DeSUMOylation of chromatin-bound proteins limits the rapid transcriptional reprogramming induced by daunorubicin in acute myeloid leukemias

Abstract Genotoxicants have been used for decades as front-line therapies against cancer on the basis of their DNA-damaging actions. However, some of their non-DNA-damaging effects are also instrumental for killing dividing cells. We report here that the anthracycline Daunorubicin (DNR), one of the main drugs used to treat Acute Myeloid Leukemia (AML), induces rapid (3 h) and broad transcriptional changes in AML cells. The regulated genes are particularly enriched in genes controlling cell proliferation and death, as well as inflammation and immunity. These transcriptional changes are preceded by DNR-dependent deSUMOylation of chromatin proteins, in particular at active promoters and enhancers. Surprisingly, inhibition of SUMOylation with ML-792 (SUMO E1 inhibitor), dampens DNR-induced transcriptional reprogramming. Quantitative proteomics shows that the proteins deSUMOylated in response to DNR are mostly transcription factors, transcriptional co-regulators and chromatin organizers. Among them, the CCCTC-binding factor CTCF is highly enriched at SUMO-binding sites found in cis-regulatory regions. This is notably the case at the promoter of the DNR-induced NFKB2 gene. DNR leads to a reconfiguration of chromatin loops engaging CTCF- and SUMO-bound NFKB2 promoter with a distal cis-regulatory region and inhibition of SUMOylation with ML-792 prevents these changes.


INTRODUCTION
Acute myeloid leukemias (AML) are se v ere hematological malignancies, which arise through the acquisition of onco genic m uta tions by hema topoietic stem or progenitor cells from the myeloid lineage.Although AML constitutes a highl y hetero genous group of diseases, most of them are treated similarly with the combination of one anthracycline, such as Daunorubicin (DNR) or Idarubicin (IDA), and the nucleoside analogue Cytarabine (Ara-C) (1)(2)(3).Most patients respond to this treatment.Howe v er, a large proportion of them relapse and become refractory to the drugs, which contributes to the dismal prognosis of this disease ( 2 , 3 ).It is ther efor e critical to better understand the mode(s) of action of these drugs to find ways to overcome chemoresistance.
The DNA-damaging properties of both Ara-C and DNR are essential for therapeutical efficacy and have been characterized e xtensi v ely ( 4 , 5 ).Howe v er, these drugs also display many other cellular effects that can both favor or counteract their ability to induce cell death.For example, anthracyclines can induce fast production of reacti v e oxygen species (ROS) that contribute to apoptosis induction by activating various signaling pathways ( 6 ).On the other hand, Ara-C and DNR also activa te, a t the same time, many pro-survival pa thways tha t mitiga te their pro-apoptotic actions.This is notable for the PI3K / AKT ( 7 ), MAPK ( 8 ) and NF-B ( 9 , 10 ) pathways, as their inhibitions potentiate genotoxicsinduced cell death in cancer cells.Finally, both anthracyclines and Ara-C have long been known to alter transcriptional programs on the mid / long term (day-range) when used at sublethal doses ( 11 , 12 ).Ho wever, ho w Ara-C and DNR contribute to gene expression changes at early times after the start of a treatment, has been poorly investigated.
We have formerly shown that one early consequence of DNR and Ara-C treatments is ROS-dependent de SUMOylation of proteins in chemosensiti v e AMLs, which participates in induction of apoptosis ( 13 ).SUMOylation consists of re v ersib le, covalent modification of proteins by the ubiquitin-related peptidic post-translational modifiers SUMO-1 to -3.SUMO-1 is 50% identical to SUMO-2 and -3, which are 95% identical and frequently r eferr ed to as SUMO-2 / 3 as their individual functions can often not be distinguished.The three SUMOs are conjugated by a conserved enzymatic cascade comprising one SUMO-activating enzyme (SAE1 / SAE2 dimer; also called SUMO E1), one SUMO-conjugating enzyme (Ubc9; also called SUMO E2) and se v eral SUMO E3s tha t facilita te SUMO transfer fr om the E2 onto its pr otein targets.SUMOylation is highly dynamic thanks to various isopeptidases (also called de SUMOylases) that remove SUMO from its substrates ( 14 ).Thousands of SUMOylated proteins involved in many cellular processes have now been identified ( 15 ).Howe v er, one of the main biological processes associated with SUMOylation is the control of gene expression.Numerous transcription factors and coregulators, as well as histones and the basal transcription machinery are SUMOylated ( 16 ).Moreover, genome-wide studies have revealed tha t SUMOyla ted proteins are highly enriched at gene r egulatory r egions, including promoters and enhancers (17)(18)(19)(20)(21). Their SUMOylation is likely to occur on chromatin as both SUMO conjugating (E1, E2 and E3s) and deconjugating enzymes can bind to the chromatin ( 17 , 22-24 ).Although SUMOylation of chr omatin-bound pr oteins has often been associated with gene silencing or gene expression limitation ( 17 , 24-26 ), it can also participate in the activation of certain genes such as ribosomal genes ( 19 , 20 ), fibroblastic genes in mouse embryonic fibroblasts (MEFs) (25), PPARg / RXR target genes during adipocyte dif ferentia tion ( 21 ) as well as RN A-pol ymer ase III ( 27 ) controlled genes.Over all, the impact of SUMOylation on transcription appears to be dependent on both genes and signaling contexts, as well as on the nature of the conjugated proteins and of the chr omatin envir onment ( 16 ).
To better understand the complex mode of action of these drugs, we explored the early effects of Ara-C and DNR on gene expression in AML cells, together with the contribution of SUMOylation to transcriptome reprogramming.We report that DNR induces rapid and broad gene expression changes that are preceded by de SUMOylation of chr omatin-bound pr oteins, in particular at acti v e promoters and enhancers, whereas the effect of Ara-C is much more limited.Intriguingly, we found that inhibition of SUMOylation limits DNR-induced changes in gene expression.Among the proteins most ra pidl y de SUMOylated in response to DNR, we identified the CTCF insulator protein, which was found highly enriched in regions of the genome marked by SUMO.This notably concerns the NFKB2 gene, whose DNR-induced expression is preceded by rearrangement of chromatin loops involving its SUMO / CTCFmarked promoter and cis -regulatory elements.

Cell culture and genotoxic treatment
HL-60 cells were obtained from the ATCC, authenticated by LGC and regularly tested for the absence of mycoplasma.They wer e cultur ed at 37 • C in the pr esence of 5% CO 2 in RPMI (Eurobio) medium supplemented with 10% decomplemented (30 min at 56 • C) fetal bovine serum (FBS) and penicillin and streptomycin.After thawing, cells were split at 0.3 × 10 6 / ml e v ery 2 to 3 da ys f or no more than 10 passages.HEK293T cells wer e cultur ed at 37 • C in the presence of 5% CO 2 in DMEM (Eurobio) medium supplemented with 10% decomplemented FBS and penicillin and str eptomycin.HL-60 cells wer e seeded at 0.3 × 10 6 / ml the da y bef or e tr ea tment with drugs a t 1 M for DNR and 2 M for Ara-C.Cells were treated for 2 h for ChIP-Seq, 4C and CUT&RUN experiments and 3 h for Affimetrix transcriptomic and RNA-Seq.For SILAC experiments, HL-60 cells were grown in SILAC medium supplemented with dialyzed serum and K0 / R0 (light condition), K4 / R6 (medium condition), K8 / R10 (heavy condition) amino acid isotopes for 21 days until incorporation of amino acids isotopes r eached 99%, as measur ed by mass spectrometry.SILAC labelled cells were then treated or not with 1 M DNR for 2 h.Hybridomas were grown in CellLine bioreactors (Integra) according to the manufacturer's protocol using RPMI in the cell compartment and RPMI + 10% FCS in the medium compartment.Antibodies were harvested from the cell compartment after 7 days of culture.

AML patients' cells and healthy donors PBMCs
Bone marrow aspirates or blood were collected after obtaining written informed consent from patients under the frame of the Declaration of Helsinki and after approval by the Institutional Re vie w Boar d (Ethical Committee 'Sud M éditerran ée 1,' ref 2013-A00260-45, HemoDiag collection).Healthy donor leukocytes were collected from blood donors of the Montpellier Etablissement Fran c ¸ais du Sang.Fr esh leukocytes wer e purified as pr eviously described ( 29 ) using density-based centrifugation using Histopaque 1077 from Sigma and directly lysed for RNA preparation or frozen and stored in liquid nitrogen.

Micr oarra y-based whole transcript expression analysis and profiling
Total RNAs were extracted using the GenEluteTM Mammalian Total RNA kit (Sigma) and treated with DNAse I according to the manufacturer's specifications.For each condition, three independent batches of RNA were prepared and controlled for purity and integrity using the Agilent 2100 Bioanalyzer with RNA 6000 Nano LabChip kits (Agilent Technolo gies).Onl y RN A with no sign of contamina tion or degrada tion (RIN > 9) were processed to generate amplified and biotinylated sense-strand cDNA targets using the GeneChip ® WT PLUS Reagent kit from Affymetrix according to the manufacturer's specifications.After fragmentation, cDNA targets were used to probe Affymetrix GeneChip ® Human Gene 2.0 ST arrays, which were then washed, stained and scanned according to Affymetrix instructions (manual P / N 702731 Rev.3).

Micr oarra y data analysis
CEL files generated after array scanning were imported into the Partek ® Genomics Suite 6.6 (Partek Inc.) for estimating transcript cluster e xpression le v els from raw probe signal intensities using default Partek settings.Resulting expr ession data wer e then imported into R ( http://www.Rproject.org/) for further analysis.First, non-specific filtering was applied to remove transcript clusters with no specified chromosome location.Then, boxplots, density plots, relati v e log expressions (RLE) and sample pairwise correlations were generated to assess the quality of the data.They re v ealed no outlier within the series of hybridizations.Principal component analysis (PCA) was also applied to the dataset.The first two components of the PCA could separate samples according to the treatment.Thus, the treatment was considered as the unique source of variability.Finally, the LIMMA package ( 31 ) was used to detect differentially expressed genes (DEG) between treated and nontreated samples.A linear model with treatment as unique factor was fitted to the data before a ppl ying eBayes function to calculate the significance of the difference in gene expression between the two groups.P -values were adjusted by Benjamin and Hochberg's False Discovery Rate (FDR) and genes with FDR < 0.05 and absolute linear Fold Change (FC) greater or equals to 2 wer e consider ed as DEG.Microarray data are available at ArrayExpress under the accession number E-MATB-4895.

RT-qPCR assays
Total mRNAs were purified using the GenElute Mammalian Total RNA kit (Sigma-Aldrich).After 1 h of DNase I (4U, NEB) treatment in the presence of RNasin (2.5U; Promega), 1 g of total RNA was used for cDNA synthesis using the Maxima First Strand cDNA kit (Thermo Fisher Scientific).qPCR assays were conducted using Taq platinum (Invitrogen) and the LightCycler 480 device (Roche) with specific DNA primers (Table 1 ).Data were normalized to the mRNA le v els of the housekeeping genes TBP and S26 or GAPDH.

RNA-seq libraries preparation and sequencing
RNA-Seq were performed as described previously ( 32 ).Total RNAs were purified using the GenElute Mammalian Total RNA kit (Sigma-Aldrich), treated with DNase I (4U; New England Biolabs) in the presence of RNasin (2.5U; Promega) and re-purified.RNA quality was assessed using a BioAnalyzer Nano 6000 chip (Agilent).Three independent experiments were performed.Libraries were pr epar ed using TruSeq ®Stranded mRNA Sample Preparation kit (Illumina).Libraries were sequenced using an Illumina Hiseq 2500 sequencer as single-end 50-base reads.Image analysis and base calling were performed using HiSeq Control Software (HCS), Real-Time Analysis (RTA) and bcl2fastq.

4C-seq experiments
Chromatin for 4C-Seq experiments was pr epar ed essentially as previously described ( 33 , 34 ).A total of 7 × 10 6 cells in 10 ml of medium were cross-linked with formaldehyde 2% for 10 min at room temperature (RT).Formaldehyde was then neutralized with 125 mM glycine for 10 min at 4 • C.After a wash with cold PBS, cells wer e r esuspended in 5 ml of lysis buffer (Tris-HCl 10 mM pH 8, NaCl 10 mM, NP-40 0.2%, aprotinin + pepstatin + leupeptin 1 g / ml each, AEBSF 1 mM) and incubated on ice for 20 min.Cells were pelleted 5 min at 380 g at 4 • C, resuspended in 1 ml of lysis buffer and snap frozen in liquid nitrogen.Lysates were thawed at 37 • C and centrifuged at 18000 g at RT for 5 min.Cell pellets were resuspended in 700 l of first enzyme manufacturer buffer 1X (NlaIII -cutsmart [NEB -R0125L]) and homogenized on ice (50 strokes in total) with a 1 ml Dounce homogenizer.Cells were permeabilized using SDS 0.3%, at 37 • C for 1 h under orbital shaking (1 krpm) on an Eppendorf thermomixer).SDS was displaced by adding TritonX100 1.65% and continuing orbital shaking at 37 • C for 1 h.A 100 l sample of the reaction mix was taken as a negati v e control for the first digestion.The digestion with NlaIII enzyme was performed at 37 • C for 24 h under orbital shaking (1 krpm) using 3 sequential additions of 300 U of enzymes at regular intervals.Before enzyme inactivation at 65 • C for 20 min, 100 l of the reaction mix was collected as a restriction enzyme digestion control.The ligation step was performed overnight at 16 • C in 8 ml of a reaction mix adjusted to 1 × of ligase reaction buffer and containing 800 l of the restriction enzyme reaction mix, 240 U of T4 DNA ligase HC (Thermo scientific, EL0013) and ATP 0.04 mM.Proteinase K (300 g) was added to ligated DNA products and the reaction was incubated an at 56 • C for 1 h.Decrosslinking was achie v ed in an incubation step of 6 h at 65 • C. The two control tubes also underwent the proteinase K and decrosslinking steps.Then, all samples were treated with 300 g of RNAse at 37 • C for 30 min.DNA purifications were performed using phenol:chloroform:Isoamyl alcohol 25:24:1 (PCI).DNAs were precipita ted a t −20 • C overnight using 2 volumes of EtOH in the presence of NaCl 250 mM and 20 g of gl yco gen (Thermo).DN As were pelleted by centrifugation (10 krpm) at 4 • C and washed using 70% EtOH.Pellets were dried at room temperature and resuspended in 50 l of water.10 l samples were collected fr om both contr ols and ligated DNA pr oducts and electr ophoresed thr ough an agar ose gel to contr ol the digestion and ligation steps.Ligation products were digested at 37 • C for 2.5 h under orbital shaking (1krpm) suing 100 U of the second restriction enzyme (DpnII from New England Biolabs, r efer ence R0543M).The second restriction enzyme was inactivated and a second ligation was performed under the same condition as above.4C libraries were purified with PCI and precipitated as described above.4C libraries were amplified using specific primers composed of P5 / P7 Illumina sequence supplemented with indexes and sequences corresponding to the NFKB2 promoter (viewpoint) (Table 2 ).The 'Expend Long Template PCR System' kit (Roche) was used using 300 ng of the 4C library following the manufacturer's instruction.The following amplification parameters were used: dena tura tion for 2 min at 94 • C followed by 30 cycles (94 • C -15 s, 58 • C -1 min and 68 • C -3 min) and 7 min a t 68 • C .4C libraries were purified with the 'Gel and PCR clean up' kit from Macherey-Nagel using NTI solution diluted 6 times and an elution buffer pre-heated at 70 • C.After 3 PCR amplification rounds, all 4C libraries for the same sample were pooled, purified and cleaned up using Agencourt AMPure XP beads (ratio 1:1) using EtOH 80% as a washing solution.The libraries were sequenced using the Illumina Hiseq 2500 sequencer as single-end 125 base reads following Illumina's instructions.Image analysis and base calling were performed using the HiSeq Control Software (HCS), Real-Time Analysis (RTA) and bcl2fastq.

Quality control of sequencing data and reads trimming
The quality of the data obtained after sequencing was assessed using the FastQC tool.When the score of the first bases of reads was lower than 30, all reads of the dataset were 5 -trimmed of the relevant number of nucleotides using the trimmomatic tool (Headcrop).All reads with more than 1 N-call were removed from datasets.

ChIP-seq reads mapping, peak calling and analysis
ChIP-seq reads were aligned on the human reference genome (hg19) using CASAVA 1.8.2 (MGX pipeline).Analysis of the aligned reads, scaling and input subtraction were performed using the R package P asha ( 35 ).Data wer e visualized using the IGB software ( 36 ).The peak calling was performed using the WigPeakCaller script, which automatizes the IGB thresholding tool ( 37 ).The SUMO-2 / 3 peak calling was done with the following parameters: by value = 32, Max Gap ≤100 and Min Run > 100.Motif search was performed using HOMER v4. 10 ( 38 ).ChIP-Seq sequencing data are available with accession GSE198986.Pub licly availab le HL-60 ChIP-seq dataset were used for H3K4me3 (GSM945222), H3K4me1 (GSM2836484), H3K27ac (GSM2836486) and RNAPII (GSM1010737).The hg19 promoter ( −2 kb to TSS) gff files have been generated with gff toolbox, using the GRCh37p13 annotation file from NCBI.The H3K4me3 histone marks, which is enriched at gene TSS, have been used as a proxy to annotate HL-60 promoter.All genomic regions presenting H3K4me1, which do not correspond to annotated promoters, wer e consider ed as candidate enhancers.Then, the activity of these regulatory elements was inferred from the presence of H3K27ac.All dataset intersects were performed using Bedtools 2.29.0 (intersect) from Quinlan laboratory ( 39 , 40 ).

RNA-seq mapping, quantification and differential analysis
RNA-seq r eads wer e mapped to Human r efer ence genome (hg19, GRCh37p13) using TopHat2 (2.1.1)( 41 ) based on the Bowtie2 (2.3.5.1) aligner ( 42 ).The reproducibility of replicates was quantified using the cufflinks v2.2.1 tool ( 43 ) with the linear r egr ession of r eads per kilobase million (RPKM) between two replicates.Read association with annotated gene regions was done using the HTseq-count tool v0.11.1 ( 44 ).The variance between replicates and conditions wer e appr eciated thanks to a principal component analysis (PCA) performed on the read count matrix.Differ ential expr ession analysis was performed using DESeq2 ( 45 ) using the normalization by the sequencing depth and the parametric negati v e binomial law to estima te da ta dispersion.All conditions wer e compar ed to the mock condition (DNR versus DMSO, ML-792 versus DMSO and ML-792 + DNR versus DMSO) and the ML-792 + DNR condition was also compared to the DNR-only condition (ML-792 + DNR versus DNR).The genes that presented a fold change ≥ or ≤ 2 and an adjusted P -value (FDR) < 0.05 wer e consider ed as differ entially expr essed genes (DEGs).RNA-seq data are available with accession GSE198982.

4C-seq mapping, trim, capture and profiling
The pipeline for the analysis of the 4C data was modified from the pipe4C pipeline ( 49 ) and is available on github ( https://github.com/Mathias-Boulanger/pipe4C).The steps are the following: Reads filtering (trim-ca pture), ma pping to r efer ence genome, assignment of reads to their restriction fragment and creation of normalized score per fragment.Only reads containing the amplification sequence (CGTGA CGCA CGGAAA CGT C) wer e kept for further analysis.Then, sequences downstream of the restriction enzyme cutting site of each selected r eads wer e mapped to GRCh37p13 human r efer ence genome with Bowtie2 aligner.Restriction fragment map was extrapolated from the r efer ence genome using the cutting sequence of restriction enzymes.The interaction peak calling has been performed with peakC and the differential profiling analysis with DESeq2 ( 45 , 50 ).4C-seq data are available with accession GSE198981.

Gene ontology and GSEA
Functional gene-annotation enrichment analyses were done using GO Panther ( 51 ) with the ID number of DEGs or proteins as input list.The gene network analyses were performed using the Cytoscape-based Cluego plugin ( 52 ).Gene Set Enrichment Analyses were performed using https: //www.gsea-msigdb.org/gsea/index.jsp (v ersion 4.0.3)( 53 ).

Coupling antibodies to protein-G beads
Hybridoma supernatants were incubated with Protein G sepharose beads (SIGMA) at room temperature for 4 h, washed 3 times with PBS (phosphate buffer 10 mM pH 7.4, KCl 2.7 mM and NaCl 137 mM) and once with Na borate 50 mM pH 9.0.Antibodies were then crosslinked for 30 min in dimethyl-pimelimidate (DMP) 20 mM diluted extemporarily in Na borate 50 mM pH 9.0.The coupling procedure was repeated a second time and the beads were washed 3 times with PBS.

Immunoprecipitation of SUMOylated proteins
For SILA C experiments, SILA C-la beled HL-60 cells were grown in spinner flasks (Nunc).5 × 10 8 cells were used for each condition.The immunoprecipitation of endogenously SUMOylated proteins was based on the protocol described in r efer ence ( 54 ).Cells wer e lysed in PBS containing SDS 2%.The final concentration of SDS after lysis was then adjusted to 1% and lysates were sonicated.Dithiotreitol (DTT) was then added at a final concentration of 50 mM.Lysates were then boiled for 10 min and diluted 10-fold in Na phosphate 20 mM pH 7.4, 150 mM NaCl, Triton X100 1%, Na deoxycholate 0.5%, EGTA 5 mM, EDTA 5 mM, NEM, 10 mM, aprotinin + pepstatin + leupeptin 1 g / ml each, filtered through 0.45 m filter and incubated with Protein G-coupled anti-SUMO-1, -SUMO-2 and -BrdU (control) antibodies at 4 • C overnight.Beads were then washed 3 times with RIPA (Na phosphate 20 mM pH 7.4, NaCl 150 mM Triton X100 1%, SDS 0.1%, Na deoxycholate 0.5%, EGTA 5 mM, EDTA 5 mM, NEM 10 mM, aprotinin1 g / ml and pepstatin 1 g / ml) and twice with RIPA containing NaCl 350 mM in Low-Bind tubes (Eppendorf).Elution of SUMOylated proteins was performed twice with peptides bearing either the 21C7 SUMO-1-(VPMNSLRFLFE) or the 8A2 SUMO-2 / 3-(IRFRFDGQPI) epitope diluted in RIPA containing NaCl 350 mM.Eluted proteins wer e pr ecipitated with 10% TCA for 1 h on ice.Pellets were then washed twice with acetone a t -20 • C , dried and resuspended in the Laemli electrophoresis sample buffer.For the identification of SUMOylated targets (SILAC1), samples were immunoprecipited with control-, anti-SUMO-1 or anti-SUMO-2 / 3 antibodies and mixed only after elution with the SUMO epitope-bearing peptides.For the identification of proteins showing DNRmodula ted SUMOyla tion, mock-and DNR-trea ted samples were mixed right after the initial lysis step and used for immunoprecipitation with SUMO-1 (SILAC2) or SUMO-2 / 3 (SILAC3) antibodies.

Mass spectrometry identification of SUMOylated proteins
Enriched SUMOylated proteins from SILAC lysates were size-separated by SDS-PAGE and in-gel digested with trypsin.The resulting peptide mixtures were extracted, desalted and concentrated on STAGE-tips with two C18 filters and eluted two times with 10 l of acetonitrile 40% in formic acid 0.5% prior to online nanoflow liquid chromato gra phytandem mass spectrometry (nano LC-MS / MS) using an EASY-nLC system (Proxeon, Odense, Denmark) connected to the Q Exacti v e HF (Thermo Fisher Scientific, Germany) through a nano-electrospray ion source.Peptides were separated in a 15 cm analytical column in-house packed with 1.9 m C18 beads (Reprosil-AQ, Pur, Dr Manish, Ammerbuch-Entringen, Germany) using an 80 min gradient from 8% to 75% acetonitrile in acetic acid 0.5% at a flow rate of 250 nl / minute.The mass spectrometers were operated in data-dependent acquisition mode with a top 10 method.For Q-Exacti v e measurements, full scan MS spectra wer e acquir ed at a target value of 3 × 10 6 and a resolution of 60 000 and the Higher-Collisional Dissociation (HCD) tandem mass spectra (MS / MS) wer e r ecorded at a target value of 1 × 10 5 and with a resolution of 60 000 with a normalized collision energy of 30%.
Raw mass spectrometry (MS) files were processed with the MaxQuant software suite (version 1.4.0.3, www.maxquant.org).All resulting MS / MS spectra were searched against the human Uniprot database ( www.uniprot.org) by the Andromeda search engine using the re v ersed database strategy a ppl ying a false discovery rate of 0.01 at both peptide and protein le v els.Ov err epr esentation of Gene Ontologies of the identified proteins were analyzed using Fisher's exact test from InnateDB ( 55 ).

Statistical analyses
Results ar e expr essed as means ± S.D. Statistical analyses were performed using Anova or paired Student's t -test with the Prism 9 softwar e. Differ ences wer e consider ed as significant for P -values of < 0.05.*, **, ***, **** correspond to P < 0.05, P < 0.01, P < 0.001, P < 0.0001, respecti v ely.ns = not significant.Statistical analyses of the transcriptomic and proteomic experiments are described in the relevant sections.

DNR r apidly induces tr anscriptional progr ams r elated to cell proliferation / death and inflammation / immunity in AML cells
To identify the genes whose expression is ra pidl y altered by Ara-C or DNR in AML cells, we performed a whole transcriptome profiling of HL-60 cells, one of the most widely used cellular model of AML ( 56 ).Cells were treated with each one of the two drugs at doses relevant to the clinical practice (2 and 1 M, respecti v ely) ( 57 , 58 ) for 3 h, i.e. before the onset of a poptosis, w hich begins after 4 h of treatment ( 13 ).Using the Affimetrix array technology, we identified 476 significant differentially expressed genes (DEGs) in DNR-treated cells, 182 being upregulated and 294 downregulated > 2-fold (Figure 1 A and Supplementary Table 1).Much less DEGs were identified in Ara-C-treated cells: 6 wer e upr egulated and 29 downregulated by a > 2-fold factor (Figure 1 B).Gene ontology (GO) enrichment analyses re v ealed that the genes identified as down-regulated upon treatment by Ara-C and / or DNR are mostly involved in nucleosome assembly (Supplementary Figure 1A).Those upregulated principally belong to functional categories linked to signal tr ansduction, tr anscription, cell prolifer ation and death (with both pro-and anti-apoptotic genes being induced) and inflammation / immunity (Figure 1 C, D, Supplementary Figure 1B, Supplementary Table 1).We confirmed the activation of four of the most DNR-induced genes ( CXCL10 , FOSB , NFKB2 and IER3 ) by RT-qPCR in HL-60 cells treated with DNR (Figure 1 E).Noteworthy, these genes were not significantly induced by Ara-C e v en a t concentra tions higher than 2 M (up to 50 M) (Figure 1 E).Taken with our Affymetrix data (Figure 1 A and B), this suggested that DNR is more potent at altering transcription than Ara-C in the HL-60 cell model.We then analyzed samples from three AML patients taken at diagnosis (Supplementary Table 2).These wer e tr eated ex vivo with DNR or Ara-C for 3 h and assayed for the expression of the same four genes.All of them were induced by DNR in the three patients tested, albeit to differ ent degr ees.Their expr ession was more induced by DNR than by Ara-C for two patients, showing that our observation in HL60 cells reflected a situation happening in primary AML cells.Howe v er, the re v erse was observed for the third patient sample, which is probably reflecting AML heterogeneity (Figure 1 F).Finally, we analyzed the effect of DNR and Ara-C on the expression of the same genes in Peripheral Blood Mononucleated cells (PBMC) from three different healthy donors.Only NFKB2 was induced in all three donors, at howe v er lower le v els than in AML patients' cells (Supplementary Figure 1C).Thus, our da ta indica te tha t one early ef fect of the chemotherapeutics used as frontline treatment of AML is tr anscriptional reprogr amming.DNR, howe v er, shows much broader effects than Ara-C and the genes most induced by DNR principally belong to two general functional categories: cell prolifera tion / dea th and inflammation / immunity.

DNR induces a fast r emo v al of SUMO from chromatin, in particular at active promoters and enhancers
We hav e pre viously shown that DNR and Ara-C induce a progressi v e de SUMOylation of proteins in AML.It is due to the inactivation of the SUMO E1 and E2 enzymes via the formation of a ROS-dependent disulfide bond between their catalytic cysteines ( 13 ).Due to the role of SUMOylation in transcription, we wondered whether DNR and Ara-C could induce fast alterations in SUMOylated protein distribution on the genome, as such changes might regulate, positi v ely or negati v ely, drug-induced transcriptional changes.This was addressed in ChIP-Seq experiments with antibodies directed to SUMO-2 / 3. HL-60 cells wer e tr eated with DNR or Ara-C for 2 h, i.e. a time point earlier than that used in our transcriptomic analysis to consider the time r equir ed between gene transcription alteration and RN A accum ulation changes in the cell.In untreated cells, and as previously shown by others ( 17 , 18 , 21 , 25 , 59 , 60 ), SUMO-2 / 3 was found distributed all along chromatin with a pproximatel y 44 000 peaks (Figure 2 A).A particular enrichment was found at both annotated gene promoters and candidate enhancer regions defined by the presence of high H3K27ac, H3K4me1 and low H3K4me3 (Supplementary Figure 2).In mocktreated cells, we identified 6861 genes showing a significant accumula tion of SUMOyla ted proteins in their promoter regions with a peak of enrichment a pproximatel y 100 bp upstream of Transcription Start Sites (TSSs).Interestingly, SUMOylated proteins were found enriched on acti v e promoter regions (those with high H3K4me3 and RN A polII) and not on inacti v e ones (those with low H3K4me3 and RNAPolII) (Supplementary Figure 2A).Along the same line, SUMOylated proteins were found localized in the center of the candidate enhancer regions and slightly more enriched on acti v e-(i.e. with high H3K27ac) than on inacti v e-(i.e. with low H3K27ac) candidate enhancers (Supplementary Figure 2B).
We then anal yzed w hether DNR and Ara-C treatments globally affected the presence and / or the distribution of SUMO-2 / 3-conjugated proteins on chromatin.At promoters (Figure 2 B) and enhancers (Figure 2 C), le v els of SUMOylation remained essentially unchanged upon Ara-C tr eatment (Figur e 2 D, cluster 1).In the other regions of chromatin, most SUMO-2 / 3 peaks disappeared upon Ara-C tr eatment (Figur e 2 D, cluster 2) and wer e r edistributed to other genomic r egions, r esulting in an incr ease in the total number of SUMO peaks (Figure 2 D, cluster 3).Howe v er, the global distribution of SUMO-2 / 3 peaks between chromatin r egions r emained similar upon Ara-C tr eatment (Figure 2 Ad) and the average SUMO-2 / 3 peaks intensity remained unchanged (Supplementary Figure 3A).By contrast, DNR treatment induced a 25% decrease in the total number of SUMO-2 / 3 peaks (Figure 2 Ac) as well as a decrease in the average SUMO-2 / 3 peak intensity (Supplementary Figure 3A).Most chromatin regions lost SUMOylation (Figure 2 D) but the decrease was particularly strong at promoters (Figure 2 B) and enhancers (Figure 2 C).Similar to Ara-C treatment, new SUMO-2 / 3 peaks appeared upon DNR treatment in regions devoid of active transcription marks (Figure 2 D, cluster 4).As mentioned earlier, the bulk of protein SUMOylation is not detectably affected at 2 h of DNR treatment (Supplementary Figure 4A).This raises the idea that chromatin-bound proteins, in particular those enriched at gene cis -r egulatory r egions ar e among the first proteins to be de SUMOylated upon DNR treatment.At this early time point, DNR has already induced some DNA damage as measured by ␥ H2AX staining (Supplementary Figure 4B).Howe v er, Anne xin-V labelling shows that cells have not yet entered into a poptosis, w hich starts after 4 h of treatment (Supplementary Figure 4C).

Inhibition of SUMOylation limits both positive and negative changes in gene expression induced by DNR
As DNR had much stronger effects on chromatin SUMOylation and gene expression than Ara-C, we continued our investigations by assessing whether inhibition of SUMOyla tion is suf ficient to induce the expression of DNRresponsi v e genes.To this aim, we performed RNA-seq analyses of HL-60 cells treated for 3 h with the highly potent and selecti v e SUMOylation inhibitor ML-792 ( 61 ).Upon ML-792 treatment, all SUMO-2 / 3 targets were deconjugated after one hour (Supplementary Figure 4D).Surprisingly, ML-792 had minimal effect on gene expression with only 21 differ entially r egulated genes (Figur e 3 A), suggesting tha t de SUMOyla tion per se is not sufficient to induce DNR-responsi v e genes.As there is no specific de SUMOylation inhibitors that could be used to pre v ent DNR-induced de SUMOylation, we used ML-792 in combination with DNR to strengthen and accelerate DNRinduced de SUMOylation.RNA-Seq being more sensiti v e than the Affimetrix array-based approach, we identified mor e DNR-r esponsi v e genes than in our former transcriptomic approach (Supplementary Figure 4E).552 genes were found up-regulated and 380 down regulated in DNR vs mock-treated cells (Figure 3 A and Supplementary Table 3).The le v el of up-or down-regulation was not correlated to the le v el of change in SUMO-2 / 3 le v els present at their promoters upon DNR treatment (Supplementary Figure 3B).Ne v ertheless, the comparison of ML-792 + DNR-to DNR only-tr eated-cells r e v ealed tha t inhibition of SUMOyla tion during the DNR treatment generally limited DNR-target genes up-or down-regulation (Figure 3 C, D).This was in particular the case for the genes, which are the most affected by DNR (Figure 3 D and E).GSEA analysis showed that all pathways enriched in DNR-treated cells were less or not enriched at all when SUMOylation was inhibited, the most pronounced effects being observed for the genes involved in inflammation (Figure 3 F and Supplementary Table 4).Thus, our data suggest that inhibition of SUMOylation counteracts the ability of DNR to alter the expression of its responsi v e genes, whether induced or down-regulated.

T r anscription factors and co-r egulators ar e the fastest and main class of de SUMOylated proteins upon DNR treatment
To better understand how de SUMOylation controls DNRresponsi v e gene e xpression, we ne xt resorted to large-scale proteomics to identify the proteins changing their SUMOylation le v els after 2 h of DNR treatment, i.e. the time point at which important changes in chromatin protein SUMOylation were detected by ChIP-seq (Figure 2 ).First, we characterized the HL-60 cell proteome conjugated to SUMO-2 / 3 but also to SUMO-1.To this aim, we immunoprecipitated and identified by quantitati v e mass spectrometry SUMO-2 / 3 and SUMO-1 modified proteins.894 SUMO targets were identified, most of them being modified by both SUMO-2 / 3 and SUMO-1 (Supplementary Figure 5A).Then, SUMO-2 / 3 or SUMO-1-conjugated proteins were immunoprecipitated and identified from HL-60 cells treated or not with DNR for 2 h.As expected from immunob lotting e xperiments (Supplementary Figure 4A), the SUMOylation le v el of most proteins did not change after 2 h treatment with DNR.Howe v er, 34 proteins (31 for SUMO-2 / 3 and 11 for SUMO-1, 8 proteins being common) showed increased modification (Figure 4 A and Supplementary Table 5).More proteins (83 for SUMO-2 / 3 and 32 for SUMO-1, 19 being common) showed a significant decrease in their SUMO conjugation upon DNR treatment (Figure 4 A and Supplementary Table 5).Finally, these changes were not due to modifications of protein abundance, as determined by sequencing of input samples in control-and DNR-treated cells (Supplementary Figure 5B and Supplementary Table 5).Interestingly, after 2 h of treatment, most of the de SUMOylated proteins (both SUMO-2 / 3 and SUMO-1 substrates) were found to be chromatinbound proteins involved in the regulation of gene expression (Figure 4 B).
Thus, our proteomic data support the idea initially raised by our SUMO-2 / 3 ChIP-seq experiments (Figure 2 ) tha t chroma tin-bound proteins are among the first to be de SUMOylated upon treatment by DNR.

CTCF colocalizes with SUMO on chromatin, in particular on active cis-regulatory regions, and is de SUMOylated upon DNR treatment
Among the SUMOylated substrates found de SUMOylated upon DNR treatment in the SILAC experiment (Figure 4 A), we noted the CCCTC-binding factor CTCF, an insulator protein known to regulate the three-dimensional architecture of chromatin ( 62 ).CTCF was formerly reported to be SUMOylatable ( 63 ) and its SUMOylation to be instrumental for activation and r epr ession of the PAX6 ( 64 ) and c-MYC ( 65 ) genes, respecti v ely.We first confirmed the SUMOylation of CTCF by the presence of a band migrating above CTCF on SDS-PAGE, w hich disa ppeared upon SUMOylation inhibition with ML-792 in both HL-60 (Figure 4 C) and primary AML patient's cells (Figure 4 D).DNR as well as the other anthracycline Idarubicin (IDA) induced a decrease in CTCF SUMOylation, whilst Ara-C had no effect (Figure 4 C and D).In addition, we found that the most r epr esented DNA-binding motif under the SUMO peaks identified in our ChIP-seq experiments (Figure 2 ) was the consensus CTCF-binding motif (Figure 4 E and Supplementary Table 6).To further confirm the link between SUMO and CTCF, we performed CUT&RUN experiments with CTCF antibodies to map CTCF binding sites in HL-60 cells.This showed a strong colocalization between SUMO and CTCF binding on the chr omatin, with ar ound one third of SUMO-bound regions being bound by CTCF (Figure 4 F).The strongest co-localization was found at chromatin regions presenting marks of active transcription (Figure 4 F).This is in particular the case around gene TSSs, which are losing SUMOylation upon DNR but not Ara-C tr eatment (Supplementary Figur e 6A).Tr eatment with DNR, ML-792 or their combination did not significantly affect CTCF distribution on the chromatin (Figure 4 G) suggesting that decreased SUMOylation of CTCF and other chr omatin-bound pr otein does not induce the offloading of CTCF from chromatin.

SUMOylation regulates DNR-induced expression of the CTCF and SUMO-bound NFKB2 gene
To further investigate the link between CTCF and SUMO in DNR-induced gene expression changes, we crossed the list of genes presenting SUMOylated proteins and CTCF in their promoters with that of genes transcriptionally affected more the 2-fold upon DNR treatment.Sixty-one genes were identified, the expression of which might be regula ted through SUMOyla tion / de SUMOyla tion of proteins bound to their promoter regions (Supplementary Figure 6B, left panel).We then crossed this list with that of the 36 genes whose DNR-induced expression changes was altered by more than 2-fold upon SUMOylation inhibition (Supplementary Figure 5B, right panel).This led to the identification of four genes ( EGR1 , ICAM1 , MYC and NFKB2 ) whose DNR-induced up-or down-regulation is reduced upon inhibition of SUMOylation and whose proximal promoters are marked by SUMO and CTCF.We then focused on the NFKB2 gene, encoding the transcription factor Nuclear Factor-kappa B2 (NF-B2), because of its involvement in the regulation of both cell death / survival and inflammation / immunity ( 66 , 67 ), processes we found associated with the response of AML to DNR.Moreover, after having formerly shown that DNR induces NFKB2 expression in AML patients' cells treated in vitro (Figure 1 F), we established the early induction of this gene in vivo using peripheral blood mononuclear cells (PBMCs) purified from 3 AML patients before and 4 h after the beginning of an induction chemotherapy comprising DNR and Ara-C (Figure 5 A).Using HL-60 cells, we could show that the DNR + Ara-C combination was howe v er not more efficient then DNR alone at inducing NFKB2 .The other anthracycline IDA was also inducing NFKB2 , at e v en higher le v els than DNR (Supplementary Figure 6C).Finally, higher induction le v els were detected when considering only NFKB2 longest isoform, which starts at the CTCF / SUMO bound site (Figure 5 B).Consistent with our RNA-Seq data (Figure 3 ), the SUMOylation inhibitor ML-792 decreased the DNR-induced expression of NFKB2 .Similar results were obtained with another SUMOylation inhibitor, TAK-981 ( 68 ) (Figure 5 B).In addition, DNR led to the accumulation of NFKB2 protein, which was limited by ML-792 (Figure 5 C).Importantly, ML-792 also pre v ented the induction of NFKB2 by DNR in primary AML cells from 2 patients treated ex vivo (Figure 5 D).
To further confirm the implication of SUMOylation inhibition in this process, we resorted to RNAi to downregulate the SUMO E2 enzyme Ubc9.This did not affect the basal le v el of NFKB2 e xpression but limited its DNRinduced up-regulation (Figure 5 E).ChIP-Seq data identified a major SUMO-2 / 3 peak colocalizing with CTCF at the most 5 promoter of NFKB2 in HL-60 cells (Figure 5 F), w hich disa ppear ed upon DNR tr eatment.Howe v er, consistent with the genome wide results, DNR did not affect the binding of CTCF to the NFKB2 gene (Figure 5 F).Thus, altogether, our results suggest that de SUMOylation limits DNR-induced expression of the CTCF-bound NFKB2 gene without modifying CTCF binding to the locus in AML cells.

De SUMOylation limits DNR-induced chromatin 3D r earrangements at the NFKB2 locus
Pub licly availab le HiC da ta indica te tha t NFKB2 is loca ted a t the center of a Topolo gicall y-Associating Domain (TAD), which extends over 500 kb on chromosome 10 (Figure 6 A).They also suggest the existence of various long-r ange inter actions between the NFKB2 gene and distant regions within this TAD.Moreover, CTCF largely colocalizes with SUMO-2 / 3 in HL-60 cells, not just at the NFKB2 locus, but also at various places covering the whole NFKB2 TAD (Figure 6 B).Together, these observations suggested that DNR-induced NFKB2 expression could be associated with changes in chromatin organization that could be regulated by SUMOylation / de SUMOylation e v ents.
To address this point, we resorted to Circularized Chroma tin Conforma tion Capture (4C) experiments in HL-60 cells, using the NFKB2 promoter as a viewpoint.In mock treated cells, we found that this promoter interacts significati v ely with two regions upstream of the NFKB2 gene (regions I and II) and two downstream of it (regions III and IV) (red domains in the upper lane of Figure 6 C).Noteworthy, they were all localized within the NFKB2 TAD in the hundred kb-range from the NFKB2 TSS and presented at least one CTCF-bound site.
The overall topology of the NFKB2 locus was not strongly affected by a 2 h treatment with DNR (compare green and orange profiles in the first two lanes of Figure 6 C).Howe v er, a differential profiling analysis (Figure 6 D) showed decreased interactions between the CTCF / SUMObound NFKB2 promoter and region IV in DNR-treated cells.Moreov er, DNR induced a ne w interaction with the region V localized at the extreme border of the NFKB2 TAD (Figure 6 C and D).Interestingly, this new interacting region is enriched for histone marks characteristic of acti v e enhancers (H3K27ac and H3K4me1), while the interacting region IV in mock-treated cells was devoid of such marks (Figure 6 B).Thus, DNR-induced up-regulation of NFKB2 is associated with changes in the frequencies of chromatin looping between its promoter region and distal regions within the NFKB2 TAD, which include a potential enhancer.
To assess whether the SUMO pathway could be involved in chromatin 3D organization changes induced by DNR at NFKB2 locus, we also conducted 4C experiments on cells treated with ML-792 alone or in combination with DNR.Treatment with ML-792 alone, which did not affect NFKB2 gene expression, did not modify the overall 4C profile of the locus (compare green and blue profiles in lanes 1 and 3 of Figure 6 C and see differential profiling in Figure 6 D).Howe v er, w hen used to gether with DNR, ML-792 pre v ented the changes observed in the presence of DNR only ( i.e. reduction of interactions with regions IV and induction of interaction with region V) and led to a new interaction with region VI surprisingly localized beyond the NFKB2 TAD border (Figure 6 C and D).Taken together, our data suggest tha t de SUMOyla tion of proteins bound a t CTCF-bound sites in the NFKB2 promoter limits NFKB2 activation by DNR by affecting the chromatin 3D ar chitectur e changes induced by DNR at this locus.

DISCUSSION
In this work, we report that an early effect of DNR, one of the two frontline chemotherapeutics used in AML treatment, is an alteration of specific transcriptional programs.DNR modifies the expression of almost 1000 genes in chemosensiti v e HL-60 cells after only 3 h of treatment.In contrast, much less genes ar e r egula ted by Ara-C .Importantly, selected DNR-up-regulated genes were also ra pidl y induced in three primary AML patient samples and one of them ( NFKB2 ) was also ra pidl y upregulated in vivo during standar d AML chemotherapy.Howe v er, besides this, the top DNR-up-regulated genes found in HL-60 cells were more induced by Ara-C than by DNR in one of the AML primary samples whereas they were hardly induced by Ara-C in the two other samples.Thus, altogether, our data indica te tha t DNR and Ar a-C induce r apid (hour-r ange) tr anscriptome changes in AML with the effect of DNR being much stronger than those of Ara-C.Howe v er, at the same time, they also suggest a certain degree of variability between AML patients that is likely explained by AML heterogeneity.
One of the main pathways we found associated with DNR-up-regulated genes is apoptosis.This suggests that the rapid gene expression changes induced by this drug set up a favorable pro-apoptotic ground that adds to the DNA damages it generates for killing chemotherapytreated AML cells at a later stage.It should, howe v er, be noted that, in addition to pro-apoptotic genes, antiapoptotic ones were also activ ated.This observ ation is consistent with those by others tha t DNR also activa tes prosurvival PI3-K / AKT-and NF-B pathways and that their targeting is considered as a potential therapeutic strategy to improve their efficiency ( 6 , 66 ).Another functional category found enriched in DNR-induced genes was inflammation and immunity-related processes.In various immunocompetent mouse models, antracyclines were described as capable of inducing the imm uno genic cell death of di v erse solid tumors, in particular through the induction of an interferon response (69)(70)(71).The genes we identified as up-r egulated in DNR-tr eated AML cells could participate in the de v elopment of an adaptati v e immune response against leukemic cells in chemotherap y-tr ea ted pa tients.Finally, downregulated genes are highly enriched for histone genes.This could result in decreased histone le v els, which might loosen chromatin and favor the genotoxic action of the chemothera peutic drugs.Alto gether, our data suggest that the fast transcriptome changes induced by DNR befor e tr ea ted cells start d ying may contribute to the response of AML to this drug.The molecular mechanisms underlying the effect of anthracyclines on gene expression are howe v er far from being understood and probably multiple.Anthracyclines induce histone eviction at open chromatin r egions, which wer e proposed to participate to the regulation of gene expression ( 72 ).DNA-damage induced by anthracyclines could also modulate gene expression.Howe v er, DNA-damages, in particular double strand breaks, are known to stall RN A-pol ymerase II at the break point and cause a global transcriptional shut down ( 73 ).It is ther efor e unlikely that DNR-induced transcriptional reprogramming, in particular gene up-r egulation, is dir ectly due to DNA damage.Ne v ertheless, DNA-damage-induced activation of specific transcription factors could participate to the activation of specific genes.Finally, anthracyclines are known to generate ROS, which functions as second messengers via the re v ersib le oxida tion of ca talytic cysteines to activate many signaling pathways ( 74 ).Although anthr acyclines-induced ROS gener ation has mostly been studied in cardiomyocytes due to their key role in anthracy clines car diotoxicity ( 75 ), it is expected that they activa te signaling pa thways in cancer cells.Among the targets of anthr acyclines-gener ated ROS are the SUMO E1 and E2 enzymes, whose respecti v e catalytic cysteines form a rev ersib le disulfide bridge upon oxidation, inhibiting their ability to activate and transfer SUMO to target proteins ( 13 , 76 , 77 ).Here, we show that DNR induces a rapid and massi v e de SUMOylation of chromatin-bound proteins, in particular at acti v e promoters and enhancers were SUMOylated proteins are highly enriched.As this occurs before massi v e de SUMOylation of other cellular proteins becomes detectable, this indicates that DNR-induced protein de SUMOylation is not random in the cell.It suggests it is kineticall y and spatiall y order ed by mechanisms that r emain to be characterized (also see below).It is howe v er worth noting that, although DNR-induced de SUMOylation affects most genomic regions where SUMO-bound proteins were found in non-treated cells, new genomic regions, mostly intergenic, gain SUMOylation.As SUMO isoforms are limiting, DNR-induced deSUMOyla tion a t promoters and enhancers could enhance the pool of unconjugated SUMO and favor the SUMOylation of other chromatin-bound proteins, such as Topoisomerase 2 and centromeric proteins (CENP-C and CENP-B), which we found up-SUMOylated upon DNR-treatment (Figure 4 A).This might be also true for Ara-C, which also leads to a redistribution of SUMOylated proteins on the chromatin.Howe v er, contrarily to DNR, SUMOylated proteins are maintained at promoters and enhancers upon Ara-C treatment.
To address if DNR-induced de SUMOylation has a role in DNR-induced gene expression alterations, we performed RNA-Seq in cells treated with DNR and the SUMOylation inhibitor ML-792.As DNR induces fast chromatin protein de SUMOylation, we first asked whether inhibition of SUMOylation alone could reproduce its effect on gene expression.This was not the case as ML-792 had very small effects on gene expression (only 18 genes up-regulated and 3 down-regulated) after 3 h or treatment.This suggests that the inhibition of SUMOylation induced by DNR is not, on its own, responsible for the fast and broad transcriptome changes.This observation is consistent with the initial report on ML-792 showing that only a few genes are activated in cultured cells by this inhibitor, e v en after longer treatments ( 61 ).We therefore wondered whether protein de SUMOylation would have a role in the regulation of gene expression only in the presence of DNR.To this aim, we used ML-792 in combination with DNR, to accelerate and strengthen the de SUMOylation induced by this drug.Although ML-792 had little effect on the nature and the number of the genes up-or down-regulated upon DNR, it limited their up-or down-regulation.Indeed, most gene signatures enriched in DNR-treated cells were no longer enriched upon inhibition of SUMOylation.This suggested that acute de SUMOylation counteracts DNR ability to activate or repr ess gene expr ession.We howe v er do not exclude that longterm and / or moderate hypo-SUMOylation could have a dif ferent ef fect on gene expression.
Our proteomic-based study of the HL-60 cell SUMOylome characterized the proteins that are de SUMOylated in response to DNR.Out of the 900 SUMOylated proteins identified in mock-treated cells, only 100 were significantly de SUMOylated after 2 h of DNR treatment.Consistent with the massi v e loss of SUMO-2 / 3 observed by ChIP-seq at promoters and enhancers at the same time point, most of these de SUMOylated proteins are transcription factors and co-regulators.This suggests that early DNR-induced de SUMOylation is spatially regulated and pr efer entially concerns proteins bound to specific chromatin regions, many of them probably being engaged in the same protein complexes.SUMOylation is indeed known to stabilize transcriptional complexes at gene regulatory regions to maintain transcription ( 16 ).For example, SUMOylation stabilizes transcription factor comple xes involv ed in the expression of somatic tr anscriptional progr ams in MEFs ( 25 , 78 ).Massi v e increase in the SUMOylation of chromatin-bound protein upon heat-shock is also r equir ed to stabilize protein complex on gene regulatory regions to maintain their transcription ( 79 ).In both cases, protein complex es ar e likely modified following a process called 'group SUMOylation' ( 80 ).According to this concept, SUMO can control the activity of protein complexes regardless of the modified protein, or the precise sites that are SUMOylated on these proteins.DNR-induced de SUMOylation could loosen interactions within transcription-r egulating complex es binding at the promoters and / or enhancers of the genes affected by DNR, thus limiting the transcription-promoting effect of DNR.If fast DNR-induced de SUMOylation at precise chromatin sites is most probably partly explained by local inhibition of chromatin-bound E1 and E2 SUMOylation enzymes, it might also involve faster deconjugation of SUMO by de SUMOylases at these same places.For example, SENP6 was reported to de SUMOylate CTCF ( 81 ), one of the proteins we found de SUMOylated by the DNR tr eatment.CT CF is a multifunctional protein involved in both the regulation of chromatin 3D ar chitectur e and the control of gene expression ( 82 ).It interacts with the cohesin complex (composed of SMC1, SMC3, RAD21 and SA1 / 2 proteins) and is involved in the formation of di v erse chroma tin regula tory loops ( 83 ).Depending on the situation, such loops can activate transcription by bringing enhancers and promoters in close proximity or r epr ess it by limiting the access of transcriptional machineries or regulators to gene promoters ( 82 ).CTCF is SUMOylated ( 63 , 65 ), its SUMOylation being decreased by various stresses including hypoxia and oxidati v e stress ( 64 ).Further links between SUMO and CTCF were described on chromatin.First, the CTCF-binding consensus sequences was found enriched at genomic loci bound by SUMOylated proteins, in particular at promoters of inacti v e genes ( 84 ).Second, heat shock was shown to induce a transient depletion of SUMOylated proteins from CTCF-bound sites in intergenic regions and their relocation at promoters of transcribed genes ( 24 ).Third, SUMOylated proteins were found enriched at CTCF-bound sites in Drosophila and associated to enhancer blocking ( 85 ).Along this line, we found that the CTCF-binding site is the most enriched motif in SUMO-2 / 3 bound chromatin regions in AML cells and CUT&RUN experiments with CTCF antibodies confirmed that the co-binding of CTCF and SUMO is highly enriched at promoters and enhancers compared to intergenic regions.Moreover, the identification of CTCF as one of the proteins ra pidl y de SUMOyla ted upon DNR trea tment, suggests that DNR-induced hypoSUMOylation of CTCF and probably of other still-to-be-identified proteins present at CTCF-bound sites could regulate the expression of specific genes through chromatin looping alteration.Hence, although we only identified four genes bound by CTCF and SUMO in their promoter and whose DNR-induced up-or down-regulation was blunted by ML-792 (more than 2-fold), we decided to explore this hypothesis.We focused on the NFKB2 gene for se v eral reasons: (i) it is one of the top-DNR-induced gene in HL-60 cells, (ii) its induction by DNR is reduced in the presence of ML-792 in HL-60 cells as well as in primary AML samples and (iii) its promoter region is both bound by CTCF and marked by SUMO and (iv) it plays important roles in the control of both cell survival and inflammation / immunity ( 66 ), two of the main gene categories ra pidl y affected by the DNR treatment.Our 4C experiments revealed that NFKB2 promoter pr efer entially contacts four distal regions located up to 200 kb upstream (2 regions) and downstream (2 regions) of the NFKB2 gene, all within the NFKB2 containing TAD and bound by CTCF in HL-60 cells.Although DNR did not markedly alter the overall ar chitectur e of the NFKB2 locus, it induced the loss of an interaction between NFKB2 promoter and a region devoid of active histone marks (region IV) and the appearance of a new interaction with a candidate enhancer (region V).This probably reflects the loss of a transcription-r epr essi v e loop and the acquisition of a transcription-stimulating one.Consistent with its limited effects on gene expression, the sole inhibition of SUMOylation by the ML-792 inhibitor alone did not affect the overall structure of the NFKB2 locus.This indicated that SUMOylation per se is not r equir ed for maintenance of the chromatin loops forming between the NFKB2 promoter and the above-mentioned interacting regions (at least for the duration of the e xperiment).Howe v er, when used with DNR to accelerate and amplify DNR-induced de SUMOylation, ML-792 pre v ented the DNR-induced interaction between NFKB2 promoter and the candidate enhancer located in region V. Instead, a new interaction with a region located beyond the TAD border (region VI) was induced.This switch might pre v ent full acti vation of NFKB2 gene.Altogether, this suggests that de SUMOylation can attenuate the transcriptional effects of DNR by contr olling chr omatin 3D structure, at least on the NFKB2 locus.Rapid and massi v e changes in the SUMO proteome associated to transcriptome alterations have already been observed in response to various external cues, including heat shock ( 24 , 79 ), oxidati v e stress ( 28 , 77 ) and genotoxics such as MMS ( 86 ).Our herein data suggest that such SUMO-dependent switches might control transcriptome changes at least in part by affecting chromatin 3D ar chitectur e or dynamics.This is all the more to be considered that inducible genes have been reported to be more enriched in CTCF-controlled chromatin loops than housekeeping ones ( 87 , 88 ).Future work will ther efor e have to elucidate whether SUMO serves as a platform, especially at CTCF-bound sites, to recruit proteins involved in chromatin remodeling or structuration and how SUMOyla tion / de SUMOyla tion cycles a t these places contributes to transcriptional changes linked to alteration of 3D chromatin organization.

DA T A A V AILABILITY
Microarray data were deposited on Arrayexpress with accession number E-MATB-4895.ChIPSeq, RNA-Seq, 4C and CUT&RUN sequencing data were deposited on Gene Expression Omnibus with accession number GSE198986.Data can be visualized on the UCSC genome browser: https://genome.ucsc.edu/s/MathiasBoul/Boulanger%20et%20al%20%2D%20HL60%20datasets .The mass spectr ometry pr oteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD032956.

SUPPLEMENT ARY DA T A
Supplementary Data are available at NAR Online.

FFigur e 1 .
Figur e 1. Chemothera peutic drugs ra pidl y alter the expression of genes involved in cell death and inflammation in AML cells.(A, B) T r ancriptome pr ofile.HL-60 cells were treated with 1 M DNR ( A ) or 2 M Ara-C for 3 h ( B ). RNAs were purified from three independent experiments and used to probe Affymetrix Human Gene 2.0 ST Genechips.The red dots on the Volcano plots r epr esent the Differ entially Expr essed genes (DEG) with an absolute Fold Change (FC) ≥ 2 (log 2 ≥ 1) and a False Discovery Rate (FDR) corrected with Empirical Bayes Statistics (EBS) ( 89 ) < 0.05.( C ) Gene Ontology enrichment analysis of the genes up-regulated ( ≥2 fold) by DNR and Ara-C.Ontologies were performed using the Panther GO database( 51 ).The main terms of each identified group are presented on the graph and classified by the number of genes present in each group.P values are corrected with Bonferroni step down.( D ) Heatmap of DEG with a FC ≥ 4 in the transcriptomic experiments presented in (A) and (B) .The data for all three replicates are represented.( E ) R T-qPCR anal ysis of selected genes .HL-60 cells wer e tr eated for 3 h with 1 M DNR or 2 M Ara-C.The le v els of the indicated mRNAs wer e measur ed by RT-qPCR, normalized to GAPDH le v els and e xpr essed as fold incr ease to mock-tr eated cells (mean ± SD, n = 7 for NF-κB2 , n = 6 for IER3 , n = 5 for FOSB , CXCL10 ). ( F ) Regulation of selected genes in primary AML cells.AML cells (bone marrow aspirate) from three patients wer e tr eated in vitro with 1 M DNR or 2 M Ara-C for 3 h.The le v els of the indicated mRNAs wer e measur ed by RT-qPCR, normalized to TBP le v els and e xpressed as fold increase to mock-treated cells.

Figure 2 .Figure 3 .
Figure 2. Treatment of AML cells with DNR depletes SUMOylated proteins from the chromatin, in particular at promoters and enhancers.(A, B) ChIP-Seq analyses of SUMO-2 / 3 distribution on the genome.HL-60 wer e tr eated with 1 M DNR or 2 M Ara-C for 2 h.ChIP-Seq experiments were carried out with SUMO-2 / 3 antibodies.( A ) a: a proportion of the different genomic regions, b −d: proportion of SUMO-2 / 3 peaks on these chromatin regions in mock-(b), DNR-(c) or Ara-C-(d) treated HL-60 cells.( B , C ) Metaprofile of the SUMO-2 / 3 ChIP-seq signal on HL-60 promoters (B) or enhancers (C) in mock-, DNR-or Ara-C-treated HL-60 cells.Promoters ( −2 kb to TSS) and enhancers as well as their activation state were defined using H3K27ac, H3K4me1 and H3K4me1 profiles as well as NCBI refseq data (see Material and methods and Supplementary Figure 2).( D ) Heat-map for the distribution of SUMO-2 / 3, H3K27ac, H3K4me1, H3K4me3.The clustering was performed on SUMO peaks present in any of the conditions (Mock, DNR, Ara-C) and the ranking was made according to SUMO-2 / 3 signal.

FFigure 4 .FFigure 5 .
Figure 4. DNR leads to de SUMOylation of chroma tin regula tors, including CTCF.( A ) Chang es in SUMO-1 and SUMO-2 / 3 proteomes upon DNR tr eatment.SUMOyla ted proteins were immunoprecipitated with SUMO-1 or SUMO-2 / 3 antibodies from SILA C-la beled HL-60 cells treated or not with DNR (1 M for 2 h).Scatterplot analysis of SUMO-1 and SUMO-2 / 3 proteome change (log 2 ratio) in cell tr eated compar ed to mock-treated cells.Doted lines r epr esent lo g 2 ratio of ±0.5.Onl y proteins found to be SUMOylated (Supplementary Figur e 4A) ar e r epr esented.( B ) DeSUMOylated pr oteins ar e mostly tr anscriptional r egulator s .Gene Ontolo gy anal ysis of the identified down-SUMOylated proteins for SUMO-1 and SUMO-2 / 3 in response to DNR (log 2 ratio < −0.5) were obtained using the Panther Protein Class da tabase ( 51 ).(C , D) CTCF is SUMOylated in HL-60 and patient cells .HL-60 ( C ) or AML patient cells ( D ) wer e tr eated with DNR (1 M), ML-792 (0.5 M), IDA (1 M) or Ara-C (2 M) for 3 h.Total cell extracts were loaded on SDS-PAGE and immunoblotted with CTCF antibodies.( E ) The CTCF motif is enriched at SUMO-2 / 3 binding sites.Motif enrichment search was performed with homer pearl script (findMotifs.pl)on the SUMO-2 / 3 ChIP-Seq data obtained for mock-treated HL-60.The three most enriched motifs are shown.( F ) SUMO / CT CF ov erlap on pr omoter s and enhancer s .HL-60 cells wer e tr eated with DNR (1 M), ML-792 (0.5 M) or the combination for 2 h.Cell extracts were then used to perform CUT&RUN with CTCF antibodies (three independent biological replica tes).Hea t-map for the distribution of SUMO-2 / 3 (ChIP-Seq, see Figure 2 ), H3K27ac, H3K4me1, H3K4me3 and CTCF (CUT&RUN).The clustering was performed on H3K4me1 and H3K4me3 and the ranking was made according to SUMO-2 / 3 signal.( G ) Metaprofile for the distribution of CTCF peaks on the whole genome in cells treated for 2 h with mock, DNR (1 M), ML-792 (0.5 M) or the combination.

Figure 6 .
Figure 6. de SUMOylation limits DNR-induced changes in the 3D conformation of the NF κB2 locus .( A ) HiC map of the TAD containing NF κB2 gene.This map was obtained using publicly available HiC data obtained in the K562 human chronic myeloid cell line( 90 ).The NFKB2 -containing TAD is underlined in blue.( B ) Distribution of SUMO and CTCF in the NF κB2 containing TAD. CUT&RUN data for CTCF and ChIP-Seq data for SUMO-2 / 3, H3K27ac and H3K4me1 ar e r epr esented by the normalized read count per 50 bp bin ( C , D ) Inhibition of SUMOylation limits DNR-induced changes in NF-κB2 locus 3D conformation .HL-60 treated for 2 h with DNR (1 M), ML-792 (0.5 M) or the combination and subjected to 4C experiment (three biological replicates).The Y axis of the 4C-seq tracks r epr esents the normalized interaction frequencies with the viewpoint ( NFKB2 promoter, VP) per 10 bp bin.Grey zones are highly reproducible interaction region in at least one condition (regions plotted in red present a P -value < 0.05 in the peakC analysis of the three replicates) and named from I to VI. (D) Differential analysis of the contact point frequency in the regions IV-VI for DNR, ML-792 and ML-792 + DNR compared to mock-treated cells.p -values for the peaks showing statistically significant differences between the conditions are indicated.

Table 1 .
Sequences of the primers used for RT-PCR experiments