A histidine cluster determines YY1-compartmentalized coactivators and chromatin elements in phase-separated super-enhancers

As an oncogenic transcription factor, Yin Yang 1 (YY1) regulates enhancer and promoter connection. However, gaps still exist in understanding how YY1 coordinates coactivators and chromatin elements to assemble super-enhancers. Here, we demonstrate that YY1 activates FOXM1 gene expression through forming liquid-liquid phase separation to compartmentalize both coactivators and enhancer elements. In the transactivation domain of YY1, a histidine cluster is essential for its activities of forming phase separation, which can be extended to additional proteins. Coactivators EP300, BRD4, MED1 and active RNA polymerase II are components of YY1-rich nuclear puncta. Consistently, histone markers for gene activation, but not repression, colocalize with YY1. Importantly, multiple enhancer elements and the FOXM1 promoter are bridged by YY1 to form super-enhancers. These studies propose that YY1 is a general transcriptional activator, and promotes phase separation with incorporation of major coactivators and stabilization by distal enhancers to activate target gene expression.


INTRODUCTION
Gene expression mediated by RNA polymerase II (Pol II) is a complex but well-regulated biological process. Transcription factors (TFs) and cofactors constitute the transcription machinery that recognizes specific elements on target promoters and coordinates concerted Pol II action (1,2).
The full-length coding sequences of YY1, HOXA1, FOXG1B, ZIC3 and HNF6 (or their variants), as well as IDRs of EP300, BRD4, MED1 and RNA Pol II, were individually subcloned into a modified version of a pGEX vector with 6×His and EGFP or mCherry at the N-terminus. A bacterial expression system was used to express these coding sequences and recombinant proteins were purified using Ni-NTA agarose. Meanwhile, the full lengths of YY1 or its mutants, and EP300 were subcloned into a eukaryotic EGFP or mCherry expression vector. YY1 or its mutants, and FOXM1 coding sequences were also cloned into a lentiviral vector with a 3×Flag-tag at the N-terminus. An shRNA, sh-YY1, targeting the 3'-UTR of YY1 mRNA, and a control shRNA (sh-Cont) were designed as previously described (28). The promoter of FOXM1 was amplified by a nest PCR method and subcloned upstream of the Gaussia luciferase (Gluc) coding sequence to generate a pFOXM1-prmt-Gluc vector (WT). Also, reporter constructs S1M, S2M, S3M, S4M, S5M, S4/5M, S1/3M and S2/4/5M with correspondingly mutated YY1 binding sites were generated using the ClonExpress ® II Recombination system (Vazyme Biotech Co., Ltd., Nanjing, China). The predicted enhancer regions were also individually amplified by the nest PCR method and subcloned upstream of the FOXM1 promoter or downstream of Gluc to generate reporter constructs.

Protein expression and purification
All His6-tagged constructs were overexpressed in E. coli BL21 (DE3) cells. Bacteria were cultured grown to an optical density at 600 nm (OD 600nm ) of 0.6, and induced overnight with 0.15 mM IPTG at 18°C. Bacteria were pelleted and resuspended in a lysis buffer (20 mM HEPES, 0.2 mM EDTA, 100 mM KCl, 20% Glycerol, 1% Triton, 2 mM PMSF, 1 mg/mL lysozyme). After sonication, the bacterial lysate was centrifuged at 12000 g for 30 min. His6-tagged proteins in the supernatant were purified by Ni-NTA agarose beads (GE Healthcare). After extensive wash by a buffer containing 20 mM imidazole, the fractions eluted by 400 mM imidazole were collected and dialyzed. The size and purity of the purified proteins were monitored by SDS-PAGE.

Cell culture, transfection, lentiviral production, and infection
The HeLa, HEK-293T, U2OS, MDA-MB-231, MCF-7 and MCF-10A cells were cultured according to the protocols of the ATCC. All culture media were purchased from Gibco and fetal bovine serum (FBS) was from ExCell Bio. Transfection of cells was carried out using Lipofectamin 2000 (Thermo Fisher 4 Scientific, Shanghai, China) according to the manufacturer's instructions. Lentiviral production and infection followed our published procedure (29).

In vitro phase separation assay
Recombinant EGFP-or mCherry-fusion proteins were diluted to appropriate concentrations using 50 mM Tris·HCl, pH 7.4. Recombinant proteins were added to solutions containing 125 mM NaCl and 10% PEG-8000 as crowding agents unless specified. Protein solution (5 μl) was immediately loaded onto a glass slide, and covered with a coverslip. Slides were then imaged with a ×60 objective using the GE Dela Vision Elite (GE, Boston, MA, USA).
Turbidity experiments were performed in tubes. Samples (60 μl) containing appropriate concentrations of proteins, NaCl and 10% PEGs with indicated molecular weights were left to stand for 10 s at room temperature. OD 600nm was measured using BioSpectrometer basic (Eppendorf, Hamburg, Germany).

Fluorescence recovery after photo-bleaching (FRAP) imaging
FRAP was performed on a Zeiss LSM880 microscope using a ×63 oil-immersion objective. Images were acquired using the ZEN software. FRAP of the central region of protein droplets, three iterations of bleaching were performed with a 488 nm Argon laser at a 100% power with 3 frames being acquired prior to the bleach pulse. Fluorescence recovery was recorded every 2 s for 400 s after bleaching. U2OS cells, which were transfected for 24 h and cultured in glass-bottom dishes (NEST, China), were analyzed using FRAP studies. Three iterations of bleaching were performed with a 488 nm Argon laser at 30% power. Fluorescence recovery was recorded every 0.8 s for 20 s after bleaching. Analyses of the fluorescence intensity of bleached region, reference region and background region were carried out using the FRAP module in the ZEN software.

Immunofluorescence staining and live-cell imaging
Cells were seeded on glass coverslips in 12-well plates and cultured overnight. Subsequently, cells were fixed and permeabilized with Immunol Staining Fix Solution (Beyotime) for 30 min at room temperature. After blocked by 10% FBS for 30 min at room temperature, cells were incubated primary antibody for 30 min at room temperature. Then, the cells were washed thrice with PBS and incubated with Alex-Fluor-488 or 594-conjugated secondary antibodies (Thermo Fisher Scientific, Shanghai, China) for 30 min at RT. Finally, the cells were washed thrice with PBS and nuclei were counterstained with DAPI (Beyotime), and images were captured by the GE Dela Vision Elite (GE, Boston, MA, USA). objective. During image acquisition, cells were incubated in a chamber at 37 °C with 5% CO 2 .

Western blot analysis
Cells were washed and lysed in a protein lysate buffer. Total protein concentrations were measured using the Bradford protein method. Protein samples were separated on the SDS-PAGE and transferred into poly-vinylidene difluoride transfer (PVDF) membranes. The membranes were blocked by 5% nonfat milk in TBST buffer at room temperature for 1 h, and incubated with primary antibodies in TBST buffer for 4°C overnight. After three washes with TBST, the membranes were incubated with corresponding secondary antibodies at room temperature for 1 h. The membranes were washed and then visualized using ECL kit (Vazyme Biotech Co., Ltd., Nanjing, China).

Reverse transcription and quantitative PCR (RT-qPCR)
Total RNAs were extracted from cultured cells using the TRIzol reagent (Thermo Fisher Scientific Inc., Shanghai, China), and cDNA was synthesized using M-MLV reverse-transcriptase (Vazyme Biotech Co., Ltd., Nanjing, China). In the reaction of reverse transcription, 1 μg of total RNA was mixed with 0.5 μg/μl of oligo(dT) primer, followed by incubation at 65C for 5 min and 4C for 2 min. The tubes were then immediately incubated at 42C for 30 min and chilled at 4℃. For qPCR, cDNA was amplified using gene specific primers and the LightCycler 480 SYBR Green PCR Master Mix (Roche, Basel, Switzerland) on Lightcycler 480 instrument (Roche). The conditions used for qPCR were as follows: 95C for 3 min, followed by 40 cycles of 95C for 15 s and 60C for 1 min. All reactions were performed in triplicate. The results were analyzed using the 2 -ΔΔCt method and normalized using βactin. Primer sequences for qPCR were as follows: β-actin (5′-TTCCTTCCTGGGCATGGAGT and 5′-

Chromatin immunoprecipitation (ChIP)
ChIP analysis was performed as previously reported (30). In brief, cross-linking was completed after cell culture, followed by nuclei preparation and chromatin digestion. DNA gel electrophoresis was used to confirm adequate digestion. Samples immunoprecipitated by a normal IgG and a specific antibody were purified and subjected to semiquantitative both PCR and qPCR. For semiquantitative PCR, PCR products were analyzed using gel electrophoresis. The sequences of primers were listed in Supplementary Table S1.

Electrophoretic mobility shift assay (EMSA)
Oligonucleotides labelled with Cy5 at the 5´-end were synthesized and purified using HPLC method by Genewiz (Suzhou, China), and sequences of oligonucleotide probes and competitors were listed in Supplementary Table S2. The EMSA was conducted as we previously described (31). Briefly, 1 μg of a purified His×6-tagged YY1 protein was incubated with 0.5 pmol of labeled double-stranded probe in a binding buffer (250 mM HEPES, 500 mM KCl, 20 mM MgSO 4 , and 10 mM DTT, pH 8.0) on ice for 30 min. In the competitive binding experiments, excessive unlabeled probes were added to the 6 binding reaction of the labeled probe and His6-YY1. After the binding reaction, the samples were separated by 8% native PAGE at 100 V for 50 min at 4°C. The fluorescent intensity of the bands was immediately determined by Typhoon FLA7000 (GE, Boston, MA, USA).

Luciferase reporter assay
The reporter plasmids were constructed as described above. Cells in 24-well plates were cotransfected by 250 ng of a reporter construct, 500 ng of shRNA or expression construct, and 20 ng of pCMV-SEAP (secreted alkaline phosphatase) as a control. The Gluc activity was measured at 48 h after transfection as described previously (31).

Chromosome Conformation Capture (3C) experiments
The 3C analysis was performed following previously described procedures with minor modifications (32). Total of 110 6  was used to digest genomic DNAs by incubating at 37℃ overnight. Then, 1.6% SDS was used to inactivate the restriction enzyme by incubation at 65℃ for 20 min. The solution was diluted by adding 7 ml of ligation buffer (NEB) containing 1% Triton X-100 and 30 Units of T4 DNA ligase (NEB), followed by ligation reaction at 16℃ for 4.5 h and then at room temperature for 30 min. Cross-linking was released by Proteinase K digestion at 65℃ for 16 h. Finally, DNA fragments were purified by phenol-chloroform extraction and ethanol precipitation. The ligation products were analyzed by PCR using primers located adjacent to EcoRI or HindIII digested sites. PCR products were cloned into a pBluescript plasmid and sequenced to verify the ligated fragments. The sequences of primers used in 3C assay are shown in Supplementary Tables S3 and S4.

Cell viability, colony formation, and wound healing assay
These assays were performed as previously reported (33). In these experiments, cells were infected by lentivirus expressing either shRNAs or cDNAs. Each experiment was performed in triplicate.

Statistical analysis
All data were derived from at least three independent experiments. Results were presented as a mean with either standard deviation (SD) or standard error of mean (SEM), and sample numbers are indicated unless otherwise noted in the figure legends. Statistical significance calculations comparing 7 two conditions were performed using a two-tailed unpaired Student's t-test. The criterion of statistical significance level was denoted as follows: *P < 0.05; **P < 0.01; ***P < 0.001.

YY1 undergoes liquid-liquid phase separation
YY1 has a highly acidic transactivation domain (TAD) at N-terminus; its N-terminal 154 amino acids contain 76 negatively charged glutamic/aspartic acid (E/D) residues, but only two positively charged residues (R109 and R122), as well as 18 histidines (Hs) and 20 glycines (Gs) (Supplementary Figure   S1A). To date, structure of full length YY1 has not been reported, although cocrystal structure of its Cterminus with a binding element was resolved (34). YY1 TAD embraces an eleven-E/D (11×E/D) cluster and an 11×H cluster flanked by two acidic regions ( Figure 1A, top row). Consistently, inspection of YY1 primary sequence by the IUPred and VSL2 algorithms (35,36) revealed a high propensity of structural disorder in its TAD ( Figure 1A, bottom row). Amino acid composition examination revealed multiple potential IDRs in YY1, and the strongest one contains an 11×E/D and an 11×H clusters separated by a G-rich stretch ( Figures 1A and 1B). Importantly, E/D and H clusters, and their proximate regions are highly conserved among different species of vertebrates ( Figure 1B).
Thus, bioinformatic analyses strongly support the presence of an IDR in YY1's TAD.
Many IDR-containing TFs form dynamic liquid-like droplets, or gel-like phase-separated condensates, due to multivalent and weak interactions among IDRs (9,25,26). Fluorescent droplets and turbidity created by gel-like condensates are used to evaluate phase separation, while polyethylene glycol (PEG) mimicking crowded cellular conditions promotes this in vitro process (9). When purified  Figure S1C). In addition, ability of EGFP-YY1 in forming droplets was higher at 37°C than that at 25°C, and markedly descended at 4°C ( Figure 1H). Importantly, EGFP-YY1 droplets were sensitive to 1,6-hexanediol, a chemical disrupting liquid-liquid phase separated condensates ( Figure 1I). We also observed fusion events between two adjacent droplets ( Figure 1J and Supplementary Video 1) and quick green fluorescence recovery of droplets after targeted photobleaching treatment ( Figure 1K), indicating a dynamic feature of EGFP-YY1 condensates.
When EGFP-YY1 was expressed in U2OS cells, we detected green fluorescent puncta that also exhibited the ability of prompt fluorescence recovery after photobleaching (Figure 2A Figure S1F). These data strongly suggested that both H and E/D clusters, especially the former one, are important elements of YY1's IDR in phase separation. We also designed (7-Methoxycoumarin-4-yl) acetic acid (Mca)-labeled peptides, with a transmembrane sequence (TAT) fused to an A/G-rich control sequence, E/D and H cluster sequences of YY1 ( Figure   2F). Importantly, both TAT-E/D and TAT-H formed fusion droplets with EGFP-YY1, suggesting their ability in promoting phase separation. Contribution of E/D clusters to liquid condensate formation has been frequently observed (9,24), but the role of H clusters in phase separation was only reported in P-TEFb (37). In addition to YY1, we also verified an indispensable role of H clusters to the phase separation of HOXA1, FOXG1B, ZIC3 and HNF6 proteins (Supplementary Figures S2A-S2D).

Phase separation capability of YY1 is essential to its cell proliferative activity
To test whether the mutations or deletions could adversely affect YY1's function, we expressed the mutants in MDA-MB-231 and MCF-7 cells, with simultaneous endogenous YY1 silencing by an shRNA targeting 3'-UTR of YY1 mRNA ( Figure 2G). YY1 depletion reduced breast cancer cell viability, consistent with our previous report (38), while ectopically expressed WT YY1 could largely restore it ( Figure 2H). In accordance to droplet and punctum formation studies, IDR deletion, H-cluster mutation or deletion completely abolished YY1's ability in restoring cell viability, suggesting indispensability of 9 the IDR and H-cluster to YY1's function ( Figure 2H). Similarly, both E/D-A and E/D mutants only partially rescued YY1 depletion-caused viability decrease, while G-A and G mutants virtually retained YY1's function in reinstating cell viability ( Figure 2H). In addition, results of YY1 mutants in promoting cell migration and clonogenicity were consistent with the data of cell viability ( Supplementary Figures S2E and S2F). Overall, the IDR and H-cluster of YY1, key components for phase separation, are essential to its role in maintaining basic cellular activities.

YY1 compartmentalizes coactivators to nuclear puncta
The name Yin Yang 1 represents its ability in mediating both repression and activation of target genes, depending on recruited cofactor (39). In the past decade, the role of YY1 in promoting gene expression has been frequently reported. Consistently, YY1 interacts with several histone acetyltransferases, including EP300, CBP and PCAF (40,41). Among them, EP300 is a wellrecognized coactivator. We analyzed the primary sequence of human EP300 using the IUPred and VSL2 algorithms, and discovered five potential IDR regions, of which the IDR3 and IDR5 are larger than the others ( Figure 3A). Using a bacterial expression system, we purified recombinant EGFP fusion proteins with these IDR regions (Supplementary Figure S3A). EGFP-EP300-IDR3 and -IDR5, but not the other three proteins, could generate droplets ( Figure 3B Consistent with in vitro data, immunostaining analyses also presented endogenous EP300 puncta with fluorescence recovery from photobleaching and sensitivity to 1,6-hexanediol ( Figures 4A and 4B), suggesting EP300 phase separation in a cellular environment. With EGFP-EP300 and mCherry-YY1 cotransfected into U2OS cells, we observed colocalization of the green and red fluorescent signal ( Figure 4C). When examining the endogenous proteins in MDA-MB-231 and MCF-7 cells, we observed that most EP300 puncta stained by its antibody were overlapped with YY1 signal ( Figure   4D). Importantly, when Flag-YY1 were transfected into the cells, endogenous EP300 showed intensely stained puncta well-overlapped with Flag tag signal ( Figure 4E). Importantly, puncta formed by Flag-YY1 were markedly larger in sizes than those by endogenous YY1 ( Figure 4F). Consistently, mCherry-YY1 could form droplets with EGFP-EP300-IDR3 and -IDR5 with overlapped color and increased sizes, especially the latter one, compared to other predicted EP300-IDRs ( Figure 4G). The data strongly suggest that YY1 is a primary TF recruiting EP300 to activate gene expression.
Furthermore, we tested whether YY1 cooperates with other coactivators. In immunostained MDA-MB-231 cells, endogenous YY1 colocalized with MED1, BRD4 and CDK9 ( Figure 5A). Consistently, Flag-YY1 also presented puncta overlapping with endogenous proteins of the three coactivators ( Figure  5B), but with significantly increased sizes versus those of endogenous YY1 ( Figure 5C). Ser2 and Ser5 phosphorylation of the C-terminal heptapeptide repeats of RNA polymerase II (Pol II S2P and S5P) are general markers of gene transcription (42). Endogenous YY1 colocalized with Pol II S2P and S5P signals ( Figure 5D), suggesting that YY1 promotes gene transcription. Similar to coactivators, Flag-YY1 showed relatively intensified puncta overlapping with Pol II S2P and S5P signals ( Figure 5E) with increased sizes versus those by endogenous YY1 (Figure 5F). In line with these data, mCherry-YY1 could form overlapped droplets with EGFP-fused IDR regions of MED1, BRD4 and Pol II in vitro ( Figure 5G). Based on these results, we propose that YY1 plays a key role in promoting gene transcription through recruiting major coactivators to form phase separation. Consistent with this hypothesis, both endogenous YY1 and Flag-YY1 showed overlapped puncta with gene activation histone markers H3K27ac, H3K4me1 and H3K4me3, but not a repressive marker H3K9me3 ( Figures   5H and 5I). In addition, Flag YY1 markedly increased sizes of the puncta overlapping with the three activation markers but not H3K9me3 ( Figure 5J). Immunostaining assays were also conducted in Figures S4A-S4I). Overall, our data strongly suggested that YY1 is a general transcriptional activator to promote gene expression.

YY1 activates FOXM1 gene expression through recruiting general coactivators
YY1 conditional knockout in mouse embryonic fibroblast cells revealed many genes as potential targets of YY1 (43). Among them, FOXM1 exhibited over 2-fold reduction in response to YY1 depletion. Both YY1 and FOXM1 play oncogenic or proliferative roles in tumorigenesis (16,44).
Consistently, they were overexpressed in breast cancer and associated with patients' poor prognosis (38,45). Analyses of a TCGA dataset of mammary samples indicated positive YY1 and FOXM1 correlations (Supplementary Figure S5A), especially in normal tissues, suggesting physiological significance of their functional interplay. In mammary cells, we detected concurrent increase of YY1 and FOXM1 expression in breast cancer cells versus nontumorigenic MCF-10A cells ( Figure 6A). At both mRNA and protein levels, ectopically expressed YY1 in MCF-10A cells increased endogenous FOXM1 expression, while shRNA-mediated YY1 knockdown reduced it in breast cancer cells ( Figure   6B). All these data strongly suggest a positive regulation of FOXM1 gene by YY1.
To examine mechanism regulating FOXM1 gene expression, we first mapped essential region of the FOXM1 promoter. Four reporter constructs were generated with Gaussia luciferase (Gluc) driven by different lengths of the upstream sequence from the TSS of the FOXM1 gene ( Figure 6C). Based on response of the reporters to cotransfected YY1, YY1-regulated essential elements reside within the 1,141-bps region upstream of the TSS ( Figure 6C). To explore potential regulation of FOXM1 expression by YY1, we examined the human FOXM1 promoter for YY1 consensus binding elements using the JASPAR (46) and Tfsitescan (http://www.ifti.org/cgi-bin/ifti/Tfsitescan.pl) databases. Within the 1,141-bps FOXM1 promoter, we identified five potential YY1 binding sites (Supplementary Figure   S5B), and mutagenesis of #2, #4 and #5 sites, especially the latter two, caused remarkable reduction of FOXM1 promoter activity in reporter assays ( Figure 6D). YY1 binding to the FOXM1 promoter through these sites was verified by chromatin immunoprecipitation (ChIP) assay ( Figure 6E). To test YY1 binding in vitro, we synthesized double-stranded (ds) oligonucleotides S1 to S5 based on corresponding YY1 binding sites in the FOXM1 promoter. In EMSA studies, a Cy5-labeled probe based on a YY1 consensus motif in the CDC6 promoter (47) could be out-competed in His6-YY1 binding by S2, S4 and S5, but not S1 and S3, or mutants of S2, S4 and S5 (Supplementary Figure   S5C). In addition, Cy5-labeled S2, S4 and S5, but not their mutants, could bind YY1 to form slowly migrated bands in EMSA ( Figure 6F). To evaluate the potential of YY1-regulated FOXM1 expression through a phase separation mechanism, we incubated Cy5-labeled oligonucleotides with a relatively low concentration (2 µM) of EGFP-YY1. Strikingly, the presence of Cy5-labeled S2, S4 and S5, but not their mutants, could associate with EGFP-YY1 to form droplets with relatively large sizes ( Figure   6G, left panel). The data not only suggested a phase separation mechanism regulating FOXM1 gene transcription, but also disclosed an important fact that YY1 phase separation can be promoted by binding to DNA containing its consensus motifs ( Figure 6G, right panel). To evaluate whether superenhancer was involved in regulating FOXM1 expression, we carried out ChIP assays for several coactivators. Both semi-quantitative and quantitative PCR analyses revealed that EP300, BRD4 and MED1 could bind to the regions containing YY1's S2 and S4/5 consensus sites, but not those of S1 and S3 sites, in the FOXM1 promoter in breast cancer cells ( Figure 6H). Consistent with the observation above, both YY1-H-A and -H mutants, deficient in phase separation, exhibited the least capability in promoting both promoter reporter and endogenous transcript of the FOXM1 gene ( Figures 6I and 6J). Interestingly, in reporter assays of Figure 6I, transfection of YY1-IDR still retained significant Gluc activity, but its expression in two breast cancers could not activate the FOXM1 gene ( Figure 6J). Noteworthily, endogenous YY1 was still present in cells of reporter assays, but knocked down when testing YY1 mutants in breast cancer cells. Thus, endogenous YY1 was likely involved in driving reporter expression, which could be functionally perturbed by YY1 mutants with deficient IDRs, but not the mutant with IDR. Nevertheless, ectopic YY1-IDR could still occupy the promoters of many reporters and thus reduced overall Gluc expression ( Figure 6I).

YY1 promotes the formation of super-enhancer to drive FOXM1 gene expression
YY1 has been reported to regulate gene expression through promoting enhancer activity (27,48,49), while phase separation is a characteristic feature of enhancer mechanism (50). To interrogate whether enhancers were involved in YY1-regulated FOXM1 expression, we surveyed the regions of overlapped YY1 binding enrichment and gene activation markers in the vicinity of the FOXM1 gene in the human genome. Within 650 kbs of the FOXM1 promoter, we identified five candidate enhancer regions, and designated them as E1 to E5 ( Figure 7A). The chromosome conformation capture (3C) approach (32) with restriction enzyme digestion, digested genomic DNA ligation, and PCR amplification of ligated DNA, was used to examine vicinal regions in forming enhancer complexes through interacting with the FOXM1 promoter. Multiple EcoRI-digested segments displayed ligationdependent PCR bands ( Figure 7B), mostly overlapping with enhancers E3, E4 and E5 ( Figure 7A).
Similarly, the genomic DNA digested by HindIII also showed positive PCR bands overlapping with E3, E4 and E5, depending on T4 DNA ligase (Supplementary Figure S6A). Furthermore, shRNAmediated YY1 knockdown eliminated these PCR bands (Supplementary Figure S6B), indicating that physical closeness of the enhancer elements to the FOXM1 promoter was contingent to the presence of YY1. Importantly, we confirmed the precise ligation between digested segments of the FOXM1 promoter and the enhancers by DNA sequencing analysis (Supplementary Figure S6C). In addition, 1,6-hexanediol treatment greatly reduced PCR products ( Figure 7C), suggesting that phase separation is prerequisite for FOXM1 promoter's proximity to enhancers. To assess potential enhancer properties of EcoRI-digested segments 19, 20, 26, 27 and 29, we generated FOXM1 promoter reporters with sub-segmented sequences of these regions located either upstream or downstream according to their natural positions relative to the FOXM1 TSS in the genome ( Figure 7D, top row). In reporter assays, several fragments from these segments showed response to YY1's ectopic expression or knockdown ( Figure 7D, bottom row), consistent with their enhancer identity. In line with these data, treatment of transfected cells by JQ1, an enhancer inhibitor, dampened reporter activities ( Figure 7E).
We further evaluated the importance of YY1-regulated FOXM1 expression in mammary cells. As we previously reported, ectopic YY1 promoted proliferation of primary mammary epithelial cells, while its depletion reduced it (38). In the current study, FOXM1 knockdown or its ectopic expression could significantly counteract cell viability changes in MCF-10A cells (by exogenous YY1) or breast cancer cells (by YY1 silencing), respectively (Supplementary Figure S7A). We reported YY1-promoted AKT activation (51); consistently, FOXM1 could reinstate AKT-T308 and -S473 phosphorylation reduced by YY1 knockdown (Supplementary Figure S7B). FOXM1 depletion attenuated YY1-promoted migration and clonogenicity of MCF-10A cells in wound healing and colony formation assays, while, in these assays, ectopic FOXM1 could significantly rescue the deficiencies of breast cancer cells caused by YY1 silencing (Supplementary Figures S7C and S7D).
Based on our data, we propose a model of YY1-regulated gene activation ( Figure 7F). In this model, a super-enhancer is formed by YY1-mediated phase separation that comprises major coactivators and is stabilized by three distal enhancers to activate FOXM1 promoter and stimulate its gene transcription.

DISCUSSION
Phase separation is a general phenomenon in polymer chemistry, but has recently been developed into a concept or mechanism of biological regulation (52). Liquid-liquid phase separation in different subcellular sections creates membrane-less condensates that compartmentalize biomolecules, such as proteins, RNAs and DNAs, with pertinent biological activities, and allows a specific biological event to be processed in a relatively undisturbed manner (13). Applications of phase separation in transcriptional regulation are the seminal discovery of Young's group through demonstrating the formation of phase-separated condensates that confine various coactivators and function as superenhancers (8,9). These discoveries largely extended our view of transcriptional regulation and revolutionized mechanism or concept of sustained gene expression through super-enhancers. 13 As a key transcription factor, YY1 interacts with numerous epigenetic writers and erasers (16). In addition to recognizing its consensus sites, YY1 binds G-quadruplex structures (53). Importantly, YY1's ability in homodimerization (27,53) allows it to bridge promoters and enhancer elements.
Weintraub et al. reported that YY1 generally occupies active enhancers and promoters in different cell types, and perturbation of YY1 binding disrupts enhancer-promoter looping (27). The promiscuous interaction of YY1 with coactivators can promote their recruitment into the enhancer complexes.
Consistently, a number of studies demonstrated YY1's participation in forming enhancers or superenhancers to regulate genes involved in various biological processes (27,49,(54)(55)(56). These studies strongly support a key role of YY1 in promoting super-enhancer assembly to activate gene expression.
However, despite these exciting indications or hints to us, molecular evidence is still lacking for detailed mechanisms of YY1-regulated enhancer formation. YY1 protein primary structure has several unique features scarcely observed in other proteins, including a highly acidic N-terminus, consecutive E/D-, G-and H-rich regions in the TAD, and four tandem zinc fingers at a basic C-terminus, as well as a high lysine content (32 versus total 414 amino acids) but none of them found in first 157 residues. When scanning YY1 sequence, we discover high propensity of structural disorder in its TAD. Subsequently, our experimental data unequivocally demonstrate YY1's competence in undertaking liquid-liquid phase separation both in vitro and in cell.
Interestingly, we identify the 11H cluster as a critical motif of YY1 to promote its phase separation, which can be extended to four additional histidine cluster-containing proteins.
The YY1-rich nuclear phase-separated condensates are likely super-enhancers based on their participation by major coactivators, including EP300, MED1, BRD4 and CDK9, as well as active Pol II.
With FOXM1 expression as a regulatory model, we also demonstrate the role of YY1 in convening three distal enhancer elements and the FOXM1 promoter to assemble a super-enhancer for gene activation. Strikingly, oligonucleotides containing YY1 consensus sites in the FOXM1 promoter can participate in and steadily facilitate YY1 droplet formation in vitro, suggesting that YY1-mediated phase-separated nuclear condensates are likely involved and promoted by genomic DNA.
Interestingly, Sigova, et al. demonstrated that YY1 could bind both gene regulatory elements and their associated RNA species transcribed from enhancers and promoters in a genome-wide manner (17).
Thus, it is logical to predict that cognate RNAs may participate YY1-coordinated phase separation, which deserves future investigation.
Despite mounting studies showing that YY1 either activates or represses gene expression, we observe YY1 colocalization with coactivators and histone markers for gene activation, but not for repression. Our data are consistent with previous studies showing YY1's general association with active promoters and enhancers (17,27). Thus, although YY1 was reported to suppress gene expression through recruiting corepressors, such as HDACs, EZH2 and DNMTs (57)(58)(59), our data support its primary role as a transcriptional activator.
FOXM1 is recognized as a critical proliferation-associated transcription factor, regulating cell proliferation, self-renewal, and tumorigenesis. As a proliferative gene, FOXM1 overexpression was 14 reported in almost all cancer types, and correlated with patients' poor prognoses (44). Consistently, FOXM1 gene is regulated by a variety of oncogenic TFs, such as E2Fs, MYC, ERα, STAT3 and CREB, as well as FOXM1 itself (44). In the current study, we provide the first evidence to demonstrate FOXM1 expression promoted by super-enhancers formed by YY1-containing phase-separated condensates, which extends molecular mechanisms driving FOXM1 overexpression in cancers.
Interestingly, FOXM1 has also been reported to participate enhancer regulation of other genes (60,61). Based on its self-regulated feature, FOXM1 itself may potentially get involved in superenhancer regulating its expression.
Our provide both in vitro and cell-based data to demonstrate capability of EP300 in forming phase separation. Based on algorithm prediction, we identify five potential IDRs in EP300 primary sequence, and verify that IDR3 (at middle region) and IDR5 (at C-terminus) form droplets in vitro. In two recent studies, EP300 C-terminus was reported to undertake phase separation (62,63), consistent with our results, but Ma, et al. showed the capability of its N-terminus (1-566 amino acids) in forming droplets in vitro, which was not the case in our study. Nevertheless, we provided ample evidence to demonstrate the competence of EP300 in forming phase-separated condensates, including the optimal parameters, and proved its role as a primary coactivator of YY1-mediated enhancer complex.

SUPPLEMENTARY DATA
Supplementary Data are available at NAR online.

ACKNOWLEDGEMENT
WW, DL and GS initiated the study, designed the experiments, analyzed data, and wrote the manuscript; WW, SQ, CY, CZ and DL carried out the research; GL also analyzed data; DBS and JS contributed conceptual suggestions and new reagents/analytic tools.    A. Schematic view across the FOXM1 gene locus (chr12:2,768,321-3,408,320 [hg19]) with genomic and epigenetic information. Graphic active regulatory regions were generated using the ENCODE database, and potential enhancers (E1 to E5) are shaded in yellow. Fragments digested by EcoRI or HindIII are numbered and shown below the graph. B. Chromatin conformation capture (3C) analysis to examine direct physical interactions between the FOXM1 promoter and enhancer elements. The 3C analyses followed the protocol in Materials and Methods, using EcoRI in genomic DNA digestion.
Ligation step was performed with or without T4 DNA ligase as indicated. Numbers on gel top correspond to "EcoRI digestion site" in Figure 8A. PCR to examine direct physical interactions between the FOXM1 promoter and each of EcoRI fragments used the primer sets in Supplementary   Table S3