolation of hepatic stellate cells (HSC), Kuppffer cells (KC), cholangiocytes (Chol) and endothelial cells (LSEC). The purity of each cell type was assessed by RT-qPCR for cell-specific markers (Supplementary Fig. 11B).

RNA extraction and RT-qPCR

Tissues were homogenized using a T10 Ultra-Turrax (Ika). The superior right lobe was systematically used for RNA extraction from mouse livers. Total RNA was extracted using Trizol (Eurobio) for cell lines and liver tissues or using the Nucleospin® RNA Plus XS kit (Macherey-Nagel) for primary cells. RNA was reverse-transcribed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystem). RT-qPCR was performed in technical triplicates using the SYBR green Brilliant II fast kit (Agilent Technologies) on an Mx3005p apparatus (Agilent Technologies) or a QuantStudio 3 (Applied Biosystems). The specificity of the amplification was checked by recording the dissociation curves, and the efficiency was verified to be above 95% for each primer pair. mRNA levels were normalized to the expression of housekeeping genes as indicated in Supplementary Data File 10, and the fold induction was calculated using the cycle threshold (ΔΔCT) method. The sequences of primers used are listed in Supplementary Data File 10.

Gene expression microarray

Purified RNA was further digested with rDNase (Macherey-Nagel). RNA integrity and quantity were evaluated using the Agilent 2100 Bioanalyser (Agilent Technologies). RNA was then processed for transcriptomic analysis using Affymetrix GeneChip arrays (HTA 2.0 or MoGene 2.0) according to the manufacturer’s instructions. All liquid handling steps were performed by a GeneChip Fluidics Station 450 and GeneChips were scanned with a GeneChip Scanner 3000-7 G using Command Console v4.1.2 (Affymetrix). Quality controls were performed using the Affymetrix expression console.

Public transcriptomic and functional genomic data recovery

Public data used in this study were downloaded from Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo/)⁷⁸, ENCODE⁷⁹, Roadmap epigenome (http://www.roadmapepigenomics.org/)⁸⁰, or CistromeDB (http://cistrome.org/db/#/)⁴¹ and are listed in Supplementary Data File 11. RLE normalized expression data (CAGE-seq) from 561 primary cells listed in Supplementary Data File 3 were downloaded from the FANTOM5 website (https://fantom.gsc.riken.jp/5/sstar)⁸¹.

Transcriptomic data analyses

RNA-seq raw data were processed using a local instance of Galaxy⁸². Data were analyzed using FastQC (http://www.bioinformatics.babraham.ac.uk/projects/fastqc) and reads were trimmed when necessary. Mapping of reads to the human genome (hg38) was achieved using Tophat 2.0.9⁸³ using the Gencode annotation (GRCh38.84). Read counting was then performed using the Htseq-count tool v1.0.0⁸⁴. Normalization taking into account gene lengths and differential expression analyses were performed using EdgeR v0.0.3⁸⁵.

Affymetrix raw data were processed through the GIANT v0.0.2 tools suite⁸⁶ on a local instance of Galaxy⁸². Normalization was performed using the apt-probeset-summarize tool of Affymetrix Power Tools (https://www.affymetrix.com/support/developer/powertools/changelog/index.html) using “gc correction+scale intensity+rma”. Normalization level was set to “probeset” and probesets assigned to the same gene (NetAffx Annotation Release 36) were averaged for human data while normalization level was set to “Core genes” for mouse data in order to use the “Collapse to gene symbols” function of the GSEA tool. For differential expression analysis, the LIMMA tool from GIANT v0.0.2 was used⁸⁷. When necessary, human gene symbols were attributed to each murine gene using the Orthologue Conversion software (https://biodbnet-abcc.ncifcrf.gov/db/dbOrtho.php).

Pathway and gene-set enrichment analyses

Enrichments for biological or molecular pathways were defined using Metascape (http://metascape.org)⁸⁸ or the ToppGene Suite⁸⁹.

Gene-set enrichment analyses (GSEA) were performed using the GSEA software (v3.0) developed at the Broad Institute⁵⁰ essentially as described previously²⁵. We used 1000 gene-set permutations and the following settings: “weighted” as the enrichment statistic and “difference of classes” or “signal to noise” as the metric for ranking genes. Ranking was performed by the GSEA software using the gene average expression when multiple probesets were present in the microarray. The following gene sets from MSigDB (https://www.gsea-msigdb.org/gsea/msigdb) were used: Biological process v7.2, Reactome v.7.2, and KEGG v7.2. The matrisome gene set comprising 1028 genes encoding for ECM and ECM-associated proteins (or selective subsets including collagen or ECM-regulator encoding genes) was retrieved from http://matrisomeproject.mit.edu/¹². The human fibrotic liver ECM gene set comprising genes encoding ECM proteins that have been detected by proteomics analyses in fibrotic livers (71 genes) were retrieved from http://matrisomedb.pepchem.org/⁵¹. In addition to enrichment plots, figures also provide NES and FDR, which are the normalized enrichment score and the false discovery rate provided by GSEA, respectively. The GSEA core enrichment comprising genes that account for the gene set’s enrichment signal was retrieved in some of the performed analyses. Random gene lists used in GSEA were obtained using the Random Gene Set Generator tool (https://molbiotools.com/randomgenesetgenerator.php). Similar terms were grouped using an in-house procedure in R⁹⁰ or using the GeneSetCluster package⁹¹.

Gene-module association determination (G-MAD)

Prediction of BNC2 functions was obtained using G-MAD analyses (https://www.systems-genetics.org/gmad)²⁸, which were performed using default parameters and “Homo sapiens” datasets. Additional runs were performed to monitor the G-MAD score obtained for “GO:0044420 extracellular matrix component” in individual organs.

Single-cell and single-nuclei RNA-seq analyses

Single-cell RNA-seq data of HSCs from control and CCl₄-treated mice have been described in³¹. Briefly, dissociated livers from CCl₄ or vehicle-treated mice were enriched by flow cytometry for LSECs, KCs/MDMs (Kupffer cells/monocyte-derived macrophages), and HSCs prior to 10x Genomics Single-Cell 3’ v2 library preparation and sequencing. Sequencing data were aligned and quantified using the Cell Ranger Single Cell Software Suite (ver. 2.1), and cell types were separated using the Scanpy implementation of the Louvain algorithm. UMAPs were generated using Seurat v. 4.0.3.

Single-nuclei RNA-seq data of liver cells obtained from mice fed a HFSC NASH diet were retrieved from https://www.livercellatlas.org/download.php³⁴. Only annotated cells were used, and cell populations with less than 50 individual cells were removed. Counts were then normalized using SCTransform from the Seurat package v. 4.0.3⁹². UMAP were generated using Seurat.

Protein extraction

Total cell extracts were obtained by washing cells with ice-cold Phosphate-Buffered Saline (PBS) and scraping in Pierce IP lysis buffer (Thermo Fisher Scientific) containing a protease inhibitor cocktail (PIC, Roche). Cell lysates were then sonicated for 5 min (five cycles of 30 s ON/30 s OFF) using Bioruptor (Diagenode), and insoluble material was removed by 10 min centrifugation at 13,000×g.

The preparation of protein from the chromatin fraction was performed according to the protocol described in ref. 25. The cytosolic fraction issued from the same samples was also saved and analyzed.

Protein immunoblotting

Protein concentration was determined using the Pierce^TM BCA protein assay kit (Thermo scientific), and samples were used for immunoblotting. Regular western blotting or Simple Western immunoassays using the Wes system (Protein Simple), performed as in ref. 25, are described hereafter. Antibodies used and their dilutions are listed in Supplementary Data File 10. Uncropped and unprocessed images are provided in the Source Data file.

Western blotting

Ïn total, 40 µg of proteins were separated by 10% SDS-PAGE and immunodetected using the primary antibodies listed in Supplementary Data File 10. Detection was achieved using HRP-conjugated secondary antibodies (Sigma-Aldrich). Images were acquired using the iBrightTM CL1500 Imaging System (Thermo Fisher Scientific). Quantifications were performed using Image Studio Litev5.2 (LI-COR Biosciences, Lincoln, USA).

Simple western immunoassays

Protein extracts (0.4 µg/mL) were run on a Wes system (Protein Simple) according to the manufacturer’s instructions. Separation was performed using the 12–230 kDa capillary cartridges. The chemiluminescence-based electrophoretogram was autogenerated, and digitally rendered bands were produced from the chemiluminescent peaks using the Compass software (Protein Simple). Quantifications were obtained using the area under the peak of the protein of interest. Since antibodies against BNC2 had been poorly characterized in the literature, we performed experiments to demonstrate the specificity of #HPA059419 (Sigma) for western blotting (Supplementary Fig. 7).

Co-immunoprecipitation assays

Co-immunoprecipitation assays were conducted as in²⁵. LX2 cells were transfected for 48 h with a plasmid coding for human BNC2 (custom construction, E-Zyvec) and a plasmid coding for constitutively active nuclear YAP1 (YAP1-S127A, Addgene). Alternatively, endogenous proteins were cross-linked by incubating LX2 cells with 1% formaldehyde for 10 min at room temperature. For nuclear extract preparation, cells were rinsed with ice-cold PBS and lysed with hypotonic buffer (20 mM Tris-HCl, pH 8.0, 10 mM NaCl, 3 mM MgCl2, 0.2% NP40) for 5 min at 4 °C. After 5 min centrifugation at 600×g, the pellet was resuspended in lysis buffer (25 mM Tris-HCl pH 8, 1 mM EDTA, 1.5 mM MgCl2) and incubated for 30 min at 4 °C. Following 10 min (30 s ON/30 s OFF) sonication with the Bioruptor, insoluble material was removed through 10 min centrifugation at 13000 g. Immunoprecipitation was performed using 2 µg of BNC2 antibody (55220-1-AP from Proteintech or HPA018525 from Sigma-Aldrich) or a control IgG and 500 µg of nuclear extract, which were incubated overnight at 4 °C in dilution buffer (25 mM Tris-HCl pH 8, 1 mM EDTA, 1.5 mM MgCl2). 10 µl of a 1:1 mixture of protein A and protein G magnetic beads (Dynabeads, Thermo Fisher Scientific) blocked overnight in PBS containing 5 mg/mL Bovine serum albumin (BSA) was then added to each sample. After incubating samples 4 h at 4 °C under agitation, beads were washed four times with cold washing buffer (25 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM EDTA, 0.2% NP40). Elution was performed in Laemmli buffer 6× (175 mM Tris-HCl pH 6.8, 15% glycerol, 5% SDS, 300 mM DTT and 0.01% Bromophenol Blue). All buffers were supplemented with PIC (Roche). Input proteins and IPed materials were analyzed by western immunoblotting.

Chromatin immunoprecipation (ChIP) and ChIP-seq

Chromatin immunoprecipitation was performed as described previously²⁵. LX2 cells were fixed for 10 min at room temperature with 1% formaldehyde (Thermo Fisher Scientific) followed by 5 min incubation with 125 mM glycine. After two washes with ice-cold PBS, cell pellets were incubated for 10 min in 0,25% Triton X-100, 10 mM EDTA, 10 mM HEPES and 0.5 mM EGTA followed by 10 min incubation in 0.2 M NaCl, 1 mM EDTA, 10 mM HEPES, 0.5 mM EGTA. Nuclei were then resuspended in Lysis buffer (50 mM Tris-HCl pH 8.0, 10 mM EDTA, 1% SDS) and sonicated for 30 min (three cycles 30 s ON/30 s OFF) using a Bioruptor (Diagenode). All buffers were supplemented with PIC (Roche). Chromatin was diluted tenfold in Dilution Buffer (20 mM Tris-HCl pH 8.0, 1% Triton X-100, 2 mM EDTA, 150 mM NaCl) and incubated overnight at 4 °C with 2 μg of H3K27ac antibody (Active Motif, #39685) or 3 μg of BNC2 antibody (Sigma, HPA018525 or Sigma, HPA059419). The next day, protein A/G sepharose beads (GE Healthcare) were added, and samples were incubated for 4 h at 4 °C under agitation in the presence of 70 μg/mL yeast tRNA (Sigma-Aldrich). Beads were washed three times with RIPA buffer (50 mM HEPES, 1 mM EDTA, 0.7% Na Deoxycholate, 1% NP40 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 cross-linking. DNA purification was performed using the MinElute PCR purification kit (Qiagen). ChIP and input samples were subjected to high-throughput sequencing and were further analyzed as described hereafter. Selected binding sites were confirmed using qPCR and primers listed in Supplementary Data File 10.

CoP-seq

The CoP (Column Purified chromatin)-seq procedure was performed essentially as described in ref. 39. In brief, LX2 cells were cross-linked, lyzed and sonicated as described for ChIP except 2% formaldehyde were used. Soluble chromatin was loaded onto MinElute PCR purification kit (Qiagen) and eluates were treated with RNAse A, proteinase K and reverse cross-linked before final DNA purification with the MinElute PCR purification kit. Inputs were processed similarly except for the column-based purification of cross-linked material which was omitted.

ChIP-seq and CoP-seq data analysis

ChIP-seq and CoP-seq raw data were processed using a local instance of Galaxy⁸² essentially as described previously⁹³. Data processing involved FastQC analysis (http://www.bioinformatics.babraham.ac.uk/projects/fastqc) and read mapping to hg38 using Bowtie2 version 1.0.0⁹⁴. Genome-wide signal tracks and enriched regions (peak calling) were obtained using model-based analysis of ChIP-seq version 2 (MACS2 v2.1.1.20160309)⁹⁵. Input DNA was used as control, duplicate tags and those mapping to ENCODE blacklisted regions v2⁹⁶ were removed and parameters used for peak calling were as described in ref. 93. Signal tracks corresponding to enrichment over input were obtained using MACS2 bdgcmp. For H3K4me3 ChIP-seq processing, peak calling was performed using the broad option for histone marks (q < 0.001). Only H3K4me3 domains which could be assigned to a gene were considered in subsequent analyses, i.e., H3K4me3 domains with predicted gene TSS (GENCODE v24 database) overlapping or within 1 kb. ChIP-seq signals and called peaks were visualized at specific loci using the Integrated Genome Browser (IGB v9.0.1)⁹⁷ or were analyzed at large scales using the deepTools v3.3.2⁹⁸. The promoters of housekeeping genes defined in Eisenberg and Levanon, 2013 were used as a control in some of these analyses. The coordinates of their TSS were retrieved using the UCSC Genome Browser⁹⁹ and RefSeq as the gene annotation.

The BNC2 cistrome was compared with that of publicly available cistromes contained within the CistromeDB database using the CistromeDB toolkit⁴¹. « All peaks in each sample » was used for these analyses. De novo motif enrichment analyses were performed using the RSAT peak-motifs tool run with default parameters (http://rsat.sb-roscoff.fr/peak-motifs_form.cgi)¹⁰⁰. BNC2 target genes were predicted by assigning peaks to genes using GREAT4.0.4⁴⁵ or FOCS⁴⁹. With regards to the latest, BNC2 peaks within 2.5 kb of a Gencode v34 TSS¹⁰¹ were defined as promoter binding sites. The remaining peaks were monitored for predicted promoter interaction within the entire FOCS datasets (http://acgt.cs.tau.ac.il/focs/download.html; enhancers without assigned genes were removed). All retrieved genes were merged defining FOCS predicted BNC2 target genes.

Super-enhancers were defined as in our previous study¹⁰². First, model-based analysis of ChIP-seq version 2 (MACS2)⁹⁵ was used to call peaks in the H3K27ac ChIP-seq data (FDR < 0.15) using input as control. Peaks matching to ENCODE blacklisted regions⁹⁶ were discarded. Super-enhancers were then called using retained MACS2-derived peaks and Rank Ordering of Super-Enhancers (stitching set at 12.5 kb)^24,103 leaving out regulatory regions mapping to promoters (TSS+/−2.5 kb) issued from GREAT⁴⁵ and using Input DNA as control.

Bioinformatical identification of MF identity TFs

The cumulative frequency distribution of H3K4me3 domain lengths was used to define the inflexion point of the curve (3017 bp), which served to separate sharp from broad H3K4me3 domains (Fig. 1A)¹⁰⁴. These broad H3K4me3 domains were assigned to genes using GREAT 2.0⁴⁵ (retrieving 1278 genes) in order to only use extremely high-confidence gene predictions as described in²⁶. TSSs were used to monitor the distance to the center of the nearest super-enhancer, which were defined using primary MF-HSC H3K27ac ChIP-seq data¹⁸ following a procedure detailed hereabove. Note that we favored the use of broad H3K4me3 over that of super-enhancers to define identity genes to overcome issues highlighted in our previous study including uncertainty about target gene assignment to distal super-enhancers¹⁰².

Next, the length of H3K4me3 domains associated with the 1278 genes in 76 normal human cell types and tissues defined by²⁶ was retrieved (Supplementary Data File 2). For each gene, the relative H3K4me3 domain length in a given cell type or tissue was obtained by dividing by the mean length of domains associated to this gene in all samples. A heatmap of relative H3K4me3 lengths was generated using the heatmap.2 function of the R package “gplots” (https://cran.r-project.org/web/packages/gplots/) and hierarchical clustering was performed using the hclust function of the R package “Stats” using ward.D2 agglomeration method¹⁰⁵. This allowed to define three main clusters of genes listed in Supplementary Data File 1. TFs within these clusters were subsequently identified by comparison with the human TFs list provided by the Animal TFDB 2.0 database¹⁰⁶.

Further analysis of TFs from cluster 1 was performed by mining the scientific literature to identify articles linking these TFs to MF or fibrosis using the easyPubMed package (https://www.data-pulse.com/dev_site/easypubmed/) in R. Articles with co-occurrence of a given TF gene name and “fibroblast”, “myofibroblast”, “fibrosis” or “extracellular matrix” in their title or abstract were retrieved. Results were visualized on a network generated using Cytoscape¹⁰⁷. First, the TF network was generated using String 11.0 (https://string-db.org/; “full network”, “Medium confidence”)¹⁰⁸. The network was then exported into Cytoscape where TFs without interactions were manually added back and where nodes were colored according to the number of previously retrieved PubMed articles.

Mouse liver histology and biochemical analyses

Histology

Liver slices were fixed with 4% paraformaldehyde for 48 h and embedded in paraffin using a STP 120 Spin Tissue Processor (Microm Microtech). Paraffin-embedded samples were cut at a thickness of 5-μm and sections were transferred on gelatin-coated slides. Fibrosis assessment was carried out by staining liver sections with a 0.1% solution of Sirius red in 1.3% saturated aqueous picric acid solution. To uncertain reliability of Sirius red staining quantifications, we followed guidelines provided in ref. 109. To avoid potential interlobular differences, we systematically used the same lobe (i.e., the right median lobe). Two large sections, including tissue margin and center but excluding the peripheral tissue and vessels, were used per mouse liver. The entire lobe sections were scanned using an Axioscan (Zeiss), and ten fields were randomly chosen (7.5-week experiment) or the entire section (12-week experiment) was used for quantification with Image J version 1.53c software (NIH, https://imagej.nih.gov/ij/). The analysis was systematically done on two different liver sections and all histological analyses were performed blinded.

Metabolic parameters

Before sacrifice, blood samples were collected from the retro-orbital sinus of mice. Plasma alanine aminotransferase (ALT) and aspartate aminotransferase (AST) activities were measured using colorimetric assays (Thermo Fisher) on a Konelab 20 (Thermo Fisher).

Measurement of liver triglycerides

A weighted piece of liver was homogenized with T10 Ultra-Turrax (Ika) in PBS. Samples were transferred into glass tubes and mixed with a 2:1 chlorofrom:methanol mixture. After centrifugation, upper- and inter-phase were discarded. The lower organic phase was evaporated under nitrogen flow and reconstituted in 1% Triton X-100. Triglyceride content was measured with the LiquiColor Triglycerides Test (Interchim). Protein concentration was measured in parallel using the Pierce^TM BCA protein assay kit (Thermo scientific).

Rapid immunoprecipitation mass spectrometry of endogenous protein (RIME)

LX2 cells were processed using the protocol described in ref. 38. Briefly, 5 × 10⁶ cells were grown for 2 days before being fixed in 1% formaldehyde (Sigma-Aldrich). Nuclear proteins were extracted in 10 mM Tris-HCl pH 8.0, 100 mM NaCl, 1 mM EDTA, 0,5 mM EGTA, 0,1% Na deoxycholate, 0.5% N-lauroylsarcosine and PIC (Roche). Immunoprecipitation was achieved using Dynabeads (Invitrogen) conjugated with 10 µg of BNC2 antibody (#55220-1-AP, Proteintech or #HPA018525, Sigma-Aldrich) or 10 μg of a non-immune control IgG (#2729, Cell Signaling). Beads were rinsed with RIPA buffer (50 mM HEPES, 1 mM EDTA, 0.7% Na Deoxycholate, 1% NP40 and 500 mM LiCl) and AMBIC solution (Sigma-Aldrich) before trypsin digestion.

Samples were analyzed by coupling a nanoflow liquid chromatography system (nanoElute, Bruker Daltonics) online to a trapped ion mobility spectrometry-quadrupole time of flight mass spectrometer (timsTOF Pro, Bruker Daltonics) equipped with a CaptiveSpray source operating in positive mode. Peptides were loaded on a trapping column (Acclaim PepMap 100, C18, 100 Å, 100 µm × 20 mm, Thermo Scientific) and separated on a reversed-phase C18 column (Aurora, 25 cm × 75 μm i.d., 1.6 μm, IonOpticks). Chromatographic separation was carried out using a gradient of 2-25% of solvent B (0.1% formic acid in acetonitrile) over 90 min, then 37% over 100 min and 95% over 110 min at a constant flow rate of 300 nl/min. The column temperature was controlled a 50 °C. Mass Spectrometry (MS) data were collected over a m/z range of 100–1700, and MS/MS range of 100–1700. LC-MS/MS data were acquired using the PASEF method with a total cycle time of 1.88 s, including 1 TIMS MS scan and 10 PASEF MS/MS scans. Ion mobility coefficient (1/K0) value was set from 0.7 to 1.25 Vs cm⁻². Singly charged precursors were excluded by their position in the m/z-ion mobility plane and precursors that reached a “target value” of 17,000 a.u. were dynamically excluded for 0.2 min. The quadrupole isolation width was set between 2 and 3 m/z depending on precursor m/z. The collision energies varied between 20 and 52 eV depending on precursor mass and charge. TIMS, MS operation and PASEF were controlled and synchronized using the control instrument software OtofControl 6.0 (Bruker Daltonik).

Raw MS files were converted into mascot generic files (mgf) and subjected to a Mascot (2.6.2) (Mascot, Matrix Science) search using a target-decoy strategy in order to evaluate the false discovery rate (FDR) of the search. The human database was created with human proteins and present in SwisProt database (created 2019-09-12, containing 20490 target sequences plus the same number of reversed decoy sequences). Search parameters were as follows: trypsin was set as the cleavage enzyme and a maximum of one miscleavage was allowed. Cysteine carbamidomethylation was set as a fixed modification, whereas methionine oxidation, was set as a variable modification. The peptide mass tolerance (tolerance of mass measurement for precursor ion) was set to 15 ppm and the MS/MS mass tolerance (tolerance of mass measurement for fragment ion) set to 0.05 Da. Proline pipeline (http://www.profiproteomics.fr/proline/) was used to validate the identification results. This statistical validation was performed using the Target Decoy approach, which consists in creating a “decoy” sequence that does not exist in nature. The false identifications thus distribute evenly between the real database (target) and the decoy one, and so the number of decoy hits can be used to estimate the false discovery rate (FDR = decoy hits/decoy hits + target hits) and eliminate false positives. Peptide and protein identification validation parameters were set as follows: minimal length of seven amino acid, score ≥20, pretty rank ≤1, and FDR ≤ 1.

Only proteins detected in the BNC2 RIME, at least two out of three biological replicates, but in none of the non-immune control IgG RIME experiments were considered. Common non-specifically detected proteins in affinity purification mass spectrometry data from the Contaminant Repository for Affinity Purification (human CRAPome2.0) database¹¹⁰ were also discarded. BNC2 interactors were further filtered using the list of human transcriptional regulators from the AnimalTFDB3.0 database¹¹¹. BNC2 interactors were clustered based on percent protein coverage in RIME assays using the heatmap.2 function of the R package “gplots” (https://cran.r-project.org/web/packages/gplots/) as described hereabove for H3K4me3 domain lengths. Biological triplicates were obtained using the anti-BNC2 antibody 55220-1-AP (Proteintech) and an additional experiment was performed using the anti-BNC2 antibody HPA018525 (Sigma-Aldrich). All non-filtered detected proteins for individual samples are provided in Supplementary Data Files 6 and 8.

RNA in situ hybridization

RNAscope assays on paraffin-embedded liver sections were performed using the RNAscope® Multiplex Fluorescent Reagent Kit v2 according to the manufacturer’s instructions (Bio-Techne). Briefly, the paraffin-embedded tissue sections were deparaffinized and pre-treated with hydrogen peroxide. Sections were incubated in RNAscope® 1× Target Retrieval Reagent for 15 min at 99 °C and treated with RNAscope® Protease Plus for 30 min at 40 °C in the HybEZ oven sequentially. Samples were then incubated with RNAscope® Probe-Hs-BNC2-C1 (#496801) and Probe-Hs_COL1A1-C2 (#401891-C2) for 2 h at 40 °C. After three steps of amplification, RNAscope® HRP-C1 was added on the sections that were incubated with Opal 650 (Akoya Biosciences) for 30 min. HRP-C2 was then added followed by incubation with Opal 570. When specified, RNA in situ hybridization was followed by immunofluorescence staining using an anti-ACTA2 antibody (ab76126, Abcam:1:50). Alcohol-related liver cirrhosis samples were analyzed using a CSU-W1 Spinning Disk (Gataca) with a 60x CFI PLAN APO LBDA objective (resolution 0.110 µM) and images were analyzed using ImageJ1.53c (https://imagej.nih.gov/ij/). For liver biopsies from NASH patients before and after bariatric surgery, the entire liver sections were scanned using the Axioscan (Zeiss) and images were analyzed using the ZEN 2 software (Zeiss). Quantification of red and white dots; illustrating the number of COL1A1 and BNC2 transcripts, respectively, was performed using ImageJ1.53c (https://imagej.nih.gov/ij/). The number of cells in each section was obtained by counting DAPI-positive nuclei.

Structural model of the BNC2-CRBN interaction and docking analyses

A structural model of the interaction between BNC2 zinc-finger 1 (BNC2ZF1) and CRBN was built using the crystallographic structure of DDB1-CRBN-pomalidomide bound to IKZF1 (Protein Data Bank ID 6h0f)⁵² and used for docking of thalidomide and its derivatives. First, we chose an experimentally resolved zinc finger-CRBN complex on the basis of its resolution and co-crystallized ligand, settling for DDB1-CRBN-pomalidomide bound to IKZF1 (Protein Data Bank ID 6h0f)⁵². Out of the six BNC2 zinc fingers, four were below a 25% identity threshold with the co-crystallized IKZF1 zinc finger. The two remaining BNC2 zinc fingers were the first (29% identity, 38.7% similarity) and fifth (25.8% identity and 38.7% similarity). We opted to keep the first (BNC2ZF1), as it was closer to the reference. BNC2ZF1 was modeled on the RMN structure of human zinc fingers and homeoboxes 1 (Protein Data Bank ID 2ghf; 20.83% identity and 50% similarity with BNC2ZF1). Interestingly, even with a higher identity percentage than the template used to build BNC2ZNF1, the model resulting from a homology with IKZF1 was marginally less interesting, with a slightly higher overall deviation from the Ramachandran optimal parameters. The BNCZF1-CRBN complex was next built by superimposing the BNC2 and IKZF1 zinc fingers and exchanging the co-crystallized IKZF1 for BNCZF1 and adjusting the side chains by a quick geometry optimization in two steps after adding hydrogens and partial charges. First, the side chains were adjusted by 1000 steps of steepest descent, then the whole zinc finger and the side chains of the DDB1-CRBN-pomalidomide complex were subjected to another 1000 steps of steepest descent. This protocol avoided a deformation of the skeletons of the proteins while minimizing the clashes between side chains. The resulting complex was used as the target for a docking study with GOLD (Genetic Optimisation for Ligand Docking)¹¹², with the binding site defined as 10 Å sphere around the co-crystallized pomalidomide. A set of ligands was docked: pomalidomide as a control, the shorter thalidomide, iberdomide which was similar in size to CC-885, and CC-885. In all, 30 poses were generated for each compound. The docking was assessed on the basis of the number of clusters of closely related poses, the score of the clusters, and their spread around their average pose. Apart from CC-885, all the compounds were built in the same configuration as the co-crystallized pomalidomide (S enantiomer).

Statistical analyses

Statistical analyses were performed using the Prism software (GraphPad) and R¹⁰⁵. The specific tests and corrections for multiple testing that were used are indicated in the figure legends. Unless specified in the figure legends, statistical significance was displayed as follow *P < 0.05, **P < 0.01, and ***P < 0.001.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.