The methyltransferase Suv39h1 links the SUMO pathway to HP1α marking at pericentric heterochromatin

The trimethylation of histone H3 on lysine 9 (H3K9me3) – a mark recognized by HP1 that depends on the Suv39h lysine methyltransferases (KMTs) – has provided a basis for the reader/writer model to explain HP1 accumulation at pericentric heterochromatin in mammals. Here, we identify the Suv39h1 paralog, as a unique enhancer of HP1α sumoylation both in vitro and in vivo. The region responsible for promoting HP1α sumoylation (aa1–167) is distinct from the KMT catalytic domain and mediates binding to Ubc9. Tethering the 1–167 domain of Suv39h1 to pericentric heterochromatin, but not mutants unable to bind Ubc9, accelerates the de novo targeting of HP1α to these domains. Our results establish an unexpected feature of Suv39h1, distinct from the KMT activity, with a major role for heterochromatin formation. We discuss how linking Suv39h1 to the SUMO pathway provides conceptual implications for our general view on nuclear domain organization and physiological functions.

H eterochromatin at pericentric domains is essential for centromere function and genome stability 1,2 . Among the major proteins conserved from fission yeast to mammals considered critical in organizing pericentric heterochromatin, the Suv39h lysine methyltransferases (KMTs) and HP1 proteins stand out. Indeed, Suv39h KMTs are responsible for the accumulation at pericentric heterochromatin of histone H3 lysine 9 trimethylation (H3K9me3) 3,4 , a mark epigenetically inherited in S. pombe (refs 5-7). In turn, HP1 recognizes and binds H3K9me3 (refs 8,9). In line with the writer/reader model in the histone code hypothesis 10,11 , it contributes to the pericentric enrichment of HP1 and heterochromatin formation. In mice, Suv39h KMTs are encoded in the genome at two loci Suv39h1 and Suv39h2 as two paralogs sharing 59% sequence identity 12 . Fluorescence recovery after photobleaching studies 13,14 showed that Suv39h1 is more dynamic at pericentric heterochromatin compared with Suv39h2. Furthermore, they are differentially expressed during early mouse development and in adult tissues 12,15 . Thus, while in the current view, Suv39h1 and Suv39h2 are thought to provide essentially redundant activities, these data hint that these paralogs might have distinct properties or a unique function beyond H3K9 methylation. We recently found that in the absence of Suv39h-dependent H3K9me3, HP1a sumoylation could promote its de novo targeting to pericentric heterochromatin in mammalian cells 16 . We thus proposed at least two distinct steps in heterochromatin formation: first 'seeding' requiring HP1 sumoylation, and second amplification and propagation exploiting H3K9me3. The importance in this scheme of a SUMO-dependent step raises the issue of the identification of the SUMO E3 ligase specific for HP1a at pericentric domains. Indeed, in the SUMO conjugation pathway, the final step is catalysed by a limited number of SUMO E3 ligases [17][18][19][20] . These SUMO ligases, by facilitating the transfer of SUMO from the E2 conjugating enzyme (Ubc9) to a given substrate, increase specificity of SUMO conjugation. By using SUMO interacting motifs to recognize SUMO and by direct interactions with Ubc9, they engage the SUMO-loaded Ubc9 such that it is primed for transfer of SUMO to substrate 19,21 . The bestcharacterized SUMO E3 ligases are the Siz/PIAS (SP)-RING family and RanBP2 (ref. 22). Interestingly, a member of the PIAS protein family is encoded by the Su(var)2-10 locus in Drosophila 23 . In this work, we investigated whether Suv39h1 and Suv39h2, known to interact with HP1a (refs 24,25), could function to enhance SUMO modification of HP1a. Using a combination of biochemical and cellular studies, we report that the Suv39h1 paralog enhances HP1a sumoylation. This capacity is independent from its H3K9 methyltransferase activity. In the N-terminal portion of the protein of key importance for this activity, we identify key residues for binding to Ubc9. Importantly, the same region promotes targeting of HP1a at pericentric domains when its binding to Ubc9 is intact. Our results unveil an unanticipated function for Suv39h1 that provides a missing link between the SUMO pathway and HP1a marking at pericentric heterochromatin.

Results
Intact Suv39h1 but not Suv39h2 enhances HP1a sumoylation. We hypothesized that the two paralogs, Suv39h1 and Suv39h2, in mice represented candidates to enhance HP1a sumoylation.
Comparison of their domain organization shows that beside a 82aa N-terminal extension in Suv39h2, they are highly similar (Fig. 1a). Using an RNA pull-down strategy described in ref. 16 that previously allowed us to successfully reveal SUMO-1-HP1a, we found that, in mouse embryonic fibroblasts (MEFs) derived from Suv39h double-null mice 4 , the proportion of SUMO-1-HP1a versus total HP1a dramatically decreased (Fig. 1b). This first indication in line with a loss of function prompted us to further explore a possible direct contribution of the Suv39h enzymes in promoting HP1a sumoylation in vivo. To further test this hypothesis, we performed a standard in vivo sumoylation assay in NIH3T3 cells (Fig. 1c). In this assay, after immunoprecipitation of HP1a-HA, sumoylation of HP1a is detected as a 70-kD band corresponding to GFP-SUMO-1-HP1a-HA. Introduction of the individual Myc-Suv39h paralogs to reach equal expression levels showed that only Suv39h1 could increase the levels of HP1a sumoylation, while Suv39h2 did not lead to any significant change ( Fig. 1d and Supplementary Fig. 1). We verified that the two proteins co-immunoprecipitated with HP1a-HA at comparable level in these experiments. The expression of a truncated Suv39h2 lacking the first 82aa (Suv39h2DN82) enabled to detect HP1a sumoylation, suggesting that the extended N-terminus could possibly interfere with the ability to promote HP1a sumoylation. However, this interference only worked on the cognate Suv39h2, since a fusion comprising the first 82aa of Suv39h2 grafted onto the N-terminus of Suv39h1 (Suv39h1 þ h2N82) could not interfere with the ability of Suv39h1 to increase the levels of HP1a sumoylation. Altogether, these data demonstrate that Suv39h1, but not Suv39h2 unless truncated, can promote HP1a sumoylation in vivo. It is thus possible that Suv39h2 might exist in an autoinhibited state as reported recently for PARKIN (refs 26,27). We next investigated whether this property in Suv39h1 can be separated from its H3K9 KMT activity and its interaction with HP1a. A H3K9 KMT catalytically dead mutant (Myc-Suv39h1-H324K) 3 used in our assay led to enhanced amounts of sumoylated HP1a, in contrast to a truncated form lacking the HP1 binding domain (Myc-Suv39h1-DN40) 24,28 ( Fig. 1e and Supplementary  Fig. 2). These data indicate that Suv39h1 promotes HP1a sumoylation regardless of its H3K9 KMT activity but absolutely requires an interaction with HP1a. Finally, given that in vivo assays could reflect indirect effects, such as the binding of a SUMO ligase activity to Suv39h1, we used an in vitro sumoylation assay with purified recombinant proteins. We could confirm that Suv39h1 on its own stimulated HP1a sumoylation in vitro in a dose-dependent manner ( Supplementary Fig. 3). Furthermore, two other HP1a-binding partners, KAP1 (ref. 29) and p150 the large subunit of CAF-1 (ref. 30) known to have a link with SUMO (refs 31,32) did not promote HP1a sumoylation, thus further underlining the specificity of this activity of Suv39h1.
Suv39h1 interacts directly with Ubc9 via a 1-167 domain. To further characterize the SUMO modification enhancing activity of Suv39h1, we thus examined whether Suv39h1 could bind directly to Ubc9 as found for the classical SUMO ligase enzymes 17,19 . Using GST pull-down and far-western assays, we found that Suv39h1 interacts directly with Ubc9 ( Fig. 2a and Supplementary  Fig. 4a,b). This is in line with classical SUMO ligases, but different from atypical SUMO ligase activities such as those associated with the SLX4 complex 33 that only bind SUMO charged Ubc9. We then sought to define which critical domain in Suv39h1 is involved. We observed that only His-Suv39h1 truncation mutants containing the linker region (92-167) between the chromodomain (CD) and the KMT catalytic domain strongly interact with GST-Ubc9 ( Fig. 2a and Supplementary Fig. 4b). We could further dissect the Ubc9 interaction domain within Suv39h1 to a region corresponding to aa114-140 ( Supplementary Fig. 4c). Interestingly, we found that this 114-140 region is sumoylated in vitro ( Supplementary Fig. 5a), consistent with a recent proteome-wide analysis in human cells that identified Suv39h1 as putative sumoylated proteins 34 . Furthermore, we found that Suv39h1 is a SUMO-1 interacting protein, where aa92-167 of Suv39h1 proved critical for this interaction ( Supplementary  Fig. 5b). Importantly, the 1-167 domain of Suv39h1 also mediates the interaction with HP1a through the first 40aa (refs 24,28). Taken together, these data indicate that the 1-167 domain could confer on Suv39h1 the ability to enhance SUMO modification of HP1a consistent with Suv39h1 possessing SUMO E3 ligase. We then examined whether the 1-167 domain on its own could be sufficient to promote HP1a sumoylation in vitro. The addition of Suv39h1-1-167 clearly stimulated sumoylation of HP1a in vitro (Fig. 2b). Notably, deletion of the regions responsible for the    HP1a-Ubc9 fusion. Based on these data, we thus propose a new function for Suv39h1, independent of its H3K9 KMT activity, involved in de novo localization of HP1a at pericentric heterochromatin.
TALE-mediated tethering of Suv39h to pericentric domains.
To directly address the potential role of Suv39h1 linked to the SUMO pathway in the 'seeding step' for the de novo targeting of HP1a to pericentric domains, we decided to assess the capacity of a  the sole 1-167 domain of Suv39h1 as described above. However, our first attempt in Suv39h double-null cells showed that Myc-Suv39h1-1-167 did not even localize at pericentric domains ( Supplementary Fig. 9a,b). We estimated by western blot, in our experimental conditions, how exogenous Myc-Suv39h1 expression compared with the endogenous Suv39h1 by quantifying the levels of Suv39h1 normalized to the proportion of transfected (expressing) cells ( Supplementary Fig. 9c). Again, since western blot analysis provides an average of the expression in the cells, we cannot formally exclude that at individual cell levels variation could occur. Yet, within these limits the levels were comparable. In fact, we found that all truncated form (N-or C-terminal) of Suv39h1 failed to localize de novo at pericentric heterochromatin, indicating that the full integrity of the protein is absolutely necessary for its targeting (Supplementary Fig. 9b). Without this localization, we could not monitor a de novo targeting of HP1a arguing for the necessity of Suv39h1 and the truncated versions to be at the site to sumoylate HP1a at pericentric domains for its effective targeting/retention required for the seeding step. Considering the importance of a local action, we thus decided to artificially tether the 1-167 domain of Suv39h1 to pericentric domain to be able to assess HP1a de novo localization to pericentric heterochromatin in Suv39h double-null cells. For this, we used transcription activator-like effectors (TALE) engineered to bind specifically to the major satellite DNA repeats found at pericentric heterochromatin 36 . We first fused this TALE with fulllength Suv39h1 and Suv39h2 (Fig. 3a). After transfection in Suv39h double-null cells, we verified by immunofluorescence that they accumulated at pericentric domains and could restore H3K9me3 enrichment (Fig. 3b). As above, we also verified expression levels of exogenous TALE-Suv39h1 compared with endogenous Suv39h1 (Supplementary Fig. 9c). Intriguingly, pericentric localization of endogenous HP1a was less efficient in the presence of tethered Suv39h2-HA compared with tethered Suv39h1-HA at early times after transfection ( Fig. 3b and Supplementary Fig. 10). This argues for an additional role for Suv39h1 not fulfilled by the Suv39h2, in line with the inability of Suv39h2 full-length to promote HP1a sumoylation (Fig. 1d). This also underlines the unique specificity of Suv39h1 to promote HP1a targeting to pericentric domains independently of its KMT activity.
The sole 1-167 domain accelerates HP1a de novo targeting. We next used TALE fusions to investigate the impact of the 1-167 domain of Suv39h1 supporting the sumoylation of HP1a. We generated TALE fusions of the entire 1-167 domain (Suv39h1-1-167-HA) and two deleted 1-167 constructs, one unable to interact with Ubc9 (Suv39h1-1-114-HA), and one unable to interact with HP1a (Suv39h1-41-167-HA). These two latter constructs proved inefficient in promoting HP1a sumoylation (Fig. 2b). All TALE-Suv39h1-HA constructs could accumulate at chromocenters after transfection in Suv39h double-null cells (Fig. 3b) Supplementary Fig. 11a,b). Moreover, we found that TALE-Suv39h1-1-167-HA and TALE-Suv39h1-1-114-HA coimmunoprecipitated equally endogenous HP1a ( Supplementary  Fig. 11c). Thus a major difference in HP1 binding could not explain our findings but rather confirms the hypothesis that the ability of Suv39h1 to interact with Ubc9 is a critical parameter to boost targeting of HP1a. We thus focused on the two following constructs, TALE-Suv39h1-1-167-HA and TALE-Suv39h1-1-114-HA, to monitor the kinetics of endogenous HP1a appearance at pericentric heterochromatin following their transfection in the Suv39h double-null cells (Fig. 3c and Supplementary Fig. 11d). At early time points, the appearance of HP1a at pericentric domains proved most efficient for Suv39h1-1-167 compared with Suv39h1-1-114 construct. This was achieved while expression of the HA-corresponding tagged proteins had reached comparable levels throughout the time-course (Fig. 3c) and showed comparable levels of accumulation at pericentric domains indicating that HP1a localization at pericentric heterochromatin cannot be simply due to high levels of Suv39H1 expression ( Supplementary  Fig. 11d). Taken together, these data demonstrate that the presence of the Ubc9 binding domain within the 1-167 domain of Suv39h1 to pericentric heterochromatin is key to accelerate HP1a targeting to these domains.
Identification of specific residues in the 1-167 domain. To delineate which amino acids within Suv39h1-1-167 were most critical for the interaction with Ubc9, we generated four triple point alanine mutants based on amino acids conservation in vertebrate species and in Drosophila ( Supplementary Fig. 12a,b). When compared with the wild-type version, we found that the KQR mutation reduced both the binding of the 1-167 domain of Suv39h1 to Ubc9 (Supplementary Fig. 12b) and HP1a sumoylation in vivo ( Supplementary Fig. 13). In contrast, the EQE mutation rather showed a slight increase for Ubc9 interaction. We then transfected TALE-1-167-KQRmut-HA in Suv39h double-null cells to assess its capacity to promote HP1a targeting to pericentric domains. Remarkably, the 1-167-KQRmut, similar to the 1-114 domain, performed less well than the 1-167-WT or 1-167-EQEmut ( Fig. 3d and Supplementary Fig. 14). Since all the TALE-constructs accumulated at pericentric domains in a similar way, the possibility of a distinct capacity in their targeting or a simple effect of expression levels can be discarded. Of note, we realized that the 1-167-EQEmut was even more potent that the WT in promoting HP1a enrichment. This enhanced capacity parallels its increased ability to bind to Ubc9. Overall, our data underline the importance of a functional Ubc9 binding domain within Suv39h1 involving specific conserved amino acids to promote HP1a sumoylation and its targeting to pericentric heterochromatin.

Discussion
While we had previously learnt that HP1a sumoylation was key for its interaction with RNA and its de novo targeting to pericentric heterochromatin 16 , how this modification on HP1a was imposed and controlled remained a mystery. Furthermore, considering the recent connection between chromatin, histone modifiers and Ubc9 (ref. 37), the issue became particularly intriguing. Here, not only we identify a new control on the HP1a sumoylation, but most remarkably it involves a major histonemodifying enzyme with a known function in regulating chromatin state through methylation of histone, a modification linked to cellular metabolism 38 . This unexpected function for Suv39h1 in addition to its KMT activity as a constitutive property contrasts with the necessity to truncate the paralog Suv39h2 to unveil the same capacity. The distinct properties of the two Suv39h paralogs open up new ways to consider the connections between Suv39h and heterochromatin organization. Importantly, two regions in the Suv39h1 protein can be distinguished, with the 1-167 domain containing the Ubc9 binding motifs to enhance sumoylation and the 167-412 domain for the H3K9me3 KMT activity. While the two activities can be separated functionally, they may not necessarily function in an exclusive manner and rather help to connect specific nuclear organization to particular metabolic sensing. We propose a working model to account for this dual role of Suv39h1 in the de novo localization of HP1a and pericentric heterochromatin formation (Fig. 4). First, Suv39h1, through its capacity to promote HP1a sumoylation, would accelerate its recruitment through the specific association between SUMO-1-HP1a and major RNAs. Second, Suv39h1 with its KMT activity that trimethylates H3K9 would then generate anchoring sites to ensure a controlled maintenance of HP1a at pericentric domains 3,4 . This is somehow reminiscent of the dual modifications in facultative heterochromatin that are controlled by the Polycomb complex with H3K27me3 and H2A ubiquitylation 39 . In this scheme, it is tempting to speculate that the activity of Suv39h1 to enhance SUMO modification could be most critical when major chromatin rearrangements and de novo establishment of pericentric domains are needed as observed during early development 40,41 or lymphocytes activation (unpublished results in collaboration with S. Amigorena), or other changes associated with metabolic stress and rewiring. The specific expression of Suv39h1, but not Suv39h2, during early mouse development 15 points to interesting avenues to explore. Thus, this unique property of Suv39h1 can have a profound impact of physiological relevance. Indeed, Suv39h1 as the first enzyme promoting sumoylation identified at pericentric heterochromatin domains could also act on other components. Precedent for this in DNA repair 42 showed that the recruitment of specific SUMO ligases at particular places often helps to target an entire group of functionally and physically connected proteins to build-up complex nuclear assemblies 21 . Suv39h1 could represent a major orchestrating activity with a broader impact on constitutive heterochromatin that one could have initially thought based only on the KTM activity. Indeed, many proteins linked to pericentric heterochromatin are known to be sumoylated, including KAP1 (ref. 31) and MeCP2 (ref. 43). Furthermore, some of them are even direct Suv39h1-interacting proteins, such as HDAC1 (refs 44,45), Dnmt1 (refs 46,47) and MBD1 (refs 48,49). Finally, the conservation of Ubc9 binding and the link with the SUMO pathway deserves to be explored in S. pombe and Drosophila to deepen our understanding of the basic properties linked to heterochromatin establishment and propagation. Clearly, the recent finding revealing Ubc9 as a major repressor of iPS cell formation further underlines the importance of SUMO pathway in the context of somatic cell identity 37 . Future work should address these issues. While we have concentrated here on sumoylation, desumoylation as another means to regulate the steady-state achieved is equally important. In this respect, the SUMO protease SENP7 localized at pericentric heterochromatin 50 will be a likely candidate to consider, in particular given its ability to bind to HP1 proteins 51 .
In conclusion, the link of Suv39h1 with the SUMO pathway discovered in this work provides a new conceptual framework for the dynamics and the stability of constitutive heterochromatin, paving the way for exciting work in the field of nuclear organization and its role during normal development and pathological situations.  The two functional domains in Suv39h1 highlighting the interaction with HP1a, Ubc9 and SUMO are depicted. During de novo HP1a targeting to pericentric heterochromatin, Suv39h1 can enhance HP1a sumoylation and accelerate its recruitment. During HP1a enrichment, Suv39h1 would function as KMT leading to H3K9 trimethylation thereby providing binding sites for HP1a allowing its accumulation/retention. While the two activities are associated with distinct subdomains, they may not necessarily function in an exclusive manner and rather provide a means to integrate a metabolic sensing to the specific targeting of a nuclear domain.
In vitro sumoylation assay and western blot. We produced recombinant proteins in Escherichia coli BL21 (DE3) corresponding to GST-HP1a, GST-KAP1, His-p150 and GST-Suv39h1 full-length protein and truncation fragments. Following SDS-PAGE, we analysed these proteins by gel staining (Imperial Protein Stain, Pierce). We used the recombinant proteins in in vitro sumoylation reactions using the SUMOlink kits (Active Motif) according to the manufacturer's instructions, except that we diluted 1:4 the E2 conjugating enzyme Ubc9. Unless stated otherwise, we added 60 ng of GST-Suv39h1 full-length or truncated versions, GST-KAP1 or His-p150. We stopped the reactions by addition of 2 Â SDS-PAGE loading buffer. Reaction mixture (25%) was run on a 4-12% Bis-Tris NuPAGE gel (Invitrogen) and transferred onto nitrocellulose membranes (Protran). We probed sumoylated and unmodified GST-HP1a using mouse monoclonal anti-HP1a antibodies and horse radish peroxidase-conjugated secondary antibodies (Jackson Immunoresearch). Before the visualization of proteins using the Super Signal West Pico chemiluminescence substrates (Pierce), we cut the membrane just above the protein marker 50 kDa and performed two different exposures: high to enable the detection of SUMO-1-GST-HP1a and a low to visualize unmodified GST-HP1a (see Supplementary Fig. 3). Western blots with high and low exposures corresponding to this cut membrane are shown in Fig. 2b. We quantified SUMO-1-HP1a and HP1a from western blots with an ImageQuant LAS 4000 mini system and the ImageQuant software (GE Healthcare). The western blot in Fig. 2a  Centromeric RNA pull-down. We prepared nuclear extracts from NIH3T3 cells or Suv39h double-null MEFs and performed RNA pull-down using published procedures 16 in the presence of 20 mM N-ethylmaleimide (Sigma).
GST pull-down. We expressed GST and GST-Ubc9 in Escherichia coli BL21 (DE3) and immobilized proteins on Glutathione Sepharose 4B (GE Healthcare). We performed GST pull-down by mixing GST or GST-Ubc9 beads with total extract of NIH3T3 cells expressing Myc-Suv39h1. We incubated for 2 h at 4°C on a rotating wheel in binding buffer (20 mM Tris-HCl pH 8, 200 mM NaCl, 2 mM EDTA, 0.5% (V/V) Nonidet P-40) supplemented with proteases inhibitors. After five washes with binding buffer, the proteins were recovered by boiling the beads in SDS-PAGE loading buffer and were analysed by western blot.
Immunofluorescence. We processed cells for immunostaining as described in ref. 35. We detected proteins of interest using the appropriate primary and secondary antibodies conjugated to Alexa-Fluor 488, 594 and 647 (Invitrogen). We used a Zeiss Imager Z1 epifluorescence microscope piloted with Metamorph software, an Â 63 oil objective lens and an HQ2 CoolSnap camera (Photometrics) for image acquisition. We obtained percentages of cells with HP1a or H3K9me3 localized at pericentric heterochromatin by counting more than 100 cells in each experiment. To quantify the enrichment of endogenous HP1a or TALE-Suv39h1-HA constructs at pericentric heterochromatin domains, we used the three-dimensionalfluorescence intensity enrichment at domains ImageJ macro 56 from Z-stacks images acquired with a Deltavision system equipped with an inverted Olympus IX71 microscope and a Â 100 objective lens ( Supplementary Fig. 11d) or with a Zeiss Z1 epifluorescence microscope and a Â 63 objective lens (Fig. 3d). The procedure first uses the 4,6-diamidino-2-phenylindole stacks to segment the nucleus and the pericentric domains to define three-dimensional masks that correspond to these regions. These masks are then applied to the fluorescent channels corresponding to HP1a and TALE-Suv39h1-HA constructs to quantify their fluorescence intensity detected within and outside pericentric domains. This fluorescence intensity is then normalized to the size of the domains and the nucleus. Enrichment for HP1a and TALE-Suv39h1-HA constructs is obtained by the ratio of these normalized fluorescence intensity within and outside of the domains. A value equal/close to 1 represents no enrichment (equal normalized fluorescence within and outside the domains) whereas increasing values quantify increasing enrichment at the pericentric domains. We performed the quantification on at least 100 nuclei for each condition. For all studies, we performed at least three independent experiments.
Data availability. The authors declare that the data supporting the findings of this study are available within the article and its Supplementary Information files, or from the authors on reasonable request.