The facultative heterochromatin of the inactive X chromosome has a distinctive condensed ultrastructure

Interphase chromatin is typically thought to be decondensed at locations in which genes are transcribed and condensed where genes are silent (Wegel and Shaw, 2005). However, this generalization is contradicted by observations that inactive genes may reside in domains of open chromatin, whereas active genes in regions of low gene density can be embedded within compact chromatin fibers (Gilbert and Bickmore, 2006; Gilbert et al., 2004). A diversity of theoretical models for the nuclear organization of active and silent chromatin (reviewed in Cremer et al., 2006; Spector, 2003) stem from our limited understanding of higher-order chromatin structure beyond the 30-nm fiber, referred to as large-scale chromatin structure. This limited understanding results from technical limitations in imaging of chromatin by light microscopy (LM) and electron microscopy (EM) (Horowitz-Scherer and Woodcock, 2006), including sensitivity of higher levels of chromatin folding to buffers and EM preparation methods (Belmont et al., 1989) and lack of suitable high-contrast DNA-specific staining methods for EM.

Traditional cytology classifies chromatin into less-condensed euchromatin and more-condensed heterochromatin. Heterochromatin has been further subdivided into permanently condensed constitutive heterochromatin and facultative heterochromatin, which becomes condensed/decondensed at some point during development (Wegel and Shaw, 2005). X inactivation in female mammals is a classic example of the formation of facultative heterochromatin. The inactive X chromosome (Xi) appears within interphase nuclei as a heteropycnotic Barr body usually positioned at the nuclear or nucleolar periphery (Barr and Carr, 1962; Belmont et al., 1986; Puck and Johnson, 1996; Zhang et al., 2007). Inactivation of one of the X chromosomes in female cells is the mechanism of mammalian dosage compensation of X-linked genes (Lyon, 1961).

A long-standing assumption has been that sequential epigenetic modifications occurring during X inactivation directly lead to Xi DNA compaction and increased condensation per se might contribute to gene silencing (Arney and Fisher, 2004; Chow and Brown, 2003). This assumption has been challenged by comparison of the Xi and the active X chromosome (Xa) conformation using 3D fluorescence in situ hybridization with X-specific whole-chromosome probes. In human amniotic cells, Xi and Xa territories were observed to occupy similar volumes but differed in shape and surface area (Eils et al., 1996), suggesting the same average chromatin compaction for both the Xi and Xa (Pollard and Earnshaw, 2004).

It has been proposed that the Xi in the interphase nucleus is organized into a core of repetitive sequences surrounded by genic regions (Chaumeil et al., 2006; Clemson et al., 2006), but a detailed ultrastructural analysis of the Xi has so far been lacking (Heard and Disteche, 2006; Straub and Becker, 2007). Here, we have combined LM and EM of the Xi in human and mouse female fibroblasts to demonstrate that the Xi has a unique ultrastructure, containing condensed, large-scale chromatin fibers/domains clearly distinct from those observed in euchromatic or constitutive heterochromatin regions. In addition, we demonstrate a distinct position of this facultative heterochromatin in the nucleus.

For ultrastructural studies, we wanted a homogeneous cell population in which the Xi was fully silenced and showed minimal variation in appearance. The original concept of increased chromatin compaction of the Xi comes from its appearance at the LM level as a distinct nuclear Barr body, which stains more intensely with nucleic-acid stains than the surrounding chromatin, implying a higher DNA compaction.

We used human female primary fibroblasts, WI-38 cells, which have a high percentage of cells showing a distinct Barr body. This percentage peaked in confluent cells, 10 days after passage. The Barr body was equated to a DAPI bright body with a unique intensity and size distinct from all other DAPI bright regions. Immunostaining against trimethylated histone H3 on lysine 27 (H3-3mK27) provided a robust, independent identification of the Xi/Barr body, with a discrete, unique, high-contrast stained body recognizable in nearly 100% of cells. In cells with a clearly defined Barr body, as defined by DAPI staining, the DAPI dense region colocalized nearly completely with the region of elevated H3-3mK27 staining (Fig. 1A).

We used H3-3mK27 staining to confirm the identification of a DAPI-dense body as the Xi, to visualize the Xi in cells without a recognizable DAPI-dense body, and to assay the variability in Xi size and conformation. Failure to recognize a Barr body exclusively by DAPI staining was either due to the presence of multiple DAPI-dense bodies elsewhere in the nucleus (Fig. 1B), or because a change in Xi conformation or orientation resulted in comparable DAPI contrast as the surrounding chromatin (Fig. 1C). Inspection of single, unprocessed optical sections by two independent observers revealed that, in ∼85% of confluent cells (n=100), the Barr-body/Xi condensed core could be identified solely by the presence of a DAPI-dense body. This number rose to 95% for confluent cells when optical section stacks were inspected (n=131).

Xi/Barr-body area, as measured by the H3-3mK27 signal, was smaller and showed a more uniform size distribution in confluent versus log-phase cells (2.7 versus 3.4 μm²) (Fig. 1D). We therefore used confluent cells for most of our ultrastructural analysis.

A major problem in viewing chromatin ultrastructure by transmission electron microscopy (TEM) is the lack of an adequate DNA-specific stain. To obtain satisfactory contrast of chromatin and to improve accessibility during pre-embedding immunogold staining, detergent extraction of the nucleoplasm prior to fixation is frequently used. However, the extreme sensitivity of large-scale chromatin structure to minor changes in buffer conditions leads to significant changes in ultrastructure dependent on the choice of permeabilization buffer (Belmont et al., 1989).

Here, we used the Barr-body appearance in live cells to optimize sample preparation and verify EM ultrastructural preservation. In cells expressing GFP-histone H2B, the GFP fluorescence distribution was similar to the DAPI staining distribution, which we assume was due to its proportional enrichment based on DNA content. In deconvolved optical sections, the Barr body visualized in live cells by GFP-H2B fluorescence displayed obvious fiber substructures measuring 200-400 nm in width (Fig. 2A). After live-cell imaging, the same samples were then fixed or permeabilized in various buffers prior to fixation in 2% glutaraldehyde (GA) and a repeat optical sectioning. Comparison of images before and after fixation (data not shown) identified buffer A, used in our previous work to preserve large-scale chromatin structures, as most suitable for Barr-body structural preservation after permeabilization (Belmont et al., 1989).

As a higher-resolution test, we used the TEM appearance of Xi large-scale chromatin fibers in unextracted cells (Fig. 2B) to guide selection of buffer and fixation conditions preserving these fibers during nucleoplasm extraction and immunostaining. To minimize buffer-induced alterations in chromatin ultrastructure, we also explored the use of a UV pre-fixation procedure prior to detergent permeabilization and aldehyde fixation. Cross-linking by UV alone or after blue-light irradiation of ethidium bromide (EtBr)-stained samples is reported to stabilize higher levels of chromatin structure (Maeshima et al., 2005; Sheval et al., 2004). By combining these two approaches, we rapidly fixed higher-order chromatin structure by UV in the presence of EtBr. This procedure provided higher-intensity pre-embedding immunogold labeling while better preserving large-scale chromatin fibers as compared to a light aldehyde-fixation step prior to immunostaining.

A distinct heavy-metal-stained heterochromatic body, of similar size and substructure to that observed for the Barr body by LM in live cells, was observed by EM in cells fixed directly in 2% GA (Fig. 2B). We observed similar ultrastructure and improved contrast of chromatin with (Fig. 2C) or without (Fig. 2D) UV/EtBr cross-linking, followed by permeabilization in buffer A^* and fixation with 2% GA. Subsequent histone H3-3mK27 immunogold staining and serial sectioning confirmed that these heterochromatin structures corresponded to the Barr bodies visualized by LM (see below).

The larger Xi examples visualized by H3-3mK27 immunofluorescence from log-phase cells often showed distinct large-scale chromatin fibers/domains near to or protruding from a denser core (Fig. 1E,F). These structures were similar in diameter to chromatin substructures observed within the more compact Barr bodies of live confluent cells (Fig. 2A). Similarly, the Xi heterochromatin substructures visualized at higher resolution by TEM (Fig. 2B-D) were comparable to those visualized by LM. The overall shape and size of the entire heterochromatin domain visualized by TEM correlated with Barr-body size and shape as defined by H3-3mK27 or DAPI staining. Individual heterochromatin fibers/domains of the Xi appeared larger in diameter compared with substructures observed in the surrounding bulk chromatin by both LM and EM.

To confirm that the heterochromatin structures visualized within individual sections by TEM corresponded to Barr bodies, we tested whether they were present exactly once per nucleus and had a higher-than-background level of H3-3mK27 immunogold staining. Serial 200-nm sections through the entire volume of 13 randomly selected WI-38 nuclei (seven confluent, six log phase) revealed that 12 nuclei contained one and only one heterochromatic body (Fig. 3A,B, supplementary material Fig. S1), similar to the examples shown in Fig. 2. The remaining nucleus was unusually large, very likely tetraploid and contained two joined heterochromatin bodies, each of similar shape to the bodies seen in the other 12 nuclei. No other condensed-chromatin region within these nuclei was comparable in size and overall chromatin compaction. Some of these 13 nuclei were from H3-3mK27-immunostained samples and each of these showed higher-than-background immunogold staining over the heterochromatic body (Fig. 3C).

In all 13 serial-section data sets, the Xi heterochromatin appeared as a non-solid volume containing spatially separated chromatin substructures that were 30-400 nm in diameter, with intervening spaces of 50-400 nm. This was also true for the Xi as visualized from single sections from an additional 189 cells prepared by various protocols (Fig. 2). When using buffer conditions preserving large-scale fibers, the Xi could be visualized as a tightly folded accumulation of large-scale chromatin fibers/domains with noticeably larger diameters than observed elsewhere throughout most of the nucleus.

We wished to compare the appearance of facultative and constitutive heterochromatin. In human cells, regions of constitutive heterochromatin are small and not easily identified. We therefore examined female mouse embryonic fibroblasts (MEFs) containing large readily identifiable chromocenters formed by coalescence of pericentric heterochromatin from several chromosomes. The Barr bodies in MEFs were identified by immunostaining against H3-3mK27. Although they showed above-background DAPI staining, the DAPI-staining intensity was considerably lower than over chromocenters, including those of similar or even smaller size (Fig. 4A).

Complete serial-section sets through 11 nuclei showed very similar MEF Barr-body ultrastructure to that observed in WI-38 cells. MEF Xi ultrastructure, with relatively loose packing of heterochromatin fibers/domains, contrasted sharply with the denser, more uniform texture of chromocenters (Fig. 4). H3-3mK27 immunogold staining clearly distinguished the facultative heterochromatin of the Xi from the constitutive heterochromatin of the chromocenters (Fig. 4C). Because of the high chromatin packing density within the chromocenters, we cannot resolve whether large-scale chromatin fibers of similar diameter as in the Xi are also present in the chromocenters, but not recognizable owing to their closer packing, from the alternative possibility that the large-scale chromatin organization in chromocenters is fundamentally different from that in the Xi.

Reconstructions of four WI-38 nuclei were assembled using nominally 40-nm-thick serial thin sections; based on the appearance of large-scale chromatin fibers in x-z and y-z orthogonal views (see below), 60 nm was a better estimate of section thickness. Within individual sections, the overall nuclear chromatin distribution appeared relatively sparse or sponge-like, with the area occupied by chromatin representing a small fraction of the total nuclear area (Fig. 5A, supplementary material Fig. S2A). Similarly, the condensed region of the Xi also appeared significantly less solid than impressions of the Barr body from LM (Fig. 1A) or thicker TEM sections (Figs 2, 3, 4, 5). Maximal intensity projections (Fig. 5B) or additive intensity projections (supplementary material Fig. S2B,C) of aligned serial thin sections, however, showed a more solid appearance of the Xi and a crowded nuclear interior, consistent with single optical sections of DAPI/H3-3mK27 staining (Fig. 1A) or H2B-GFP fluorescence of live cells (supplementary material Fig. S2D). With ∼20 sections per image stack, projections corresponded to an ∼1.0- to 1.5-μm depth of volume, close to the expected LM depth of field for a high-NA objective lens (±0.75 μm).

EM images of single thin sections showed that Xi heterochromatin is not a solid structure but rather an open structure of chromatin fibers/domains of variable diameters, from 30 to 600 nm (Fig. 5A). In some regions, actual fiber segments could be visualized as spatially distinct structures, with no overlap with other chromatin domains, in adjacent serial sections. In other regions, fiber-like regions/domains had multiple contacts with neighboring chromatin condensed regions so the connectivity of DNA between neighboring domains could not be determined unambiguously. The most common structural motif was ∼200 nm in diameter, larger than the characteristic diameter of large-scale chromatin fibers in the surrounding euchromatin. Interchromatin regions up to 400 nm in dimension separated Xi heterochromatin substructures. These chromatin-free subvolumes were contiguous with nuclear pores and with the surrounding nucleoplasm (Fig. 5A, supplementary material Movie 1). Deep extensions of nucleoplasm into the Barr-body core could be clearly demonstrated by orthogonal views of interpolated stacks (Fig. 5C). Moreover, orthogonal views revealed a connection of nuclear pores with intrachromosomal spaces found within the Barr body (Fig. 5C; xz and yz slices, NP). A solid model of the Xi formed from density-threshold segmentation of the serial section reconstruction was calculated (Fig. 5D). 3D visualizations of the porous sponge-like substructure of the Xi is provided by different views of the reconstructed solid model (Fig. 5D′,D″, supplementary material Movie 2) as well as by stereo pairs (Fig. 5D′″).

Consistent with previous findings in the literature (Belmont et al., 1986; Bourgeois et al., 1985; Zhang et al., 2007), we observed ∼60% of Barr bodies in WI-38 cells associated with the nuclear envelope, based on analysis of single optical sections. However, our initial TEM experiments showed connections of the Xi with the nuclear envelope in all examples examined. Therefore, we investigated the relationship of the Xi to the nuclear envelope using a combination of 3D deconvolution LM and 3D EM analysis.

Using H3-3mK27 staining to mark the Xi and nuclear-pore staining to mark the nuclear envelope, we found that 21/23 of Xi examples that were scored as `interior' based on single optical sections showed some association with the nuclear envelope after analysis of 3D LM data. In the remaining two cells, the Xi position relative to the nuclear envelope was unclear because of the extreme flatness of the nuclear envelope adjacent to the coverslip and poor resolution of LM in the z direction. In total, 14/23 cells showed the Xi adjacent to the top surface of the nuclear envelope, facing away from the coverslip (supplementary material Fig. S3A). Deep invaginations of nuclear envelope reaching the Xi territory were found in 6/23 cells (supplementary material Fig. S3B). Two cells showed a connection to the nuclear envelope through the attachment of extended, H3-3mK27-stained large-scale chromatin fibers protruding from the Xi (supplementary material Fig. S3B). One cell showed both an attachment to a nuclear envelope invagination and an extended fiber reaching the main nuclear periphery.

3D analysis by TEM using 200-nm serial sections revealed an attachment of the Xi to the nuclear envelope in essentially 100% of cells – 24/24 complete serial-section data sets and 7/7 incomplete serial-section data sets. These 31 randomly selected cells included examples from confluent and log-phase WI-38 cells and MEFs. In total, 20/31 cells showed close association of the Xi with the nuclear envelope in sections parallel to the nuclear equatorial plane (Figs 2, 3, 4). 8/31 cells showed the Xi attached to the nuclear envelope in sections far from the equatorial plane, mostly on the top of the nucleus (Fig. 6A-C). The remaining 3/31 cells showed an internally located Xi attached to the nuclear envelope via an invagination of the nuclear envelope (Fig. 6D).

The observed association of the Xi with the nuclear envelope does not appear to be specific to the Xi facultative heterochromatin, because a similar association was observed for constitutive heterochromatin of mouse chromocenters. 3D TEM analysis of serial sections through MEF nuclei (n=11) revealed that all observed chromocenters were also attached to the nuclear envelope (supplementary material Fig. S4 and Movie 3).

Consistent with a reported ∼50% of Xi association with the nucleolus in G0 and log-phase MEFs (Zhang et al., 2007), we found a 40% attachment (n=108) of the Xi to nucleoli by LM and 38% by TEM using serial sections through entire nuclei (9/24) in confluent, quiescent WI-38 cells (supplementary material Fig. S5). All Xi in contact with the nucleolus maintained some contact with the nuclear envelope.

The mammalian Xi has been a classic model for facultative heterochromatin for nearly 50 years. Throughout most of this time, an increased chromatin compaction, as inferred by the appearance of the pycnotic Barr body, has been assumed. Moreover, during the past 10 years, a set of repressive epigenetic marks has been associated with X inactivation, correlating changes in chromatin histone composition with X inactivation (Chow and Brown, 2003; Heard and Disteche, 2006). However, a FISH analysis revealed similar volumes for the Xa versus Xi, emphasizing instead differences in shape rather than overall compaction as the major distinguishing feature of the Xa versus Xi (Eils et al., 1996). Surprisingly, no careful ultrastructural analysis of the Barr body was available to address this apparent contradiction. Here, we identified buffer and sample-preparation conditions preserving Xi large-scale chromatin structure close to that visualized in live cells. A novel procedure was developed to provide improved preservation of Xi large-scale chromatin structure during immunogold-staining procedures.

Using these methods, here we show for the first time that the condensed region of the Xi assumes a unique ultrastructure, distinct from surrounding euchromatin or constitutive heterochromatin, containing large-scale chromatin fibers/domains with noticeably larger diameter than observed in the surrounding euchromatin regions. In contrast to its appearance as a solid mass by wide-field LM without deconvolution, the Barr body at the ultrastructural level clearly shows a relatively porous structure, formed by the tight folding of large-scale chromatin fibers/domains within a compact volume. The mean density of chromatin compaction is significantly higher than euchromatic chromosome regions but lower than that observed for mouse chromocenters. 3D ultrastructural analysis also revealed that essentially all Barr bodies maintain some contact with the nuclear periphery, which in mammalian cells is correlated with repression of gene expression (Taddei et al., 2004), with ∼40% of Barr bodies also in contact with the nucleolus in confluent WI-38 cells.

Our results showing a distinctive, condensed ultrastructure for the mouse and human Barr bodies appear to contradict a FISH analysis that found similar volumes for both the Xa and Xi in human amniotic cells (Eils et al., 1996), implying similar mean compaction for both chromosomes (Pollard and Earnshaw, 2004). Several possibilities may account for this apparent contradiction. The limited resolution of LM relative to the size of the Barr body, combined with chromosomal volume changes induced by the FISH procedure, might have led to errors in Xi and Xa volume estimation. Alternatively, our study examined the ultrastructure of the condensed region of the Xi, corresponding to the Barr body visualized by DAPI staining, whereas the FISH study measured the total Xi volume, which includes both the Barr body plus the chromosome regions and active genes that escape silencing. In fact, the authors of the FISH study noted that the Xi chromosome territory appeared frequently to have a denser core FISH signal, whereas the Xa territory showed a more uniform FISH density throughout the territory. The focus of the FISH study on volume comparison might therefore have missed the significant differences in chromatin structure assumed by the condensed region of the Xi.

How higher levels of chromatin folding and chromosome territory shape influence gene expression remains unclear. Based on low-resolution FISH chromosome paints, an interchromosome domain (ICD) model was proposed in which decondensed active genes are positioned on the surface of compact chromosome domains, bordering an interchromosome space containing macromolecular complexes and nuclear bodies important for transcriptional activity (Zirbel et al., 1993). Additional LM analysis of chromosome territories have instead shown less-solid chromosome territory structures with more substructure, while also localizing transcription and active genes within chromosome territories (Mahy et al., 2002b; Osborne et al., 2004). This has prompted several refinements of the original ICD model, so that more or less solid chromosome territories contain extensive invaginations, leading to more-extensive contact with an interchromatin space (Cremer et al., 2006; Williams, 2003). In all of these models, however, the positioning of active or potentially active genes to the exterior of a dense chromosome territory such that they are accessible to the interchromosome space remains a key concept. Based on these models, the rounder, smoother shape of the Xi versus Xa might have functional significance in gene repression by reducing the territory surface area in contact with the interchromosome space.

By contrast, our results show that chromatin occupies a relatively low fraction of the nuclear volume in mouse and human fibroblasts, with the possibility for extensive space, not occupied by chromatin, for movement of large macromolecular complexes and even nuclear bodies. Even the relatively condensed Barr body still shows considerable intrachromosomal space separating distinct large-scale chromatin fibers/domains and contiguous with the nuclear pores and nuclear interior. Recent experiments have revealed looping of Mbp-sized gene-rich regions outside of chromosome territories accompanying gene activation (Mahy et al., 2002a; Ragoczy et al., 2003; Volpi et al., 2000). Therefore, the rounder shape reported for the Xi versus Xa (Eils et al., 1996) might be at least partly a consequence rather than a cause of Xi gene silencing, with the elimination of the looping of active regions outside of the core Xi chromosome territory.

In considering how Barr-body ultrastructure might influence gene silencing, we suggest the proper focus instead should be on distinctly higher-level compaction of large-scale chromatin fibers/domains observed within the Barr body as compared to surrounding euchromatin regions. This region of the Xi has been linked with transcriptional silencing (Chaumeil et al., 2006; Clemson et al., 2006), with a movement of genes towards the interior of the Xist-coated Xi core region accompanying gene silencing during embryonic stem (ES) cell differentiation (Chaumeil et al., 2006). Our results show that this core also corresponds to the region of H3K27 trimethyl modification and assumption of a distinctive, large-scale chromatin-folding motif. These correlations suggest a probable functional link between the altered large-scale chromatin packing and gene silencing.

Based on the biased localization of X-linked inactive genes towards the periphery of the condensed Xist-coated Barr body by FISH, together with the localization of Cot-1 DNA largely within the Barr body, Clemson and colleagues have suggested extensive looping of both active and inactive genes outside or on the edge of the Barr body, which they propose is formed largely by transcriptionally repressed repetitive DNA (Clemson et al., 2006). However, Chaumeil and colleagues observed X-linked genes protruding from the Xist-coated Xi core very early during Xi silencing accompanying ES cell differentiation, with genes undergoing silencing shifting later during differentiation to a more interior position largely at the edge of the Xist-coated Xi core (Chaumeil et al., 2006).

In the Clemson et al. model, all X-linked genes, not just those that escape silencing, loop out from the dense Barr-body core (Clemson et al., 2006). Moreover, in this model, only centromeric and Cot-1 repetitive DNA localizes within the dense Barr-body core, implying extensive looping of repetitive DNA sequences located in intergenic regions back into the Barr-body core. Therefore, this model would predict extensive looping on the same DNA distance scale as gene spacing – typically tens to hundreds of kb. Previously, we estimated a compaction of approximately 3 Mbp per μm for ∼100 nm large-scale chromatin fibers (Tumbar et al., 1999), with the expected compaction of the thicker, ∼200 nm diameter, large-scale chromatin fibers/domains in the Xi condensed regions described in this paper expected to be substantially higher. Therefore, with compactions corresponding to greater than 300 kb per 100-nm lengths of large-scale chromatin fibers, we should expect to see many looped-out genes protruding from the surfaces of the condensed, large-scale chromatin fibers that make up the Barr body. However, in our serial thin sections, we did not see the extensive looping of decondensed chromatin protruding outwards from the Barr body as would be expected by the Clemson et al. model (Clemson et al., 2006). However, we did see chromatin domains packaged similarly to surrounding euchromatin – and therefore estimated as being 100's to 1000's of kb in size – in close contact with but extending outwards from the condensed, H3K27-trimethyl-labeled Barr body. Without any specific label, we could not unambiguously determine whether these chromatin domains were part of the Xi, but it is possible that these chromatin regions correspond to X-linked genes, or more likely clusters of X-linked genes, that escape silencing.

Clearly, a large problem in extrapolating from the above-cited FISH studies to our own work is the large mismatch in structural preservation between the associated protocols. We have used sample-preservation conditions that preserve the Barr-body size and internal structure so that they are very close to that observed by live-cell imaging, as well as by EM of unextracted cells under optimal chemical-fixation methods used in conventional EM. Unfortunately, these fixation methods are too stringent to allow localization of specific genes or chromosome regions by FISH methods. By contrast, the best 3D FISH procedures produce near-complete loss of large-scale chromatin ultrastructure as visualized by EM (Solovei et al., 2002) and, at least in some cases, these structural perturbations are quite apparent even by LM (Robinett et al., 1996). Therefore, it is difficult at this time to reconcile differences in possible models of Xi organization created by FISH methods versus ultrastructural analysis.

In conclusion, we have demonstrated a distinct condensation of Xi chromatin and association of the Xi in all cells with the nuclear envelope. These results are consistent with the idea that spatial segregation of Xi chromatin helps maintain Xi silencing by limiting access to transcription factors (Heard and Disteche, 2006). Future experiments directly comparing the ultrastructure of X-linked active versus inactive genes, and correlating these differences with regulation of gene silencing, will require the development of methods for visualizing the location of specific genes and components of the transcriptional machinery without perturbing chromatin ultrastructure.

Human female fibroblasts (WI-38) obtained from the American Type Culture Collection were maintained in minimum essential medium (Invitrogen) supplemented with 10% FBS (HyClone Laboratories) at 37°C in 5% CO₂. WI-38 cells at cumulative population doublings between 20 and 40 were used 3 days (log-phase cells) or 10 days (confluent cells) after plating. Female mouse embryonic fibroblasts (MEFs) were provided by Edith Heard (Curie Institute, Paris, France). MEFs were maintained in Dulbecco's modified Eagle's medium (Invitrogen), 10% FBS (HyClone Laboratories) and 0.001% 2-mercaptoethanol (Sigma) at 37°C in 8% CO₂. MEFs were cultured for 3 days before they were placed in low serum (0.1% FBS) for an additional 72 hours to obtain quiescent cells.

WI-38 cells were transfected by FuGene6 (Roche) with plasmid pH2BGFP-N1 (Kanda et al., 1998). Cells were plated in delta-T dishes (Biotechs) or coverslips 2 days after transfection and cultured for another 10 days before live imaging.

Staining was performed essentially as described previously (Tumbar and Belmont, 2001) with the following modifications. Cells were permeabilized in 0.1% Triton, fixed in 1.6% formaldehyde (Polysciences) at room temperature for 10 minutes, and blocked for 10 minutes with 5% NGS and 5% donkey serum. For Xi labeling, we applied mouse anti-tri-methyl histone H3Lys27 (H3-3mK27) antibody (Danny Reinberg, NYU College of Medicine, NY) diluted 1:2000. For double-staining of the Xi and nuclear membrane, rabbit anti H3-3mK27 antibody (Upstate) diluted 1:1000 was used together with mouse monoclonal antibody RL1 against nuclear pore O-linked glycoprotein (ABR-Affinity BioReagents) diluted 1:750. For double-staining of the Xi and nucleoli, mouse anti-H3-3mK27 antibody diluted 1:2000 was used together with rabbit polyclonal anti-fibrillarin (Santa Cruz Biotechnology) antibody diluted 1:100. Secondary antibodies (Texas-red-labeled donkey anti mouse, FITC-labeled goat anti rabbit) antibodies were diluted 1:500 and obtained from Jackson Laboratory. Cells were counterstained by DAPI.

Immunostained samples were observed with a 60× objective (NA=1.4) using an Olympus IMT-2 microscope. Live cells were imaged with a 100× objective (NA=1.40) using a Zeiss Axiovert 100M within 10 minutes after removal from the incubator or maintained at 37°C in a closed chamber system (FCS2; Bioptechs). Data acquisition and 3D deconvolution methods were processed as previously described (Chuang et al., 2006; Tumbar et al., 1999).

Cells were grown on glass coverslips. Several procedures designed to preserve large-scale chromatin fibers were used: (a) live cells were fixed directly in 0.1 phosphate buffer, pH 7.4, in 2% GA (Polysciences) overnight at 4°C; (b) cells were permeabilized with 0.1% Triton X-100 (Pierce) in buffer A^* (80 mM KCl, 20 mM NaCl, 2 mM EDTA, 0.5 mM EGTA, 15 mM Pipes, 0.5 mM spermidine, 0.2 mM spermine, 10 μg/ml turkey egg white inhibitor, pH 7.0) for 30-180 seconds followed by fixation in 2% GA (25% solution added dropwise) overnight at 4°C; (c) cells on ice in media were UV irradiated (9000 J/m²) for 8 minutes (Stratalinker 1800 UV crosslinker, Stratagene). EtBr (20 μg/ml) was added to media before UV treatment. After UV fixation, coverslips were permeabilized with 0.1% Triton X-100 in buffer A^* for 5 minutes and then post-fixed overnight at 4°C in 2% GA in buffer A^*.

Two UV/EtBr cross-linking fixation methods were used prior to immunostaining: cells were permeabilized for 30 seconds with 0.1% Triton X-100 in buffer A^* in the presence of 20 μg/ml EtBr followed by 4 minutes of UV irradiation on ice (4500 J/m²). Alternatively, live cells in media were placed on ice and 20 μg/ml EtBr was added prior to 8 minutes of UV irradiation (9000 J/m²).

Samples fixed by either method were washed in 0.1% Triton X-100 buffer A^*. All following steps were performed in the same buffer unless specified otherwise. Cells were blocked in 1% BSA for 1 hour at room temperature and then rabbit antibody (Upstate) against histone H3-3mK27 diluted 1:1000 in blocking buffer was applied at 4°C overnight. Specimens were washed three times followed by a second blocking step for 30 minutes at room temperature with 0.1% fish gelatin (Sigma). Incubation with secondary nanogold anti-rabbit antibody (Nanoprobes) was performed overnight at 4°C (antibody diluted in 0.1% fish gelatin, 1% BSA, 1:500). Specimens were washed three times and post-fixed in 2% GA for 1 hour at room temperature. Samples were quenched with 150 mM glycine (3× 5 minutes) and then blocked with 0.1% fish gelatin in blocking buffer for 10 minutes at room temperature. Specimens were washed (5× 2 minutes) in double-distilled H₂O before silver or gold enhancement of nanogold by HQ Silver (Nanoprobes) for 6-10 minutes or by Gold Enhancement (Nanoprobes) for 2 minutes.

All EM samples were en bloc stained with uranyl acetate (1% in H₂O) for 1 hour at room temperature, followed by dehydration in an ethanol series, and infiltration and embedding in Epon 812 (Polysciences). Embedded cells were removed from glass coverslips after boiling the Epon samples in H₂O for several minutes. Sections of 60 or 200 nm on formvar support films were stained with uranyl acetate (2% in H₂O) for 15 minutes and lead citrate (0.02% in 0.01 M NaOH) for 10 minutes. Carbon coating was applied after staining to grids used for serial section reconstructions.

For Fig. 2B, Fig. 6D and supplementary material Fig. S5C, samples were fixed in 2% GA in 0.1 M phosphate buffer. For Fig. 2C, the sample was fixed by UV/EtBr cross-linking for 8 minutes prior to permeabilization and GA post-fixation. For Fig. 2D, Fig. 3A,B, Fig. 4B, Fig. 6, and supplementary material Fig. S1, Fig. S2A-C, Fig. S4 and Fig. S5B, samples were fixed by 2% GA in buffer A^* after partial permeabilization for 30-90 seconds in buffer A^*. For Fig. 3C, sample was fixed by 10 minutes of UV irradiation with EtBr in medium. For Fig. 4C, Fig. 6A-C and supplementary material Fig. S5D, samples were fixed after quick permeabilization in EtBr/buffer A^* by 4 minutes of UV irradiation.

Sections were examined with a Philips CM 200 electron microscope operating at 120 kV. Images were acquired using a Tietz Video and Image Processing Systems GmbH 2k × 2k Peltier-cooled CCD camera and software.

LM and EM images were adjusted for brightness/contrast, superimposed, pseudocolored and assembled using Adobe Photoshop CS. Alignment of 200-nm serial sections was done manually in Photoshop and movies were created in WCIF Image J.

60-nm serial sections were aligned for 3D reconstructions using a cross-correlation alignment procedure including rotation, translation and a uniform magnification adjustment as described previously (Belmont and Bruce, 1994; Kireeva et al., 2004). Aligned and normalized images were further processed in ImageJ. A 2D median filter with a box size of 5×5 was applied to each section of a 512×512 cropped image stack. The resolution was then reduced by a factor of two by summing 2×2 pixels to produce 256×256 image sections. A cubic spline interpolation was used to produce a uniform 14-nm pixel size in x, y and z. A final 3D smoothing filter (2×2×2) in the Ortview plugin was applied and orthogonal images and stereo projections were then generated. Xi solid models were calculated using density-threshold segmentation in Matlab 7.0 using the DIPimage toolbox (Quantitative Imaging Group, Department of Imaging Science and Technology, Delft University of Technology, Delft, The Netherlands). AVI movies were created using a Camera Sequence MATLAB script (Olivier Salvado, CSIRO). ImageJ was used to convert movies to a compressed QuickTime format using QT Movie Writer plugin.

We thank Edith Heard (Curie Institute, France) for her comments on the manuscript as well as for providing MEFs. We thank Teru Kanda and Geoffrey Wahl (The Salk Institute for Biological Studies, USA) for providing the H2B-GFP expression vector (pH2BGFP-N1), as well as Danny Reinberg (NYU, College of Medicine, USA) for the anti-H3-3mK27 mouse antibody. Electron microscopy was performed in The Visualization, Media and Imaging Laboratory at the Beckman Institute, UIUC. This work was supported by HFSP RGP0019/2003 and grant number R01 GM42516 from the National Institute of General Medical Sciences (A.S.B.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of General Medical Sciences or the National Institutes of Health.