The CUE1 domain of the SNF2-like chromatin remodeler SMARCAD1 mediates its association with KRAB-associated protein 1 (KAP1) and KAP1 target genes

Chromatin in embryonic stem cells (ESCs) differs markedly from that in somatic cells, with ESCs exhibiting a more open chromatin configuration. Accordingly, ATP-dependent chromatin remodeling complexes are important regulators of ESC homeostasis. Depletion of the remodeler SMARCAD1, an ATPase of the SNF2 family, has been shown to affect stem cell state, but the mechanistic explanation for this effect is unknown. Here, we set out to gain further insights into the function of SMARCAD1 in mouse ESCs. We identified KRAB-associated protein 1 (KAP1) as the stoichiometric binding partner of SMARCAD1 in ESCs. We found that this interaction occurs on chromatin and that SMARCAD1 binds to different classes of KAP1 target genes, including zinc finger protein (ZFP) and imprinted genes. We also found that the RING B-box coiled-coil (RBCC) domain in KAP1 and the proximal coupling of ubiquitin conjugation to ER degradation (CUE) domain in SMARCAD1 mediate their direct interaction. Of note, retention of SMARCAD1 in the nucleus depended on KAP1 in both mouse ESCs and human somatic cells. Mutations in the CUE1 domain of SMARCAD1 perturbed the binding to KAP1 in vitro and in vivo. Accordingly, an intact CUE1 domain was required for tethering this remodeler to the nucleus. Moreover, mutation of the CUE1 domain compromised SMARCAD1 binding to KAP1 target genes. Taken together, our results reveal a mechanism that localizes SMARCAD1 to genomic sites through the interaction of SMARCAD1's CUE1 motif with KAP1.


Chromatin in embryonic stem cells (ESCs) differs markedly from that in somatic cells, with ESCs exhibiting a more open chromatin configuration. Accordingly, ATP-dependent chromatin remodeling complexes are important regulators of ESC homeostasis. Depletion of the remodeler SMARCAD1, an
ATPase of the SNF2 family, has been shown to affect stem cell state, but the mechanistic explanation for this effect is unknown. Here, we set out to gain further insights into the function of SMARCAD1 in mouse ESCs. We identified KRAB-associated protein 1 (KAP1) as the stoichiometric binding partner of SMARCAD1 in ESCs. We found that this interaction occurs on chromatin and that SMARCAD1 binds to different classes of KAP1 target genes, including zinc finger protein (ZFP) and imprinted genes. We also found that the RING B-box coiled-coil (RBCC) domain in KAP1 and the proximal coupling of ubiquitin conjugation to ER degradation (CUE) domain in SMARCAD1 mediate their direct interaction. Of note, retention of SMARCAD1 in the nucleus depended on KAP1 in both mouse ESCs and human somatic cells. Mutations in the CUE1 domain of SMARCAD1 perturbed the binding to KAP1 in vitro and in vivo. Accordingly, an intact CUE1 domain was required for tethering this remodeler to the nucleus. Moreover, mutation of the CUE1 domain compromised SMARCAD1 binding to KAP1 target genes. Taken together, our results reveal a mechanism that localizes SMARCAD1 to genomic sites through the interaction of SMARCAD1's CUE1 motif with KAP1.
Embryonic stem cells (ESCs) 3 are pluripotent and thus able to respond to inductive signals leading to differentiation into every cell type of the body while simultaneously balancing the ability to undergo self-renewal in each replication cycle. Apart from transcription factors, the importance of ATP-dependent chromatin remodeling enzymes for supporting and reinforcing stem cell homeostasis is increasingly recognized (1)(2)(3)(4)(5)(6)(7). ATPdependent chromatin remodeling enzymes modulate interactions between DNA and histones, driving changes in nucleosome position, structure, or composition, thereby regulating the access of other proteins, such as transcription factors or chromatin modifiers. As a result, these enzymes are central to the control of chromatin structure and genome function. Interfering with their normal expression frequently results in impairment of embryonic development and pluripotency (1,2,8). For instance, brahma-associated factor (BAF) is an example of a remodeling complex that directly interacts with components of the pluripotency network to regulate the expression of ESC regulators (1,2,5). More recently, knockdown of Smarcad1 (SWI/SNF-related, matrix-associated actin-dependent regulator of chromatin, subfamily A, containing DEAD/H box 1) was recognized to elicit a change in the pluripotent state (9 -11).
The SMARCAD1 family of remodelers belongs to the evolutionarily most conserved remodeling complexes and includes Fun30 in Saccharomyces cerevisiae, Fft3 in Schizosaccharomyces pombe, and SMARCAD1/Etl1 in mammals. One conclusion from studies of the individual family members is that they are important for the regulation and maintenance of functional chromatin domains (12)(13)(14)(15)(16)(17)(18)(19). Depletion of this enzyme results in an open chromatin conformation in normally transcriptionally repressed regions and can affect gene expression, suggesting that SMARCAD1 ensures correct chromatin structure of silent domains. SMARCAD1 can also activate transcription together with co-activator p300/CBP (CREB-binding protein), and interactions with the transcription machinery have recently been reported in S. pombe (20,21). In addition, SMARCAD1 is involved in double-strand break repair (22). Biochemically, this remodeler has been best characterized in budding yeast. Fun30 is capable of binding chromatin and DNA in vitro with a preference for single-stranded chromatin and exhibits activity in ATP-dependent chromatin remodeling assays (19,23,24). A homozygous mutation of this remodeler in the mouse results in growth retardation, prenatal and perinatal lethality, reduced fertility, and skeletal abnormali-ties (25). Whereas SMARCAD1 is expressed throughout development, its function is best characterized in adult cells, yet SMARCAD1 levels are particularly high in the inner cell mass of the blastocyst embryo and in ESCs (26 -28). ESCs depleted for SMARCAD1 lose the typical morphology and show defects in exit from self-renewal (9 -11). Despite its importance in ESCs, little is known about SMARCAD1 function and regulation in the context of the chromatin environment in pluripotent cells.
SWI/SNF proteins typically function together with accessory proteins that help to direct these enzymes to specific genomic loci, modulate their activity, and integrate chromatin remodeling with distinct cellular pathways. Besides, changes in the composition and stoichiometry of these complexes during mammalian development confer unique roles to remodelers (2,8,29). For instance, specialized assemblies of the BAF remodeling complex with cell type-specific subunits were found to be critical for progression from pluripotency to multipotency to committed neurons. We and others have previously identified candidate accessory factors of SMARCAD1 in human somatic cells (13,30). Prominent among them was the KRABassociated protein 1, KAP1 (TRIM28; TIF1␤). Conversely, KAP1 purifications from HEK293 contain SMARCAD1 (31). KAP1 is an important regulator of normal development and differentiation. It has transcriptional and non-transcriptional roles and, like SMARCAD1, is involved in DNA repair and chromatin replication (13,(32)(33)(34)(35). How KAP1 functions in the context of the human SMARCAD1 remodeling complex is not known. Open questions also concern whether KAP1 is a tissue-specific or constitutive interaction partner of SMARCAD1.
In this study, we present evidence for a physical and regulatory link between SMARCAD1 and KAP1 in mouse ESCs. Our results reveal that KAP1 target genes are bound by SMARCAD1, and we provide mechanistic insights into how they are recognized.

KAP1 is a stoichiometric component of SMARCAD1 mouse ESC complexes
To identify the major functional component(s) of SMARCAD1 remodeling complexes in pluripotent cells, we generated mouse ESCs stably expressing FLAG-tagged SMARCAD1 protein (Fig. S1A) to utilize in an unbiased purification coupled to a proteomic analysis. Tagged SMARCAD1 was nuclear and localized indistinguishably from the endogenous protein (Fig.  S1B). FLAG affinity purification was carried out under conditions where nucleic acids were degraded to rule out proteinprotein interactions bridged by DNA/RNA (Fig. S1, C and D). Silver stain of proteins purified from FLAG-SMARCAD1 nuclear extracts identified two major protein bands not present in purifications from control extracts; the protein migrating at about 130 kDa corresponds to tagged SMARCAD1 (Fig. 1A). The protein at 100 kDa was identified by mass spectrometry analysis as KAP1 in three independent purification experiments (Fig. 1B). We applied the exponentially modified protein abundance index (emPAI; (36)) to the proteomic data sets. SMARCAD1 and KAP1 have comparable emPAI scores, indicating that both proteins are of similar abundance and thus stoichiometric in the FLAG purification (Fig. 1B). Next, we sought to confirm this association with endogenous proteins. Soluble and chromatin fractions were prepared from ES cellular extracts to further localize SMARCAD1 and KAP1 (Fig. 1C). Endogenous proteins co-eluted and were found predominantly in the soluble fraction. About 20% of the total pool of SMARCAD1 and KAP1 was found to be in the chromatin fraction in an asynchronous ESC population. A SMARCAD1 antibody efficiently co-immunoprecipitated KAP1 from ESC extracts (Fig. 1D). Collectively, these data show that SMARCAD1 forms a stable complex with KAP1 in mouse ESCs. This extends previous observations made in human somatic cell lines (13,30) and indicates that the association between SMARCAD1 and KAP1 is conserved in multiple cell lines, human and mouse, of different developmental stages, pluripotent and differentiated.

The RBCC domain of KAP1 is required for SMARCAD1 interaction
KAP1 is a modular protein and harbors several functional domains ( Fig. 2A (34,35)). The RING B-box coiled-coil (RBCC) domain has been characterized as a protein interaction interface and binds KRAB-ZNF proteins involved in KAP1 recruitment to the genome. A centrally located motif recognized by the heterochromatin protein HP1 and the C-terminal part of KAP1 fulfill silencing functions. To establish which domain of KAP1 mediates the interaction with SMARCAD1, glutathione S-transferase (GST) pull-down experiments were performed with a series of recombinant GST-KAP1 deletion mutants (Fig.  2, A and B). SMARCAD1 bound to GST-full-length KAP1 but not to GST (Fig. 2B, compare lanes 1 and 7). Deletion of the C-terminal plant homeodomain and bromodomain (PB), which has been previously shown to recruit the chromatin remodeler CHD3 and the histone methyltransferase SETDB1 (34,35), did not abolish the interaction with SMARCAD1 (Fig.  2B, lane 2). However, binding was abolished when the N-terminal RBCC domain was absent (Fig. 2B, lane 3). Notably, the 376-amino acid RBCC domain alone can mediate the interaction with SMARCAD1 (Fig. 2B, lane 4). Neither the HP1 binding domain (lane 5) nor the C-terminal part of KAP1 (lane 6) bound SMARCAD1. These results show that the RBCC domain of KAP1 is both necessary and sufficient for its interaction with SMARCAD1 in vitro.

The CUE1 domain of SMARCAD1 mediates the interaction with KAP1
The signature motif of ATP-dependent chromatin remodeling enzymes is a conserved SNF2-like ATPase domain. In addition, mammalian SMARCAD1 contains tandem copies of a CUE domain located near the N terminus ( Fig. 2C) (18). CUE domains are predicted ubiquitin-binding motifs stretching over ϳ40 amino acids (37). We previously found that a recombinant GST-SMARCAD1 double CUE domain construct (amino acids 131-367) robustly associated with KAP1 from HeLa nuclear extracts (13). The double CUE domain is thus a candidate for mediating a direct physical interaction with KAP1. We corrob-

SMARCAD1 genomic localization via KAP1
orated this in a series of GST pulldown experiments in vitro (Fig. 2C); full-length GST-SMARCAD1 and GST-double CUE but not GST alone interact with baculovirus-expressed KAP1. Interestingly, the CUE1 domain is sufficient to mediate KAP1 binding, whereas CUE2 is not required (Fig. 2C, compare lanes 4 and 5). To verify the CUE1 KAP1 interaction, we mutated key residues in the CUE1 domain. CUE domains typically fold into three-helix bundles (Fig. S2) in which two highly conserved sequence motifs are involved in stabilizing the hydrophobic core and in the efficient interaction with ubiquitin (37). These are a dileucine motif in ␣ helix 3 and an N-terminal FP motif ( Fig. S2) (37). We converted the invariant proline at position 170 in the FP motif in SMARCAD1 to a glycine. Independently, we replaced Phe-169 and Leu-196, which contribute to the hydrophobic core, with polar amino acids (Fig. S2). Recombinant proteins with these CUE1 mutations were tested for their ability to pull down KAP1. Whereas the P170G exchange had little effect on binding (Fig. 2C, lane 6) the double mutation targeting the two highly conserved sequence motifs characterizing CUE domains completely abolished the interaction with KAP1 (Fig. 2C, lane 7). This suggests that preserving a CUE1 protein fold with a hydrophobic core is necessary for stable KAP1 interaction.
The involvement of a CUE domain in the interaction with KAP1 implies that SMARCAD1 may interact with ubiquity-lated KAP1. However, posttranslational modification of KAP1 by ubiquitylation is unlikely to be important for the interaction with SMARCAD1 because pure recombinant proteins are able to interact. To investigate this further, we considered whether ubiquitin can compete for the interaction with KAP1 (Fig. 2D). We estimated the binding affinity of SMARCAD1 CUE1-KAP1 to be in the micromolar or submicromolar range, as the binding reactions routinely contained low micromolar KAP1 and GST-CUE1 proteins. Pure ubiquitin was titrated into the GST pulldown assays up to a concentration of a 1000-fold molar excess over KAP1 (Fig. 2D, middle). No detectable competition occurred, demonstrating that the CUE1 domain binds to KAP1 with significantly higher affinity than to ubiquitin (Fig. 2D, top). In summary, whereas CUE domains are recognized as canonical ubiquitin binding domains, the CUE1 domain in SMARCAD1 displays unusual characteristics in that it recognizes unmodified KAP1.

Steady-state levels of SMARCAD1 depend on KAP1
Our experiments so far showed a direct physical interaction between SMARCAD1 and KAP1. To investigate the interdependence of these proteins, ESCs were depleted for either KAP1 or SMARCAD1 by RNAi, and the protein and RNA levels of the other partner were assessed in immunoblots and by quantitative RT-PCR. Depletion of KAP1 resulted in concomi- , and the emPAI from three purifications are shown. The Mascot score is the probability that the observed match is a random event. The emPAI (36) is a semiquantitative analytical value indicating the relative abundance of peptides within the sample. Scores for the corresponding control purification are shown in parentheses. C, Western blot analysis of ESCs biochemically fractionated into chromatin-enriched (C) and soluble fractions (S). Endogenous SMARCAD1 and KAP1 proteins from E14 cells are readily extractable with 0.1% Triton (S); ϳ20% of each protein remains associated with the chromatin (C). Controls, GAPDH, and histone H3 separate into the appropriate subfractions. D, endogenous SMARCAD1 was immunoprecipitated from J1 ESC extracts in the presence of Benzonase and ethidium bromide. Immunoprecipitates (IP) and 5% of the input were separated by SDS-PAGE and KAP1 was detected by immunoblotting as co-immunoprecipitating protein.

SMARCAD1 genomic localization via KAP1
tant reduction of SMARCAD1 protein levels without affecting lamin B1 ( Fig. 3A (left), compare lanes 1 and 2). This was seen in multiple independent ESC lines (E14, J1, and PGK12.1) with different methods of KAP1 depletion (shRNA or esiRNA) and occurred progressively; cells depleted for longer time periods exhibited lower SMARCAD1 levels than at earlier time points ( The Smarcad1 gene is bound by the core pluripotency transcription factors NANOG, OCT4, and SOX2, and its expression is down-regulated upon differentiation (26,27,38). We therefore considered two possible explanations for reduced SMARCAD1 levels: KAP1 directly affects SMARCAD1 stability in ESCs and/or the reduction is an indirect consequence of cell differentiation. ESCs are known to progressively lose pluripotency upon KAP1 removal (3, 39 -41), and consistent with this, we detected reduced levels of the pluripotency protein NANOG following KAP1 depletion ( Fig. 3A (left), compare lanes 1 and 2). To test whether the observed drop in SMARCAD1 levels is due to reduced self-renewal capability of KAP1 knockdown cells, we overexpressed epitope-tagged SMARCAD1 driven by a chicken ␤-actin promoter, which is not regulated by the pluripotency network. When we depleted KAP1 under these conditions, SMARCAD1 levels were also severely diminished ( Fig. 3A (left), compare lanes 3 and 4), suggesting that the effect on SMARCAD1 steady-state levels is not solely a direct consequence of differentiation. We also observed that mRNA levels of SMARCAD1 were affected upon KAP1

SMARCAD1 genomic localization via KAP1
knockdown ( Fig. 3A and Fig. S3, right-hand panels), albeit to a lesser extent than the protein levels. Taken together, these results indicate that KAP1 plays a role in the regulation of SMARCAD1 steady-state levels by yet unknown mechanisms.

KAP1 facilitates nuclear retention of SMARCAD1
Classically, histones and non-histone chromosomal proteins were characterized by their different extractability from cells and nuclei. We therefore examined the interdependence of KAP1 and SMARCAD1 at the individual cell level by different extraction methods and indirect immunofluorescence. Both proteins co-localize throughout the ESC nucleus and accumulate at DAPI-dense pericentric heterochromatin in a similar proportion of cells (Fig. 4). Knockdown of SMARCAD1 had no apparent effect on the KAP1 localization pattern in the ESC nucleus (Fig. 4, A and B). Reciprocally, we analyzed KAP1-depleted cells and KAP1-expressing cells in parallel and performed double-staining of SMARCAD1 and KAP1 (Fig. 4C).
We found that depletion of KAP1 alters SMARCAD1 binding in the nucleus (Fig. 4C, dotted circle). SMARCAD1 could be detected when KAP1 knockdown cells were fixed before permeabilization and antibody staining (Fig. 4C, Not Pre-extracted). In contrast, in KAP1-depleted cells, SMARCAD1 was largely washed out by pre-extraction of soluble proteins compared with cells that express KAP1 (Fig. 4C, Pre-extracted). The release of SMARCAD1 from the nucleus by detergent extraction indicates that it is normally stably associated with nuclear components, but this association is dependent on KAP1. We obtained identical results in different ECSs treated with  Representative examples of indirect immunofluorescence for KAP1 and SMARCAD1 proteins upon shRNA-mediated knockdown of either SMARCAD1 or KAP1. A and B, stable control knockdown (Ctrl) and SMARCAD1 knockdown (KD) mouse E14 ESCs were permeabilized, and soluble proteins were extracted before fixation. A, clear knockdown of SMARCAD1. B, KAP1 localization is not affected when SMARCAD1 is depleted. Similar results were obtained when the experiment was repeated in a different ESC line. DNA was counterstained with DAPI. C, KAP1 depletion alters SMARCAD1 binding in the nucleus. Double-staining in PGK12.1 ESCs transfected for 4 days with control or Kap1 shRNA plasmids without antibiotic selection. Conditions chosen allow detection of untransfected and transfected cells in parallel. Two fixation and permeabilization protocols were used. Top, cells were formaldehyde-fixed before permeabilization (not preextracted). Bottom, extraction of soluble proteins before fixation (pre-extracted). Dotted circle, example of SMARCAD1 localization in nuclei depleted for KAP1. Scale bar, 10 m.

SMARCAD1 genomic localization via KAP1
shRNAs or esiRNAs against Kap1 and detecting SMARCAD1 with a different antibody (data not shown). That SMARCAD1 can be readily detected in formaldehyde-fixed KAP1 knockdown ESCs by immunofluorescence when protein levels appear reduced by Western blotting is probably due to the differential sensitivity of these assays. These extraction experiments demonstrate that KAP1 is necessary to stably retain SMARCAD1 in the nucleus of mouse ESCs.
Because a relaxed chromatin architecture with a hyperdynamic chromatin protein population is a distinctive feature of pluripotency (1,7), this ready release of SMARCAD1 could be a property specific to ESCs. To establish the generality of our observation, we turned to human cells. We performed double-staining in HeLa cells depleted for KAP1 (Fig. S4). As in ESCs, when soluble proteins were pre-extracted before fixation, SMARCAD1 staining was lost in HeLa nuclei that lack KAP1.

SMARCAD1 nuclear retention requires an intact CUE1 domain
KAP1 is thus required for stable maintenance of SMARCAD1 in nuclei of both undifferentiated mouse and differentiated human cells. Given the stable interaction of these proteins in vitro, a direct mechanism is likely. In this scenario, specific disruption of the KAP1-SMARCAD1 interaction would have consequences similar to those of the complete removal of KAP1. To test this, we investigated whether the SMARCAD1 CUE1 domain mutant, which abolishes KAP1 binding in vitro, directly affects the retention of this remodeler in the nucleus and ultimately bound to the genome. Epitopetagged wildtype and CUE1 F169K/L196K SMARCAD1 constructs were introduced into ESCs carrying a doxycyclineinducible shRNA targeting the 3Ј-UTR of Smarcad1. After treatment with doxycycline, the levels of endogenous SMARCAD1 were diminished (Fig. 5A, compare lanes 1 and 2), whereas exogenous SMARCAD1 was successfully expressed (Fig. 5A, V5). Importantly, exchanging two hydrophobic amino acids in the CUE1 domain (mt) does not affect the expression, stability, or nuclear localization of SMARCAD1, as determined by Western blotting and indirect immunofluorescence using an antibody that recognizes the C-terminal V5 tag (Fig. 5, A and  C). Nor does it affect the stability or localization of KAP1 (Fig. 5, A and C) or the cell cycle profile (Fig. S5). Next, we wanted to establish that the CUE1 mutation disrupts the interaction with KAP1 in vivo. We confirmed that the triple FLAG/V5-tagged SMARCAD1 protein correctly assembled into KAP1 protein complexes by immunoprecipitating the WT from cell extracts and examining the amount of co-precipitated KAP1 (Fig. 5B,  lanes 1-3). Western blot analyses indicated that equivalent amounts of wildtype and mutant FLAG-SMARCAD1 were recovered in immunoprecipitations (Fig. 5B, FLAG, compare  lane 3 with lane 6). However, the association of KAP1 with the mutant SMARCAD1 protein was severely reduced compared with the wildtype protein (Fig. 5B, KAP1, compare lane 3 with lane 6). Thus, an intact CUE1 domain is important for the efficient interaction of SMARCAD1 with KAP1 in cells. To further strengthen this observation, we performed indirect immunofluorescence. In double labeling experiments, exogenous SMARCAD1 protein was detected using an antibody that rec-ognizes the C-terminal V5 tag (Fig. 5C). Tagged wildtype SMARCAD1 showed co-localization with KAP1 under all conditions tested. Cells expressing SMARCAD1 protein mutated in the CUE1 domain displayed different staining patterns, depending upon whether cells were immediately fixed or soluble proteins were removed by detergent treatment before fixation. SMARCAD1 carrying a CUE1 mutation was readily washed out from cells if they were not fixed before extraction (Fig. 5C, Pre-extracted). This demonstrates that tethering of SMARCAD1 in the nucleus requires an intact CUE1 domain.

SMARCAD1 CUE1 domain facilitates its binding to KAP1 target genes
Having established that the physical interaction between SMARCAD1 and KAP1 is essential for the retention of SMARCAD1 in the nucleus, we set out to determine whether SMARCAD1 binds to KAP1 target genes in the ESC genome. To this end, we conducted ChIP in ESCs expressing FLAGtagged wildtype SMARCAD1 protein using a FLAG antibody. In addition, ChIP was performed in untransfected ESCs using antibodies against endogenous SMARCAD1. We selected known KAP1-binding sites in the ESC genome from a published ChIP-seq data set and previous studies for investigation by qPCR (see Fig. 7) (42,43). The genes chosen represent different categories of KAP1-binding sites, including imprinted control regions (Peg13; Peg3), promoters (Fkbp6), introns (Pank4), and 3Ј-coding exons (ZNF genes, Ezr). First, we determined whether SMARCAD1 binds to all or a subset of these KAP1 target sites. A link between SMARCAD1 and KAP1 occupancy has not previously been reported in any cell type. Here, we identified all nine loci tested as direct targets of endogenous SMARCAD1 (Fig. 6, Ctrl) as well as of FLAG-tagged wildtype SMARCAD1 (Fig. 7, left panels, Ctrl). SMARCAD1depleted cells exhibited reduced levels of SMARCAD1 occupancy (Fig. 6, KD; Fig. 7 (left), KD), whereas levels of H3 were unchanged (Fig. S6B), confirming the specificity of the assay. An intergenic control region was not enriched in SMARCAD1 (Fig. 7, bottom). This establishes for the first time that SMARCAD1 occupies main categories of KAP1 target genes in the ESC genome.
Next we sought to determine whether the SMARCAD1 CUE1 mutation that disrupts its interaction with KAP1 would affect the ability of SMARCAD1 to bind to KAP1 target genes. We carried out FLAG-ChIP in ES cells expressing tagged SMARCAD1 protein, either WT or a CUE1 mutant, and found that SMARCAD1 occupancy at the selected KAP1 target genes is significantly reduced in the CUE1 mutant (Fig. 7A, right). This is a specific reduction, not observed in immunoprecipitation with the H3 antibody (Fig. 7B). Hence, an intact CUE1 domain is required for efficient binding of SMARCAD1 to these genomic sites. Collectively, our results suggest a model whereby SMARCAD1 localization to different categories of KAP1 target genes is dependent on KAP1.

Discussion
In this work, we provide mechanistic insights into the regulation of the ATP-dependent chromatin remodeler SMARCAD1.  4). To avoid possible artifacts that can arise when comparing single cell clones, we chose to work with pools of transfectants. Tagged WT and mutant proteins are expressed to similar levels. KAP1 and pluripotency marker OCT4 levels are not affected by depletion or overexpression of SMARCAD1. Lamin B1 serves as a loading control. B, a FLAG-specific antibody readily co-immunoprecipitates KAP1 from ESCs expressing FLAG-tagged SMARCAD1 WT protein (lane 3). The addition of ethidium bromide and Benzonase demonstrated that the interaction of SMARCAD1 and KAP1 is DNA-independent. KAP1 association with FLAG SMARCAD1 is significantly reduced when the CUE1 domain is mutated at F169K and L196K (lane 6). Lanes 1 and 4, 3% input; lanes 2 and 5, IgG. PRMT5 served as a negative control. Dotted line, discontinuous lanes from the same gel. Shown is a co-immunoprecipitation performed under conditions where endogenous SMARCAD1 protein was depleted by 2-day doxycycline treatment. Similar results were observed in independent experiments when SMARCAD1 WT and CUE1 mutant constructs were transiently transfected in ESCs expressing endogenous SMARCAD1. C, SMARCAD1 remodeler harboring the F169K/L196K mutation in the CUE1 domain is readily extractable from ESC nuclei. Co-staining of KAP1 and V5-tagged SMARCAD1 in E14 cells expressing the CUE1 mutant (CUE1 mt V5). ESCs with V5-tagged WT protein served as control (WT-V5). Cells were either stained directly or after extraction of soluble proteins. Similar results were obtained in the presence (not shown) or absence of doxycycline (shown). Scale bar, 10 m.

SMARCAD1 genomic localization via KAP1
A key finding is that KAP1 is a major functional co-factor of SMARCAD1 in mouse ESCs. Our results lead us to propose a model whereby the interaction with KAP1 is required for efficient nuclear retention and tethering of SMARCAD1 to genomic binding sites. In direct support for this, we have shown that SMARCAD1 occupies KAP1 target genes in the ESC genome, and this association is disrupted when SMARCAD1 cannot bind KAP1. SMARCAD1 and KAP1 are expressed throughout development, and both display particularly high levels in ESCs, where they are required to maintain the ESC state (3, 9, 10, 26-28, 40, 41, 44, 46). Our biochemical experiments identify KAP1 as a core component of SMARCAD1 complexes in pluripotent cells. Association between these proteins has been described previously in HeLa, Hct116, and HEK293 cells (13,30). We have therefore established that this interaction is conserved and has important functions beyond transformed human cells. KAP1 and SMARCAD1 interact directly with each other via their RBCC and CUE1 domains, respectively. They behave similarly in biochemical fractionation experiments and co-localize throughout the ESC nucleoplasm as well as in DAPI-dense staining heterochromatin. This suggests a model in which SMARCAD1-KAP1 complexes operate in both eu-and heterochromatin contexts. Accordingly, KAP1 facilitates the retention of the majority of SMARCAD1 protein in the nucleus in both mouse ESCs and human somatic cells. We therefore consider the physical interaction between SMARCAD1 and KAP1 to be the primary means for tethering SMARCAD1 in the genome. These findings do not exclude the possibility that there are sites in the genome that are bound by SMARCAD1 independently of KAP1. For example, additional mechanisms could involve histone modifications; one potential candidate is citrullinated histone H3 Arg-26, which SMARCAD1 interacts with in ESCs (10).
Steady-state levels of SMARCAD1 are dependent on KAP1 in ESCs. Depletion of KAP1 leads to progressive loss of SMARCAD1. Although the precise mechanisms underlying this are not clear, it implies that previous studies on the effect of KAP1 depletion in ESCs may have not considered the contribution of SMARCAD1. KAP1 function has many facets, including destabilization of heterochromatin at DNA damage sites, stabilizing polymerase II pausing thereby controlling transcriptional elongation, and heterochromatin establishment to silence genes and retrotransposons (34,35). One proposed function for KAP1 is to protect imprinted genes against the wave of DNA demethylation that affects the mammalian genome during early embryogenesis (43). It remains to be determined whether these or other processes are affected by the SMARCAD1-KAP1 partnership in pluripotent cells. Our ChIP experiments provide the first data consistent with a contribution of SMARCAD1 to KAP1-mediated heterochromatin establishment/maintenance. We observe binding of SMARCAD1 to sites in the ESC genome where KAP1 was previously shown to maintain H3K9me3 and DNA methylation, such as Fkbp6, Zfp629, and the imprinted control regions of Peg3 and Peg13 (43). Thus, SMARCAD1 may be part of the complex comprising KAP1 and chromatin modifiers that preserves the histone modification and DNA methylation status during early embryogenesis. We also identified 3Ј exons of zinc finger genes as SMARCAD1 targets. These are generally considered strong KAP1-binding sites and contain repeated DNA sequences encoding zinc finger motifs, 19 in Zfp629 and 8 in Zfp13. It has been proposed that heterochromatinization of ZNF 3Ј-coding exons may prevent inappropriate recombination between these regions (35). It is achieved by KAP1 acting as a scaffold for the assembly of silencing factors such as the H3K9 methyltransferase SETDB1 and the heterochromatin protein 1 (HP1). Polrmt and Pank4 represent examples of genes marked by KAP1 and H3K9me3 that we find to be co-occupied by SMARCAD1 (42,48,49). A SMARCAD1 CUE1 mutant protein displayed reduced binding to all KAP1 targets analyzed, highlighting a critical role for this motif in either the initial targeting of SMARCAD1 to KAP1 target genes or in the stability of the chromatin-bound complex.
We speculate that the importance of the CUE1 motif in facilitating stable localization of SMARCAD1 in the genome is not limited to those selected sites, given that in immunofluores-

Figure 6. SMARCAD1 binds KAP1 target genes in ESCs.
Shown is the measurement of SMARCAD1 occupancy by ChIP-qPCR at KAP1 binding sites in cells that lack exogenous SMARCAD1. After induction of SMARCAD1 knockdown with doxycycline for 2 days (see Fig. 5A, lanes 1 and 2), a reduction of SMARCAD1 enrichment was apparent at all KAP1 target genes examined (KD). One representative experiment of either three (for Ezr and Fkbp6) or two biological replicates is shown. Percentages of input values are the mean Ϯ S.E. (error bars) of three technical replicates.

SMARCAD1 genomic localization via KAP1
cence, the vast majority of SMARCAD1 is readily extracted from the nucleus if it carries a CUE1 mutation.
CUE motifs can bind mono-and polyubiquitin and are found in many eukaryotic proteins with various, often cytoplasmic functions, but are not common in ATP-dependent chromatin remodelers (18,37,(50)(51)(52). A possible role for the CUE motifs in SMARCAD1 and its yeast homologs in recognizing ubiquitylated histones has previously been debated. In support, an increase of ubiquitylated H2B was observed in Fun30 target genes, and SMARCAD1 was shown to bind H2A-ubiquitin dependent on intact CUE1 and CUE2 domains (19,53). However, in vitro binding experiments could not detect a preference of Fun30 or SMARCAD1 for ubiquitylated histones compared with unmodified H2 (24, 54). Our in vitro binding studies were carried out with recombinant KAP1 protein that was not ubiquitylated, suggesting that the KAP1 SMARCAD1-CUE1 interaction can occur independently of ubiquitin. It remains possible that (i) ubiquitin affects the affinity of this interaction in vivo, (ii) a ubiquitin-like mimic within KAP1 mediates the interaction, or, alternatively, (iii) that the CUE1 domain of SMARCAD1 recognizes a motif not related to ubiquitin. A putative homology with ubiquitin in KAP1 was not apparent using standard in silico prediction programs. Nor did we detect competition in KAP1-CUE1 binding experiments upon the addition of an excess of pure ubiquitin. Therefore, we postulate the possible existence of a novel, atypical interaction mode of the CUE1 domain of SMARCAD1 with KAP1. Figure 7. Association of SMARCAD1 with KAP1 target genes depends on the CUE1 domain. A, FLAG ChIP followed by real-time PCR analysis to assess FLAG-SMARCAD1 enrichment at KAP1 target genes, including imprinted genes and zinc finger protein (ZFP) genes. Screenshots show KAP1 peaks called from the data set of (42). The scale varies between 2.3 and 6.9 kb. The direction of genes and the primers used to detect binding are indicated with an arrowhead and a P, respectively. Depicted are qPCR data collected for each gene upon SMARCAD1 depletion (left) or of SMARCAD1 mutation in the CUE1 domain (right). Left, E14 ESCs expressing tagged wildtype SMARCAD1 after 2 days of FLUC (Ctrl) or Smarcad1 esiRNA treatment (KD); the latter conditions ensure that both endogenous and tagged SMARCAD1 protein levels are diminished (Fig. S6A). IgG control is shown, and H3 control is shown in Fig. S6B. Right, stable SMARCAD1 association with genes bound by KAP1 requires an intact CUE1 domain. Inducible SMARCAD1 knockdown ESCs expressing FLAG SMARCAD1 wildtype (WT) or CUE1 mutant protein (mt, F169K/L196K) were treated with doxycycline for 2 days to deplete endogenous SMARCAD1. ChIP results represent the mean Ϯ S.E. of biological duplicate experiments. B, binding patterns of H3 on KAP1 target genes in E14 ESCs expressing tagged SMARCAD1. Chromatin analyzed corresponds to samples from Fig. 7. Right, FLAG WT SMARCAD1 (WT) and FLAG SMARCAD1 CUE1 domain mutant F169K/L196K (mt).

Cell lines and plasmids
Mouse ES cells (PGK12.1; J1; E14 and derivatives thereof) were grown on gelatin-coated dishes without feeders in Dulbecco's modified Eagle's medium (Life Technologies, Inc.) supplemented with 15% FBS, 0.1 mM 2-mercaptoethanol, 0.1 mM non-essential amino acids, 2 mM L-glutamine supplemented with leukemia inhibitory factor (1000 units/ml) (Cell Guidance System). Medium with doxycycline (0.5 g/ml final concentration) was prepared fresh and changed during induction. HeLa cells and KAP1 knockdown clones thereof were described (13). Cell lines tested negative for mycoplasma contamination.
Generation of cell lines for affinity purification-The coding sequence of SMARCAD1 was amplified from mouse ESC cDNA and inserted into a CAG-driven expression vector, with an N-terminal single FLAG tag. Tagged Smarcad1 construct along with the empty vector control were transfected into PGK12.1 ESCs using Lipofectamine 2000 (Invitrogen), and clonal lines were selected with 1.7 g/ml puromycin. Expression of the tagged SMARCAD1 in selected clones was analyzed by indirect immunofluorescence and Western blot analysis with a FLAG antibody.
Generation of FLAG-SMARCAD1 ES cell lines for ChIP-Wildtype Smarcad1 and CUE1 double mutant (F169K/L196K) were inserted into a pCAG-driven expression vector with an N-terminal triple FLAG tag and a C-terminal V5 tag. These plasmids were stably integrated into E14 ESCs carrying a doxycycline-inducible shRNA targeting the 3Ј-UTR of Smarcad1 as described below before knocking down endogenous SMARCAD1. Cells with stable integrants were selected using 1 g/ml puromycin. Pools of transfectants were characterized by Western blotting and indirect immunofluorescence.

Knockdown (KD) experiments by RNAi
SMARCAD1 and KAP1 depletion was accomplished with esiRNAs and shRNA vectors using non-targeting controls in parallel. Knockdown efficiency was determined by Western blotting. Synthetic esiRNAs used to achieve transient depletion of SMARCAD1 and KAP1 were from Sigma, MISSION esiRNA targeting mouse Smarcad1 EMU209081 and mouse Kap1 EMU024081. As a non-targeting control, we used esiRNA against FLUC (Sigma, MISSION esiRNA EHUFLUC) as described in the manufacturer's instructions. For ChIP experiments, 2 ϫ 10 7 E14 ESCs were transfected in suspension with 39 g of esiRNA using Lipofectamine 2000 and analyzed 48 h later. For protein stability analysis, 1 ϫ 10 6 J1 ESCs were transfected as above with 800 ng of esiRNA twice, on day 0 and day 1, and analyzed 72 h after the first transfection.
An shRNA targeting the 3Ј-UTR of Smarcad1 was selected from the predictions in Fellmann et al. (55) and used in both constitutive and inducible KD experiments. An shRNA targeting Renilla luciferase served as a control (56). The constitutive KD vector has eGFP-miRE shRNA backbone and a puromycin resistance driven by a CAG promoter (CAG-eGFP-miRE-IRES-Puro). For the inducible KD, the shRNA eGFP-miRE cassette was cloned into the doxycycline-inducible vector pINDUCER (57). Several features of the original pINDUCER were modified, including changing the constitutive promoter to a CAG promoter and upgrading the inducible promoter to TRE3G. Stable clones of the inducible KD cells were generated by transducing ES cells before inducing knockdown by treatment with 0.5 g/ml doxycycline. Smarcad1 shRNA oligo sequence used: 5Ј-TGCTGTTGACAGTGAGCGACAGCAGAACAGATTAA-CTTAATAGTGAAGCCACAGATGTATTAAGTTAATCTG-TTCTGCTGGTGCCTACTGCCTCGGA-3Ј.

Purification of FLAG-SMARCAD1 and mass spectrometry
SMARCAD1 was purified from nuclear extracts of PGK12.1 ESCs expressing 1ϫ FLAG-tagged mouse SMARCAD1 protein. Nuclear extract preparations from cells transfected with an empty vector lacking FLAG-SMARCAD1 were used as a negative control. Immunoprecipitation from nuclear extracts using the mouse M2 FLAG antibody (Sigma) was performed for 3 h or overnight (4°C) in the presence of 150 units/ml Benzonase (Novagen) essentially as described (13,58). Affinity resin was loaded on spin columns (Thermo Scientific Pierce) and washed twice in 20 mM HEPES, pH 7.6, 100 mM KCl, 0.2 mM EDTA, 1.5 mM MgCl 2 , 0.5 mM DTT, 20% glycerol, 0.02% Nonidet P-40 and then washed three times in the same buffer without Nonidet P-40. Bound proteins were eluted in the final wash buffer supplemented with 0.3 mg/ml 1ϫ FLAG peptide (Sigma). Eluted proteins were run on polyacrylamide gels, gel slices were trypsinized, and peptides were analyzed by mass spectrometry, as described (58).

Co-immunoprecipitation
Precleared extracts, either 150 g of nuclear extract or 300 g of whole cell extract, were incubated with 3-3.5 g of specific antibody or IgG. 150 units/ml Benzonase (Novagen) and 0.1 g/l EtBr were added throughout the procedure to reduce interactions mediated by nucleic acid. Immune complexes were captured by Protein G Dynabeads (Novex). Dynabeads-Ab-Ag complexes were washed four times in 20 mM HEPES, pH 7.6, 100 mM KCl, 0.2 mM EDTA, 1.5 mM MgCl, 0.5 mM DTT, 20% glycerol, 0.05% Triton X-100 containing protease inhibitors and then once in the same buffer but with additional 50 mM NaCl. Immune complexes were eluted with SDS sample buffer.
Images were acquired on a LEICA DMR fluorescent microscope and processed using Adobe Photoshop CS3 or FIJI. In experiments where different cell lines were compared, these were grown on the same chamber slide and stained and processed simultaneously. Images were collected with the same exposure time, and identical post-processing was used.

Chromatin immunoprecipitation
ChIP assays were performed using a one-day ChIP kit (Diagenode C01010080) and double-cross-linking. About 5 ϫ 10 7 E14 ES cells/15-cm plate were cross-linked with 2 mM disuccinimidyl glutarate (Thermo Scientific, 20593) in PBS for 45 min at room temperature, washed three times with PBS, and crosslinked with 1% formaldehyde in PBS (Polysciences, 04018) for 10 min at room temperature. Formaldehyde was quenched using 125 mM glycine. After cell lysis, chromatin fragmentation was performed using a Diagenode Bioruptor (20 -25 cycles of 30 s on/off) to produce fragments of ϳ200 -600 bp. Immunoprecipitation was performed following the manufacturer's instructions using 100 g of chromatin and 3 g of specific antibody (SMARCAD1 HPA016737, Sigma; IgG C15410206, Diagenode; IgG 5381, Sigma; FLAG F1804, Sigma; H3 ab1791, Abcam). Pulldowns using IgGs were used as a negative control. ChIP-enriched DNA was analyzed in triplicate using real-time PCR with SYBR Green (Bio-Rad) on a CFX96 Connect (Bio-Rad) or Agilent MX3000P system. Enrichment values for specific factors were normalized to input. Primer sets used for qPCR are available in Table S1.

Extract preparations, cell fractionations, and Western blot analysis
Nuclear Extracts were prepared according to Dignam et al. (59). Total whole-cell extracts were prepared as described previously (60).
Cell fractionations were based on the method presented by Zhang et al. (47). ES cell lines stably expressing epitope-tagged SMARCAD1 were treated for 2 days with doxycycline to reduce endogenous SMARCAD1 levels. Asynchronously growing ESCs were trypsinized and washed in ice-cold PBS supplemented with 1 mM benzamidine and 0.5 mM PMSF. The cell pellets were resuspended in 1ϫ packed cell volume (PCV) of CSK buffer (10 mM PIPES, pH 6.8, 300 mM sucrose, 100 mM NaCl, 3 mM MgCl 2 , 1 mM benzamidine, 0.5 mM PMSF) with 0.1% Triton X-100 and incubated on ice for 3 min. Ten percent of the extract was retained as the "total fraction" and diluted in an equal amount of CSK buffer supplemented with 300 units/ml Benzonase (Novagen), and after 10 min on ice, 5 mM EDTA was added to stop the reaction. The remaining 90% of the lysate was centrifuged at 12,000 rpm for 5 min at 4°C. The supernatant was retained as the "soluble fraction." The pellet was washed once briefly with 1ϫ PCV CSK buffer and then taken up in 1ϫ PCV CSK buffer supplemented with 300 units/ml Benzonase. After 10 min on ice, 5 mM EDTA was added to stop the reaction. Extract amounts corresponding to the equal numbers of starting cells were analyzed by SDS-PAGE and immunoblotting.

Recombinant proteins
His-tagged human SMARCAD1 and KAP1 baculovirus expression constructs were provided by Dr. P. Varga-Weisz (13). Proteins were prepared from infected Sf9 cells by four freeze-thaw cycles in 250 mM NaCl, 20 mM HEPES, pH 7.5, 10% glycerol, and protease inhibitors. Proteins were bound to a nickel-agarose column (Qiagen 30210). After washing, bound protein was eluted in the same buffer containing 5 mM imidazole. Fractions containing pure protein were dialyzed against the same buffer plus 1 mM 2-mercaptoethanol. V5-tagged human SMARCAD1 was produced from the pCDNA3.1/nV5-DEST hSMARCAD1-wt plasmid (13) in vitro in a T7 TNT coupled transcription-translation reaction (Promega).

SMARCAD1 genomic localization via KAP1 GST pulldown assays
Binding experiments with recombinant proteins utilized 25 l of packed glutathione-Sepharose 4B beads (GE Healthcare) with 10 -15 g of GST or GST fusion protein to give an effective concentration in the micromolar range in the binding reactions. A blocking step was regularly incorporated into the assay in which beads were blocked with 1% fish gelatin and 1 mg/ml BSA in binding buffer at 4°C for 1 h followed by three washes in binding buffer. Pure recombinant target protein was mixed with GST fusion protein beads to give a final concentration of 0.2-0.5 M in a final volume of 150 l and incubated for 1.5 h at 4°C. Beads were washed three times in binding buffer, and bound proteins were eluted from the beads with two sequential extractions in SDS sample buffer at room temperature. Bound proteins were detected by Western blotting.
Binding reactions using GST fusion proteins and target proteins generated in the T7 TNT reaction were carried out in 20 mM HEPES, pH 7.5, 75 mM KCl, 0.1 mM EDTA, 2.5 mM MgCl 2 , 1 mM DTT, 0.05% Triton X-100 as recommended by the supplier.

Real-time PCR
Total RNA was extracted with TRIzol (Invitrogen), and cDNA was synthesized using SuperScript II reverse transcriptase (Invitrogen) following the supplier's recommendations. qPCRs were performed on the CFX Connect real-time PCR detection system (Bio-Rad) or Agilent MX3000P, using iTaq Universal SYBR Green (Bio-Rad). Pairs of primers were evaluated for generating single-peak melting profiles and for linear amplification over a range of DNA template dilutions. qPCR assays were performed in technical triplicates. Three housekeeping genes (Gapdh, Hsp90ab1, and Atp5b) were used as references for normalization. Primers are listed in Table S1.

Flow cytometry
ESCs resuspended in 200 l of PBS were fixed by the dropwise addition of 1.3 ml of ice-cold 70% ethanol. After fixation at 4°C, cells were taken up in PI/RNase staining buffer (BD Pharmingen, 550825), incubated for 15 min at room temperature, and analyzed using a BD LSR II cytometer and FlowJo version 10.2 software.