Chromatin-association of the Polycomb group protein BMI1 is cell cycle-regulated and correlates with its phosphorylation status

Despite the recognition that Polycomb-group (Pc-G) proteins play an important role in sustaining developmentally established expression boundaries of homeotic genes in Drosophila and Hox gene clusters in mammals, relatively little is known about the molecular principles of this maintenance function. In Drosophila, Pc-G dependent silencing is mediated via polycomb responsive elements (PREs) (Zink and Paro, 1989; Rastelli et al., 1993; Simon et al., 1993; Chan et al., 1994; Chiang et al., 1995; Müller, 1995; Strutt and Paro, 1997). Several lines of evidence predict an extensive interaction between known Pc-G members in both flies and mammals. Their colocalization on polytene chromosomes and in interphase nuclei of cultured insect cells as well as mammalian cells provides a visual indication of physical association (Franke et al., 1995; Alkema et al., 1997a; Gunster et al., 1997; Buchenau et al., 1998). Furthermore, copurification and coimmunoprecipitation of several Pc-G protein classes yields direct evidence for their biochemical interaction (Franke et al., 1992; Alkema et al., 1997a; Gunster et al., 1997; Kyba and Brock, 1998). Studies in Drosophila have shown that, when mutated, the Pc gene product seems to dislodge Pc-G complexes in embryonal cells; this correlates with a severely

disturbed larval development (Franke et al., 1995). Taken together, these data have prompted the idea that multiple binding sites with varying binding affinities for Pc-G members may assist in recruitment of multiple factors into larger nucleation sites for Pc-G repression (Zink and Paro, 1995; Pirotta, 1997, 1998; Strutt and Paro, 1997; Van Lohuizen, 1998). How Pc-G proteins recognize and interact with target genes is still largely unclear. Most Pc-G proteins fail to recognize specific DNA sequences, although recently the Drosophila Pc-G gene pleiohomeotic was found to encode a protein with homology to a mammalian transcriptional regulator with sequence-specific DNA binding properties (Brown et al., 1998). Moreover, a conserved sequence motif was reported to occur in Polycomb-responsive elements (Mihaly et al., 1998) suggesting that at least some Pc-G group members may be in direct DNA contact.

Polycomb function has been conserved in mammals. The Bmi1 proto-oncogene was the first functional mammalian Pc-G identified (Van Lohuizen et al., 1991b; Brunk et al., 1991). Subsequent gain- and loss-of-function mutational analysis in the mouse showed that morphological transformations of the vertebra along the antero-posterior axis of the skeleton are associated with Hox gene expression boundary shifts (Van der Lugt et al., 1994, 1996; Alkema et al., 1995). Several Pc-G gene null-mutant studies in mice have revealed comparable, yet not fully identical, morphological transformations, accompanied by expression boundary shifts of distinct Hox genes, again suggesting the existence of multiple complexes with diverse target gene specificity (reviewed in van Lohuizen, 1998). Pc-G function is likely to be more complex in mammals than in insects. This increased complexity is partly due to the multiplication of Hox gene clusters in mammals as well as the duplication of many Drosophila Pc-G homologues (reviewed by Gould, 1997; van Lohuizen, 1998). To date, no cis-acting polycomb responsive elements have been identified in mammalian DNA.

Despite increasing knowledge on processes controlled by Pc-G proteins, relatively little is known about regulation of Pc-G protein complexes themselves. In human cell lines, a non-uniform Pc-G protein distribution is visible in interphase nuclei, which disappears at mitosis (Alkema et al., 1997a; Gunster et al., 1997). In the present study, we examined the subnuclear distribution of BMI1 in primary human and tumor cells in relation to cell cycle progression. We demonstrate that chromatin-association of Pc-G proteins in primary and tumor cell lines is subject to a strict cell cycle-dependent regulation. In addition, chromatin association of BMI1 throughout the cell cycle inversely correlates with its phosphorylation status, indicating a pronounced cell cycle-dependent post-translational modification. By employing more refined immunohistological techniques, we show that detection of the endogenous BMI1 protein is possible, also on mitotic metaphase chromosomes, albeit at significantly reduced levels. The implications of our findings with respect to recent reports on the biological significance of Pc-G chromatin association by others and to mechanistic models of Pc-G function in mammalian cells are discussed.

A (myc)3H6-epitope-tagged expression vector was generated from pKW2T (a derivative of pRK7; Genentech). The resulting CMV (cytomegalo virus)-(myc)3H6[NotI] vector contains a unique NotI cloning site immediately following the amino-terminal tag. A PCR-amplified DNA fragment was inserted in-frame at the NotI site encoding the human BMI1 protein (Alkema et al., 1993). The CMV enhancer/promoter directs BMI1 overexpression in mammalian cells (Aagaard et al., 1999). An LZRS-Bmi1-IRES-EGFP viral vector (Jacobs et al., 1999) was used to obtain high polyoma-tagged (2PY) BMI1 expression in primary human cells.

HeLa, U2-OS and MCF-7 cells were grown at 37°C, 5% CO₂in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal calf serum (FCS). Primary amniotic cell cultures and primary peripheral white blood (PWB) cells were obtained at the University Hospital of Vienna (Austria). Confluent low passage number diploid primary human TIG3 fibroblast cultures (Fukami et al., 1995) were split 1:4 (one passage equals two population doublings, PDL). PWB cells were cultured in PB-max medium (Gibco) for 72 hours. Cells were either grown on multiwell slides (Alkema et al., 1997a) or spun onto glass slides in a Cytospin (Shandon). For transient expression, HeLa cells (1×10⁵) were transfected with 3 mg of the CMV-driven (myc)3-Bmi1 plasmid using Lipofectase (Gibco); transfection efficiencies varied between 2-15% (not shown). 48 hours after transfection, cells were processed for indirect immunofluorescence. Isolation of recombinant Bmi1 and control virus and viral infection of primary human TIG3 cells were carried out essentially as described (Jacobs et al., 1999). Infection efficiency was estimated at approximately 100% by GFP-positivity of cells. To attain synchronization, primary and tumor cell cultures were arrested in early G₁using lovastatin (40 nM, 48 hours; Keyomarsi et al., 1991), or accumulated in G₁either by serum starvation or contact inhibition. S-phase enrichment was achieved by double thymidine block (2×2 mM; 12-16 hours); G₂/M specific cell suspensions were obtained by treatment with either nocodazole (50 ng/ml; 16 hours) or colcemid (0.02-0.05 mg/ml; 16 hours) followed by mitotic ‘shake-off’.

Cell suspensions from a spontaneous BMI1-overexpressing tumor (derived from Bmi1-transgenic mouse) were starved for 1.5 hours, and subsequently grown for 3.5 hours in minimal medium supplemented with 10% dialyzed FCS, in the presence of 10 mCi ³²P-orthophosphate. Cell extracts were made in ELB buffer (Alkema et al., 1997a). Total lysates were incubated with preformed protein-sepharose/antiserum complexes for 1 hour at 4°C. Other cell extracts for western analysis were made in RIPA buffer (Alkema et al., 1997a). Separation of soluble and chromatin-bound nuclear fractions was essentially carried out as previously described (Muchardt et al., 1996).

Indirect immunofluorescence detection of the endogenous BMI1 protein in HeLa cells, and of (myc)3-tagged BMI1 protein in transiently transfected HeLa cells, respectively, essentially followed a cytospin procedure that is outlined elsewhere (Earnshaw et al., 1984; Haaf et al., 1990) with some minor modifications. Cells were incubated for 3 hours with colcemid (0.1 mg/ml; Gibco), harvested by trypsinization or by ‘shake off’ and hypotonically swollen for 30 minutes at 37°C in RBS buffer (10 mM Tris, pH 7.4, 10 mM NaCl, 5 mM MgCl²). Cytospin-spread nuclei and chromosomes were fixed in 2% formaldehyde in PBS, washed, permeabilized with 0.5% NP-40, blocked in culture medium containing 10% FCS for 30 minutes at room temperature, after followed by incubation with the respective primary and secondary antibodies. Alternatively, cells grown on multiwell coverslips were formaldehyde-fixed as described before (Alkema et al., 1997a). Primary TIG3 cells harboring LZRS/IRES-EGFP viral integrations were grown on 0.1% gelatin coated multi-well slides and extracted either in 0.5% Triton-X in KCM buffer (Jeppesen et al., 1992) and fixed in formaldehyde or directly in 100% methanol at -20·C for 3-5 minutes. Antisera used are indicated in the figure legends. Immunofluorescently labeled mitotic cells were counterstained with the DNA dye 4′-6-diamidino-2-phenylindole (DAPI) alone or in combination with distamycin A (Schweizer et al., 1978; Schweizer and Ambros, 1994). Preparations were mounted in Vectashield AntiFade (Vector Laboratories).

To enhance detection of endogenous BMI1 protein on M-phase chromatin, a reversed fixation/staining technique was employed (i.e. ‘post-fixation’; Jeppesen, 1994; Wreggett et al., 1994). A metaphase-enriched cell fraction was harvested by mitotic ‘shake-off’ (modified after Jeppesen et al., 1992). Cells were hypotonically swollen for 30 minutes at room temperature in 75 mM KCl, spun onto microscope slides and immediately immersed in KCM-buffer (120 mM KCl, 20 mM MgCl₂, 10 mM, Tris pH 8.0, 0.5 mM EDTA, 0.1% Triton X-100). For interphase and mitotic chromosome immunofluorescence, preparations were blocked for 30 minutes with 10% FCS in KCM, and sequentially incubated with the appropriate primary and secondary antisera (see above). Chromosomes were then fixed in 4% formaldehyde in KCM for 10 minutes, counterstained with DAPI, or with DAPI plus distamycin A, and mounted in Vectashield AntiFade.

Microscopic analysis and documentation was done as described below.

Following immunofluorescence microscopic analysis, coverslips were removed by floating-off in PBS at ambient temperature. Preparations were again fixed in buffered paraformaldehyde (4%) for 15 minutes (Schweizer and Ambros, 1994). Paracentromeric constitutive heterochromatin on human chromosome 1 (1qh) was visualized using a previously described FISH procedure (Strehl and Ambros, 1993). A human chromosome 1-specific subregional probe pUC1.77 (Cooke and Hindley, 1979) was used to probe for region 1q12. The DNA probe was labeled by nick-translation either with biotin-11-dUTP (Sigma) or with digoxigenin-11-dUTP (Boehringer, Mannheim). For detection of biotin-labeled probes a mouse anti-biotin antibody in conjunction with a secondary rabbit anti-mouse tetramethyl-rodamine isothiocyanate (TRITC)-conjugated antibody (1:20 and 1:30, respectively; DAKO, Glostrup, Denmark) were used. Detection of digoxigenin-labeled probes was achieved using a combination of sheep anti-digoxigenin (1:100; Dakopatts no. R270; or Sigma, St Louis, Missouri) and FITC-conjugated rabbit anti-sheep antibodies (1:100; DAKO, Glostrup, Denmark). Following FISH, the preparations were fixed with 4% paraformaldehyde, counterstained with DAPI and mounted in Vectashield AntiFade solution.

Chromosome analyses and karyotyping of cultured cells were done following conventional methods for mitotic stimulation, metaphase accumulation and for metaphase chromosome spreading upon methanol/acetic acid (3:1) fixation. Fluorescent chromosome R-banding and C-banding of chromosome spreads was achieved using the chromomycin/distamycin-A/DAPI tristaining technique (Schweizer and Ambros, 1994). Chromosome 1-identification by banding techniques was confirmed by sequential application of FISH using the human chromosome 1-specific satellite probe pUC1.77 (Cooke and Hindley, 1979).

Fluorescence microscopical examination was performed on a Zeiss Axioplan or Axioskop. Pictures were taken both on black and white and on color films. Alternatively, pictures were digitalized by means of a cooled CCD camera (Photometrics). For this purpose, images were acquired separately for each fluorochrome using IPLab software, pseudocolored (Gene Join) and, when required, processed and merged using Adobe Photoshop 4.0 software.

The nuclear localization of the BMI1 protein was studied in detail in three tumor cell lines. The BMI1 protein is exclusively seen in nuclei; the protein is excluded from the nucleolus. In HeLa cervical carcinoma cells, in addition to a fine-grained pattern, the majority of BMI1-positive nuclei exhibits three (54%) or four signals (36%; Fig. 1A). These subnuclear domains are associated with distamycin-DAPI-positive heterochromatin, marking chromosomes 1, 9 and 16, and the short arm of chromosome 15. Only a fraction of all DA-DAPI positive chromocenters are BMI1-positive (Fig. 1A). Conversely, however, all bright BMI1 fluorescent signals are associated with DA-DAPI positive heterochromatin. Consecutive immuno-FISH using satellite probe pUC1.77 reveals a FISH labeling pattern fully congruent with the BMI1 staining pattern (Fig. 1B), proving BMI1 protein concentration at heterochromatic chromosome region 1q12 (1qh). Similar results were obtained with U2-OS osteosarcoma cells and MCF-7 breast epithelial carcinoma cells. FISH data in HeLa and U2-OS cells were confirmed by conventional cytological analysis (not shown). Since the three aneuploid cell lines exhibit significant differences in dosage of chromosome region 1q12, it is concluded that this variation is a reflection of differences between cell lines in karyotype rather than expression levels (see Discussion). The above data are in agreement with a recently published report (Saurin et al., 1998).

Close to mitosis, the intensity of the nuclear BMI1 signal decreases significantly (Alkema et al., 1997a). To establish in detail at what cell cycle phase Pc-G-chromatin dissociation occurs, synchronized and asynchronously growing U2-OS cells were simultaneously immunostained for BMI1 and either cell-cycle marker PCNA or phosphorylated histone H3. The PCNA staining pattern in nuclei going through S phase displays characteristic alterations (Celis and Celis, 1985; Celis et al., 1986). Likewise, phosphorylation of histone H3 can be used as an immunocytochemical marker for late S, G₂/M phase (Hendzel et al., 1997). In G₀/ G₁, nuclear PCNA staining is virtually absent in U2-OS nuclei; at these stages, Pc-G-paracentromeric heterochromatin association is maximally visible (Fig. 2A). As cells move into S phase, the PCNA signal is initially detectable as a granular pattern, but subsequently redistributes into a punctate pattern, revealing foci close to the nuclear membrane. At late S phase histone H3 phosphorylation is initiated at clearly visible foci (Hendzel et al., 1997). Interestingly, at this transition, detection of 1qh-Pc-G association becomes increasingly difficult in U2-OS nuclei: in 32.8% of nuclei phosphorylation of histone H3 has been initiated, while approximately the same number of nuclei (33.8%) displays a clearly diminished or absent Pc-G-heterochromatin association (Fig. 2A, asterisks, B). When histone H3 phosphorylation has spread over the entire genome and the PCNA signals have dissipated, the Pc-G signal is virtually undetectable and remains low throughout G₂/M, but is re-established during the ensuing G₁phase. The latter is demonstrated by the re-occurrence of Pc-G domains in early G₁-arrested cells (Fig. 2A, bottom, C). These findings suggest a remarkable cell cycle-dependent regulation of Pc-G-paracentromeric heterochromatin association.

Absence of BMI1 staining at mitosis could, besides dissociation from chromatin, be due to cell cycle-dependent inaccessibility of the protein for the antibody, or to damage or loss of epitopes during the preparation procedure. We therefore applied a more sensitive immunocytological technique aimed at detection of less abundant endogenous proteins, in which chromatin immuno-staining precedes fixation (Jeppesen, 1994; Wreggett et al., 1994). In control experiments, antibodies against phosphorylated histone H3 (see Fig. 2B) or a CREST scleroderma patient serum (not shown) in conjunction with a BMI1 specific antiserum was used. Clearly detectable staining with these antisera proves accessibility of the respective antigens. Using the more sensitive ‘post-fixation’ technique a fine-grain BMI1 protein staining was visible over metaphases, albeit significantly weaker compared to interphase nuclei (Fig. 3A). Importantly, BMI1-concentration at 1q12 on metaphase chromosomes was only seen when ‘post-fixation’ was employed. As depicted in Fig. 2A, all FISH signals colocalized with BMI1 signals and, vice versa, all BMI1 signals merged perfectly with a FISH signal indicating association with chromosome 1 paracentromeric heterochromatin. The mitotic chromatin staining and colocalization were confirmed in U2-OS (Fig. 3B) and MCF-7 cells (not shown).

To substantiate the concentration of endogenous BMI1 at 1qh, we ectopically expressed epitope-tagged BMI1 in tumor cell lines. This approach serves as an immunocytological control and allows an assessment of the influence of BMI1 expression levels on staining patterns. Besides a fine-grain overall chromatin labeling, all antigen-positive interphase nuclei of HeLa cells expressing myc-tagged BMI1 exhibit distinct bright signals (on average, four). These results demonstrate that the staining pattern in cells ectopically expressing tagged BMI1 protein (1-2 orders of magnitude above the level of the endogenous protein; L. A., unpublished) is qualitatively similar to that of the endogenous protein. The findings are also in line with our previous observations that haemagglutinin (HA)-tagged BMI1 perfectly colocalizes with endogenous proteins in transfected U2-OS cells (Alkema et al., 1997a). In addition, analysis of antigen-positive ‘post-fixed’ metaphase chromosomes reveal an immunostaining pattern similar to that of endogenous BMI1: a pronounced enrichment of myc-BMI1 within the paracentromeric heterochromatic region in the long arm of human chromosome 1 next to a weaker distribution along the chromosome arms (Fig. 4A,B). Consecutive immuno-FISH results on metaphase and interphase chromatin were fully congruent (Fig. 4Band not shown). These data confirm the specific concentration of BMI1 at 1qh during interphase and, at a lower intensity, at mitosis. In addition, this demonstrates that ectopic expression of the BMI1 protein does not alter its overall subnuclear distribution pattern.

Interphase and metaphase chromatin of human primary cells was examined immunohistologically for BMI1 distribution. Interphase nuclei of primary peripheral white blood cells show a fine-grained pattern, qualitatively comparable to that of long-term cultured cell lines, although considerably weaker in intensity (Fig. 5A). Metaphase spreads of these cells did not reveal BMI1 positive staining, although counter-staining with antiserum against phosphorylated histone H3 was positive (not shown). Similar findings were made in primary amniotic cells and in an EBV-immortalized normal B-cell derived cell line (Fig. 5A; not shown). Again a weak, fine-grained pattern was visible in these cells. Finally, diploid, primary human TIG3 fibroblasts show the same fine-grained staining pattern: the large nuclear domains as seen in tumor cells are absent (Fig. 5A). The low intensity of BMI1 staining most likely relates to low levels of gene expression in vivo. Indeed, infection of primary TIG3 cells at low PDL with retroviral expression vectors carrying the Bmi1 gene, gave rise to higher, although varying, levels of BMI1 expression throughout the culture. However, in synchronized cultures only a fine-grained staining pattern was observed; none of the low passage primary cells showed the nuclear Pc-G domains characteristic of tumor cell lines (Fig. 5A). The nuclear staining pattern of BMI1 was confirmed with two independent antisera (anti-BMI1 and anti-2PY epitope tag; not shown). Lack of BMI1 domains in nuclei of primary cells suggests that preferential BMI1 staining at human chromosome region 1q12 is tumor-specific or alternatively, might be associated with changes in local chromatin structure as a result of long term establishment in culture. To test the latter possibility, synchronized TIG3 cells expressing high levels of BMI1 (TIG3/bmi) were examined at low (<30) and high PDL (>60). Remarkably, a low percentage (5-10%) of high passage number TIG3/bmi cells show clearly visible nuclear domains, while none were found in low passage number TIG3/bmi; this finding was confirmed with unrelated antisera (Fig. 5B). However, consecutive immuno-FISH analysis indicated that these nuclear sites were not paracentromeric a-satellite repeats at position 1q12 (data not shown). The above data argue that the nuclear distribution of BMI1 in primary human cells is fine-grained and that formation of distinctive Pc-G domains is secondary to prolonged culturing.

We next examined BMI1-chromatin interaction in primary cells in relation to cell cycle progression. As described above, primary TIG3 cultures were counterstained at different cell cycle stages for PCNA or phosphorylated histone H3. Primary TIG3 cells are subject to replicative senescence at higher PDL (Fukami et al., 1995). We have previously shown that overexpression of BMI1 yields a proliferative advantage to primary mouse and human fibroblasts and delays the Hayflick limit in TIG3 cells (Jacobs et al., 1999). As a result, at high passage, more than 80% of cells in G₀/ G₁display a strong nuclear staining for BMI1 versus 40-50% of TIG3/bmi cells at low passage. This enabled us to trace the global nuclear BMI1 staining intensity throughout the cell cycle in primary cells. As shown in Fig. 6A, the BMI1 signal drops significantly at late S phase, when the PCNA pattern has become located in distinct foci close to the nuclear membrane and subsequently begins to break down. BMI1-chromatin association is virtually undetectable by standard methods throughout G₂/M, at which time phosphorylation of histone H3 has become widespread (Fig. 6Band not shown). These data indicate that, like in tumor cells, in primary human cells the biochemical interaction between Pc-G proteins and chromatin is also regulated in a cell cycle-dependent fashion.

Many cellular proteins undergo cell cycle-related post-translational modifications, which in turn can affect molecular interactions. Interestingly, cell cycle-specific chromatin binding was also reported for trx-G proteins (Muchardt et al., 1996). We asked whether BMI1 is subject to cell cycle-regulated post-translational modification and whether this correlates with chromatin-association. Endogenous BMI1 protein is readily detectable by western analysis in many tumor-derived cell lines and is represented in multiple bands (Fig. 7A; see also Alkema et al., 1997a). Comparison of cellular extracts at distinct phases of the cell cycle reveals a difference in migration speed of BMI1: the majority of BMI1 appears as faster migrating bands during G₁and early S phases (Fig. 7A). At mitosis, when paracentromeric heterochromatin association of BMI1 in tumor cells is microscopically undetectable, the slow migrating from of BMI1 is most abundant. In U2-OS extracts, the polyclonal (pAb) bmi38 and the monoclonal (mAb) bmiF6 antisera display a slight preference for the differentially migrating forms of BMI1, which complicates assessment of protein levels throughout the cell cycle. We therefore also studied epitope-tagged BMI1 in primary cells at low PDL. When expressed in primary TIG3 cells, PY-tagged BMI1 shows similar cell cycle-dependent migration differences as those in U2-OS cells (Fig. 7C). Total cellular BMI1, detected with an mAb, a pAb or the PY-tag-specific antibody, shows a similar abundance of slower migrating BMI1 in M-phase arrested cells. Importantly, this demonstrates that ectopically expressed BMI1 protein is also subject to cell cycle-dependent regulation. Furthermore, these data suggest that BMI1 is not subject to massive degradation during mitosis (Fig. 7C,E). Direct immunoprecipitation following orthophosphate labeling identifies BMI1 as a phospho-protein in vivo (Fig. 7D). Treatment of the M-phase specific form of BMI1 with calf intestinal phosphatase shows that, by removal of phosphate groups, faster migration of BMI1 is restored to that seen in G₁/S phase extracts (Fig. 7E). The enzymatic removal of phosphate groups is concentration dependent and specifically prevented by phosphatase inhibitors (Fig. 7E). When chromatin-association of BMI1 is most pronounced, from G₁onward into S phase, it would follow that BMI1 in U2-OS cells is relatively hypophosphorylated. An intermediate phosphorylation degree is obvious in extracts of early G₁-arrested cells (Fig. 7B), confirming the observation that Pc-G domains in U2-OS cells are being visibly restored at this time (Fig. 2C). When subjected to differential nuclear extraction, most of the hypophosphorylated BMI1 (G₁phase) is retained in the chromatin-bound fraction (Fig. 7F). Conversely, at M phase BMI1 is readily extracted in the soluble nuclear fraction. Thus, the microscopically visible dissociation of Pc-G complexes at mitosis is supported by biochemical data, indicating actual detachment of phosphorylated BMI1 from chromatin. MPH-chromatin interaction follows similar dynamics (Fig. 7F). In addition, at least some Pc-G protein-protein interactions appear to have been retained at M phase, since MPH is readily immunoprecipitated with antibodies against BMI1 (Fig.7G) or M33 (data not shown). Taken together, the above findings demonstrate that cellular BMI1 is phosphorylated in a cell cycle-dependent manner, and that phosphorylated BMI1, likely in association with other Pc-G proteins, is physically dislodged as a complex from chromatin at G₂/M phase in both primary and tumor cell lines.

In an earlier study, we reported on accumulation of endogenous BMI1 in distinct subnuclear domains in human U2-OS cells (Alkema et al., 1997a). Several other tumor cell lines show a comparable BMI1 distribution pattern (e.g. K562, HeLa, MCF-7, SOAS-2 and SW480). Although a possible connection to specific chromatin regions remained unclear, the exactly overlapping nuclear patterns of several other Pc-G proteins (Alkema et al., 1997a; Gunster et al., 1997; Satijn et al., 1997; Schoorlemmer et al., 1997; Saurin et al., 1998) suggested that these domains reflect sites at which these Pc-G proteins interact in large Pc-G complexes.

Here we show that the domains at which BMI1 is concentrated in human tumor cell lines are at paracentric heterochromatin on chromosome 1 (1qh). These results are in full agreement with recently published findings of an independent study (Saurin et al., 1998). We find that primary human cells do not display the 1qh concentration of nuclear Pc-G protein typical of human tumor cell lines, but instead a fine-grained nuclear Pc-G staining pattern. The 1qh-associated nuclear Pc-G domains therefore most likely represent an acquired feature of tumor cell lines. Immuno-FISH data and karyotype analysis presented here for HeLa, U2-OS and MCF-7 cells, strongly suggest that the variation in size and number of heterochromatin associated Pc-G domains results from karyotypic differences between tumor cell lines, and further underwrite this notion. Also, we have observed that local subnuclear Pc-G accumulation can in fact be evoked at low frequency in primary cells by prolonged culturing (see Results) or by exposure to demethylating agents (data not shown), but this association is 1qh-unrelated. This effect on nuclear Pc-G redistribution is most likely due to acquired changes in chromatin structure (see also below). Endogenous BMI1 levels tend to be higher in tumor-derived human cell lines compared to primary cells and may contribute to Pc-G-accumulation at 1qh. However, overexpression of BMI1 per se does not alter its staining pattern: the nuclear distribution of ectopically expressed BMI1 in tumor cell lines and, more importantly, in primary human cells at low passage, is identical to that of the endogenous protein in the respective culture types. The combined data suggest that a fine-grained distribution of BMI1 in interphase nuclei of primary human cells is the default situation, as has been reported for Drosophila embryonal cells (Buchenau et al., 1997), murine fibroblasts (Wang et al., 1997) and primary murine cells (our unpublished observations).

Specific satellite DNA binding proteins, such as the chromatin-binding factor GAGA (Platero et al., 1998) and the product of proliferation disrupter (prod) (Török et al., 1997), have been identified in Drosophila. We suggest that the preferential 1qh-association of BMI1 is mediated by specific protein interaction(s) with a repeated sequence motif occurring in chromosome 1-specific a-satellite DNA. Alternatively, it is possible, because of satellite DNA instability and high divergence rate of alphoid tandem repeats, that chromosome region 1q12 by chance has acquired and amplified potential Pc-G binding sites (Waye et al., 1987; Csink and Henikoff, 1998). Since we see no specific DNA binding with BMI1 (our unpublished results), it is likely that the putative protein-satellite DNA interaction is conferred by human BMI1-interacting protein(s) or the complex as a whole rather than BMI1 itself. We and others have recently identified proteins that bind the RING-finger motif of BMI1, thus confirming its role in protein-protein rather than in protein-DNA interaction (Hemenway et al., 1998; M. van Lohuizen et al., unpublished results). Interestingly, the RING-finger domain is required for proper subnuclear localization of BMI1 (Alkema et al., 1997a). Specific in vivo DNA-binding had not been reported for any of the thus far known Pc-G proteins, although recently, the Drosophila Pc-G gene pleiohomeotic was found to encode a protein with homology to the mammalian sequence-specific DNA-binding transcriptional regulator YY1 (Brown et al., 1998). Furthermore, a common sequence motif occurs in polycomb responsive elements (PREs; Mihaly et al., 1998) and in the YY1 binding consensus. As yet unknown Pc-G complex proteins may recognize analogous sequence elements in a-satellite repeats or mediate binding of PcG-complexes to DNA. An interesting candidate protein with the latter potential may be the recently identified RYBP1 (Garcia et al., 1999).

The absence of Pc-G domains in primary cells (see above) argues that these subnuclear domains in fact represent an acquired feature of tumor cells in culture. Of interest in this respect is that in many types of human cancers pericentromeric rearrangements of chromosome 1 occur at high frequency (Brito-Babapulle and Atkin, 1981). Pericentromeric chromosomal anomalies involving rearrangements between satellite repeat sequences on chromosomes 1, 9 and 16 have recently been linked to DNA hypomethylation (Weizhen et al., 1997and references therein). However, global demethylation by itself appears not sufficient to induce Pc-G/1qh-association in primary cells (our unpublished observations). Most established cell lines studied to date for nuclear Pc-G distribution are tumor derived (Alkema et al., 1997a; Gunster et al., 1997; Satijn et al., 1997; Saurin et al., 1998). Paracentric satellite repeat sequences in these cells are likely to be hypomethylated and possibly rearranged and may have acquired PcG binding capacity. Of interest, in this context, is the correlation between BMI1 expression levels and its tumorigenicity (van Lohuizen et al., 1991a; Alkema et al., 1997b) and the apparent expressional regulation of Pc-G genes during normal haematopoietic development (Lessard et al., 1998). In addition, hBmi1 gene amplification was reported in human malignancy recently (Beà et al., 1999). Recent mechanistic insights in Polycomb and trithorax-Group function connect both deregulation of Pc-G and trx-G genes with cell cycle control and tumorigenesis (reviewed in van Lohuizen, 1999). Given these correlations, it will be of considerable interest to explore a-satellite/Pc-G-association as a possible diagnostic marker for tumorigenicity.

A model recently put forward (Brown et al., 1997; Marshall et al., 1997) suggests that centromeric heterochromatin forms a transcriptionally repressive subnuclear domain and that repressed genes are selectively recruited into such domains. It is possible that BMI1 target gene(s) are silenced through Pc-G-association at heterochromatin. Active recruitment of genes in a trans-fashion into repressive genomic domains is precedented in telomere proximity silencing in yeast (Gotta et al., 1997) and somatic pairing as a result of the PEV brown dominant mutation in Drosophila (Dernburg et al., 1996; Martin-Morris et al., 1997; Henikoff, 1997). In addition, Drosophila polycomb responsive elements have been recently found to exhibit trans-interactions (Sigrist and Pirrotta, 1998; Pirotta, 1998). The preferential association of BMI1 with chromosome region 1q12 in human cell lines provides an experimental basis to further test fragments from 1qh-specific a-satellite DNAs in DNA-binding assays or for ectopic action as PREs.

We here report a remarkable cell cycle-dependent association of Pc-G proteins to chromatin in both primary and tumor cells, which correlates well with changes in its phosphorylation status. Parallel-staining studies on growing cells with cell cycle-specific markers (PCNA, phosphorylated histone H3) demonstrate that at late S phase the majority of Pc-G proteins visibly dissociate from chromatin. This dissociation occurs in both tumor cell lines and in primary cells expressing BMI1 ectopically, arguing the observation is not an artifact or the consequence of establishment in culture or immortalization/ transformation. At G₂/M, BMI1 is phosphorylated. Hyperphosphorylated BMI1 does not bind chromatin, as established by differential nuclear fractionation. These data provide direct biochemical evidence that BMI1-chromatin-association is lost during mitosis and thus corroborate the immunohistochemical findings. Importantly, our findings also indicate that cell cycle-dependent regulation of Pc-G-chromatin association takes mainly place at level of nuclear redistribution. At least some of the protein interactions within Pc-G complexes remain during mitosis, suggesting that phosphorylation determines Pc-G complex-chromatin interaction: Pc-G proteins are released from chromatin as complexes rather than as single units. This may have direct bearing on the rapid re-establishment of Pc-G-chromatin binding following nuclear division.

Using more sensitive detection methods, we find that a small fraction of the BMI1 protein is detectable as a fine dotted layer over metaphase chromosomes and at heterochromatic region 1q12 of HeLa, U2-OS and MCF-7 cells. Chromosomes would be stripped of most Pc-G-protein complexes at mitosis, leaving a small, but biologically significant amount of transcriptional repressors in place, thereby epigenetically marking relevant loci for transcriptional repression in ensuing cell generations. Our combined data support this view and are in full agreement with an analogous mechanism proposed for Pc-G-mediated transcriptional memory function in Drosophila, where a small percentage of Pc-G proteins was shown to remain physically attached to chromatin (Buchenau et al., 1998). Taken together, these data confirm that the biochemical interaction of Pc-G proteins with chromatin and its cell cycle-dependent character have been conserved throughout phylogenetically distinct taxa, and add to the importance of phosphorylation in the regulation of chromatin-association.

The phosphorylation-linked removal of Pc-G complexes at mitosis may simply be required to facilitate chromosomal condensation and subsequent segregation, a process in which phosphorylation of histone H3 at Ser10 was recently shown to play a crucial role (Wei et al., 1999). Significantly, however, members of the trithorax group (trx-G) proteins, a diverse protein family that functions as counter-actors of Pc-G repression, are also subject to cell cycle-dependent post-translational modification: hbrm and BRG-1 are specifically phosphorylated at mitosis (Muchardt et al., 1996). This modification correlates well with the chromosomal exclusion of the human SNF/SWI complex at G₂-M. It was speculated that this dissociation could be part of a mechanism leading to transcriptional arrest at mitosis. Based on the observation that Pc-G complexes undergo a similar cell cycle-dependent displacement, we suggest an alternative explanation. Since chromatin dissociation coincides in time with (completion of) de novo DNA-synthesis, the dissociation of Pc-G and trx-G complexes from chromatin during late S-/G₂-M phase would present a dividing cell with an opportunity for specific changes in gene expression patterns. This notion is in line with recent reports that PCNA directs CAF-1-mediated nucleosome assembly onto newly synthesized DNA (Shibahara and Stillman, 1999and references therein), during a window in the cell cycle at which changes in heritable gene expression can be implemented through modulation of chromatin structure (reviewed by Wolffe, 1991).

Elegant studies using purified Pc-G complexes have yielded important insights in how Pc-G and trx-G proteins may counteract each other (Shao et al., 1999). When positioned on nucleosomes prior to exposure to the chromatin remodeling factors of the SWI/SNF family, Pc-G complexes appear to act as a molecular lock. Conversely, once exposed to SWI/SNF factors, repression by Pc-G complexes does not occur. In addition, both chromatin-remodeling activities appear independent of histone tails in vitro (Shao et al., 1999). Pc-G-mediated reporter gene silencing in Xenopus oocytes (Strouboulis et al., 1999) and in mammalian cells (our unpublished observations) is trichostatin A-insensitive, and various Pc-G protein-specific antisera fail to coprecipitate HDAC activity from U2-OS and E12.5 embryo extracts (our unpublished observations). These combined findings suggest that histone (de)acetylation probably plays a minor role, if at all, in these expressional-maintenance processes by Pc-G and trx-G proteins. It is conceivable that phosphorylation-dependent chromatin interaction is crucial for proper regulation of Pc-G and trx-G action and provides coupling to cell cycle progression. Histone H3 was recently identified as a downstream target of a MAPKAP-kinase, which acts in a signaling cascade independent of mitosis (Sassone-Corsi et al., 1999). This exciting finding indicates a potential link between extracellular signaling and chromatin remodeling. In the context of Pc-G and trx-G function, this finding raises the intriguing possibility that as yet unidentified protein kinases and phosphatases play crucial roles during development to direct commitment and differentiation-associated gene expression. Importantly, the here described phosphorylation-linked chromatin dissociation of Pc-G complexes gives a first insight into how Pc-G proteins are dynamically regulated, possibly by extracellular cues.

In summary, we conclude that Pc-G proteins are distributed in a fine-grained pattern along interphase chromosomes of human primary cells, as seen in primary cells of other species. In addition, tumor cells have acquired concentrated Pc-G binding at 1qh, a feature that may be a consequence of cellular transformation. Our data support the notion that the biochemical interactions between Pc-G complexes and chromatin and its cell cycle-dependent regulation have been conserved throughout evolution. The observation that BMI1 is phosphorylated in a strictly cell cycle-dependent manner should pave the way to a better understanding of how Pc-G group complexes are regulated.

We thank P. F. Ambros, H. Willard, B. Earnshaw and T. Jenuwein for expert help and advice. We thank T. Ide for TIG3 cells and G. Nolan for providing phoenix packaging cell lines and retroviral vectors. We are grateful to J. Fuchs, F. Klein, M. Lambrou and L. Oomen for help in image acquisition. G. Steiner kindly provided the CREST scleroderma serum; the anti-phosphorylated histone H3 antiserum was a kind gift of D. Allis. J.W.V. was supported by a grant from the Dutch Organization for Scientific Research (NWO).