{"paper_id":"1cd78d4d-a10d-441c-a876-de01b016157d","body_text":"1 \nLoss of SUMOylation drives aberrant PRC1 clustering and 3D genome rewiring independent of \nH3K27me3 \n \nNazli Akilli1, Paul-Swann Puel2, Marco Di Stefano1, Fernando Muzzopappa3, Lauriane Fritsch1, Fabien \nErdel3, Daniel Jost2, Thierry Cheutin1, Giacomo Cavalli1 \n \n \n1 IGH, UMR9002, University of Montpellier, CNRS, INSERM , 141 Rue de la Cardonille, 34396, \nMontpellier Cedex 5, France \n2 Laboratoire de Biologie et Modélisation de la Cellule, École Normale Supérieure de Lyon,CNRS, \nUMR5239, Inserm U1293, Université Claude Bernard Lyon 1 \n3 MCD, Center for Integrative Biology (CBI), University of Toulouse, CNRS, Toulouse, France \n \n* Contact Information \ngiacomo.cavalli@igh.cnrs.fr  \n  \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n2 \nAbstract \n \nPolycomb Repressive Complex 1 (PRC1) forms nuclear condensates that organize target chromatin \ndomains. SUMOylation modulates PRC1 clustering, but its impact on condensate properties and 3D \ngenome architecture remains unclear. Here, we show that depletion o f SUMO in Drosophila wing \nimaginal discs transforms PRC1 condensates into large structures  with reduced molecular dynamics. \nStrikingly, this biophysical reorganization occurs without global loss of the H3K27me3 mark. Instead, \nHi-C reveals widespread rewiring of topologically associating domain (TAD) interactions. PRC1-bound \nTADs lose specific l ong-range contacts with each other while gaining ectopic interactions with active \nchromatin. These topological shifts correlate with gene misregulation independently of changes in \ncanonical Polycomb histone modifications. Our results establish SUMOylation as a critical regulator of \nPRC1 condensates, demonstrating that post -translational control of biomolecular condensation \ndictates 3D genome architecture and transcriptional output through mechanisms separable from \nhistone mark deposition. \n \n \n \n  \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n3 \nIntroduction \nPolycomb group (PcG) proteins are evolutionarily conserved chromatin regulators that establish \ntranscriptional silencing through the formation of repressive chromatin compartments. Canonical \nPolycomb repressive complex 1 (cPRC1), composed of core proteins Pc, Ph, Psc and Sce, contributes \nto chromatin compaction, histone H2A ubiquitylation, and long -range chromatin interactions that \ncoordinate gene repression 1. Recent studies have revealed that PRC1 can assemble into nuclear \ncondensates, potentially via phase separation, thereby influencing genome topology and \ntranscriptional memory2-6. PRC1 condensates coincide with clustered genomic targets, highlighting a \nclose link between condensate formation and PRC1-mediated chromatin organization7-9. \nIn vitro studies on reconstituted nucleosome arrays have shown that both Drosophila and mammalian \nPRC1 can compact chromatin into dense aggregates that compete remodeling by Swi/Snf complexes \n10. Chromatin compaction is mediated by intrinsically disordered regions (IDRs) of PRC1 subunits, \nincluding Psc in Drosophila and CBX2 in mammalian cPRC1 1,11-13. While these protein domains are \ncentral to phase separation, PRC1 is also subject to post -translational modifications such as \nglycosylation and SUMOylation, which can regulate the dynamics of membraneless nuclear \ncondensates14-16. However, the contribution of such modifications to PRC1 phase behavior remains \npoorly understood. \nSUMOylation involves the covalent attachment of a small ubiquitin -like modifier (SUMO) polypeptide \nto lysine residues of target proteins, altering their interactions and nuclear organization17-18. Pc, a core \nsubunit of PRC1, has been shown to be SUMOylated in the nucleus, and reduced SUMOylation leads \nto the formation of unusually large Pc foci of ~1 μm compared with less than 500 nm under normal \nconditions15. Similarly, the PRC1 subunit Scm is SUMOylated, and perturbation of SUMO levels \ninduces Scm-dependent changes in target gene expression14. Despite these observations, the precise \ncontribution of SUMOylation to PRC1 phase behavior and the resulting impact on chromatin \narchitecture and transcriptional outcomes remains unclear. \nHere, we investigate how integrity of the nuclear SUMOylation pathway modulates the phase behavior \nand functional output of PRC1. We hypothesized that if hypo-SUMOylated PRC1 condensates remain \nfunctionally competent, then SUMOylation may act as a molecular switch controlling PRC1 clustering \nand chromatin organization.  Using RNAi-mediated depletion of SUMO in Drosophila wing discs, we \nexamined how this global perturbation impacts PRC1 foci formation, chromatin topology, and gene \nexpression. Our results reveal that SUMO -dependent regulation of PRC1 clustering is a key \ndeterminant of PcG -mediated chromatin organization, highlighting the profound architectural \nconsequences of disrupting the SUMO pathway, which correlates with changes in gene expression \nwithin PcG target domains.  These findings have  potential implications for developmental gene \nregulation and the dynamic control of nuclear architecture during cell  physiology and differentiation. \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n4 \nResults \nLoss of SUMOylation transforms PRC1 condensate morphology and dynamics \nTo decipher the mechanism of Polycomb foci formation and address the role of PRC1 phase \nseparation, in vivo, we took advantage of the previously reported observation that Pc-GFP accumulates \nin large, intense nuclear foci upon SUMO inactivation15. We compared the formation of Polycomb foci \nin fly lines that constitutively expressed Pc -GFP where we either induced a SUMO -KD or \noverexpressed Pc-GFP in the pouch of wing imaginal discs. Compared to control discs, Airy Scan \nimaging confirmed that SUMO-KD induces the formation of few, large (frequently a single one), intense \nPolycomb foci per cell nucleus. On the other hand, overexpression of Pc -GFP increases the intensity \nof all Polycomb foci ( Fig. 1a) without merging them into fewer and larger ones. This result indicates \nthat the formation of large Polycomb foci in SUMO -KD does not simply result from an increase in the \nnuclear concentration of Pc -GFP, but suggests that SUMO -KD specifically affects the  condensation \nproperties of Pc-GFP.  \nSince phase behavior is often concentration -dependent19,20 , we asked whether SUMOylation affects \nPc protein abundance. To estimate the free amount of nuclear Pc-GFP, i.e. the fraction of Pc-GFP not \nlocated in Polycomb foci, we measured the median intensity of Pc -GFP in the cell nuclei of control, \nSUMO-KD and Pc -GFP overexpressing wing discs. In addition to the formation of few large intense \nPolycomb foci, SUMO -KD also sign ificantly reduced the level of Pc -GFP outside foci, compared to \ncontrol conditions (Fig. 1b), suggesting that SUMO inactivation favors the association of Pc-GFP with \nfoci. On the other hand, Pc-GFP overexpression increases the level of free Pc-GFP in cell nuclei (Fig. \n1b). To quantify the accumulation of Pc -GFP in Polycomb foci, we computed local maxima to detect \nPc-GFP foci and then ranked their intensity in descending order to compare the control, SUMO -KD \nand overexpression conditions. The intensity of Pc-GFP in all nuclear foci is higher than that observed \nin control wing discs when Pc -GFP is overexpressed. Conversely, only a few Pc -GFP foci \n(approximately 5%) show a higher intensity in SUMO -KD wing discs than in control discs. The \nremaining foci show a lower intensity in SUMO-KD, suggesting that most of the Pc-GFP accumulates \nin a few intense foci ( Fig. 1a-c). Thus, SUMO loss alters condensate properties rather than simply \nstabilizing Pc protein. To verify that the localization of Pc-GFP reflects the nuclear distribution of PRC1, \nwe performed immunolabelling experiments using the PRC1 subunit Ph and counterstained them with \nDAPI. Super-resolution Airyscan imaging of both control and SUMO-KD wing discs shows a strong co-\nlocalization between Ph and Pc subunit in both conditions, demonstrating that the large, intense \nPolycomb foci observed upon SUMO -KD correspond to PRC1 structures ( Fig. 1c). In addition, DAPI \nsignal also co -localizes with the large PRC1 foci upon SUMO -KD, confirming their chromatin -\nassociated nature rather than off-chromatin aggregation (Fig. 1c-d, S1b-c). \nPRC1 condensates have been proposed to form via liquid–liquid phase separation (LLPS)2-7,12,13,21,22. \nTo specifically test this point, we next asked whether SUMOylation regulates Pc-GFP phase behavior \nby performing calibrated half -FRAP (MOCHA-FRAP)23. The purpose of this assay is to distinguish \nbetween liquid -like condensates with a significant interfacial barrier (characterized by a dip in the \nunbleached half curve due to intermixing) and condensates where exchange with the nucleoplasm \ndominates, reflecting assemblies formed by simple chromatin binding or by polymer -polymer phase \nseparation. Analysis of PRC1 foci revealed that the large majority lacked the characteristic dip of liquid-\nlike condensates (Fig. 1e), suggesting restricted internal dynamics and a shift toward a gel-like or solid-\nlike state upon SUMO loss. Consistent with this interpretation, FRAP analysis revealed a mobile \nfraction of only ~30% in PRC1 condensates ( Fig. 1e ), indicative of a stable scaffold with limited \nmolecular exchange. However, ~14% of foci did show a small dip, suggesting that a sub-population of \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n5 \nlarge Pc-GFP foci do have liquid -like properties upon SUMO KD. These experiments could only be \nperformed on the wing discs where SUMO RNAi is induced and not in control wing discs because Pc-\nGFP foci are too small to be half -bleached in this tissue. Howeve r, upon SUMO depletion PRC1 \ncondensates transform into clearly larger, morphologically distinct structures with a lower internal \ndynamics than control15, without the surface tension that would be expected from a phase -separated \nliquid structure (Fig. 1e). The observation of  liquid behavior in few cases suggests the possibility of a \ndynamic regulation of the properties of PRC1 foci during the accumulation of Pc-GFP upon SUMO KD.    \n \nFigure 1: SUMO RNAi leads to unusually big PRC1 foci formation. (a)  2 µm thick projections showing the \nnuclear distribution of Pc -GFP in win g discs of control, SUMO -KD and overexpressing Pc -GFP flies. The \npseudo-color images show that few Pc -GFP foci become very intense in the SUMO -KD, while the intensity of \nall foci increases when Pc-GFP is overexpressed. Scale bar: 1 µm. (b)  Box plots comparing median Pc-GFP \nintensities in cell nuclei of control (grey), SUMO -KD (red) and overexpressing Pc-GFP (blue) wing discs (left). \nCompared to control wing discs, the intensity of nucleoplasmic Pc -GFP significantly decreases in SUMO -KD, \nwhereas it strongly increases when Pc-GFP is overexpressed. Cumulative histograms comparing the intensity \nof Pc-GFP foci in control (grey), SUMO-KD (red), and Pc-GFP-overexpressing (blue) wing discs (right). Pc-GFP \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n6 \nfoci are ranked in descending order of intensity, and the points corresponding to the first, second, fifth, tenth, \ntwentieth, and fiftieth percentiles are shown.(c) Airyscan images of Immuno-staining against Ph, Pc and DAPI \nstaining from left to right, in control sample (top) and SUMO RNAi sample (bottom).  (d) Quantification of the \nimmunostaining in Control and SUMO RNAi samples Pc (green), Ph (red), DAPI (blue). Normalized intensity \nindicates the intensity of the signal within the largest focus in the nucleus, divided by the total intensity inside \nthe nucleus for the respective signal. (e)  Average MOCHA-FRAP curve of all the analyzed foci (n=34, dip=0.11, \np.value=0.44). The dip depth with a p-value above 0.05 indicates that the majority of the PRC1 foci do not show \nliquid-like behavior with an interfacial barrier (left). MOCHA-FRAP curves of a subpopulation of 5 foci (that are \nincluded in the average curve shown on the left) that exhibit liquid-like behavior with an interfacial barrier (n=5, \ndip=0.37, p.value=0.01) (right). Solid lines represent the theoretical recovery for an internally mixed condensates \nwith a Pc-GFP enrichment of Γ=10 and a vanishing permeability of κ*=0.001. In both panels, the curve in purple \nindicates the change in intensity of the non -bleached half of the focus, and the curve in green indicates the \nrecovery of the intensity of the bleached half of the focus. The fraction of immobile molecules in the focus c an \nbe detected by the final distance between the purple and the green curves, as indicated in the panel. \n \n \nChanges in PRC1 self-interactions can explain the aberrant clustering behaviour of PRC1 upon \nSUMO-RNAi.  \n \nTo investigate how SUMO depletion can lead to the formation of large  PRC1 condensates, we \ndeveloped a generic biophysical framework coupling the Liquid -Liquid Phase Separation (LLPS) \nproperties of PRC1 molecules with chromatin mechanics (see Methods). We considered a coarse \ngrained polymer representation of Drosophila chromosome 3R at 1 kbp resolution (29 Mbp), featuring \nmultiple H3K27me3 marked domains that can specifically interact with 𝑁! diffusible self -attracting \nPRC1 molecules ( Fig.2a). Using this framework, we systematically investigated multiple parameter \nsets to characterize the formation of PRC1 foci and its impact on chromosome organization, in control \nand perturbed conditions (Fig.2b-c, Fig.S2).  \n \nIn particular, we tested the impact of the strengths of homotypic attractions between PRC1 complexes \n(𝐸!\"!) and of heterotypic interactions between PRC1 and H3K27me3 -marked loci (𝐸!\"#) for different \nconcentrations of PRC1 (given by the ratio 𝑅!/# between 𝑁! and the number of H3K27 -marked \nmonomers). As 𝐸!\"! increases (Fig.2b, left), the average number of foci decreases (top panel) with \nthe appearance of larger foci at the expense of smaller ones (middle), while the free nucleoplasmic \nfraction of PRC1 is strongly reduced (bottom). As 𝐸!\"# decreases, foci are in average less numerous \nand larger and the free nucleoplasmic fraction of PRC1 increased ( Fig.S2). All these effects are \nconserved regardless of the values of the other parameters (Fig.S2).   \n \nTherefore, the observations made by microscopy in SUMO RNAi conditions, i.e. the emergence of \nfewer but larger foci in a darker nucleoplasmic background (see Fig 1) are consistent with an increase \nof the strength of self-attraction 𝐸!\"! (Fig.2c). To further validate the model, we investigate the effect \nof an overexpression of PRC1 by varying 𝑅!/#(Fig.2b,c, right panels and Fig.S2).  We predict an \noverall increase in the free nucleoplasmic fraction and in the number of condensates that become \nlarger homogeneously, in perfect accordance to what was experimentally observed  ( Fig. 1). Taken \ntogether, polymer simulations indicate that homotypic interactions between PRC1s are necessary to \nexplain the formation of large, intense PRC1 foci observed upon SUMO inactivation, providing indirect \nevidence that PRC1 phase separation also occurs in vivo. \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n7 \nIn addition, all this suggests that SUMOylation may act as a solubility factor buffering the interactions \nbetween PRC1 molecules. Its removal  leads to increased PRC1 -PRC1 interactions, aberrant PRC1 \nclustering and the formation of large PRC1 foci. \n \n \n \nFigure 2: PRC1 self-interaction recapitulates the effect of SUMO RNAi. (a)  Scheme of the chromatin and \nPRC1 simulation. Self -interacting PRC1 complexes, (in green, 𝐸!\"!) exhibit interaction for H3K27 -marked \nmonomers (𝐸!\"#, in red). Steric hindrance ( 𝐸%#) between PRC1 and monomer (both black and red) is taken \ninto account. The 3R chromosome model is modeled from approximately 5 Mbp to 30 Mbp with methylation \nprofiles approximating the size and location of PcG region in chromosome 3R of Drosophila. (b) Dependence \nof the number of foci, their volume distributions (in number of PRC1) and the PRC1 free nucleoplasmic fraction \non the PRC1 energy of self-interaction 𝐸!\"! and on the ratio between PRC1 complexes and to H3K27 marked \nmonomer 𝑅!/#. The ratio 𝑅!/# is first fixed to 50% for the mapping of 𝐸!\"!. The energy 𝐸!\"! is fixed to 1 kBT \nfor the mapping of 𝑅!/#. The teal configuration is considered to be representative of the wild type situation. The \npurple and the orange configuration respectively are examples of the SUMO RNAi and the over -expressed \nconfiguration. (c) Example of the final frame for each configuration described in (b). \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n8 \nLoss of SUMOylation leads to changes in gene expression independent of H3K27me3. \n \nGiven the dramatic reorganization of PRC1 condensates, we next asked whether SUMO depletion \nalters gene expression. We performed bulk RNA -seq on SUMO RNAi wing discs. Because SUMO is \nrequired for diverse cellular processes essential for viability 24-25, we first confirmed that the tissues \nused were not undergoing widespread apoptosis. RNA levels of canonical apoptotic markers in \nDrosophila, including rpr, Dronc, Dcp-1, AIF, Alg-2, and hid, showed no significant upregulation at the \ndevelopmental stage analyzed ( Fig. S3b )26,27. Consistently, SUMO -depleted wing discs appeared \noverall healthy, although flies with SUMO RNAi either failed to hatch after the pupal stage or emerged \nwith severe wing defects, suggesting that apoptosis might occur later in development. A modest \nincrease in rpr expression is consistent with these tissues being primed for apoptosis while remaining \nhealthy and viable at the time of analysis 26,27. RNA -seq analysis further  revealed extensive \ntranscriptional changes, with 700 significantly differentially expressed (DE) genes, including known \nPolycomb targets such as vg, Hr51 and ftz (Fig. 3a). We did not detect deregulation of the canonical \nPcG target gene Ubx in the wing discs, however, gene set enrichment analysis (GSEA) indicated a \ncoordinated shift toward an active state of cellular growth and developmental reprogramming (Fig. 3b \nand S3), consistent with a deregulation of the PcG machinery 28.  \nWe next asked whether these expression changes were associated with alterations in chromatin \nfeatures. To address this point, we performed CUT&RUN profiling for Pc, H3K27me3, H2Aub118, and \nH3K27ac. Pc occupancy was mildly decreased across the genome and the Pc signal appeared slightly \nmore diffusely distributed around TSSs in the SUMO-depleted condition (Fig. 3c). This reduction in Pc \nbinding was observed even at PcG target genes that were transcriptionally downregulated (Fig. 3e). \nOn the other hand, the mild reduction in Pc occupancy was not accompanied by widespread changes \nin canonical Polycomb histone modifications. H3K27me3 enrichment was largely stable genome-wide, \nwith only three novel sites appearing upon SUMO depletion (Fig. 3c-d and Fig. S4). PcG target genes \nalso maintained their H3K27me3 signal ( Fig. 3e). H2Aub118 showed a slight reduction around PcG \ntarget TSSs, but this trend was not statistically significant (Fig. 3d-e). Instead, H2Aub118 levels were \nincreased near the TSSs of non-differentially expressed genes (Fig. 3d). As expected, H3K27ac levels \nwere increased in SUMO RNAi compared to Control at TSSs of upregulated genes, but not at \ndownregulated genes. However, we also detected a slight increase at unchanged genes ( Fig. 3c-e) \nand importantly, no differences were found when we focused specifically on PcG target genes. Of note, \nthe changes described above were modest, ar guing against a major instructive role for histone \nmodifications to drive changes in gene expression. \nTogether, these results indicate that the transcriptional changes triggered by SUMO depletion, \nespecially down-regulation, do not follow canonical correlations with Polycomb histone marks. This \nsuggested the hypothesis  that altered PRC1 clustering may contribute changes in gene expression \nthrough mechanisms not directly dependent on H3K27me3 or H2Aub118 levels. \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n9 \n \nFigure 3: SUMO RNAi leads to gene expression changes without changes on H3K27me3 and H3K27ac \nmarks. (a) Volcano plot of the differential gene expression (DE) analysis between SUMO RNAi and control \nsample; (p-adjust < 0.05 and absolute Fold-Change > 1.5).  The PcG target genes are highlighted. (b) GSEA of \nall the DE genes. Top 10 enriched processes are represented. (c) Heatmaps of Pc, H2Aub118, H3K27me3 and \nH3K27ac marks around the TSSs of differentially expressed genes. Up TSSs= Transcription start site of up \nregulated genes; Down TSSs = Transcription start site of down regulated genes. (d-e) Comparisons of the \nnormalized C&R signals around the TSSs of all (d) and PcG-target (e) genes between Control and SUMO RNAi. \nGrey indicates the TSSs of the genes which didn’t show a gene expression change; blue indicates the TSSs of \nthe genes which were u p-regulated; pink indicates the TSSs of the genes which were down -regulated. Dash \nlines indicate the median of the control sample.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n10 \nInter-TAD contacts are rewired upon loss of SUMOylation \n          \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n11 \nFigure 4: SUMO RNAi leads to genome-wide re-organization of the genome. (a) Heatmap of  Hi-C scores \nin Control and SUMO RNAi samples for the genomic region around the dac locus. Colors towards red indicate \nhigh contact scores, colors towards dark blue indicate low contacts. C&R tracks for H3K27me3, Pc, H3K27ac, \nand H3K9me3 marks in control and the SUMO RNAi sample shown in the bottom. We used ChromHMM27 to \nassign chromatin states (colors) to each TAD in the genome ( Methods). Chromatin states (colors) include: \nActive (red) chromatin enriched by H3K27ac, Null (black) chromatin not enriched in any features used for the \nanalysis, HP1 (green) chromatin enriched by H3K9me3, and Polycomb (blue) chromatin enriched by \nH3K27me3. Each square indicates a TAD called and the color indicates the state. An example of increased \nTAD-TAD interaction between a PcG and Null TAD highlighted by the dashed lines. (b) Heatmap of the average \ninter-TAD score between the four chromatin states in Control and SUMO RNAi samples . (c) Quantification of \nthe change of inter and intra-TAD contacts of each TAD type pair between control and SUMO samples. P-values \nare shown above each violin plot pair. p-values were corrected for multiple testing using Benjamini -Hochberg \n(BH) method. \n \nThe lack of major changes in H3K27me3 and H2Aub118, even at differentially expressed genes, \nsuggested that the altered transcription upon SUMO depletion cannot be explained only by the altered \nlevels of PRC1 or changes on SUMOylation of transcription facto rs. We hypothesized that SUMO \nRNAi-dependent changes in the nuclear organization of PRC1 might also alter  the higher -order \norganization of PcG targets and contribute to gene expression changes. PcG domains are known to \nengage in long -range interactions between repressive TADs, often forming so -called “TAD cliques” \nthat facilitate coordinated repression of target genes 29-30. These PcG-associated TADs are typically \nenriched for H3K27me3 and depend on PRC1 for their long -range connectivity. We therefore asked \nwhether SUMOylation contributes to 3D chromatin architecture. \nTo address this, we performed Hi -C experiments to obtain a genome -wide view of chromatin folding \nupon SUMO depletion ( Fig. 4a). To systematically assess changes between the two conditions, we \nsegmented the genome into physical domains (TADs) . We classified them in different classes based \non histone mark enrichment using ChromHMM ( Methods and Fig. 4a-c): Active TADs (H3K27ac -\nenriched), Polycomb TADs (H3K27me3 -enriched), constitutive heterochromatic TADs (H3K9me3 -\nenriched), and null TADs lacking specific marks ( Fig. 4b)31. We then quantified intra - and inter-TAD \ncontacts in control and SUMO RNAi samples. \nThis analysis revealed striking rewiring of inter -TAD interactions upon SUMO depletion. All TAD \nclasses exhibited increased interactions with active chromatin, including Active–Heterochromatic and \nActive–Null contacts ( Fig. 4b-c). In contrast , homotypic interactions of Null chromatin regions were \nreduced, while PcG-PcG and Het -Het ( e.g., Heterochromatin -Heterochromatin) contacts were \npreserved and, finally, Active–Active interactions became significantly more frequent . These results \nindicate that the effect of SUMO depletion extends beyond Polycomb chromatin and induces ectopic \ninteractions with active chromatin, as observed between Active-PcG domains. \nThus, the deregulation of gene expression in SUMO -depleted wing discs in the absence of major \nchanges to histone marks can be associated with the  rewiring of TAD–TAD interactions. Concerning \nPolycomb chromatin, loss of SUMOylation may compromise the insulated chromatin environments \nnormally maintained by PRC1 condensates, leading to exposure of PcG targets to neighboring active \ndomains and a reconfigured regulatory landscape. \n \nSUMO KD leads to changes in PcG TAD interactions that correlate with PcG target gene \nexpression \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n12 \nTo further quantify and characterize the differential 3D chromatin interactions TADs, we analyzed \ncontacts that each TAD makes with all other TADs.  \nWe first focused our analysis on the PcG -PcG contacts. In order to  analyze the re lative changes in \nPcG chromatin contacts upon SUMO RNAi, we first divided PcG TAD pairs into quartiles based on the \nfrequency of contacts made by the underlying PcG TADs. We then compared PcG TAD pair contacts \nbetween the control and the SUMO RNAi conditions, focusing on three categories  (Fig. 5a): the entire \ndistribution of inter-TAD contacts (All), the pairs that are weakly interacting , which are the pairs with \nlittle or no mutual contacts in controls (Q1 - 1st quartile), and the pairs that normally engage in strong \ninteractions (Q4 - 4th quartile). \nWhile the entire distribution of inter-PcG contacts showed no significant change, both the bottom and \ntop quartiles displayed significant shifts (Fig. 5b). In particular, TAD pairs that had low numbers of PcG \ncontacts in control tissues increased their contact frequencies upon SUMO RNAi, whereas TAD pairs \nwith high PcG contact frequencies in controls decreased their contacts upon SUMO RNAi.  This \nsuggests that subsets of PcG domains alter their connectivity upon SUMO loss, potentially reshaping \nco-regulated gene groups. We then extended our analysis to heterotypic contacts between PcG TADs \nand other chromatin states (PcG –Active, PcG –Het, PcG –Null). PcG –Active contacts increased \nsignificantly (Fig. S5), consistent with the broader gain of interactions between Active domains and \nnon-Active TADs observed in the genome -wide analysis ( Fig. 4b and c). This reorganization likely \nperturbs the repressive environment normally required for PcG target gene silencing. \nTo further analyze the potential role of PcG-PcG contact changes on gene expression, we defined four \ncategories to distinguish the differential the PcG inter-TADs contacts: gained (red), de novo contacts, \nwhich represents the TAD contacts which were not pr esent in the Control but detected in the SUMO \ncondition (ContactScoreControl = 0 & ContactScoreSUMO > 10); lost contacts which are not detectable \nin SUMO but detectable in Controls (ContactScoreControl > 10 & ContactScoreSUMO = 0), increased \ncontacts which represent the TAD pairs which contact in both conditions but stronger in SUMO control \n(ContactScoreControl>0 & ContactScoreSUMO>0 &  Δscore>10); and the decreased contacts which \nrepresent the weakening of the contacts that are present in both conditions (ContactScoreControl>0 & \nContactScoreSUMO>0 &  Δscore <−10). We show that TADs that gained contacts are further away \nfrom each other in linear (1D) distance1D genomic range, highlighting that loss of SUMOylation leads \nto longer-range contact increase, whereas the closer TADs decrease their contacts ( Fig. 5c). Finally, \nthe PcG TADs with decreased interactions in SUMO showed higher increase in the gene expression \ncompared to the ones with increased contacts ( Fig. 5c). In summary, these data indicate that SUMO \ndepletion rewires PcG dependent genome contacts, affecting underlying gene expression patterns. \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n13 \n \n \n \n \n      \n \n \nFigure 5: PcG-PcG TADs networks are re-wired upon SUMO RNAi. (a) Cartoon representing the stratification \nof inter-PcG-PcG TAD contacts. All indicates contacts without any filtering; Q1 represents the PcG TADs that \nengage in low contacts (below quartile 1) in the control sample; Q4 represents the PcG TADs that engage in  \nhigh contacts (above quartile 4) in the control sample. (b) Quantification of the inter -PcG-PcG TAD contact \nscores (All, Q1 and Q4). (c) Comparison of different features of each TAD pair in different categories (increased, \ndecreased, gained, lost). The left plot shows the log2 distance between each TAD, and the right plot shows the \naverage gene expression change within each TAD. Average gene expression is calculated by the sum of log2 \nfold change of each gene in a particular TAD, divided by the number of the genes found in the respective TAD. \n     \nPcG-Active TAD interaction changes correlate with PcG target gene expression \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n14 \nThe significant overall increase in the PcG -Active contacts prompted us to investigate the contact \npatterns between these two types of TADs. First, we classified TAD pairs in quartiles based on the \nfrequency of PcG-Active contacts made by each pair (Fig. 6a). The analysis of these data showed that, \nwhile there is a general increase in PcG-Active contacts when analysing all pairs, the increase is much \nstronger and more significant when restricting the analysis to the first quartile (Q1), the one with \nweakest contract frequency. In contrast, the analysis of Q4, the one with the highest contact frequency \nin controls, shows a significant reduction of contacts ( Fig. 6b). These data suggest that the nuclear \nreorganization of PRC1 into large condensates induces a drastic remodeling of PcG TADs that \ndisorganizes their overall contact pattern and exposes them to new subsets of contacts with active \nchromatin. \nTo link structural changes in PcG TAD contacts with transcriptional outcomes, we stratified PcG TADs \ninto three groups: those containing only down -regulated genes (onlyDown), those with only up -\nregulated genes (onlyUp), and those with a mixture of both kin ds of genes. For each category, we \ncompared SUMO RNAi and control samples by calculating the change in contact scores for every TAD \npair and summing the differences by TAD class (e.g., PcG^onlyUp –active, PcG^onlyDown–null; Fig. \n6c). All three PcG domain types exhibited increased contacts with active TADs, but the effect was most \npronounced for onlyUp domains ( Fig. 6c). By contrast, PcG –PcG interactions decreased in onlyUp \ndomains but increased in onlyDown domains, a difference that was statistically significant (Fig. 6c-d). \nThese findings suggest that SUMO depletion promotes exposure of PcG domains to active chromatin, \nwhile losing PcG-PcG interaction, inducing upregulation of genes in these PcG TADs.  \nFocusing on the relation between linear distances of TADs and changes in contact patterns, we found \nthat Active–PcG TAD pairs with increased contact frequency are closer to each other on the linear \nscale along the chromosome than those with decreased conta cts, indicating that heterotypic TAD \ninteraction increases preferentially occur between nearby TADs ( Fig. 6e). Furthermore, the average \ngene expression change was slightly lower in pairs of PcG -Active TADs that increase interaction, \nsuggesting that an increase of Active TAD contacts with PcG TADs might reduce transcription, albeit \nmarginally.  \nFinally, we analyzed whether SUMO depletion affects intra-TAD interactions within PcG domains. Pile-\nup plots and distributions revealed no significant changes in intra-PcG contact strength (Fig. S6a and \nb). Stratification by initial contact levels (low vs. high in control) similarly showed no differences ( Fig. \nS6b). Thus, PcG domains maintain their internal insulation and contact patterns upon SUMO loss, \nshowing that the main effect of PRC1 clustering upon SUMO depletion is to modify long -range \nchromatin interactions of Polycomb TADs rather than interfering with sh ort-range chromatin fiber \nfolding. \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n15 \n \n \n \n \nFigure 6: PcG -Active TADs networks are re -wired upon SUMO RNAi. (a) Cartoon representing the \nstratification of PcG -Active TAD contacts. All indicates contacts without any filtering; Q1 represents the PcG \nTADs that engage in low contacts (below quartile 1) in the control sample; Q4 represents the PcG TADs that \nengage in high contacts (above quartile 4) in the control sample. (b) Quantification of the contact scores between \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n16 \nPcG and Active TADs in each category (All, Q1 and Q4). (c) Heatmap of the sum of the changes on each type \nof inter-TAD contacts. PcG TADs were split into three categories: those that contain only upregulated genes \n(onlyUp); only down regulated genes (onlyDown), or both down and up regulated genes (common). The to tal \nscore changes for each of the three classes with all other TADs (Active, Het, Null, PcG) is indicated in each box \nin the plot. The score change is calculated for the TADs at the same chromosome arms only. (d) Distribution of \nthe differences of contact scores of PcG -PcG TADs in each category. (e) Comparison of different features of \neach TAD pair in different categories (increased, decreased, gained, lost). The left plot shows the log2 distance \nbetween each TAD, and the right plot shows the average gene expression change within each TAD. Average \ngene expression is calculated by the sum of log2 fold change in each gene under a particular TAD, divided by \nthe number of the genes. \n \nDiscussion \nIn this study, we demonstrate that global reduction of nuclear SUMOylation profoundly reorganizes the \nmaterial properties of Polycomb Repressive Complex 1 (PRC1) condensates and their associated \nchromatin domains. Remarkably, these architectural changes an d the resulting transcriptional \nmisregulation occur without major alterations in PRC1 binding and H3K27me3 distribution, highlighting \nthe importance of higher -order chromatin organization as a separable layer of Polycomb -mediated \nregulation. \nSUMOylation as a modulator of PRC1 condensate material state \nPRC1 functions as a key architectural organizer, forming nuclear foci that facilitate long -range \ninteractions between H3K27me3 -enriched domains 1,6-9. Our data show that SUMO depletion \ntransforms these foci into enlarged, less mobile structures with properties consistent with a gel -like or \nsolid-like state, exhibiting reduced molecular dynamics and limited internal mixing ( Fig. 1e). In our \nbiophysical model, we found that increasing the attraction between PRC1 complexes (E P−P), is key to \nmimicking a hypo-SUMOylated state, resulting in a transition from many small, local clusters to a few \nsignificantly larger foci (Fig. 2b). These large clusters could act as larger topological hubs, pulling more \ndistant H3K27me3-marked regions closer into physical space ( Fig. 5b). This could lead to a global \nincrease in longer -range chromosomal contacts, especially among PcG -enriched domains and \nbetween PcG -enriched and Active domains ( Figs. 5 -6). This PRC1 reorganization correlates with \nwidespread rewiring of TAD interactions, suggesting that the regulation of PRC1 condensate properties \nis essential for maintenance of the appropriate contact pattern between specific Polycomb domains \n(Figs. 5-6). SUMOylation provides a reversible, tunable mechanism to regulate condensate size and \nmaterial properties. While previous work has emphasized protein -protein interactions among PRC1 \nsubunits in foci formation 7,10,22,32, our findings demonstrate that SUMOylation critically modulates \ncondensate dynamics and that the formation of aberrant, gel -like condensates upon SUMO loss can \ndrive ectopic 3D TAD interactions that compromise Polycomb domain insulation. \nSUMO gates proper 3D genome organization \nWe observed that PcG-enriched TADs normally engage in specific long-range interactions, consistent \nwith previous reports 29,30. Upon SUMO depletion, these networks are extensively rewired, with PcG \nTADs relatively close along the genomic sequence losing contacts with each other while gaining \nectopic interactions with neighboring active chromatin ( Figs. 5-6). This reorganization suggests that \nSUMOylation helps constrain PRC1-mediated interactions to appropriate chromatin partners in normal \nconditions, preventing promiscuous cross-talk between repressive and active compartments. \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n17 \nOur findings complement prior work showing that chemical disruption of PRC1 condensates abolishes \ncontacts among PcG target regions 9. Here, SUMO loss drives the opposite extreme, excessive \nclustering and rigidity, which similarly disrupts proper topology by creating aberrant bridges between \nnormally segregated domains. This non -monotonic relationship highlights the fact that both the \npresence and the proper material state of PRC1 condensates are essential for correct genome folding. \nMechanistic uncoupling from histone marks \nA striking finding is the stability of H3K27me3 despite a slightly reduced Pc binding and 3D \nreorganization of TAD -to-TAD interactions. This demonstrates that H3K27me3 maintenance and \nhigher-order chromatin interactions are mechanistically separable. The p ersistence of H3K27me3 in \nmany domains, even as their spatial relationships change, suggests that this mark may be more stable \nthan the condensates that organize it.  \nThe global increase in H3K27ac upon SUMO depletion ( Fig. 3d ) suggests broader effects on \nchromatin-modifying activities, potentially through misregulation of other SUMOylated factors. \nHowever, the specific correlation between PcG -Active contact increases and upregulation of PcG \ntarget genes (Fig. 6) implicates architectural changes that are driven by aberrant PRC1 clustering as \nthe dominant mechanism disrupting Polycomb-mediated repression in this context. \nLimitations and future directions \nA key consideration is that we employed global SUMO depletion rather than PRC1 -specific SUMO \nmutants. While this approach affects hundreds of SUMOylated proteins, the coherence of the \nphenotype-specific changes in PRC1 condensate morphology, coupled with s pecific rewiring of PcG-\nassociated TAD interactions, suggests a direct functional link. Supporting this, overexpression of a Pc-\n3KR mutant (lacking three SUMOylation sites) shows partially enlarged nuclear foci15. However, when \nwe replaced the endogenous w t Pc gene with a Pc -3KR mutant form by CRISPR -dependent \nmutagenesis we failed to detect phenotypic effects as strong as in SUMO RNAi condition on imaginal \nwing discs or adult flies, suggesting that other SUMOylation sites in Pc and/or SUMOylation of other \nPRC1 subunits might complement for the reduction in Pc SUMOylation. Dissecting the specific \nSUMOylation sites within PRC1 remains an important aim for future research.  \nSUMOylation is dynamically regulated during the cell cycle and in response to stress 33-36, suggesting \nthat PRC1 condensate properties may be tuned in specific biological contexts. SUMO may also alter \nprotein interaction networks by recruiting partners with SUMO -interacting motifs (SIMs); for example, \nCtBP localizes to Polycomb bodies in a SUMO -dependent manner37,38. Future profiling of the PRC1 \ninteractome under SUMO -depleted conditions might reveal how PTMs reshape condensate \ncomposition and function. \nConclusions and implications \n \nSpatial organisation of chromatin together with its associated proteins, is strongly interlinked with their \nphysical properties39,40. Our findings establish SUMOylation as a critical regulator of PRC1 condensate \nmaterial properties, 3D genome architecture, and transcriptional fidelity. By demonstrating that \naberrant condensate clustering leads to ectopic inter -TAD contacts and gene mis regulation \nindependently of canonical histone marks, we reveal a broader paradigm in which reversible PTMs \nfine-tune the mat erial state and regulatory capacity of nuclear condensates. This work links phase -\nseparation mechanics directly to chromatin topology and gene expression control, suggesting that \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n18 \npost-translational regulation of biomolecular condensation represents a general strategy for \ncoordinating nuclear architecture with transcriptional programs in metazoans. Future studies are \nneeded to explore how fluctuating SUMO levels during development, cell cycle progression, and stress \nresponses dynamically regulate nuclear condensates to maintain genome organization and function. \n \nFunding \nThis work was supported by grants from the European Research Council (Advanced Grant 3DEpi and \nWaddingtonMemory), the European CHROMDESIGN ITN project (Marie Skłodowska -Curie grant \nagreement No 813327), by the Fondation ARC (EpiMM3D), by the Agence Nationale de la Recherche \n(Cell-ID grant from the France 2030 program with reference numbers ANR-24-EXCI-0002 and     ANR-\n24-EXCI-0004; LIVCHROM, grant N. ANR -21-CE45-0011), by the Fondation pour la Recherche \nMédicale (EQU202303016280), and by the MSD Avenir Foun dation (Project EpiMuM-3D). Research \nin the D.J. lab is supported by Agence Nationale de la Recherche (Grants No. ANR-21-CE45-0011, \nANR-21-CE13-0037, ANR-22-CE12-0035, ANR-23-CE12-0039, ANR-23-CE12-0014). This work was \nsupported by government grants managed by the Agence Nationale de la Recherche under the France \n2030 program, with the reference numbers ANR -24-EXCI-0002 and ANR -24-EXCI-0004. Nazli Akilli \nwas funded by EPiGenMed, Montpellier University and La Ligue Contre le Cancer. \n \nAcknowledgments \nWe would like to thank the MRI and Drosophila facilities (BioCampus Montpellier), CNRS, INSERM \nand the University of Montpellier. We are grateful to the Genotoul bioinformatics platform Toulouse \nOccitanie (BioinfoGenotoul,  https://doi.org/10.15454/1.5572369328961167E12) and the Pôle \nScientifique de Modélisation Numérique (PSMN) of the ENS de Lyon for computational resources. \n \nAuthor Contributions \nG.C., N.A. and T.C. conceptualized the study. N.A. performed the experiments. L.F and N.A. performed \nH2Aub118 CUT&RUN, and crosslinked Pc CUT&RUN. N.A. performed the remaining CUT&RUN \nexperiments. P.S.P and D.J performed the biophysical modelling. M.D.S. performed the bioinformatics \nanalysis of Hi-C, CUT&RUN, and RNA-seq experiments and N.A. performed downstream analysis of \ngenomics experiments. F.E. and F.M. supervised MOCHA -FRAP experiments. T.C. performed the \ndirect fixing and imaging of Pc -GFP expressing wing discs. N.A. wrote the manuscript. N.A., M.D.S., \nT.C. and G.C. edited the manuscript with inputs from all other authors. \n \nDeclaration of Interests \nThe authors declare no competing interests. \n \n \n \n \n \n \n \n \n \n \n \nMethods \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n19 \nFly lines and husbandry \nAll Drosophila melanogaster stocks were maintained on standard cornmeal agar medium at 25 °C \nunder a 12 h light/dark cycle. Gene knockdown and transgene expression were achieved using the \nGAL4/UAS system (Brand and Perrimon, 1994) with the nub -GAL4 driver obtained from the \nBloomington Drosophila Stock Center (BDSC). \nRNAi lines were sourced from the Vienna Drosophila RNAi Center (VDRC) and the Transgenic RNAi \nProject (TRiP; smt3 RNAi, stock JF02869). The following genotypes were used in this study: \n● w[1118]; P{w[+mC]=UAS -Dcr-2.D}1; P{w[+mW.hs]=GawB}nubbin -AC-62 (nub -GAL4 driver \nline) \n● PcGFP/CyO; Smt3RNAi/TM6B (SUMO RNAi) \n● Actin5C-PcGFP/FM7 (overexpression of Pc-GFP) \nCrosses were performed at 25 °C, and F1 progeny was used for downstream experiments. All fly lines \nused in this work are the same as the ones used in Gonzales et.al,2014.  \nImmunostaining \nHome made antibodies pre -cleaned prior to the staining procedure.For pre -cleaning of antibodies \nagainst PRC1 components, at least 20 larvae were dissected, retaining imaginal discs but removing \nfat tissue. Carcasses were transferred to 1.5 -ml microcentrifuge tubes containing 750 µl PBS, and \nfixation was performed by adding 250 µl 16% paraformaldehyde (4% final) for 20 min at room \ntemperature on a rotating wheel. Samples were washed three times with PBS, then permeabilized in \n0.5% Triton X-100 in PBS (PBTr) for 1 h at room temperature with gentle rotation; the PBTr solution \nwas replaced 2 –3 times during this step. Tissues were subsequently blocked for 1 h in 3% BSA \nprepared in 0.1% PBTr at room temperature. \nPrimary antibodies were diluted 1:500 (Ph and Pc antibodies) in PBTr containing 1% BSA. Carcasses \nwere incubated with the antibody solution for 2 h at room temperature, after which the antibody \nsolutions were recovered and used for staining newly dissected samples. \nFor staining, newly dissected larvae were processed as above (4% paraformaldehyde fixation for 20 \nmin, 0.5% Triton X-100 permeabilization for 1 h, and blocking in 3% BSA for 1 h) and then incubated \novernight at 4 °C with the pre-cleaned antibody solutions. The following day, tissues were washed 4–\n6 times with 0.1% PBTr at room temperature, and then incubated with secondary antibodies (1:200 \ndilution in PBTr containing 1% BSA) for 2 h at room temperature. Samples were washed at least three \ntimes for 15 min each in 0.1% PBTr. \nNuclei were counterstained with 1 µg ml⁻¹ DAPI in 0.1% PBTr for 20 min at room temperature, followed \nby three 15-min washes in 0.1% PBTr and three quick PBS washes without detergent. Wing discs were \nthen dissected from carcasses in PBS and mounted in Vectashield (Vector Laboratories). \n \n \nRNA preparation for RNA-sequencing  \nThird instar larvae were dissected in Schneider medium. For SUMO RNAi and for respective control \nonly male larvae were dissected. Total RNA was extracted using TRIzol reagent and extracted RNA \nwere purified using RNA Clean & Concentrator kit (Zymo Research , #R1015). The sequencing was \ndesigned for mRNAs, thus, poly-A RNA enrichment, library preparation and illumina sequencing were \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n20 \nperformed by BGI. 6 GB of data was produced for each sample by BGI using paired end 150bp \nsequencing. \n \n \nRNA-seq analysis \nRNA-seq samples were aligned using STAR aligner (v2.7) 41. Subread (v2.0.6) 42 was used to create \nthe count table with the command featureCounts -p --countReadPairs -s 2. This count table served as \nthe input for DESeq2 (v1.40.2) 45. The volcano plots were obtained using the R library \nEnhancedVolcano (v1.20.0). \n \nCUT&RUN experiments \nCUT&RUN experiments were performed as described by Kami Ahmad in protocols.io \n(10.17504/protocols.io.umfeu3n) with minor modifications. We dissected 20 Wing Discs for histone \nmarks and 40 WDS for indirect DNA-binding proteins in Schneider medium centrifuged them for 3 min \nat 2000g and washed them twice with wash+ buffer before addi ng concanavalin A-coated beads. For \nexperiments involving Pc, we included a 5 -minute crosslinking step with 0.1% formaldehyde on the \nwheel at room temperature. MNase digestion (pAG -MNase Enzyme from Cell Signaling) was \nperformed for 30 min at 4°C on shaker . After Proteinase K digestion, DNA was recovered using \nSPRIselect beads and eluted in 15 μl of Tris 10 mM. DNA libraries for sequencing were prepared using \nthe NEBNext Ultra II DNA Library Prep Kit for Illumina. Sequencing (paired -end sequencing 150 bp, \nroughly 2 Gb per sample) was performed by Novogene (https://en.novogene.com/). The following \nantibodies were used: H3K27me3 (1:100, Active Motif, catalogue no. 39155), H2AK118Ub (1:100, Cell \nSignaling, catalogue no. 8240), H3K27ac (1:100, Active Motif, catalogue no. 39135), H3K9me3 (1:100, \nActive Motif, catalogue no. 39161 ), Pc (home made by G.Cavalli lab). All experiments were performed \nin biological duplicates. \n \nCUT&RUN analysis \nCUT&RUN samples for H3K27me3, Pc, H2Aub118, H3K9me3, H3K27ac, and IgG in Control and \nSUMO RNAi imaginal discs were aligned using Bowtie2 (v2.4.2) with the options -p 8 --local --very-\nsensitive-local --no-unal --no-mixed --no-discordant --phred33 -I 10 -X 700. Low-quality reads were \nfiltered with samtools (v1.9) using the command “samtools view -F 4 -h -q 30.” The resulting BAM files \nwere sorted, indexed and de-duplicated using sambamba (v1.0.0) with commands sambamba sort and \nsambamba markdup --remove-duplicates respectively. To generate normalized BigWig files, the \nbamCoverage command from deepTools2 (v3.5.5) was used with parameters --normalizeUsing RPKM \n--ignoreDuplicates. Macs3 (v3.0.0b1) was used to call peaks on each replicate with commands macs \ncallpeak -t -f BAMPE -g 142573017 -q 0.01 -c ${controlBam} for Pc and H3K271c  and macs3 callpeak -t \n${sortedDedupBamPerReplicate} -c ${controlBam} -f BAMPE -g 142573017 -q 0.01 --broad --broad-cutoff \n0.01 for H3K27me3, H2Aub118,  and H3K9me3. Per each condition, the peaks per target were defined as \nthe intersection between the peaks called in each of the replicates using the command bedtools \nintersect (v2.31.1). R library DiffBind (v3.10.1 on R v4.3.3) was used to find differential binding regions \non the union of the peaks obtained for the SUMO RNAi and Control conditions. To define chromatin \nstates in the Control condition, ChromHMM analysis was applie d by first binarizing the signal in the \nreplicate-merged .bam files of H3K27me3, Pc, H2Aub118, H3K9me3, and H3K27 ac using IgG as a \ncontrol track (command: java -mx40G -jar ChromHMM.jar BinarizeBam -b 200 -p \n0.0001 ./${bamFilesDir}/ cellmarkfiletable.tsv ./${binarizedBamFilesDir}) and next by defining a 4-state \nmodel based on the binarized signal (command: java -mx40G -jar ChromHMM.jar LearnModel -b \n200 ./${binarizedBamFilesDir} ./${modelDir} 4 dm6 ). The resulting four ChromHMM emissions were \nassigned to chromatin states: Active (red) chromatin enriched by H3K27ac, Heterochromatin (green) \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n21 \nenriched by H3K9me3, Polycomb (blue) chromatin domains enriched by H3K27me3, and Null (black) \nchromatin not enriched in any features used for the analysis. \n \nHi-C experiments \nHi-C was performed using the EpiTech Hi-C Kit (Qiagen) according to the manufacturer’s instructions \nwith minor modifications. \nSixty wing discs were dissected in Schneider’s medium and transferred to 200 µl HBSS containing \n10.8 µl 37% formaldehyde (2% final). Samples were homogenized using a Biomasher II (Funakoshi, \n320103) and fixed for 20 min at room temperature. Crosslinking was quenched for 5 min with 100 µl 3 \nM Tris -Cl (pH 7.5), followed by addition of 650 µl HBSS and gentle inversion. The Collagenase \ntreatment from the manufacturer’s protocol was omitted, as it reduced DNA yield. Samples were \ncentrifuged (800 × g, 5 min, 4 °C), and 250 µl QIAseq Beads (equilibrated to room temperature) were \nadded to the remaining ~250 µl containing nuclei. After 10 min at room temperature, beads were \nmagnetically separated and washed sequentially with 50 µl ice -cold PBS, 150 µl cold RNase -free \nwater, and 50 µl cold Buffer C1. The mixture was incubated on ice for 10 min and then at room \ntemperature for an additional 10 min before a final wash with 500 µl cold RNase-free water. \nFor chromatin digestion, beads were resuspended in 40 µl Hi -C Digestion Solution and incubated at \n65 °C for 10 min, followed by cooling on ice. Subsequently, 4.4 µl 10% Triton X -100 and 4 µl Hi -C \nDigestion Enzyme were added, and samples were incubated at 3 7 °C for 2 h (600 rpm) and then at \n65 °C for 20 min. Samples were placed on ice and could be frozen at this stage. \nEnd labeling was performed by sequential addition of 6 µl Hi -C End Labeling Mix and 1 µl Hi -C End \nLabeling Enzyme, with gentle mixing and incubation at 37 °C for 30 min. For ligation, 350 µl Hi -C \nLigation Solution was added, and samples were gently inverted and incubated at 16 °C for 2 h. Samples \nwere cooled on ice before proceeding or frozen for later use. \nFor de-crosslinking, 10 µl Proteinase K was added, and samples were incubated at 56 °C for 30 min \nand at 80 °C for 90 min, then cooled to room temperature. DNA was precipitated by adding 40 µl 3 M \nsodium acetate (pH 5.2) and 280 µl isopropanol, vortexed br iefly, and applied (including beads) to a \nMinElute® column (Qiagen). The column was washed with 0.75 ml Buffer PE, centrifuged (17,900 × g, \n1 min), and eluted with 35 µl Buffer EB pre-warmed to 65 °C. \nRecovered DNA was diluted to 130 µl and sonicated for 90 s using a Covaris S220 (140 W peak power, \n10% duty factor, 200 cycles per burst). Sonicated DNA was purified by adding four volumes of Buffer \nSB1, loading onto a MinElute column, washing twice with 7 00 µl 80% ethanol, and eluting with 50 µl \nBuffer EB pre-warmed to 65 °C. \n \nHi-C analysis methods \nRaw data from Hi -C sequencing were processed using the ‘scHiC2’ pipeline. Valid interactions were \nstored in a database using the ‘misha’ R package ( https://github.com/msauria/misha-package). \nExtracting the valid interactions from the misha database, the ‘shaman’ R package \n(https://bitbucket.org/tanaylab/shaman) was used for computing the Hi -C expected models, Hi -C \nscores with parameters k = 250 and k_exp = 500. Specifically, Hi -C scores quantify the contact \nenrichment (positive values) or depletion (negative values) of each bin of the map with respect to a \nstatistical model used to evaluate the expected number of counts. To generate this expected model, \nwe randomized the observed Hi -C contacts using a Markov chain Monte Carlo -like approach per \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n22 \nchromosome. Shuffling was conducted such that the marginal coverage and decay of the number of \nobserved contacts with the genomic distance were preserved but any features of genome organization \n(for example, TADs or loops) were not. These expected maps wer e generated for each biological \nreplicate separately and contained twice the number of observed cis contacts. Next, the score for each \ncontact in the observed contact matrix was calculated using the k nearest neighbors (kNN) strategy. In \nbrief, the distributions of two-dimensional Euclidean distances between the observed contact and its \nnearest k_exp neighbors in the pooled observed and pooled expected (per cell type) data were \ncompared, using Kolmogorov –Smirnov D statistics to visualize positive (higher density in observed \ndata) and negative (lower density in observed data) enrichments. These D scores were then used for \nvisualization (using a scale from −100 to +100) and are referred to as Hi -C scores in the text. \nAccordingly, the color scale of the Hi -C scores comprises both positive and negative values. When \ncomputing the differential Hi-C scores maps of, the reference dataset was used as the expected model. \nThe box plots show the median (central line), the 75th and 25th percentiles (box limits) and 1.5 × IQR \n(whiskers). .mcool were obtained using cooler44 (v0.10.2) and cooltools45 (v0.7.0) and normalized via \nthe Iterative Correction and Eigenvector decomposition algorithm (ICE) with default parameters \n(command “cooler zoomify -r 5000 file.cool -o file.mcool --balance”). The physical domains (TADs) \nwere called in the Control condition using the TopDom  algorithm (v0.10.1)46 as in the HiCExperiment \nR-package (v1.4.0) 47 using a window size parameter of 5 on the .mcool files at 5 kilo -bases (kb) \nresolution. Each of the 5 -kb TAD borders were then refined at 1 kb by looking at the 1kb poin t of \nmaximum insulation.  \nImage acquisition \nFor Confocal and the half -FRAP experiments LSM980 microscope was used. For half -FRAP, time \nseries were recorded every 150,200,300,350 or 500 ms. Bleaching performed at the 3rd time point. \nThe pixel size was 30nm for the images. The pin hole was increased to 150µm. AiryScan microscopy \nwas performed using a LSM 980 microscope (Carl Zeiss Microscopy, Iena, Germany) with a AiryScan2 \ndetector and a 63× PlanApo objective having a numerical aperture (N.A.) 1.4. Pc -GFP was excited \nwith a 488 nm laser diode and we used the mode AiryScan SR to collect images with an optimal pixel \nsize of 42 nm in x, y and 170 nm in z. Processing of raw images to produce AiryScan images was done \nwith the Zen software controlling the microscope with a wiener filter automatically adjuste d by the \nsoftware. \nImage analysis \nTo characterize Pc foci inside cell nuclei ( Fig. 1 ), we quantified 3D images acquired in Airy Scan \nmicroscopy by using the software Imaris 9.8.0 (Oxford Instruments, UK). We applied the spots option \nof the Imaris software to identify nuclear foci and measure their Intensity. To estimate the intensity of \nPc-GFP inside the nucleoplasm, we used the median intensity of Pc-GFP inside cell nuclei.  \n \nPolymer model \nPRC1 complexes and chromatin are modeled following the framework developed in previous work 48. \nBriefly, PRC1s molecules are represented as a lattice gas with 𝑁! particles, i.e. as beads of effective \nsize 20\t𝑛𝑚 residing on the vertices of a face-centered cubic lattice49. Similarly, chromatin is described \nas a semi-flexible, self-avoiding polymer chain composed of 𝑁& monomers evolving on the same lattice \nas the PRC1s. Each monomer encompasses 1\t𝑘𝑏𝑝 of chromatin (5 nucleosomes) and can be in one \nof two epigenetic states: H3K27me3 -marked (Polycomb-regulated regions) or H3K27me3 -unmarked \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n23 \n(the rest of the chromatin). The Hamiltonian H of the system describing its dynamics is composed of \nfour terms: 𝐻 = 𝐻'()* + 𝐻!\"! + 𝐻!\"# + 𝐻%#. \n𝐻'()* = ∑+!\"1\n,-2 𝜅(1 −𝑐𝑜𝑠 𝜃,) accounts for the bending rigidity of the chromatin where 𝜃, is the angle \nbetween monomers (𝑖 − 1, 𝑖, 𝑖 + 1) and 𝜅 = 3.2\t𝑘.𝑇 is the bending modulus of the polymer chain, \ncorresponding to a Kuhn length of 100\t𝑛𝑚50,51. \n𝐻!\"! = −\n/\"#\"\n2 ∑,∈𝒫 ∑2∈𝒱(,) 𝜎2 accounts for the self-attraction between PRC1 molecules where 𝐸!\"! \nis the strength of self-attraction, 𝒫 is the ensemble of PRC1s, 𝒱(𝑖) are the lattice sites that are nearest-\nneighbor of the one where PRC1 𝑖 is seated, and 𝜎2 = 1 if a PRC1 is on lattice site 𝑙 (𝜎2 = 0 otherwise). \n𝐻!\"# = −𝐸!\"# ∑,∈𝒫 ∑2∈𝒱(,) 𝜅2 accounts for the cross -interaction between PRC1 molecules and \nH3K27-marked monomers where 𝐸!\"# is the strength of cross -interaction and 𝜅2 = 1 if an H3K27 -\nmarked monomer is on lattice site 𝑙 (𝜅2 = 0 otherwise). \n𝐻%# = +𝐸%# ∑,∈𝒫 𝜎,𝜋, accounts for steric hindrance interactions between PRC1 molecules and any \nmonomers where 𝐸%# is the strength of steric hindrance and 𝜋, is the number of monomer on the lattice \nsite 𝑖. \nThe lattice is made of 𝑁 vertices, with periodic boundary conditions. Each vertex has 12 nearest \nneighbors and can contain at most one PRC1 molecule and 2 chromatin monomers. Note that double \noccupancy by monomers is allowed if and only if monomers are consecutive along the chain , \naccounting for bond fluctuations while ensuring correct self-avoidance48. \nMonte-Carlo simulation \nSimulations were performed starting from random, unknotted initial configurations for the chromatin \nchain and a uniform random distribution for PRC1 molecules. The system is evolved via a kinetic \nMonte-Carlo scheme, as detailed in previous work48. Briefly, each Monte Carlo step (MCS) consists of \n𝑁& monomer trial moves and 𝑁! PRC1 trial moves. During a trial move, a PRC1 molecule or a monomer \nis randomly selected and an attempt to move it to one of randomly -chosen nearest neighbor vertex is \nattempted. The new configuration is accepted via the Metropolis criterion based on the energy \ndifference between the trial and current configurations. \nFor each investigated parameter set, we simulated 𝑁6789 = 40 stochastic trajectories, each composed \nof a first warm up of 10⁶\t𝑀𝐶𝑆 in absence of self- and cross-interactions to relax the system, followed \nby 10⁷\t𝑀𝐶𝑆 with the full Hamiltonian, where snapshots were extracted every 10⁵\t𝑀𝐶𝑆.  \nWe simulated chromosome arm 3R ( 𝑁& = 29184 monomers) including 2859 H3K27-marked \nmonomers distributed in several domains across the polymer based on ChIP-seq data in the wing disk. \nThe ratio 𝑅!/# between 𝑁! and the number of H3K27-marked monomer was varied from 50% to 100%, \n𝐸!\"! from 0.2 to 2.2\t𝑘.𝑇, 𝐸!\"# from 0.8 to 1.2\t𝑘.𝑇 and 𝐸%# from 8 to 10\t𝑘.𝑇. The set of parameters \n(𝐸!\"! = 1\t𝑘.𝑇, 𝐸!\": = 0.8\t𝑘.𝑇, 𝐸%# = 8\t𝑘.𝑇 and 𝑅!/# = 50%) is representative of the wild -type \ncondition, capturing the strength of typical interactions of architectural proteins having LLPS \nproperties2,48,52 and typical in vivo concentration53. Our results are not qualitatively dependent on this \nchoice.  \nData analysis of the simulations \nFor each snapshot, PRC1 foci were defined similarly as in previous work 48. Briefly, for each PRC1 \nmolecule 𝑖, we defined its local PRC1 density 𝜂, =\n1\n12 ∑2∈𝒱(,) 𝜎2 and its local H3K27-marked monomer \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n24 \noccupancy ratio 𝜁, =\n1\n26 ∑2∈𝒱(,) 𝜅2. The ensemble of PRC1s belonging to a condensate is defined as \n𝒟\t = \t {𝜂, + 𝜁, ≥ 0.5, 𝑖 ∈ 𝒫}. \nThe connected components of 𝒟 are the foci. The volume of each focus is defined as the number of \nPRC1s inside it. The PRC1 free nucleoplasmic fraction is computed as J𝑁! − 𝐶𝑎𝑟𝑑(𝒟)N/𝑁!. The PRC1 \nfree nucleoplasmic fraction, the number of foci and the distribution of foci volume were averaged over \nthe 10 last frames of each simulation. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \nReferences \n \n1. Schuettengruber B, Bourbon HM, Di Croce L, Cavalli G. Genome Regulation by Polycomb and \nTrithorax: 70 Years and Counting. Cell. 2017;171(1):34-57. doi:10.1016/j.cell.2017.08.002 \n2. Plys, Aaron J., et al. “Phase Separation of Polycomb-Repressive Complex 1 Is Governed by a \nCharged Disordered Region of CBX2.” Genes & Development , vol. 33, no. 13 –14, 2019, pp. \n799–813, doi:10.1101/gad.326488.119. \n3. Seif, Elias, et al.  “Phase Separation by the Sterile Alpha Motif of Polyhomeotic \nCompartmentalizes Polycomb Group Proteins and Enhances Their Activity.” bioRxiv, 2020,  \ndoi:10.1101/2020.08.20.259994. \n4. Brown, Kyle, et al.  “Principles of Assembly and Regulation of Condensates of Polycomb \nRepressive Complex 1 through Phase Separation.” Cell Reports, vol. 43, no. 3, 2024, p. 113997, \ndoi:10.1016/j.celrep.2024.113997. \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n25 \n5. Niekamp, Stefan, et al.  “Modularity of PRC1 Composition and Chromatin Interaction Define \nCondensate Properties.” Molecular Cell , vol. 84, no. 9, 2024, pp. 1651 -1666.e12, \ndoi:10.1016/j.molcel.2024.03.001. \n6. Akilli, Nazli, et al.  “Phase Separation and Inheritance of Repressive Chromatin Domains.” \nCurrent Opinion in Genetics & Development , vol. 86, no. 102201, 2024, p. 102201,  \ndoi:10.1016/j.gde.2024.102201. \n7. Isono, Kyoichi, et al. “SAM Domain Polymerization Links Subnuclear Clustering of PRC1 to \nGene Silencing.” Developmental Cell , vol. 26, no. 6, 2013, pp. 565 –577, \ndoi:10.1016/j.devcel.2013.08.016. \n8. Cheutin, Thierry, and Giacomo Cavalli. “Loss of PRC1 Induces Higher -Order Opening of Hox \nLoci Independently of Transcription during Drosophila Embryogenesis.” Nature \nCommunications, vol. 9, no. 1, 2018, p. 3898, doi:10.1038/s41467-018-05945-4. \n9. Williamson, Iain, et al. “Dispersal of PRC1 Condensates Disrupts Polycomb Chromatin Domains \nand Loops.” Life Science Alliance, vol. 6, no. 10, 2023, doi:10.26508/lsa.202302101. \n10. Shao, Z., et al. “Stabilization of Chromatin Structure by PRC1, a Polycomb Complex.” Cell, vol. \n98, no. 1, 1999, pp. 37–46, doi:10.1016/S0092-8674(00)80604-2. \n11. Francis, Nicole J., et al . “Chromatin Compaction by a Polycomb Group Protein Complex.” \nScience (New York, N.Y.) , vol. 306, no. 5701, 2004, pp. 1574 –1577, \ndoi:10.1126/science.1100576. \n12. Grau, Daniel J., et al. “Compaction of Chromatin by Diverse Polycomb Group Proteins Requires \nLocalized Regions of High Charge.” Genes & Development , vol. 25, no. 20, 2011, pp. 2210 –\n2221, doi:10.1101/gad.17288211. \n13. Tatavosian, Roubina, et al.  “Nuclear Condensates of the Polycomb Protein Chromobox 2 \n(CBX2) Assemble through Phase Separation.” The Journal of Biological Chemistry , vol. 294, \nno. 5, 2019, pp. 1451–1463, doi:10.1074/jbc.ra118.006620. \n14. Smith, Matthew, et al.  “Small Ubiquitin -like Modifier (SUMO) Conjugation Impedes \nTranscriptional Silencing by the Polycomb Group Repressor Sex Comb on Midleg.” The Journal \nof Biological Chemistry , vol. 286, no. 13, 2011, pp. 11391 –11400, \ndoi:10.1074/jbc.M110.214569. \n15. Gonzalez, Inma, et al.  “Identification of Regulators of the Three -Dimensional Polycomb \nOrganization by a Microscopy-Based Genome-Wide RNAi Screen.” Molecular Cell, vol. 54, no. \n3, 2014, pp. 485–499, doi:10.1016/j.molcel.2014.03.004. \n16. Gambetta, Maria Cristina, and Jürg Müller. “O -GlcNAcylation Prevents Aggregation of the \nPolycomb Group Repressor Polyhomeotic.” Developmental Cell, vol. 31, no. 5, 2014, pp. 629–\n639, doi:10.1016/j.devcel.2014.10.020. \n17. Cheng, Xiaodong. “Protein SUMOylation and Phase Separation: Partners in Stress?” Trends in \nBiochemical Sciences, vol. 48, no. 5, 2023, pp. 417–419, doi:10.1016/j.tibs.2022.12.003. \n18. Gutierrez-Morton, Emily, and Yanchang Wang. “The Role of SUMOylation in Biomolecular \nCondensate Dynamics and Protein Localization.” Cell Insight, vol. 3, no. 6, 2024, p. 100199,  \ndoi:10.1016/j.cellin.2024.100199. \n19. Mittag, Tanja, and Rohit V. Pappu. “A Conceptual Framework for Understanding Phase \nSeparation and Addressing Open Questions and Challenges.” Molecular Cell, vol. 82, no. 12, \n2022, pp. 2201–2214, doi:10.1016/j.molcel.2022.05.018. \n20. Bolognesi, Benedetta, et al. “A Concentration-Dependent Liquid Phase Separation Can Cause \nToxicity upon Increased Protein Expression.” Cell Reports, vol. 16, no. 1, 2016, pp. 222 –231, \ndoi:10.1016/j.celrep.2016.05.076. \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n26 \n21. Eeftens, Jorine M., et al. “Polycomb Condensates Can Promote Epigenetic Marks but Are Not \nRequired for Sustained Chromatin Compaction.” Nature Communications, vol. 12, no. 1, 2021, \np. 5888, doi:10.1038/s41467-021-26147-5. \n22. Uckelmann, Michael, et al. “Dynamic PRC1-CBX8 Stabilizes a Porous Structure of Chromatin \nCondensates.” Nature Structural & Molecular Biology , vol. 32, no. 3, 2025, pp. 520 –530, \ndoi:10.1038/s41594-024-01457-6. \n23. Muzzopappa, Fernando, et al. “Detecting and Quantifying Liquid -Liquid Phase Separation in \nLiving Cells by Model-Free Calibrated Half-Bleaching.” Nature Communications, vol. 13, no. 1, \n2022, p. 7787, doi:10.1038/s41467-022-35430-y. \n24. Hay, Ronald T. “SUMO: A History of Modification.” Molecular Cell, vol. 18, no. 1, 2005, pp. 1 –\n12, doi:10.1016/j.molcel.2005.03.012. \n25. Huang, Chien-Hsin, et al. “Mechanisms and Functions of SUMOylation in Health and Disease: \nA Review Focusing on Immune Cells.” Journal of Biomedical Science , vol. 31, no. 1, 2024, p. \n16, doi:10.1186/s12929-024-01003-y. \n26. Xu, Dongbin, et al. “Genetic Control of Programmed Cell Death (Apoptosis) in Drosophila.” Fly, \nvol. 3, no. 1, 2009, pp. 78–90, doi:10.4161/fly.3.1.7800. \n27. Steller, H. “Regulation of Apoptosis in Drosophila.” Cell Death and Differentiation, vol. 15, no. 7, \n2008, pp. 1132–1138, doi:10.1038/cdd.2008.50. \n28. Schuettengruber, Bernd, and Giacomo Cavalli. “Recruitment of Polycomb Group Complexes \nand Their Role in the Dynamic Regulation of Cell Fate Choice.” Development (Cambridge, \nEngland), vol. 136, no. 21, 2009, pp. 3531–3542, doi:10.1242/dev.033902. \n29. Paulsen, Jonas, et al. “Long-Range Interactions between Topologically Associating Domains \nShape the Four -Dimensional Genome during Differentiation.” Nature Genetics, vol. 51, no. 5, \n2019, pp. 835–843, doi:10.1038/s41588-019-0392-0. \n30. Liyakat Ali, Tharvesh M., et al. “TAD Cliques Predict Key Features of Chromatin Organization.” \nBMC Genomics, vol. 22, no. 1, 2021, p. 499, doi:10.1186/s12864-021-07815-8. \n31. Ernst, Jason, and Manolis Kellis. “Chromatin -State Discovery and Genome Annotation with \nChromHMM.” Nature Protocols , vol. 12, no. 12, 2017, pp. 2478 –2492, \ndoi:10.1038/nprot.2017.124. \n32. Gemeinhardt, Tim M., et al. “A Disordered Linker in the Polycomb Protein Polyhomeotic Tunes \nPhase Separation and Oligomerization.” Molecular Cell , vol. 85, no. 11, 2025, pp. 2128 -\n2146.e15, doi:10.1016/j.molcel.2025.05.008. \n33. Vijayakumaran, Shamini, and Dean L. Pountney. “SUMOylation, Aging and Autophagy in \nNeurodegeneration.” Neurotoxicology, vol. 66, 2018, pp. 53 –57, \n doi:10.1016/j.neuro.2018.02.015. \n34. Heaton, Phillip R., et al. “Analysis of Global Sumoylation Changes Occurring during Keratinocyte \nDifferentiation.” PloS One, vol. 7, no. 1, 2012, p. e30165, doi:10.1371/journal.pone.0030165. \n35. Zhao, Xiaolan. “SUMO -Mediated Regulation of Nuclear Functions and Signaling Processes.” \nMolecular Cell, vol. 71, no. 3, 2018, pp. 409–418, doi:10.1016/j.molcel.2018.07.027. \n36. Celen, Arda B., and Umut Sahin. “Sumoylation on Its 25th Anniversary: Mechanisms, Pathology, \nand Emerging Concepts: Emerging Concepts in Sumoylation.” The FEBS Journal, vol. 287, no. \n15, 2020, pp. 3110–3140, doi:10.1111/febs.15319. \n37. Merrill, Jacqueline C., et al. “A Role for Non-Covalent SUMO Interaction Motifs in Pc2/CBX4 E3 \nActivity.” PloS One, vol. 5, no. 1, 2010, p. e8794, doi:10.1371/journal.pone.0008794 \n38. Poortinga, G., et al.  “Drosophila CtBP: A Hairy -Interacting Protein Required for Embryonic \nSegmentation and Hairy-Mediated Transcriptional Repression.” The EMBO Journal, vol. 17, no. \n7, 1998, pp. 2067–2078, doi:10.1093/emboj/17.7.2067. \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n27 \n39. Salari, Hossein, et al.  “Spatial Organization of Chromosomes Leads to Heterogeneous \nChromatin Motion and Drives the Liquid- or Gel-like Dynamical Behavior of Chromatin.” Genome \nResearch, vol. 32, no. 1, 2022, pp. 28–43, doi:10.1101/gr.275827.121. \n40. Erdel, Fabian. “Phase Transitions in Heterochromatin Organization.” Current Opinion in \nStructural Biology, vol. 80, no. 102597, 2023, p. 102597, doi:10.1016/j.sbi.2023.102597. \n41. Dobin, Alexander, et al. “STAR: Ultrafast Universal RNA-Seq Aligner.” Bioinformatics (Oxford, \nEngland), vol. 29, no. 1, 2013, pp. 15–21, doi:10.1093/bioinformatics/bts635. \n42. Liao, Yang, et al. “The R Package Rsubread Is Easier, Faster, Cheaper and Better for Alignment \nand Quantification of RNA Sequencing Reads.” Nucleic Acids Research, vol. 47, no. 8, 2019, p. \ne47, doi:10.1093/nar/gkz114. \n43. Love, Michael I., et al. “Moderated Estimation of Fold Change and Dispersion for RNA-Seq Data \nwith DESeq2.” Genome Biology, vol. 15, no. 12, 2014, p. 550, doi:10.1186/s13059-014-0550-8. \n44. Abdennur, Nezar, and Leonid A. Mirny. “Cooler: Scalable Storage for Hi -C Data and Other \nGenomically Labeled Arrays.” Bioinformatics (Oxford, England), vol. 36, no. 1, 2020, pp. 311 –\n316, doi:10.1093/bioinformatics/btz540. \n45. Open2C, et al.  “Cooltools: Enabling High -Resolution Hi -C Analysis in Python.” PLoS \nComputational Biology, vol. 20, no. 5, 2024, p. e1012067, doi:10.1371/journal.pcbi.1012067. \n46. Shin, Hanjun, et al. “TopDom: An Efficient and Deterministic Method for Identifying Topological \nDomains in Genomes.” Nucleic Acids Research , vol. 44, no. 7, 2016, pp. e70 –e70, \ndoi:10.1093/nar/gkv1505. \n47. Serizay, Jacques, et al.  “Orchestrating Chromosome Conformation Capture Analysis with \nBioconductor.” Nature Communications, vol. 15, no. 1, 2024, p. 1072, doi:10.1038/s41467-024-\n44761-x. \n48. Tortora, Maxime M. C., et al. “HP1-Driven Phase Separation Recapitulates the \nThermodynamics and Kinetics of Heterochromatin Condensate Formation.” Proceedings of the \nNational Academy of Sciences of the United States of America , vol. 120, no. 33, 2023, p. \ne2211855120, doi:10.1073/pnas.2211855120. \n49. Amith Zafal Abdulla. Painters in Chromatin : Theoretical models for 3D propagation of epigenetic \nmarks. Biophysics. Ecole normale supérieure de lyon - ENS LYON, 2022. English.  ⟨NNT : \n2022ENSL0044⟩. ⟨tel-03968006⟩ \n50. Socol, Marius, et al. “Rouse Model with Transient Intramolecular Contacts on a Timescale of \nSeconds Recapitulates Folding and Fluctuation of Yeast Chromosomes.” Nucleic Acids \nResearch, vol. 47, no. 12, 2019, pp. 6195–6207, doi:10.1093/nar/gkz374. \n51. Ghosh, Surya K., and Daniel Jost. “How Epigenome Drives Chromatin Folding and Dynamics, \nInsights from Efficient Coarse-Grained Models of Chromosomes.” PLoS Computational Biology, \nvol. 14, no. 5, 2018, p. e1006159, doi:10.1371/journal.pcbi.1006159. \n52. Strom, Amy R., et al. “Phase Separation Drives Heterochromatin Domain Formation.” Nature, \nvol. 547, no. 7662, 2017, pp. 241–245, doi:10.1038/nature22989. \n53. Bonnet, Jacques, et al. “Quantification of Proteins and Histone Marks in Drosophila Embryos \nReveals Stoichiometric Relationships Impacting Chromatin Regulation.” Developmental Cell, \nvol. 51, no. 5, 2019, pp. 632-644.e6, doi:10.1016/j.devcel.2019.09.011. \n \n \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n28 \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \nSupplementary Figures: \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n29 \n \n \n \n \n \n \n \n       \nFig. S1. SUMO RNAi induced on the imaginal wing disc pouch leads to reorganization of PRC1 foci. (a) \nCartoon representing the fly lines used to induce SUMO RNAi on Drosophila wing discs. (b) Airyscan images of \nImmuno-staining against Ph, Pc and DAPI staining, in control wing discs. (c) Airyscan images of Immuno -\nstaining against Ph, Pc and DAPI staining, in SUMO RNAi wing discs. \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n30 \n \n \nFig. S2. Table of simulation results of the biophysical framework. (Columns) Each column shows \nsimulations of varying strength of steric hindrance ( 𝐸!\"). (Rows) Each row shows simulations of varying ratio \nbetween PRC1 complexes and to H3K27 marked monomer ( 𝑅#/\"). (Heatmaps) Each heatmap shows \nsimulations of varying strength of PRC1 complexes self -interaction ( 𝐸#%#, horizontal) and of varying PRC1 \ninteraction strength for H3K27-marked monomers (𝐸#%\", vertical). (Blue) Each blue heatmap shows the number \nof foci at the end of the simulations. (Red) Each red heatmap shows the average volume of foci at the end of \nthe simulations in number for PRC1 complexes. (Green) Each green heatmap shows the average PRC1 free \nnucleoplasmic fraction. The trends shown in Fig.2b are not dependant on the values of 𝐸!\" and 𝑅#/\". Indeed, \nan increase of 𝐸#%# leads to an in, the number of foci and the PRC1 free nucleoplasmic fraction are decreasing \nwhereas the volume of foci is increasing. \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n31  \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n32 \nFig. S3. Gene expression profiles upon SUMO RNAi. (a) FPKM values of the SUMO gene ( smt3) in the \ncontrol and SUMO RNAi samples. P.adjusted= 6.49e-22. (b) Volcano plot showing the DEseq2 results for the \napoptotic markers. Absolute value of the Fold change > 1.5, p.adjusted < 0.05. GO term enrichment analysis for \nthe all differentially expressed genes upon SUMO RNAi. (d) GO term enrichment analysis for the differentially \nexpressed PcG target genes upon SUMO RNAi. \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n33 \n  \nFig. S4. SUMO RNAi does not lead to major changes on the distribution of the H3K27me3 mark compared \nto the control. (a) Scatter plot of the enrichment of H3K27me3 mark, x axis indicates SUMO RNAi and y axis \nindicates the control. (b) Tracks of the differentially bound sites for the H3K27me3 mark in each condition.  \n \n \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n34 \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n35 \n \n \nFig. S5. Heterotypic inter-PcG TADs networks are re-wired upon SUMO RNAi. (a) Cartoon representing the \nstratification of inter-PcG TAD contacts. All indicates contacts without any filtering; Q1 represents the PcG TADs \nthat engage in low contacts (below quartile 1) in the control sample; Q4 represents the PcG TADs that engage \nin high contacts (above quartile 4) in the control sample(left). (b)Quantification of the contact scores between \nPcG TADs with other types of TADs in each category (all;Q1 and Q4).  \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n36 \n \n \nFig. S6. Intra-domain interactions of PcG domains don’t show significant changes upon SUMO RNAi. (a) \nCartoon representing the intra-domain contacts that are measured (top); pile -up plots of PcG TADs in control \nand SUMO RNAi samples (bottom). (b) Cartoon representing the stratified intra-domain contacts (all intra-PcG \nTADs; Q1 intra-PcG TADs; Q4 intra -PcG TADs) that are measured (top). ; quantification of the stratified intra \nTAD contacts (all intra-PcG TADs; Q1 intra-PcG TADs; Q4 intra-PcG TADs) (bottom).   \n \n \n \n \n \n \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint \n\n37 \n  \n \n \nFig. S7. SUMO RNAi leads to genome-wide re-organization of the genome. (a) ICE normalised Hi-C maps \nfor the genomic region around the dac locus in Control and SUMO RNAi samples. Colors towards black and red \nindicate high contact values, colors towards yellow and white indicate low contacts. C&R tracks for Pc, \nH3K27me3, H3K27ac, H2Aub118 marks in control and the SUMO RNAi sample shown in the  bottom. Each \nsquare indicates a TAD called and the color indicates the enriched histone mark type: blue for H3K27me3 -\nenriched, red for H3K27ac-enriched, and black for no enrichment of any used histone mark.  \n \n \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted February 6, 2026. ; https://doi.org/10.64898/2026.02.05.704038doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}