Live-cell single-molecule tracking reveals co-recognition of H3K27me3 and DNA targets polycomb Cbx7-PRC1 to chromatin

The Polycomb PRC1 plays essential roles in development and disease pathogenesis. Targeting of PRC1 to chromatin is thought to be mediated by the Cbx family proteins (Cbx2/4/6/7/8) binding to histone H3 with a K27me3 modification (H3K27me3). Despite this prevailing view, the molecular mechanisms of targeting remain poorly understood. Here, by combining live-cell single-molecule tracking (SMT) and genetic engineering, we reveal that H3K27me3 contributes significantly to the targeting of Cbx7 and Cbx8 to chromatin, but less to Cbx2, Cbx4, and Cbx6. Genetic disruption of the complex formation of PRC1 facilitates the targeting of Cbx7 to chromatin. Biochemical analyses uncover that the CD and AT-hook-like (ATL) motif of Cbx7 constitute a functional DNA-binding unit. Live-cell SMT of Cbx7 mutants demonstrates that Cbx7 is targeted to chromatin by co-recognizing of H3K27me3 and DNA. Our data suggest a novel hierarchical cooperation mechanism by which histone modifications and DNA coordinate to target chromatin regulatory complexes. DOI: http://dx.doi.org/10.7554/eLife.17667.001


Introduction
Chemical covalent modification of histones and DNA regulates the chromatin structure states that play central roles in chromatin-templated biological processes (Jenuwein and Allis, 2001;Li et al., 2007a;Luco et al., 2011;Ruthenburg et al., 2007). This is exemplified by Polycomb group (PcG) proteins that function as histone-modifying enzymes and regulate gene expression via modulating higher order chromatin structures (Simon and Kingston, 2013). PcG proteins were initially identified as a body structure specification in Drosophila (Lewis, 1978). In mammals, PcG orthologs are essential for normal embryonic development and disease pathogenesis (Helin and Dhanak, 2013). For example, PcG subunits are frequently overexpressed or mutated in cancer, and perturbing PcG interactions can suppress cancer growth (Helin and Dhanak, 2013). Because of their clinical significance, enormous efforts have been devoted to develop drugs for targeting PcG subunits (Helin and Dhanak, 2013). However, the molecular mechanisms by which PcG proteins establish and maintain repressive Polycomb domains are still incompletely understood.
Several mechanisms underlying the targeting of PRC1 to chromatin have been documented (Blackledge et al., 2015;Simon and Kingston, 2013). Initial studies of Drosophila PcG (dPcG) proteins have suggested a mechanism of the PRC2-mediated recruitment of PRC1 (Cao et al., 2002;Min et al., 2003;Wang et al., 2004b). dPRC2 is recruited to Polycomb response elements (PRE) by its interaction with sequence-specific DNA-binding proteins and then modifies chromatin with H3K27me3 that recruits dPRC1. Consistent with the notion, genetic analyses have demonstrated that dPRC1 and dPRC2 co-regulate PcG target genes and dPRC1 is displaced from chromatin in dPRC2 mutants (Cao et al., 2002;Wang et al., 2004b). Genome-wide studies have shown that dPRC1 and dPRC2 co-occupy many PcG target genes (Schwartz et al., 2006).
In mammals, the recruitment of PRC1 is enigmatic and complicated, and has been broadly defined as H3K27me3-dependent and -independent recruitment mechanisms (Blackledge et al., 2015;Farcas et al., 2012;He et al., 2013;Tavares et al., 2012). An additional layer of complexity is added when considering that PRC1, in some cases, recruits PRC2 Cooper et al., 2014;Kalb et al., 2014). The H3K27me3-dependent recruitment of mammalian PRC1 originates from the Drosophila model and is based on the facts that the Cbx family members and dPc both contain a conserved chromodomain (CD) (Blackledge et al., 2015). The model is consistent with studies demonstrating a link between H3K27me3 and PRC1 recruitment (Agger et al., 2007;Boyer et al., 2006;Lee et al., 2007;Mujtaba et al., 2008). Although the model for the mammalian Cbx-PRC1 recruitment is prevalent, several lines of evidence argue against the proposed model as a general mechanism of action. First, unlike to the dPc CD, the Cbx CDs have a much weaker affinity for H3K27me3 (Bernstein et al., 2006;Kaustov et al., 2011;Tardat et al., 2015). The Cbx2 CD shows preference for H3K27me3 while the Cbx4 and Cbx7 CDs exhibit preference for H3K9me3 (Bernstein et al., 2006;Kaustov et al., 2011;Tardat et al., 2015). The affinity of the Cbx6 and Cbx8 CDs for H3K27me3 is nearly undetectable (Bernstein et al., 2006;Kaustov et al., 2011). One question is whether the recognition of H3K27me3 by the Cbx CDs is required for the targeting of Cbx proteins to chromatin. Likewise, genome-wide approaches have demonstrated that H3K27me3 forms a broad domain and the binding PRC1 is sharply localized within the H3K27me3 domain, and that a subset of H3K27me3 domains corresponds to PRC1 binding sites (Ku et al., 2008). Thus, there are missing molecular links between genetic, biochemical, and genome-wide analysis for our understanding of how the Cbx-PRC1 complexes are targeted to chromatin.
Single-molecule techniques have been widely applied to study DNA-and chromatin-templated processes in vitro and provide insights into genetic information flow in vivo (Bell and Kowalczykowski, 2016;Dangkulwanich et al., 2014;Duzdevich et al., 2014;Geertsema and van Oijen, 2013;Harada et al., 2016;Herbert et al., 2008;Li et al., 2004;Ngo et al., 2015;Ren et al., 2003;Ren et al., 2006;Tatavosian et al., 2015). Recent advances in single-molecule imaging allow measuring the quantitative kinetics of gene control in living mammalian cells Coleman et al., 2015;Gebhardt et al., 2013;Grimm et al., 2015;Izeddin et al., 2014;Katz et al., 2016;Knight et al., 2015;Liu et al., 2015Liu et al., , 2014Mazza et al., 2012Mazza et al., , 2013Morisaki et al., 2014;Normanno et al., 2015;Swinstead et al., 2016;Zhang et al., 2014). Here, we combine live-cell SMT and genetic engineering to determine whether H3K27me3 is required for the targeting of Cbx proteins to chromatin and to dissect the targeting mechanisms. Single-molecule quantitative measurement is used to determine the kinetics and dynamics of the Cbx protein interactions with chromatin in living mouse embryonic stem (mES) cells. The analyses demonstrate a new functional role of the Cbx-PRC1 complex formation in the targeting of Cbx7 to chromatin and uncover the molecular mechanism underlying the targeting of Cbx7 to chromatin and fill in the knowledge gap between genetic, biochemical, and genome-wide analyses. These results contribute significantly to our quantitative understanding of kinetics and dynamics of the Cbx-PRC1 proteins in living cells, allowing us to suggest the molecular mechanisms underlying how the Cbx-PRC1 complexes are targeted to chromatin.

Results
Validation of live-cell SMT using HaloTag and histone H2A fused to HaloTag To investigate the Cbx proteins binding dynamics at endogenous genomic loci, we performed SMT to determine diffusion and chromatin binding properties of individually fluorescently labeled Cbx molecules within living mES cells. HaloTag was fused at the N-terminus of Cbx proteins under an inducible, tetracycline response element (TRE)-tight promoter ( Figure 1A). These fusion genes were stably integrated into the genome of wild-type (PGK12.1) mES cells. We used highly inclined thin illumination (HILO) to avoid stray-light reflection and to reduce background from cell auto-fluorescence (Tokunaga et al., 2008) ( Figure 1B). The HaloTag ligand of the bright, photostable fluorophore Janelia Fluor 549 (JF 549 ) allowed for visualization of single HaloTag-Cbx molecules at their basal expression level without doxycycline induction (Grimm et al., 2015) ( Figure 1C).
The above D m analysis involves averaging over independent pairs of the squared jump distance of a single trajectory with a 30-ms interval. Such averaging might obscure transitions between chromatin-binding, confined, and Brownian motion for the single trajectory of a particle within the observation time. To investigate whether the averaging affects resolving the kinetic fractions, we calculated D f 1 based on the squared jump distance between the initial position r 0 and the first position D m of a single-trajectory with a 30-ms interval, and constructed the logD f 1 distribution (Figure 1-figure supplement 1B and Figure 1-source data 1). Counting only the first displacement of each track has been reported previously for studying of transcription factors binding to DNA (Gebhardt et al.,  Distinct chromatin-binding behaviors among the Cbx proteins At the single-molecule level, we quantitatively measured diffusion constants and chromatin-binding levels of the Cbx proteins in mES cells (Video 3-7). We fitted the histograms with a three-component Gaussian function and calculated the diffusion constants and the fractional sizes of the individual populations ( Figure 1E and Figure 1-source data 1). Since the peak centers of the CB populations for the HaloTag-Cbx proteins were almost the same as that for H2A-HaloTag, we fixed logD m1 to be the value À1.5 (D m1 = 0.032 mm 2 s -1 )(see Materials and methods). We measured F 1 = (30 ± 1)%, F 2 = (61 ± 1)%, and F 3 = (9 ± 4)% for HaloTag-Cbx2, F 1 = (10 ± 1)%, F 2 = (50 ± 1)%, and F 3 = (40 ± 2)% for HaloTag-Cbx4, F 1 = (17 ± 1)%, F 2 = (47 ± 2)%, and F 3 = (36 ± 3)% for HaloTag-Cbx6, F 1 = (29 ± 1)%, F 2 = (49 ± 1)%, and F 3 = (22 ± 3)% for HaloTag-Cbx7, and F 1 = (26 ± 1)%, F 2 = (50 ± 2)%, and F 3 = (24 ± 4)% for HaloTag-Cbx8 ( Figure 1F and Supplementary file 1). To complement . These data provided a few novel observations: (1) the Cbx proteins exhibit distinct chromatin-associating capacities, (2) Cbx2, Cbx7, and Cbx8 exhibit the highest chromatin-bound level while Cbx4 has the lowest one, (3) the fractional sizes of the FD components are distinct among the Cbx family proteins, (4) except for Cbx2, the fractional sizes of the ID components are similar among the Cbx proteins, and (5) among the Cbx proteins, the diffusion constants are distinct for the ID components, but similar for the FD components. Altogether, our results demonstrate that the Cbx proteins employ distinct ways to interact with chromatin and to explore the nucleus.
The above SMT experiments were performed in wild-type mES cells where the endogenous and exogenous (fusion) proteins co-exist. HaloTag may make the fusion proteins less-equal competition with their endogenous counterparts. Given that Cbx7 is the major Cbx protein within mES cells (Morey et al., 2013;Morey et al., 2012), we integrated HaloTag-Cbx7 to the genome of Cbx7 À/À mES cells and performed SMT (Video 8). We measured F 1 = (30 ± 1)%, F 2 = (46 ± 1)%, and F 3 = (24 ± 2)% for HaloTag-Cbx7, which are comparable to those obtained from HaloTag-Cbx7 in wild-type mES cells ( Figure 1-source data 1). Next, we performed biochemical analysis of Cbx7. Immunoblotting indicated that the level of HaloTag-Cbx7 protein was less than that of its endogenous counterpart (Figure 1-figure supplement 3A). Chromatin immunoprecipitation (ChIP) analysis indicated that HaloTag antibody greatly precipitated promoters of Polycomb target genes from HaloTag-Cbx7/Cbx7 À/À mES cells, but much less from wild-type mES cells (Figure 1-figure supplement 3B), suggesting that the HaloTag-Cbx7 protein binds to Polycomb target genes.

Effects of the Cbx7 CD on the residence time of Cbx7 at chromatin
To determine the binding kinetics, we measured the in vivo residence time of Cbx7 molecules bound to chromatin. To reduce the photobleaching of JF 549 , we performed time-lapse experiments at an integration time, t int , of 30 ms interspersed with a dark time, t d , of 170 ms ( Figure 4A and Video 26).
We calculated diffusion coefficients of individual HaloTag-Cbx7 molecules and considered molecules to be bound to chromatin if their D m was < 0.10 mm 2 /s. The dwell times of individual stationary Cbx7 molecules were directly measured as the lifetime of the fluorescence spots. The cumulative frequency distributions of dwell times were fitted with a two-component exponential decay function (see Materials and methods) ( Figure 4B and . We then investigated the residence times of the Cbx7 variants at chromatin (Video 27-29). The residence times of the stable chromatin-bound populations were determined for HaloTag-CD Cbx7 (t sb = 4.7 ± 0.1 s), HaloTag-Cbx7 F11A (t sb = 5.8 ± 0.1 s), and HaloTag-Cbx7 4CD (t sb = 4.7 ± 0.1 s) ( Figure 4C and Supplementary file 2). Thus, the residence times of the stable chromatin-bound molecules of HaloTag-Cbx7 F11A was longer than that of HaloTag-CD Cbx7 and HaloTag-Cbx7 4CD , but shorter than that of HaloTag-Cbx7 ( Figure 4C and Supplementary file 2). The F 1sb level of HaloTag-Cbx7 F11A was higher than that of HaloTag-CD Cbx7 and HaloTag-Cbx7 4CD , but was less than that of HaloTag-Cbx7 ( Figure 4D and Supplementary file 2). To allow visual comparison among Cbx7 and its variants, we plotted their survival probability in the same figure ( Figure 4E and Figure 4-source data 1). We observed that HaloTag-Cbx7 F11A stays a longer time at chromatin than HaloTag-CD Cbx7 and HaloTag-Cbx7 4CD , but a shorter time than HaloTag-Cbx7 ( Figure 4E). Altogether, our results suggest that the interaction of H3K27me3 and CD Cbx7 is not enough for the stabilizing of Cbx7 at chromatin.

CD Cbx7 and ATL Cbx7 together constitute a DNA-binding unit
Since the Cbx-PRC1 components tested are not required for the targeting of Cbx7 to chromatin, we turned our attention into Cbx7 itself. In addition to the conserved CD, Cbx7 harbors an ATL motif adjacent to CD (Senthilkumar and Mishra, 2009) ( Figure 6A). Since the Cbx7 ATL (ATL Cbx7 ) contains 6 basic amino acids out of 16, we postulated that the ATL motif may be involved in nucleic acid-binding. To test this hypothesis, we generated Cbx7 variants ( Figure 6-figure supplement 1) and performed electrophoretic mobility shift assay (EMSA) ( Figure 6B). EMSA analysis indicated that CD Cbx7 has no DNA-binding activity, consistent with early studies (Bernstein et al., 2006). Although ATL Cbx7 contains a high content of basic amino acids, EMSA analysis demonstrated that ATL Cbx7 has undetectable DNA-binding activity, consistent with previous report that ATL does not bind to DNA (Reeves and Nissen, 1990). However, under the same conditions, CD-ATL Cbx7 showed clear DNA-  binding activity, suggesting that the DNA-binding activity requires both CD Cbx7 and ATL Cbx7 . To test whether the basic amino acids of ATL Cbx7 affect the DNA-binding activity, we substituted these basic amino acids with alanine or glycine to generate CD-ATLm Cbx7 . EMSA analysis showed that the substitution abolishes the DNA-binding activity of CD-ATL Cbx7 . As a control, GST did not bind to DNA. The DNA-binding capacity of CD-ATL Cbx7 is concentration-dependent ( Figure 6C). The K d was determined to be~1.0 mM, which is much smaller than the CD Cbx7 binding to H3K27me3 peptide (Bernstein et al., 2006;Kaustov et al., 2011;Tardat et al., 2015).
Previous studies have shown that CD Cbx7 binds to RNA with affinity of~50 mM (Bernstein et al., 2006;Yap et al., 2010). To compare the relative affinity of CD-ATL Cbx7 binding to DNA versus RNA, we performed competitive assays. 0.5 mM of double-stranded DNA-2 (dsDNA-2), with sequence being different from dsDNA-1, completely dissociated the fluorescently labeled dsDNA-1 (0.1 mM) from CD-ATL Cbx7 ( Figure 6D). At 4.0 mM, single-stranded DNA (ssDNA) could not completely dissociate dsDNA-1 from CD-ATL Cbx7 ( Figure 6E). Under the identical conditions, 10 mM of double-stranded RNA could not completely dissociate dsDNA-1 from CD-ATL Cbx7 ( Figure 6F). Likewise, 40 mM of single-stranded RNA had no noticeable effects on the association CD-ATL Cbx7 with dsDNA-1 ( Figure 6G). Altogether, our results demonstrate that CD-ATL Cbx7 preferentially recognizes dsDNA rather than ssDNA, dsRNA and ssRNA. Source data 1. Source data for Figure 4B and

CD Cbx7 and ATL Cbx7 together control the targeting of Cbx7 to chromatin
To investigate whether the in vitro capacity of the binding of CD-ATL Cbx7 to DNA has a functional role in the targeting of Cbx7 to chromatin in vivo, we made Cbx7 variants lacking DNA-binding ability and stably expressed these variants in wild-type mES cells ( Figure 7A and Figure 3-figure supplement 1). The histograms for HaloTag-Cbx7 4ATL were fitted with three populations ( Figure 7B, Video 34, and Figure 7-source data 1). We measured F 1 = (16 ± 1)%, F 2 = (44 ± 2)%, and F 3 = (40 ± 3)% ( Figure 7C and Supplementary file 1), indicating that the fractional size of the CB component of HaloTag-Cbx7 4ATL is about half of that of HaloTag-Cbx7. The histograms for HaloTag-Cbx7 ATLm indicated three components ( Figure 7B, Video 35, and Figure 7-source data 1). We measured F 1 = (16 ± 1)%, F 2 = (60 ± 1)%, and F 3 = (24 ± 3)% ( Figure 7C and Supplementary file 1), indicating that the fractional size of the CB component of HaloTag-Cbx7 ATLm is comparable to that of Halo-Tag-Cbx7 4ATL . Thus, our data indicate that the ATL Cbx7 motif is required for the efficient targeting of Cbx7 to chromatin.
Given that CD Cbx7 and ATL Cbx7 constitute an H3K27me3-and DNA-binding unit, we tested the role of CD-ATL Cbx7 in the targeting of Cbx7 to chromatin. We deleted both CD Cbx7 and ATL Cbx7 to generate HaloTag-Cbx7 4CD-ATL that was stably expressed in wild-type mES cells ( Figure 7A Figure 7-source data 1). We measured F 1 = (8 ± 3)%, F 2 = (46 ± 1)%, and F 3 = (46 ± 1)% for HaloTag-Cbx7 4CD-ATL ( Figure 7C and Supplementary file 1), indicating that the F 1 level was lower than that for HaloTag-Cbx7 4CD and HaloTag-Cbx7 4ATL (Supplementary file 1). Thus, our data suggest that CD Cbx7 and ATL Cbx7 function together as a unit to mediate the targeting of Cbx7 to chromatin.

Discussion
In this study, we have elucidated the recruitment mechanism for the Cbx-PRC1 complexes by integrating approaches from live-cell SMT, genetic engineering, and biochemistry. We have demonstrated that H3K27me3 has a central and a direct role in the recruitment of Cbx7 and Cbx8 to chromatin in vivo, while plays a less important role in the targeting of Cbx2, Cbx4, and Cbx6 to chromatin. We have identified that the CD-ATL Cbx7 cassette functions as a unit that co-recognizes H3K27me3 and DNA and regulates the targeting of Cbx7 to chromatin. These results challenge the prevailing view that all Cbx family members require H3K27me3 for the targeting of them to chromatin and provide new insights into the genetic, biochemical, and genome-wide analysis for our understanding of the Cbx-PRC1 targeting mechanisms. We propose that a hierarchical cooperation between a low-affinity H3K27me3-binding CD Cbx7 and a high-affinity DNA-binding CD-ATL Cbx7 targets Cbx7-PRC1 to chromatin.   Targeting the Cbx family members with and without dependence on PRC2 and H3K27me3 During evolution, the number of genes encoding Cbx proteins has increased, which has resulted in structural and functional diversification Klauke et al., 2013;Morey et al., 2012;Ren and Kerppola, 2011;Ren et al., 2008;Tatavosian et al., 2015;Vincenz and Kerppola, 2008;Whitcomb et al., 2007;Zhen et al., 2014). At the single-molecule level, we quantified the kinetic fractions of the Cbx proteins within living mES cells and revealed that~30% of Cbx2, Cbx7, and Cbx8 associate with chromatin at a given time period while~10-15% of Cbx4 and Cbx6 bind to chromatin. The fractional sizes and diffusion constants of the ID populations among the Cbx family members are distinct, suggesting that the Cbx proteins employ distinct mechanisms to explore the nucleus of the cell.
At the single-molecule sensitivity, we demonstrated that Cbx7 and Cbx8 are displaced from chromatin in Eed À/À and Ezh2 À/À mES cells. The introduction of Eed into Eed À/À mES cells and of Ezh2 into Ezh2 À/À mES cells restored the Cbx7 and Cbx8 association with chromatin and the H3K27me3 level. Thus, it is likely that H3K27me3 directly controls the association of Cbx7 and Cbx8 with chromatin. Consistent with this notion, previous genome-wide ChIP-Seq analysis demonstrated that Cbx7 is displaced from chromatin in Eed À/À mES cells (Morey et al., 2013). We found that the removal of H3K27me3 has no or small effects on the association of Cbx2, Cbx4, and Cbx6 with chromatin. No effects on the Cbx6 association with chromatin is consistent with previous studies where Cbx6 does not interact with Ring1b and only 5% of Cbx6 target genes are occupied by H3K27me3 in mES cells (Morey et al., 2012). In contrast to Cbx6, Cbx2 and Cbx4 form the Cbx-PRC1 complex and overlap with H3K27me3 Polycomb domains (Gao et al., 2012;Mardaryev et al., 2016). Cbx4 is a SUMO E3 ligase and can function as the H3K27me3-dependent or -independent way (Kagey et al., 2003;Li et al., 2014;Mardaryev et al., 2016;Roscic et al., 2006). Our recent study has shown that Cbx2 is targeted to mitotic chromosomes independently of PRC1 and PRC2, and directly recruits the canonical PRC1 components to mitotic chromosomes . Another study has demonstrated that Cbx2 targets the canonical PRC1 to constitutive heterochromatin by directly recognizing pericentromeric chromatin during early mouse development (Tardat et al., 2015). Additionally, in vitro study has shown that Cbx2 can directly bind to and compact reconstituted nucleosomes (Grau et al., 2011). Thus, these studies suggest additional mechanisms exist to target Cbx2, Cbx4, and Cbx6 to chromatin.
Targeting of the Cbx7 protein to chromatin by co-recognition of H3K27me3 and DNA We observed that the level of CD Cbx7 at chromatin is less than 30% of Cbx7 and that the residence time of the stable chromatin-bound population of CD Cbx7 is about 65% of Cbx7. Mutational analysis demonstrated that Cbx7 F11A and Cbx7 4CD both remain associating with chromatin. These data imply that additional factor(s) exist(s) to target the Cbx7 protein to chromatin. Cbx7 contains two conserved domains: CD and ATL (Senthilkumar and Mishra, 2009). Consistent with previous reports, our data showed that CD Cbx7 and ATL Cbx7 do not bind to DNA, respectively (Bernstein et al., 2006;Reeves and Nissen, 1990). Interestingly, our results demonstrated that CD Cbx7 and ATL Cbx7 together function as a DNA-binding unit. CD-ATL Cbx7 exhibited much higher affinity for dsDNA than for ssDNA, dsRNA, and ssRNA. The DNA-binding capacity of CD-ATL Cbx7 was functionally significant. Perturbation of the DNA-binding capacity of CD-ATL Cbx7 impaired the level of Cbx7 at chromatin and reduced the Cbx7 residence time. Deletion of both CD Cbx7 and ATL Cbx7 results in the significantly reduced level of the stable chromatin-bound population. Thus, our data demonstrate that the co-recognition of H3K27me3 and DNA by the CD-ATL Cbx7 module contributes significantly to the targeting of Cbx7 to chromatin.
Since histone-modifying enzymes typically reside in protein complexes, components within the protein complexes often contribute to targeting of them to chromatin by multivalent engagement of chromatin (Lalonde et al., 2014;Rando, 2012;Ruthenburg et al., 2007). Given that previous studies have shown that Mel18 binds DNA directly in vitro (Akasaka et al., 1996)  in the increased chromatin-bound levels and the increased residence times of Cbx7. Further studies are needed to understand whether the Cbx-PRC1 complex formation is required for the targeting specificity of Cbx7.

Hierarchical cooperation between DNA and H3K27me3
Hierarchical cooperation within chromatin regulatory proteins or complexes between unmodified DNA and histone markers is emerging as a mechanism for gene control. For example, SWR1 is recruited to promoter regions containing nucleosome free region > 50 bp and an adjoining nucleosome by the nanomolar DNA-binding affinity of Swc2, a subunit of SWR1. Once bound, the micromolar affinity of Bdf1 bromodomains for acetylated histones directs SWR1 binding to the +1 nucleosome over the -1 nucleosome (Ranjan et al., 2013). Thus, hierarchical cooperation between DNA and histone modifications could underpin the SWR1's role in promoting H2A.Z replacement. Another example is that the Rpd3S histone deacetylase complex binds to H3K36-methylated dinucleosome with 100 pM affinity by multiple engagements of histone modifications and DNA (Huh et al., 2012;Li et al., 2007b). The DNA-and histone-binding abilities of Eaf3, a subunit of Rpd3S, are self-contained and allosterically regulated by Rco1, another subunit of Rpd3S (Ruan et al., 2015).
Our results suggest that the mechanism of targeting of Cbx7 to chromatin is dependent on hierarchical cooperation via co-recognition of DNA and H3K27me3 by the CD-ATL Cbx7 entity ( Figure 8). We propose that Cbx7-PRC1 is recruited to chromatin by the CD Cbx7 recognition of H3K27me3. We hypothesize that the interaction between H3K27me3 and CD Cbx7 triggers conformational changes of the Cbx7-PRC1 complex, which drive the high-affinity interaction between DNA and CD-ATL Cbx7 . This hypothesis is consistent with our observation that the removal of H3K27me3 significantly reduces the targeting of Cbx7 and Cbx8 to chromatin. Implicit in this model is that the binding of CD-ATL Cbx7 to DNA is auto-inhibited by unknown mechanisms and allosterically regulated by the CD Cbx7 interaction with H3K27me3. Previous studies have shown that H3K27me3 allosterically activates the methyltransferase activity of the PRC2 complex by its interaction with the C-terminus of Eed (Jiao and Liu, 2015;Margueron et al., 2009)

Generation of transgenic mES cell lines by lentivirus transduction
Establishing the mES cell lines stably expressing the Polycomb and H2A genes was performed as described previously Zhen et al., 2014). HEK293T cells at 85-90% confluency were co-transfected with 21 mg pTRIPZ (M) containing the fusion gene, 21 mg psPAX2, and 10.5 mg pMD2.G by using calcium phosphate precipitation. At the time of 12 hr after transfection, the medium was replaced with 10 ml DMEM supplemented with 10% FBS, 2 mM L-glutamine, 100 units/ml penicillin G sodium, and 0.1 mM b-mercaptoethanol. At the time of 48-50 hr after medium change, the medium was harvested to transduce mES cells in the presence of 8 mg/ml polybrene (H9268; Sigma-Aldrich, St Louis, MO) and LIF. For co-transducing multiple genes, lentiviruses were produced separately and mixed at the time of transduction. At the time of 72 hr after transduction, infected cells were selected by using 1.0-2.0 mg/ml of puromycin (P8833; Sigma-Aldrich, St Louis, MO). Unless otherwise indicated, for live-cell single-molecule imaging experiments, the fusions were expressed at the basal level without administrating doxycycline.

EMSA
Alexa Fluor 488-labelled dsDNA-1, dsDNA-2, and ssDNA were purchased from IDT. dsRNA and ssRNA were kindly provided by Dr. Marino Resendiz (University of Colorado Denver). The GST-Cbx7 fusion proteins were mixed with Alexa Fluor 488-labelled dsDNA-1 in binding buffer (20 mM HEPES pH 7.9, 100 mM KCl, 1 mM EDTA pH 8, 5 mM DTT, 0.05 mg/ml bovine serum albumin, and 0.1% NP-40). For competitive assay, DNA and RNA were added to the reaction mixture. After incubation at room temperature for 15 min, 20% glycerol was added to the reaction. The mixtures were then loaded to the wells of Novex 10% Tris-Glycine Mini Protein Gels (EC6075BOX; Life Technologies, Carlsbad, CA). The gels were run for 90 min at 100 V and 400 mA at 4˚C in the dark. The gels were imaged using ChemiDoc XRS system (Bio-Rad). The intensities of bands were quantified using ImageJ (http://imagej.nih.gov/ij/).

Labelling HaloTag fusion proteins with HaloTag ligand in living cells
Labelling HaloTag Fusion Proteins is described in more detail at Bio-protocol (Duc and Ren, 2017). 24 hr prior to imaging, mES cells stably expressing HaloTag fusion proteins were seeded to gelatin-coated cover glass dish. Several concentrations (5 nM, 15 nM, and 30 nM) of Janelia Fluor 549 (JF 549 ) HaloTag ligand were used to treat cells for 15 min at 37˚C in 5% CO 2 . Cells were washed with the mES cell medium once and then incubated in the mES cell medium at 37˚C in 5% CO 2 for 30 min. After replacing with the live-cell imaging medium (A1896701, FluoroBrite DMEM, Life Technologies, Carlsbad, CA), cells were maintained at 37˚C using a heater controller (TC-324; Warner Instrument, Hamden, CT) during imaging. Each dish was used for a maximum of 1.5 hr after placing them on the microscope. The number of individual fluorescent spots was typically~10-50 spots per nucleus by controlling the HaloTag ligand concentration.

Single-molecule optical setup and image acquisition
Live-cell single molecule tracking was conducted by using a Zeiss Axio Observer D1 Manual Microscopy (Zeiss, Germany) equipped with an Alpha Plan-Apochromatic 100Â/1.46 NA Oil-immersion Objective (Zeiss, Germany) and an Evolve 512 Â 512 EMCCD camera (Photometrics, Tucson, AZ). Additional magnification of 2.5Â was placed on the emission pathway and thus the overall magnification was 250Â. The pixel size of the EMCCD was 16 mm. A laser beam from solid state laser (Intelligent Imaging Innovations, CO) was focused on a rotating mirror, which allows to choose wild-field or inclined excitation configuration. The inclined excitation was used to avoid stray-light reflection and reduce background from cell auto-fluorescence (Tokunaga et al., 2008). JF 549 was excited at 552 nm. A Brightline single-band laser filter set (Semrock; excitation filter: FF01-561/14, emission filter: FF01-609/54, and dichroic mirror: Di02-R561-25) was used to filter the excitation and emission wavelength. The microscope and the EMCCD camera were controlled by Slidebook 6.0 software. A laser power intensity of~15 mW was used to study diffusion components and a power intensity of 5 mW for residence times (dissociation constants).

Single-molecule localization and tracking
U-track algorithm was used for tracking and linking single particles (Jaqaman et al., 2008). Before analysis, stacks of images were visually checked and stacks with movement and drift were discarded. About two-thirds of stacks were discarded. The particle localization (x, y) was obtained through 2D Gaussian fitting based on a u-track algorithm using Matlab. A 10-pixel search radius upper limit was allowed for frame-to-frame linking. The detailed localization and tracking parameters were listed in the Supplementary file 3. A Matlab script was developed to process the output of 2D tracking from the u-track and to convert the trajectories into a matrix form.

Extraction of diffusion components
Our SMT was the 2-dimensional projection of the 3-dimensional motion of HaloTag labelled molecules. We assumed that the HaloTag-labelled molecules diffuse isotopically along the three-dimensional axes X, Y, and Z. Thus, the XY projection data reflect the 3-dimensional motion of the molecules. We performed 30-ms integration time without interval. To count labelled molecules from short tracks and to avoid bias toward slowly moving particles that remain visible for longer times, we calculated two kinds of diffusion coefficients: the maximum likelihood diffusion coefficient ðD m Þ per track and the diffusion coefficient of the first step ðD f 1 Þ per track.
where r 2 i and r 2 f 1 are the mean squared step size and the squared first-step size, respectively, and t equals 30 ms. The underlying assumption for this analysis was that particles undergo the lateral Brownian motion. An R script was developed to calculate D m and D f 1 diffusion coefficients from SMT data (https://gist.github.com/dododas/fb34dc8d9ee5f7d30ebc). The resulting distributions of the logarithm of diffusion coefficients logD m were pooled from data generated from three independent imaging dishes. We assumed that the chromatin-bound HaloTag-Cbx7 molecules are stationary at chromatin. Thus, the diffusion constant of the chromatin-bound population of the HaloTag-Cbx proteins approximately equals that of the nucleosomal H2A-HaloTag. To estimate the diffusion coefficient of the chromatin-bound component of the HaloTag-Cbx proteins, the distributions of logD m from the control H2A-HaloTag in wild-type mES cells were fitted with a three-component Gaussian function by OriginLab (OriginLab Corporation).
where logD m is offset, x i is the center of the peak, A i is the area of the peak, and w i is the full width at half maximum. The diffusion coefficient of the nucleosomal H2A-HaloTag was determined to be D m1 = 0.032 mm 2 s -1 . To systematically compare the CB levels, the subsequent distributions of the HaloTag-Cbx proteins and their variants were fitted with a three-component Gaussian function using the fixed value D m1 = 0.032 mm 2 s -1 while other parameters were set free. There was no convergence if the distributions for HaloTag-NLS in PGK12.1 mES cells and HaloTag-Cbx7 in Eed À/À mES cells were fitted with a three-component Gaussian function. Thus, a two-component Gaussian function was used for the two distributions. The distributions of the logarithm of diffusion coefficient have previously been used to separate individual populations and to estimate their diffusion coefficients and relative abundance (Liu et al., 2014;Normanno et al., 2015;Saxton, 1997). Fractions of diffusion components were calculated as follows.
We denoted the F 1 component as the chromatin-bound (CB) population, F 2 as the intermediate diffusion (ID) population, and F 3 as the fast diffusion (FD) population. Errors were calculated as the s.d. of parameters obtained from fits.

Determination of residence time
To calculate residence time and survival probability of molecules on chromatin, we performed 30-ms integration time and 170-ms dark time. The track lengths and diffusion coefficients were calculated as described above. We selected molecules for at least two consecutive frames with the maximum likelihood diffusion coefficient logD m < 0.10 mm 2 /s as chromatin-bound molecules. 97% of H2A-Halo-Tag molecules had diffusion coefficient below this threshold. The duration of individual tracks (apparent residence time) was directly calculated based on the track length. We estimated the residence times of Cbx7 and its variants using the cumulative frequency distribution of dwell times as described in (Mazza et al., 2012;Mazza et al., 2013;Morisaki et al., 2014). To determine the photobleaching rate of JF 549 , mES cells stably expressing H2A-HaloTag were incubated with 500 nM JF 549 as described above. Live-cell image stacks were taken using the same power and integration and dark time as that for the studying residence times. 9 curves have been obtained. The curves were normalized to 1 and averaged. The averaged curve of photobleaching decay was better described with a two-component exponential decay function based on the F-test implemented in OriginLab.
where y 0 is offset, f b1 and f b2 are amplitude, and 1 tb1 and 1 tb2 are photobleaching rates. The cumulative frequency distributions of dwell times were normalized for photobleaching by dividing by B t ð Þ as described in (Mazza et al., 2012;Mazza et al., 2013;Morisaki et al., 2014). The normalized cumulative frequency distributions were better fitted with a two-component exponential decay function based the F-test implemented in OriginLab.
y ¼ y 0 þ B 1 e Àt=ttb þ B 2 e Àt=tsb where y 0 is offset, B 1 and B 2 are amplitude, and t tb and t sb are residence times of the transient chromatin-bound component and the stable chromatin-bound component, respectively. Among the chromatin-bound population, fractions of the transient chromatin-bound component (F 1tb ) and the stable chromatin-bound component (F 1sb ) were calculated as follows.
where F 1 is the chromatin-bound fraction obtained from fitting the distribution of the logarithm of diffusion coefficient.