The Polycomb Repressive Complex 1 Protein BMI1 Is Required for Constitutive Heterochromatin Formation and Silencing in Mammalian Somatic Cells*

Background: BMI1 silences the expression of genes located at the facultative heterochromatin. Results: BMI1 is abundant at repetitive genomic regions, including the pericentromeric heterochromatin (PCH), where it is required for compaction and silencing. Conclusion: BMI1 is essential for PCH formation. Significance: BMI1 function at PCH is important to understand how BMI1 regulates genomic stability. The polycomb repressive complex 1 (PRC1), containing the core BMI1 and RING1A/B proteins, mono-ubiquitinylates histone H2A (H2Aub) and is associated with silenced developmental genes at facultative heterochromatin. It is, however, assumed that the PRC1 is excluded from constitutive heterochromatin in somatic cells based on work performed on mouse embryonic stem cells and oocytes. We show here that BMI1 is required for constitutive heterochromatin formation and silencing in human and mouse somatic cells. BMI1 was highly enriched at intergenic and pericentric heterochromatin, co-immunoprecipitated with the architectural heterochromatin proteins HP1, DEK1, and ATRx, and was required for their localization. In contrast, BRCA1 localization was BMI1-independent and partially redundant with that of BMI1 for H2Aub deposition, constitutive heterochromatin formation, and silencing. These observations suggest a dynamic and developmentally regulated model of PRC1 occupancy at constitutive heterochromatin, and where BMI1 function in somatic cells is to stabilize the repetitive genome.

The polycomb repressive complex 1 (PRC1), containing the core BMI1 and RING1A/B proteins, mono-ubiquitinylates histone H2A (H2A ub ) and is associated with silenced developmental genes at facultative heterochromatin. It is, however, assumed that the PRC1 is excluded from constitutive heterochromatin in somatic cells based on work performed on mouse embryonic stem cells and oocytes. We show here that BMI1 is required for constitutive heterochromatin formation and silencing in human and mouse somatic cells. BMI1 was highly enriched at intergenic and pericentric heterochromatin, co-immunoprecipitated with the architectural heterochromatin proteins HP1, DEK1, and ATRx, and was required for their localization. In contrast, BRCA1 localization was BMI1-independent and partially redundant with that of BMI1 for H2A ub deposition, constitutive heterochromatin formation, and silencing. These observations suggest a dynamic and developmentally regulated model of PRC1 occupancy at constitutive heterochromatin, and where BMI1 function in somatic cells is to stabilize the repetitive genome.
Chromosomes are structurally organized in distinct subcompartments as determined by the local DNA sequence and chromatin organization. Euchromatin defines "relaxed" chromatin regions containing actively transcribed genes. In contrast, heterochromatin defines "compacted" chromatin regions containing tissue-specific and developmental genes (the facultative heterochromatin) or gene-poor regions (the constitutive heterochromatin) (1). The constitutive heterochromatin is found at the center (centromere) and ends (telomeres) of chromosomes and is mostly constituted of repetitive DNA sequences (1). Numerous (about 10,000) repetitive A/T-rich DNA elements of 231 bp are also found in the pericentromeric heterochromatin (PCH) 4 of mouse chromosomes. Because constitutive heterochromatin regions contained repetitive DNA sequences, the maintenance of chromatin compaction is essential to preserve genomic stability (2). During mitosis, repetitive elements can recombine, resulting in non-homologous recombination between different chromosomes or different regions of paired chromosomes and thus chromosomes deletion, translocation, and fusion (3). Repetitive DNA sequences can be also transcribed, resulting in aberrant noncoding RNA. Stabilization of telomeres and centromers is also essential for chromosome end capping and kinetochore attachment during mitosis (4). Finally, about 40% of the mammalian genome is constituted of "parasitic" retro-element located in intergenic regions of chromosomes. Active repression of these elements is important to maintain genomic stability because some of these can self-replicate and randomly integrate the genome (5,6).
Nucleosomes are the basic building unit of chromatin and are constituted of a 147-bp DNA wrapped against a histone octamer containing two molecules of each of the four histones H2A, H2B, H3, and H4 (the nucleosome core particle) (1). Addition of linker histones, such as histone H1, increases the amount of associated DNA by 20 bp to elicit higher levels of chromatin compaction and high order chromatin structure. The chromatin is also attached at multiple points to the nuclear envelope and spatial organization of the chromatin in the nucleus is important for the regulation of gene transcription (7,8). Post-translational modifications of histones, such as methylation, acetylation, and ubiquitylation can modify chromatin compaction and stability. For example, silent or compact chromatin is associated with trimethylation of histone H3 at lysine 9 (H3K9 me3 ) or 27 (H3K27 me3 ), whereas open chromatin is associated with histone H3 trimethylation at lysine 4 (H3K4 me3 ) or acetylation at lysine 9 (H3K9 ac ) (9). Acetylation brings in a negative charge, acting to neutralize the positive charge on histones and decreases the interaction of the N termini of histones with the negatively charged phosphate groups of DNA. In pathological conditions, histones hyper-acetylation can result in chromatin and chromosomes de-condensation (10).
A core of proteins is involved in establishment and maintenance of constitutive heterochromatin. Most of these proteins are conserved in Drosophila and were identified as modifiers of position effect variegation (1). In mammals, the buildup of these proteins to heterochromatic DNA follows a relatively well characterized sequence where zinc finger proteins recognize and bind repetitive DNA sequences. This is followed by enrichment for Histones H1 and H2a/z, accumulation of Hmga1/2, attachment of KAP1/Trim28 (a SUMO E3 ligase), and of the ATP-dependent chromatin remodeler ATRx, deacetylation of histones by HDAC2, and trimethylation of histone H3 at lysine 9 by SUV39h1 and SUV39h2. Association of SUMOylated HP1a and HP1b to a non-coding RNA results in increased binding affinity for H3K9 me3 and this is further enhanced and stabilized by the suppressor of variegation protein DEK1 (11,12). The DEK1-HP1-SUV39 complex then propagates the H3K9 me3 marks on the chromatin, resulting in heterochromatin formation (13)(14)(15)(16)(17). Interestingly, ATRx localizes at both telomeric and PCH, and germline mutations in ATRx are associated with the ␣-thalassemia with mental retardation X-linked syndrome (18 -20). Surprisingly, it was showed that the BRCA1 protein, which possesses histone H2A monoubiquitin ligase activity when in complex with BARD1, is also enriched at PCH and required for H2A ubiquitinylation, heterochromatin compaction, and silencing (21)(22)(23). This novel BRCA1 function was proposed to explain the severe genomic instability phenotype of BRCA1-deficient cells (22).
Although PRC1 proteins have not been directly implicated in constitutive heterochromatin formation or maintenance, at least some line of evidences support this possibility: 1) immuno-gold localization of BMI1 by electron microscopy in U-2 OS cells revealed high enrichment in electron-dense heterochromatin; 2) BMI1 immunolocalization was found at PCH in transformed human cell lines (46 -49). However, based on work performed on mouse embryonic stem cells and oocytes, it is generally assumed that PRC1 proteins are excluded from PCH in normal mammalian somatic cells (30, 50 -52). In contrast with this model, we found that BMI1 is abundant at constitutive heterochromatin in mouse and human somatic cells, and required for heterochromatin formation/maintenance and silencing. In Bmi1-null mice, cortical neurons showed loss of heterochromatin compaction and activation of intergenic retro-elements and satellite repeats. Consistently, Bmi1 co-localized with H3K9 me3 and was highly enriched at PCH in mouse neurons. BMI1 was also enriched at constitutive heterochromatin, including PCH, in normal human neural precursors. Furthermore, BMI1 co-purified with architectural heterochromatin proteins and with histone H3K9 me3 . BMI1 localization and H2A ub deposition at constitutive heterochromatin were EZH2 and H3K27 me3 independent. In both transformed and normal primary somatic cells, BMI1 inactivation resulted in loss of heterochromatin and alteration in the architecture of the nuclear envelope. Notably, BRCA1 localization was unaffected upon BMI1 deficiency, and both proteins showed partial functional redundancy for H2A ubiquitinylation, heterochromatin formation, and silencing. These findings reveal an essential function for BMI1 in constitutive heterochromatin formation and silencing in mammalian somatic cells.

Experimental Procedures
Animals-Mice were used in accordance with the Animal Care Committee of the Maisonneuve-Rosemont Hospital Research Center (approval ID numbers 2009-40, 2009-42, and 2011-23).
Chromatin Immunoprecipitation (ChIP) Assay-ChIP was performed using the ChIP Assay kit (Upstate). Cells were homogenized at room temperature according to the manufacturer's protocol and sonicated on ice for 10 s at 30% amplitude to shear the chromatin (Branson Digital Sonifier 450, Crystal Electronics, On. Canada). Sonicated materials were immunoprecipitated using 2 g of mouse anti-BMI1, mouse anti-H2AK119ub clone E6C5, mouse anti-RING1B, and mouse anti-HP1 (Millipore), rabbit anti-H3K9me3, and rabbit anti-H3K27me3 (Abcam), rabbit anti-BRCA1 (Santa Cruz Biotechnology), and rabbit anti-mouse IgG (Upstate) antibodies. Fragments were then amplified by real-time PCR in triplicates. Human primers sets used were as described in Ref. 22. ChIPquantitative PCR (qPCR) data were analyzed according to the Percent Input method. First, the raw C t of the diluted 1% input fraction is adjusted by subtracting 6.64 cycles (i.e. log2 of the dilution factor 100). Subsequently, the percent input of each immunoprecipitation (IP) fraction is calculated according to this equation: 100 ϫ 2 (Adjusted Input CtϪCt(IP) .
Real-time RT-PCR-Mouse cortices or human cells were diced and RNA was isolated using TRIzol reagent (Invitrogen). Reverse transcription (RT) was performed using 1 g of total RNA and the Moloney murine leukemia virus reverse transcriptase (Invitrogen). Real-time PCR was carried in triplicates using Platinum SYBR Green Supermix (Invitrogen) and Realtime PCR apparatus (ABI Prism 7002).
Plasmid Constructs and Viruses-Sequence-specific oligonucleotides stretch shRNA designed to target the BMI-1 ORF (accession number BC011652) were synthesized. Oligo#1 (nt 1061-1081) 5Ј-CCTAATACTTTCCAGATTGAT-3Ј, and oligoScramble (nt 573-591) 5Ј-GGTACTTCATTGATGC-CAC-3Ј were used in this study. These sequences are followed by the loop sequence (TTCAAGAGA) and finally the reverse complements of the targeting sequences. The double stranded shRNA sequences were cloned downstream of the H1P promoter of the H1P-UbqC-HygroEGFP plasmid using Age1, SmaI, and XbaI cloning sites. The shRNA-expressing lentiviral plasmids were cotransfected with plasmids pCMVdR8.9 and pHCMV-G into 293FT packaging cells using Lipofectamine (Invitrogen) according to the manufacturer's instructions. Viral containing media were collected, filtered, and concentrated by ultracentrifugation. Viral titers were measured by serial dilution on 293T cells followed by microscopic analysis 48 h later. For viral transduction, lentiviral vectors were added to dissociated cells prior to plating. Hygromycin selection (150 g/ml) was added 48 h later. shBRCA1 constructs (MISSION shRNA) were from Sigma, and siRING1B (FlexiTube siRNA) from Qiagen. The RNAi-resistant BMI1-Myc construct (BMI1 mycϪR ) was generated by synthesis (GenScript) and where the nucleotide sequence of the human BMI1 cDNA (5Ј-CCTAATAC-TTTCCAGATTGAT-3Ј) was changed to (5-CCCAACA-CATTTCAAATAGAC-3), thus preserving the original amino acid sequence of BMI1.
Fixation, Sectioning, and Immunolabeling-Tissues were fixed in 10% buffered formalin and embedded in paraffin according to standard protocols. 5 to 7-m thick sections were mounted on Super-Frost glass slides (Fisher Scientific) and processed for immunohistochemistry staining. Formalin-fixed paraffin-embedded slices were analyzed by using the Vectastain ABC kit (Vector) according to the manufacturer's instructions. Peroxidase substrate DAB (brown) was used (Sigma). Observations were made under a fluorescence microscope (Leica DMRE, Leica Microsystems) and images were captured with a digital camera (Retiga EX; QIMAGING; with OpenLab, version 3.1.1 software; Open-Lab, Canada). Antibodies used in this study were mouse anti-BMI1 and anti-HP1 (Millipore), and rabbit anti-H3K9Ac and anti-H3K9me3 (Abcam). Secondary antibodies used were FITC-conjugated donkey anti-mouse and rhodamine-conjugated donkey anti-rabbit (Chemicon).
Immunoprecipitation and Western Blot-For BMI1/Myc immunoprecipitation experiments, 293T cells were transfected with EFv-/CMV-GFP or EFv-BMI1 Myc /CMV-GFP plasmids using Lipofectamine according to the manufacturer's instructions. Whole cell extracts were collected in IP buffer (100 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% Tween 20, protease inhibitors Complete (Roche Applied Science)). Following the determination of protein concentration, lysates were subjected to immunoaffinity purification. Briefly, protein extracts (4 mg) were incubated with continuous rotation for 3 h at 4°C with 50 l of affinity matrix carrying mouse monoclonal anti-c-Myc IgG (clone 9E10; Covance). The matrix was washed four times with the wash buffer (50 mM Tris-HCl, pH 7.5, 150 mM NaCl, 0.1% Tween 20, protease inhibitors Complete). The bound proteins were eluted by treating the beads twice with 1 bead volume (50 l) of c-Myc peptide solution (Covance) (400 g/ml in 20 mM Tris-HCl, pH 7.5, 150 mM NaCl) for 15 min and used in Western blot experiments. Detection and identification of immunoprecipitated proteins were performed by Western blot and LC-MS/MS (liquid chromatography-tandem mass spectrometry). For LC-MS/MS analysis, proteins were revealed in silver-stained gels, according to standard protocol. Protein band cutting, trypsin-based in-gel protein digestions, and subsequent LCMS/MS procedures were performed in the Innovation Centre at Genome Quebec.
Primer Sequences-All primer sequences used in this study are shown in the supplemental Experimental Procedures.
Statistical Analysis-Statistical differences were analyzed using Student's t test for unpaired samples. Two-way analysis of variance test was used for multiple comparisons with one control group. In all cases, the criterion for significance (p value) was set as mentioned in the figures.

Bmi1 Is Required for Constitutive Heterochromatin Formation and Silencing in Mouse Cortical
Neurons-We performed transmission electron microscopy on cortical slices from WT and Bmi1 Ϫ/Ϫ mice at postnatal day 30 (P30). Notably, electrondense chromocenters were smaller and the nuclear envelope was generally irregular in Bmi1 Ϫ/Ϫ neurons (Fig. 1A). By immunohistochemistry on cortical sections and using antibodies against H3K9 me3 and H3K9 ac , a mark of open chromatin, we observed reduced H3K9 me3 labeling in Bmi1 Ϫ/Ϫ neurons together with increased H3K9 ac labeling (Fig. 1B). Immunoreactivity for HP1, KAP1, HDAC1, and ATRx was also reduced in Bmi1 Ϫ/Ϫ neurons, suggesting heterochromatin anomalies (Fig.  1B). Quantitative analysis revealed that the number of H3K9 me3 -positive chromocenters was reduced in Bmi1 Ϫ/Ϫ neurons, whereas the nuclear diameter of the neuron was increased (Fig. 1C). Because postnatal neurodegeneration may account for the observed chromatin anomalies, we analyzed cortical sections from WT and Bmi1 Ϫ/Ϫ embryos at e18.5. We found that H3K9 me3 and HP1 staining were reduced in Bmi1 Ϫ/Ϫ neurons, whereas that of H3K9 ac was unaffected, suggesting that histone hyperacetylation is secondary to defective heterochromatinization (Fig. 1D). Likewise, cultured cortical neurons from Bmi1 Ϫ/Ϫ embryos showed reduced H3K9 me3 labeling when compared with Bmi1 ϩ/Ϫ littermates (Fig. 1E). By immunofluorescence (IF) on P30 brain sections, we observed reduced immunolabeling for Lamin A/C at the neuron nuclei center of Bmi1 Ϫ/Ϫ when compared with WT, suggesting anomalies in the nuclear envelope (Fig. 1F). Deficiency in constitutive heterochromatin formation can affect repeat-DNA sequences expression. By qRT-PCR analyses, we found increased expression of Major pericentromeric repeats in Bmi1 Ϫ/Ϫ mouse cortices and of intergenic LINE elements and major and minor pericentromeric repeats in Bmi1 Ϫ/Ϫ cultured e18.5 neurons, when compared with WT ( Fig. 1, G and H). To test if Bmi1 was enriched at PCH in mouse neurons, we performed ChIP-qPCR FIGURE 1. Bmi1-deficient mouse cortical neurons present heterochromatin anomalies. A, transmission electron microscopy analysis of cortical neurons in P30 WT and Bmi1 Ϫ/Ϫ mice. Note the reduction in electron dense chromocenters and the anomalies in nuclear membrane architecture in Bmi1 Ϫ/Ϫ neurons (arrows). B, paraffin-embedded brain sections from P30 WT and Bmi1 Ϫ/Ϫ mice were analyzed by immunohistochemistry. Labeled cells are neurons located in the upper cortical layers of the cerebral cortex. Scale bar, 10 m. C, quantification of the total number of H3K9 me3 -positive chromocenter and number of large H3K9 me3 -positive chromocenters. Note that the nuclear diameter of the neuron is increased in Bmi1 Ϫ/Ϫ mice. Where n ϭ 3 brains for each genotype. *, p Ͻ 0.05, **, p Ͻ 0.01. D, paraffin-embedded brain sections from e18.5 WT and Bmi1 Ϫ/Ϫ embryos were analyzed as in B. E and F, cultured embryonic cortical neurons (E) and P30 cortical sections (F) were analyzed by immunofluorescence, revealing reduced H3K9 me3 and Lamin A/C labeling (arrows) in Bmi1 Ϫ/Ϫ neurons. Scale bars: 40 (E), 10 (F), and 5 m (FЈ). G, whole cortices or (H) e18.5 neurons from WT and Bmi1 Ϫ/Ϫ mice were analyzed by qPCR for satellite repeats and intergenic retroelements expression. P16 Ink4a was used as positive control. Note the up-regulation of minor and major satellite repeats in Bmi1 Ϫ/Ϫ neurons. Where n ϭ 3 independent samples for each genotype. *, p Ͻ 0.05; **, p Ͻ 0.01. on cultured WT and Bmi1 Ϫ/Ϫ e18.5 neurons after 7 days in vitro. We found that Bmi1 specifically accumulated at all repeat-DNA sequences and at the Bmi1-target gene Hoxa7 ( Fig. 2A). Notably, whereas enrichment for H3K27 me3 was observed at Hoxa7 in WT neurons (and slightly reduced in Bmi1 Ϫ/Ϫ neurons), H3K27 me3 enrichment at repeat-DNA was negligible in both conditions ( Fig. 2A). Bmi1 deficiency in mouse neurons also resulted in depletion of RING1B, HP1, H3K9 me3 , and H2A ub at repeat-DNA sequences, whereas accumulation of BRCA1 was unaffected or increased ( Fig. 2A). Bmi1 co-localization with H3K9 me3 in mouse cortical neurons was confirmed by IF on brain sections at P30 (Fig. 2B). Bmi1 antibody specificity was further validated by IF on cultured e18.5 cortical neurons (Fig. 2C). These results revealed that Bmi1 is required for heterochromatin formation and repeat-DNA silencing in mouse cortical neurons and enriched at PCH together with RING1B and BRCA1.
BMI1 Is Highly Enriched at Repetitive Sequences in Human Neural Precursors-To investigate BMI1 distribution on the chromatin genome-wide, we took advantage of publicly available BMI1 ChIP-Seq raw data on normal human neural precursors (53). Using MACS statistical peak calling, we identified 21,525 BMI1 binding sites. The majority of the peaks (56%) were located at intergenic regions, which are highly enriched for constitutive heterochromatin (Fig. 3A). As expected, we observed BMI1 enrichment at the canonical BMI1 targets CDKN2A (p16 INK4A ) and HOXC locus (Fig. 3E, top). Among 9,471 gene-associated BMI1 peaks, 714 peaks were also enriched by at least 2-fold for either H3K9 me3 or H3K27 me3 . Notably, 565 were marked for both H3K9 me3 and H3K27 me3 (Fig. 3B). We further annotated BMI1 peaks surrounding DNA repetitive sequences using a repeat masker. Surprisingly, 81% of the total pool of BMI1 was located at repetitive sequences ( Fig.  3C) with 985 peaks co-enriched with H3K9 me3 and 1,067 peaks co-enriched with H3K27 me3 (Fig. 3D). If considering repetitive sequences containing at least one BMI1 peak, LINE, SINE, and LTR were the most represented families of repeats. Among satellite repeats, BMI1 was mainly enriched at PCH regions (Fig.  3F, inset). A closer look at PCH regions on chromosome 9 revealed a "pocket-like" disposition of BMI1 peaks surrounding the H3K9 me3 deposition (Fig. 3E). A similar pattern of BMI1 peak distribution was also found at PCH regions in human chromosomes 1-10. Notably, repetitive sequences containing 3 or more BMI1 peaks were largely represented (80%) in centromeric satellite repeats (Fig. 3G). In contrast, although 20% of LINE L1 repeats contained 3 or more BMI1 peaks, no other LINE subfamilies were highly enriched for BMI1 (Fig. 3G). We concluded that in human neural precursors, BMI1 is enriched at constitutive and facultative heterochromatin with prevalence for repetitive sequences.

BMI1 Co-purifies with Architectural Heterochromatin
Proteins-To identify new BMI1 partner proteins, we infected 293T cells with a lentivirus expressing a Myc-tagged BMI1 fusion protein and GFP (EFv-BMI1 Myc /CMV-GFP) or a control virus only expressing GFP. After IP with an anti-Myc antibody, samples were separated on a one-dimensional gel and sequenced by LC-MS (Fig. 4A). We identified several unique peptides in BMI1 Myc samples corresponding to proteins involved in heterochromatin organization, including histone H1x, HP1a (also called CBX5), LAMIN A/C and LAMIN B, DEK (also called DEK1), and CENP-V (Fig. 4A) (7,11,54,55). We also identified 2 members of the ISWI family, BAZ1a (also called ACF1) and BAZ1b, which can promote heterochromatin formation and transcription silencing by generating spaced nucleosome arrays (56). To validate some of these findings, we performed IP experiments on control and BMI1 Myc virus-infected cells. As expected, we observed that RING1B, but not EZH2, co-precipitated with BMI1 (Fig. 4B). Co-precipitation of ATRx, KAP1, DEK1, and HP1 with BMI1 was also observed, with a notable enrichment of ATRx when compared with input ( Fig. 4B), and where the full-length (ϳ280 kDa) and truncated (ϳ180 kDa) ATRx isoforms were present, together with a lower molecular mass isoform of ϳ115 kDa. Notably, whereas co-precipitation with histones H3K9 me3 , H3 (total), H1, and H2A ub was robust, co-precipitation was not observed with histones H3K9 me2 , H3K27 me2 , and H3K27 me3 (Fig. 4B). We use FPLC to separate protein complexes and found that BMI1 was present in one fraction of very large molecular weight and in several other fractions of lower molecular weight all also containing HP1 and ATRx (Fig. 4C). By IF studies, we confirmed that BMI1 largely co-localized with H3K9 me3 in interphase nuclei (Fig. 4D).
BMI1 Is Required for Heterochromatin Compaction and Silencing-To evaluate BMI1 activity in heterochromatin silencing, we measured gene expression in loss-and gain-offunction experiments. Upon BMI1 deficiency, 293T cells showed reduced proliferation and underwent cell proliferation arrest after 3 passages (Fig. 5A). BMI1 overexpression had, however, no apparent adverse effect on cell proliferation (not shown). In BMI1 knockdown cells, expression of the canonical BMI1 target gene p16 Ink4a as well as that of McBox and SATIII was increased (Fig. 5B). Conversely, BMI1 overexpression resulted in transcriptional repression of p16 Ink4a and of all tested repeat-DNA sequences (Fig. 5C). By ChIP-qPCR experiments on shScramble and shBMI1-treated cells, we found that BMI1 and RING1B were highly enriched at repeat-DNA sequences and HOXC13 in control cells (Fig. 5D). In shBMI1 cells, BMI1, RING1B, HP1, H3K9 me3 , and H2A ub were reduced at all chromatin regions tested (Fig. 5D). In contrast, BRCA1 enrichment at repeat-DNA sequences was independent of BMI1 function. Increased BRCA1 enrichment was even observed at ALU sequences upon BMI1 deficiency (Fig. 5D). We tested if RING1B knockdown mimicked the BMI1-deficient phenotype. Although RING1B accumulation at constitutive heterochromatin and HOXC13 was highly reduced in siRING1B-treated cells (clone #4, 80% RING1B knockdown), H2A ub reduction was only detected at HOXC13 (data not shown). Notably, RING1B knockdown had no effect on BMI1, HP1, and H3K9 me3 on all tested regions. No significant effect on repeat-DNA sequence expression was observed (data not shown), revealing that RING1B knockdown is not sufficient to reproduce the BMI1-deficient heterochromatin phenotype. BMI1 enrichment at repeat-DNA sequences was also EZH2 and H3K27 me3 independent (data not shown). Nuclease hypersensitivity is a common phenotype of cells  525. B, proportion of gene-associated BMI1 peaks co-enriched for H3K9 me3 or H3K27 me3 . C, proportion of BMI1 peaks surrounding a repetitive sequence. D, proportion of repeat-associated BMI1 peaks co-enriched for H3K9 me3 or H3K27 me3 . E, example of canonical BMI1 target genes (CDKN2A and HOXC) and the pericentromeric region of human chromosome 9. Red arrowheads indicate BMI1 peaks. Top: physical map on the chromosome. F, families of repeat containing at least one BMI1 peak. G, families of repeat containing three or more BMI1 peak. BMI1 peaks determined by MACS peak calling; p value Ͻ0.05. At least 2-fold enrichment for H3K9 me3 and H3K27 me3 were considered. deficient in heterochromatin condensation (11,(57)(58)(59). We used native chromatin extracts isolated from control and shBMI1 293T cells in MNase and DNase I experiments and found that cells knocked down for BMI1 were hypersensitive to both nucleases (Fig. 5E), thus suggesting globally reduced chromatin compaction.

Severe Heterochromatin and Nuclear Envelope Alterations in
Human Cells Deficient for BMI1-To further characterize the BMI1-deficient phenotype, we analyzed cells by confocal IF using heterochromatin and nuclear envelope markers. We found severe depletion of the H3K9 me3 and H3K27 me3 histone marks in BMI1-knockdown cells together with dramatic elevation of the H3K9ac mark (Fig. 6, A and C). Likewise, DEK1 and HP1 heterochromatic nuclear foci were lost upon BMI1 knockdown and positive cells for LAMIN A/C were significantly reduced (Fig. 6, B and C), suggesting perturbation of the nuclear envelope architecture.
To test whether this correlated with alterations in the subnuclear distribution of heterochromatin proteins, we performed cellular fractionation experiments and where the SDSsoluble fraction is thought to be highly enriched for constitutive heterochromatin proteins (11,60). In control cells, BMI1 was detected in the 450 nM NaCl and SDS fractions (Fig. 6D). ATRx, HP1, BRCA1, and H3K9 me3 were also highly enriched in the 100 -450 nM NaCl-and SDS-soluble nuclear fractions. Modest HP1 distribution was also found in the nucleosol fraction (Fig.  6D). In shBMI1 cells, which underwent premature cell proliferation arrest (thus explaining the overall reduced total protein loading), ATRx, HP1, and H3K9 me3 were highly reduced in the SDS fraction and displaced in the other fractions. In contrast, the distribution of BRCA1 in chromatin fractions and its overall expression were unaffected upon BMI1 knockdown (Fig. 6D and data not shown). To test BMI1 function in primary human cells, human dermal fibroblasts were infected with the lentiviruses and analyzed by IF. We observed that in contrast to control cells where robust H3K9 me3 labeling was widespread throughout the interphase nucleus, H3K9 me3 labeling in BMI1 knockdown cells was highly reduced and present at the nuclear periphery where it did not co-localize with DAPI (Fig. 6E). Notably, co-localization of H2A ub and H3K9 me3 with DAPI in shBMI1-infected cells could be rescued by an RNAi-resistant BMI1-Myc fusion protein (BMI1 mycϪR ), thus excluding possible off-target effects (Fig. 6F). Bubbling of the nuclear envelope and loss of DEK1 nuclear labeling were also observed (Fig. 6E), revealing BMI1 requirement for constitutive heterochromatin maintenance in both transformed and primary human cells.
BMI1 and BRCA1 Display Partial Functional Redundancy in Heterochromatin Compaction-Because BMI1/RING1A/ RING1B and BRCA1/BARD1 display H2A monoubiquitinylation activities and that BRCA1 enrichment and distribution at heterochromatin is not affected upon BMI1 depletion, we tested whether BMI1 and BRCA1 displayed functional redundancy. For this, we first inactivated BRCA1 to test the impact on BMI1 localization. In control cells, both BRCA1 and BMI1 were enriched at repeat-DNA sequences (data not shown). In BRCA1 knockdown cells, BRCA1, HP1, H2A ub , and H3K9 me3 levels were reduced and transcription of repeat-DNA sequences was increased (data not shown), altogether confirming previous findings (22). Notably, however, BMI1 and RING1B were enriched at all tested regions upon BRCA1 knockdown (data not shown). Next, stably infected shBMI1 cells were transfected with an shBRCA1 plasmid, generating double knockdown (DKN) cells. Although H3K9 me3 enrichment was reduced by 55-70% in DKN cells at all tested chromatin regions, enrichment for HP1 and H2A ub was further reduced by 80 -90% (Fig. 7A), suggesting additive effects in DKN cells when compared with single BRCA1 or BMI1 knockdown cells. To test whether BMI1 could compensate for BRCA1 deficiency, we overexpressed the BMI1 Myc construct. In control cells, BMI1 overexpression could not displace endogenous BRCA1 localization on the chromatin (Fig. 7B). However, chromatin accumulation of both endogenous and ectopic BMI1 proteins was highly increased at all tested regions following BRCA1 knockdown (Fig. 7B). A similar but less dramatic trend was also observed for RING1B. Most notably, whereas BMI1 overexpression could increase H2A ub and H3K9 me3 deposition as well as HP1 accumulation at all tested regions in control cells, it could also rescue the corresponding heterochromatin anomalies in shBRCA1 cells (Fig. 7B). BMI1 overexpression in shBRCA1 cells also resulted in normalization of ALU, McBox, Sata, and SatIII expression (Fig. 7C), altogether suggesting functional redundancy in constitutive heterochromatin compaction and silencing between BMI1 and BRCA1.

Discussion
We showed here that Bmi1-deficient mouse neurons and BMI1 knockdown human cells displayed severe anomalies at the constitutive heterochromatin. These anomalies were accompanied by transcriptional activation of repeat-DNA sequences and correlated with robust accumulation of BMI1 at constitutive heterochromatin. Genome-wide analysis of BMI1 distribution on the chromatin further revealed predominant enrichment at repetitive DNA sequences. BMI1 co-purified with architectural heterochromatin proteins, co-localized with H3K9 me3 , and was required for HP1, DEK1, and ATRx localization at constitutive heterochromatin. In contrast, BRCA1 localization was BMI1-independent, and both proteins displayed partial functional redundancy for H2A ub deposition, heterochromatin formation, and silencing.
The recruitment mechanisms of PcG proteins are complex and not fully understood. In mouse ES cells and oocytes, evidences suggest that recruitment of PcG proteins at PCH is prevented by a high level of DNA methylation (30,52). Likewise, observations of Polycomb bodies at PCH in transformed human cells are thought to occur following loss of DNA methylation (30,47,52). Based on this, it was proposed that PcG proteins are excluded from PCH in normal somatic cells (52).
Using several methods, we demonstrated BMI1 enrichment at PCH and other repetitive elements in mouse neurons, human neural precursors, and immortalized human cells. Our cell fractionation assays further demonstrated that about 50% of the BMI1 pool was bound to the SDS-soluble chromatin fraction, which is enriched for constitutive heterochromatin. Interestingly, we also observed that: 1) BMI1 did not co-purify with EZH2 or H3K27 me3 ; 2) EZH2 and H3K27 me3 were not enriched at constitutive heterochromatin; and 3) BMI1 accumulation at constitutive heterochromatin and HOXC13 was EZH2 independent. Although apparently surprising, these results are in agreement with numerous findings showing that PRC1 recruitment can be PRC2 independent, or that the PRC1 can work upstream of the PRC2 (29,61,62). It is also notable that RING1B knockdown could not mimic the BMI1-deficient phenotype. More specifically, whereas H2A ub levels were reduced at HOXC13 in RING1B knockdown cells, this was not accompanied by a corresponding reduction in HP1 and H3K9 me3 levels, as observed in BMI1 knockdown cells. Furthermore, there was no apparent effect on H2A ub levels at heterochromatin. This finding leaves us open with many explanations, one being functional compensation by RING1A for H2A ub deposition, as shown in other context (50,(63)(64)(65). It is also possible that in addition to promote H2A ub at silenced developmental genes and heterochromatin, BMI1 stimulates chromatin compaction and H3K9 me3 loading through interactions and activities not shared by RING1A or RING1B (66).
Based on our findings and previously published work, we propose that PRC1 recruitment to PCH is highly dynamic and developmentally regulated (Fig. 7D). The highly variable DNA methylation states between ES cells/oocytes and progenitor cells/somatic cells, in combination with the distinct histone tail modifications and chromatin compactions levels, would explain the re-localization of PRC1 components to PCH in progenitor and somatic cells (67)(68)(69). Consistently, BMI1 expression levels are extremely low in human ES cells when compared with human neural progenitors and post-mitotic neurons. 5 Although PRC1 proteins are excluded from PCH in ES cells and oocytes, they would start to accumulate at PCH during midembryonic development coincidently with H3K9 me3 deposition and progressive de novo DNA methylation (69). This is supported by the ChIP-Seq data showing that BMI1 is moderately enriched at PCH when compared with HOX and p16 INK4A canonical sites in human neural progenitors (Fig. 3). Interest-ingly, both H3K9 me3 and H3K27 me3 marks were also present at PCH but did not clearly overlap (Fig. 3), similarly as reported in DNA methylation-deficient mouse ES cells (30). The histone modification pattern of human neural progenitors at PCH is also distinct from that of mouse neurons where H3K27 me3 is excluded (Fig. 2). This could be best explained by the robust accumulation of H3K9 me3 at PCH in post-mitotic neurons, because H3K9 me3 can prevent PRC2, but not PRC1 recruitment (30). Indeed, Bmi1 is highly enriched in mouse neurons at PCH when compared with canonical sites (Fig. 2). Taken together, these observations suggest a dynamic and developmentally regulated model of PcG occupancy at PCH (Fig. 7D). Because constitutive heterochromatin is intrinsically instable, we further propose that in somatic cells, the main biological function of BMI1 is to stabilize the repetitive genome by promoting chromatin compaction and silencing.
We observed that the Bmi1-null neuronal phenotype was associated with increased nuclear diameter and an irregular nuclear envelope. Human cells knockdown for BMI1 also presented anomalies in nuclear envelope architecture (Fig. 6, B-E). These anomalies are particularly interesting considering that loss of heterochromatin foci can result in disruption of the nuclear lamina (7). Perturbations of the nuclear envelope architecture is also a prominent feature of Hutchinson-Gilford Progeria cells carrying mutations in LAMIN-A and normal aging human cells (70). In most eukaryotes, constitutive heterochromatin perturbations result in genomic instability and premature aging or reduced lifespan (10,16,(71)(72)(73)(74). It is thus notable that Bmi1-deficient mice show reduced lifespan, genomic instability, neurodegeneration, and progeria features (38,(42)(43)(44)(45)75). Similar anomalies were also reported for ATRxdeficient mice (18,20). Taken together, this raises the possibility that BMI1 requirement for constitutive heterochromatin formation and silencing could underlie the premature aging/ senescence and genomic instability phenotypes observed in Bmi1-null mice and cells.
The BRCA1/BARD1 complex is required for heterochromatin formation and silencing through monoubiquitinylation of H2A at PCH, and the genomic instability phenotype of BRCA1 Ϫ/Ϫ cells could be rescued by overexpression of a histone H2A protein fused to an ubiquitin moiety in C terminus (22). How BMI1 or BRCA1-mediated H2A ub deposition at repetitive DNA sequences can translate into H3K9 me3 loading and heterochromatin spreading is unknown. One possibility is that H2A ub induces allosteric changes in the histone H3 lysine trimethyltransferases SUV39H1/2 to promote their activity, such as proposed for H2B ub and H3K4 methylation (76,77).  . BRCA1 and BMI1 display redundant activities in constitutive heterochromatin formation and silencing. A, 293T cells were infected with shScramble or shBMI1 viruses. After selection with hygromycin, cells were transfected or not with an shBRCA1-encoding plasmid and analyzed by ChIP. Note the severe reduction for HP1, H3K9 me3 , and H2A ub at human Satellite repeats in shBMI1/shBRCA1 cells. All data where normalized to shScramble (black horizontal bars). B, 293T cells stably expressing BMI1 Myc or not were transfected with shScramble or shBRCA1 plasmids and analyzed by ChIP (B) and qPCR (C). B, endogenous and exogenous BMI1 was enriched in shBRCA1-treated cells at all tested loci. BMI1 overexpression also rescued HP1, H3K9 me3 , and H2A ub depletion in BRCA1 knockdown cells at human Satellite repeats. All data were normalized to shScramble (black horizontal bars). C, BMI1 overexpression rescues repeat-DNA sequences expression in BRCA1 knockdown cells; *, p Յ 0.05; **, Յ0.01. D, model depicting the dynamic and developmentally regulated PRC1 (BMI1) occupancy at PCH. In ES cells and oocytes, BMI1 is expressed at low levels and binding to PCH is prevented by DNA methylation. At pre-implantation stages, DNA methylation is erased coincidently with low levels of BMI1. During development, high BMI1 levels in progenitor cells stimulate PCR1-mediated H2A ub (1), which promotes H3K9 me3 deposition (2), and propagation (3). In turn, H3K9 me3 prevents PRC2-mediated H3K27 me3 activity on nucleosomal histones (4). PRC2 occupancy at PCH prior to PRC1 would allow deposition of the H3K27 me3 mark (right) and the bivalent histone signature. In mature somatic cells, the PRC1 and BRCA1/BARD complexes are highly enriched at PCH, leading to heterochromatin compaction and silencing. DNA methylation at PCH may occur after spreading the H3K9 me3 mark.
This would be consistent with previous observations that PRC1 components can interact with SUV39H1 (48). Alternatively, BMI1 may directly or indirectly regulate the transcription of H3 lysine methyltransferases or demethylases, thus operating in trans. We showed here that co-inactivation of BMI1 and BRCA1 induces more severe heterochromatin anomalies than individual BMI1 or BRCA1 deficiencies, and that BMI1 overexpression could rescue the BRCA1-deficient heterochromatin phenotype. These results suggest that BMI1 and BRCA1 are at least partially redundant for H2A ub deposition at constitutive heterochromatin, although the BRCA1/BARD1 complex targets H2A at lysines 127-129 (21). This would indicate that the commonly used anti-H2A ub antibody recognizes both the H2AK119 ub and H2AK127-129 ub motifs. The observation that BMI1 and BRCA1 proteins accumulation is mutually independent and that BMI1 levels are increased in BRCA1-deficient cells (and reciprocally) also suggests that both protein complexes possibly bind to very close substrates to catalyze H2A monoubiquitinylation. Structural analyses have indeed revealed a high degree of conservation between the nucleosome-binding loop of BRCA1 and the corresponding domain of RING1B (78). Notably, the reduced neuronal chromocenter number and size phenotype observed in Bmi1-null neurons is about identical to that reported for mouse cortical neurons conditionally deficient for BRCA1, thus further supporting our findings (22). Taken together, this suggests that although BMI1 and BRCA1 protein complexes target distinct lysine residues on histone H2A, the resulting biological effects on heterochromatin compaction and silencing are highly similar.
In conclusion, we demonstrated for the first time that BMI1 is highly enriched at intergenic repetitive elements and PCH of the mouse and human genomes in normal somatic cells, and required for constitutive heterochromatin formation and silencing. Because BM1 is also present at PCH in cancer cell lines and that several cancer cells were shown to be sensitive to BMI1 inhibition, this raises the possibility that BMI1 may be important to stabilize the transformed heterochromatic genome. BMI1 function at constitutive heterochromatin may be even more critical in BRCA1-deficient tumors, thus opening possibilities for the development of synthetic lethal strategies.