IntroductionSeveral regulatory layers control gene expression during cell differentiation. While transcription factors act on discrete elements such as promoters and enhancers to induce transcription1,2, polycomb components are able to maintain the silencing of most developmental genes outside their normal spatiotemporal expression domains3. Polycomb group (PcG) proteins belong mainly to two types of polycomb repressive complexes (PRCs), both of which are required for gene silencing. PRC2 catalyzes the methylation of H3K27, whereas PRC1 can ubiquitylate H2AK119 (H2AK118 in flies), and each of the two groups of complexes can reinforce silencing through their association with the histone mark deposited by the other group4,5. PcG proteins bind to specific sites, called polycomb response elements (PREs), in Drosophila3. PRC1 forms several variant complexes depending on the presence of alternative subunits: so-called canonical PRC1 maintains gene silencing, whereas binding of noncanonical PRC1 starts direct repression of transcription, notably by recruiting PRC26. Once recruited, PRC2 deposits H3K27me3 in the chromatin surrounding PREs, generating chromatin domains coated with H3K27me3. This mark is then recognized by canonical PRC1, which, in flies, contains the polyhomeotic (Ph) and polycomb (Pc) subunits. Transgenesis experiments have shown that PREs are DNA sequences that are necessary and sufficient for the recruitment of PcG subunits and the silencing of their flanking genes3,7. However, studies attempting to predict the genomic localization of PcG proteins and analyses of endogenous deletions of PREs have provided a complex picture, emphasizing the role of neighboring chromatin in determining the actual binding of PcG proteins to PREs8,9,10. Furthermore, H3K27me3, the histone mark deposited by PRC2, spreads across large chromatin domains, strongly suggesting that polycomb function is linked to the higher-order organization of the linear genome11,12,13. In the three-dimensional (3D) nuclear space, PRC1 subunits accumulate in structures called polycomb foci, which have been shown to colocalize with repressed chromatin covered with H3K27me314,15,16. Additionally, long-range interactions between H3K27me3 chromatin domains involve polycomb foci17,18,19,20. Although there is evidence that polycomb-dependent silent chromatin is more compacted than the same genomic regions in cells in which the corresponding genes are expressed18,21, the mechanism of compaction is more controversial22,23. Indeed, the effect of PcG proteins on chromatin compaction could be direct or indirect, as the release of gene repression could also lead to chromatin decompaction. We previously demonstrated that the deletion of PRC1 subunits affects homeobox (Hox) cluster chromatin compaction prior to any ectopic Hox gene expression during early Drosophila embryogenesis, suggesting that PRC1 directly condenses Hox clusters21. However, how the mild effect of PRC1 on Hox cluster compaction maintains Hox gene silencing, as well as the PRC1-induced compaction mechanism, remain enigmatic. Recent studies have reported that some canonical PRC1 subunits can make liquid-like droplets in vitro, suggesting that polycomb foci correspond to condensates formed by phase separation24,25,26,27,28,29. Interestingly, this biochemical property would explain the formation of higher-order structures and suggests that polycomb function does not rely solely on the direct binding of its components to chromatin but may be driven by self-association properties. However, the lack of optical microscopy resolution has hampered the fine characterization of these nuclear structures in the past, making it impossible to link the linear distribution of PcG proteins along the genome with their precise 3D localization relative to the associated chromatin.The polycomb machinery acts during cell differentiation when chromatin is extensively reorganized by modifications of histone marks and changes in chromatin 3D organization8,30. Here, we analyzed the higher-order organization of canonical PRC1 components in the nuclear space in relation to two of their best characterized Drosophila target loci, the BX-C and Antennapedia complex (ANT-C), clusters of Hox genes31,32. These two gene clusters host the two largest H3K27me3 domains, contain many PREs and appear under microscopy as the most intense polycomb foci in the cell nuclei of Drosophila embryos21,32. Interestingly, H2AK118Ub was shown to be dispensable for the function of polycomb on Hox genes33, which allows one to focus specifically on the relationship between H3K27me3 and polycomb components when these loci are being studied. A further advantage is the large size of both clusters (over 300 kb each), which reduces the challenge posed by the limited spatial resolution that characterizes even superresolution microscopy methods. Here, we used stimulated emission depletion (STED) microscopy to study the 3D nuclear localization of Ph and Pc relative to repressed Hox chromatin in Drosophila embryos. Our data show that the largest polycomb foci are composed of multiple substructures approximately 70 nm in diameter, which we named “PRC1 nanoglobules”. We confirmed their presence in living embryos using an orthogonal superresolution microscopy method called AiryScan microscopy and showed that PRC1 nanoglobules move rapidly and reshape constantly. Immunofluorescence in situ hybridization (immuno-FISH) experiments revealed that PREs are more often associated with PRC1 nanoglobules than the remaining H3K27me3-coated Hox chromatin. Importantly, through biophysical chromatin modeling using polymer simulations, we linked the genomic position of PcG proteins and H3K27me3 to the localization of polycomb target chromatin relative to PRC1 nanoglobules and predicted Ph-dependent compaction of the BX-C locus. Taken together, the results of this study demonstrate that PRC1 subunits form higher-order structures that fold polycomb-associated chromatin.ResultsIntense polycomb foci are composed of multiple PRC1 nanoglobulesThe observation of large H3K27me3 genomic domains and the ability of several PRC1 subunits to phase separate suggest that the organization of PRC1 in the 3D nuclear space is essential for PRC1 function. In particular, the Ph subunit of the Drosophila PRC1 complex has been shown to form large condensates in vitro24,25,26,27,28, suggesting that it may play a crucial role in the formation of nuclear foci. To test this hypothesis, we generated Drosophila lines in which a GFP tag was added to the endogenous Psc, Sce or Pc genes by CRISPR-Cas9 knock-in, and we imaged the nuclear distribution of Psc-GFP, Sce-GFP and Pc-GFP in ph null mutant embryos. As expected, the accumulation of the three PRC1 subunits within nuclear foci was greatly reduced in mutant embryos (Supplementary Fig. S1a), demonstrating that the Ph subunit is necessary for the formation of PRC1 foci. To further investigate the ability of PRC1 proteins to form large, liquid-like condensates inside the cell nucleus, we imaged Drosophila embryos using superresolution microscopy. If the large foci observed by confocal microscopy correspond to standard liquid-like droplets, STED microscopy should reveal large foci that should not separate into disconnected substructures. Instead, Ph or Pc immunostaining imaged by STED microscopy revealed that many polycomb foci are composed of substructures ~70 nm in diameter and characterized by a heterogeneous signal intensity, which suggests that they contain a variable number of PcG molecules. We named them “PRC1 nanoglobules” (Fig. 1a). To describe their distribution within the cell nucleus, we defined clusters of nanoglobules as sets of signals that were less than 140 nm apart. While the majority (approximately 70%) of the nanoglobules were isolated, the others were grouped into clusters of several nanoglobules (Supplementary Fig. S1b).Fig. 1: Large polycomb foci are composed of several PRC1 nanoglobules.Full size imagea Two examples of confocal, STED, and merged Ph and Pc immunostaining images of the head of Drosophila embryos. A single optical section is shown in 2D images, whereas the 3D view corresponds to a projection of 5 sections. The bars measure 1 µm. b Scatterplot showing the average maximum distance between nanoglobules of one focus and its number of nanoglobules for both Ph and Pc foci. c Scatterplot showing the number of nanoglobules and the minimum distance between nanoglobules of one focus for both Ph and Pc foci. d Five examples illustrating the effect of resolution on images of large polycomb foci. Centers of nanoglobules identified by processing STED images (first row) can be used to compute images using an artificial p.s.f. of 70 (same resolution as the STED images), 140 (same resolution as the AiryScan images), and 300 nm (same resolution as the confocal images). The bars represent 500 nm. e Two time-lapse imaging showing one picture every 0.1 s of large Ph foci imaged in the head of Drosophila embryos expressing Ph-GFP with AiryScan microscopy. The shape and internal organization of large Ph foci change rapidly over time, indicating that they are composed of several mobile substructures (arrowheads). The bars represent 500 nm. f Confocal and STED images of a cell nucleus located in the head of a Drosophila embryo illustrating the colocalization of Ph and Pc. The bar represents 1 µm. g Three examples showing the colocalization of Ph and Pc in large polycomb foci imaged in confocal and STED microscopies. The bars represent 500 nm.To characterize the internal structure of the most intense polycomb foci, we quantified the STED signals contained within the areas previously defined by the segmentation of large polycomb foci in confocal images (Supplementary Fig. S1c). In the heads of embryos, where all Hox genes are repressed, the most intense polycomb foci, which generally correspond to the largest polycomb domains in the genome, namely, the BX-C and ANT-C Hox clusters15, showed high morphological variability, ranging from 4 to 20 nanoglobules separated by a maximum distance between them of 200 to 800 nm (Fig. 1b). Interestingly, the minimum distance between nearest-neighbor substructures was most often less than 200 nm, i.e., lower than the spatial resolution of confocal microscopy, which explains why superresolution is required to distinguish them (Fig. 1c). To model the effect of spatial resolution on the visualization of large polycomb foci, we used the coordinates of each nanoglobule identified after the quantification of the STED images and artificially reconstructed the corresponding images that such substructures would produce when using simulated point spread functions (p.s.f.) of 70, 140, and 300 nm (Fig. 1d). Modeling using a p.s.f. of 70 nm produced virtual images that well recapitulated experimental STED images, whereas a p.s.f. of 300 nm mimicked confocal images. Interestingly, an intermediate resolution of 140 nm could still capture some substructures within polycomb foci (Fig. 1d), suggesting that imaging at an intermediate resolution would be sufficient to identify the composite nature of large polycomb foci. To exploit this feature and track the dynamics of polycomb foci in vivo, we generated a Drosophila line in which a GFP tag was added to the endogenous ph-d gene (one of the two paralogous genes coding for the Ph subunit of PRC1) by CRISPR-Cas9 knock-in. Timelapse AiryScan microscopy experiments on live embryos revealed that intense polycomb foci do not form homogenous and immobile structures. Instead, they are composed of several substructures moving rapidly relative to one another (Fig. 1e). Individual substructures could be observed in each frame of the timelapse experiment, suggesting the presence of fusion and demixing events between them. Importantly, the observation of substructures by live microscopy ruled out that PRC1 nanoglobules are artifacts due to the formaldehyde fixation or the immunostaining process, nor are they microscopic aberrations due to the procedure of STED imaging.We observed nanoglobules inside the most intense foci for multiple subunits of PRC1, namely, Ph, Pc and Psc (Supplementary Fig. S1d). To test whether different PRC1 subunits colocalize within these nanostructures, we performed two-color immunostaining to detect both the Ph and Pc subunits by STED microscopy upon imaging of embryo heads. As previously shown15, the two proteins were found to colocalize by confocal microscopy (Fig. 1f, g; Supplementary Fig. S1e). Similarly, the Ph and Pc nanostructures showed a high degree of overlap by STED microscopy (Fig. 1f, g; Supplementary Fig. S1e), indicating that the intense polycomb foci are composed of nanoglobules containing multiple subunits of PRC1. Taken together, these observations demonstrate that intense polycomb foci are composed of multiple PRC1 nanoglobules in Drosophila embryos.PRC1 nanoglobules associate more frequently with PREs than with intermediate regionsIn Drosophila, chromatin immunoprecipitation (ChIP) experiments showed that polycomb-associated chromatin is composed of discrete PREs, where PRC1 and PRC2 proteins bind strongly, flanked by large regions covered with H3K27me3 with weaker PcG protein binding11,12, which we call “intermediate regions” in this study. However, the 3D organization of PREs and flanking regions relative to the nuclear localization of polycomb complexes is unknown. To test whether these two types of chromatin regions have a different localization with respect to PRC1 nanoglobules, we designed ten 6 kb probes targeting 6 PREs (3 in BX-C and 3 in ANT-C) and 4 intermediate regions (Fig. 2a and Supplementary Table S1). We used them to perform Ph or Pc immuno-FISH experiments with Drosophila embryos and examine the embryo head (Fig. 2b). The observation of 2 or 3 FISH spots (Fig. 2c) and rarely 4 (Supplementary Fig. S5b) associated with a single polycomb focus suggested that paired homologous chromosomes and sister chromatids can be distinguished by STED microscopy. To quantify the localization of PREs and intermediate regions relative to Ph- or Pc-immunolabeled regions, we calculated the minimum distance between the center of FISH spots and the boundary of Ph or Pc substructures: distances are negative when the FISH spot centers were inside immunolabeled structures and positive when they were outside. PREs located in the middle of Hox clusters were significantly more associated with Ph nanostructures than intermediate regions (Fig. 2d, f, h). Similar results were observed with Pc nanostructures, with the exception of the intermediate region int4, for which the data indicate a lower interaction frequency, but the difference was not significant (Fig. 2e, g, h). Moreover, bx and dAntp, two PREs located near the H3K27me3 domain boundary, exhibited an intermediate distribution (Fig. 2f–h), indicating that the PRE position in Hox clusters might also affect their localization relative to PRC1 nanoglobules.Fig. 2: Compared with intermediate regions, PREs interact more with Ph/Pc substructures.Full size imagea Genomic maps showing the H3K27me3, Ph and Pc profiles observed during mid-embryogenesis37 and the localization of FISH probes in BX-C and ANT-C. PRE probes are shown in blue, whereas probes for intermediate regions are shown in green. b Scheme of a fly embryo representing the pattern of Hox gene expression along the antero-posterior axis. c Confocal, STED, and segmented (quant.) images of a PRE (green) compared with Ph foci (red) acquired in the head of Drosophila embryos. STED images show that a single FISH signal (arrow) acquired by confocal microscopy can be composed of two spots (arrowheads). Moreover, STED images of immunolabeling can be robustly segmented (red objects in quant. pictures). The bars represent 500 nm. d, e Confocal, STED, and segmented (quant.) images illustrating the localization of one PRE (bxd) or one intermediate region (int2) relative to Ph (d) or Pc (e) immunolabeling. STED and segmented images show that int2 is less associated with Ph or Pc immunolabeling. The bars measure 500 nm. f, g Violin plots showing the distribution of minimum distances measured between the center of FISH spots and the border of Ph (f) or Pc (g) substructures in embryos from stages 9 to 13 (mid-embryogenesis). Distances are negative when FISH spots are inside Ph (or Pc) substructures and positive when they are outside. With the exception of bx and dAntp (light blue), which have an intermediate distribution, PREs located in the middle of the Hox clusters (dark blue) are more associated with Ph or Pc substructures than intermediate regions (green). h Table of statistical significance from the pairwise comparisons of the distance distributions shown in panels f (Ph: upper triangle of the matrix) and g (Pc: lower triangle of the matrix). Relative to Ph substructures, PREs located in the middle of Hox clusters (dark blue) and intermediate regions (green) form two groups with different distance distributions (red). Compared with Pc substructures, intermediate regions show more variation. Nevertheless, with the exception of int4, the 3 other intermediate regions are significantly less associated (red) with Pc staining than are the PREs located in the middle of the Hox clusters. ns, not significant; *P