A Role for Widely Interspaced Zinc Finger (WIZ) in Retention of the G9a Methyltransferase on Chromatin*

Background: G9a-GLP lysine methyltransferases mono- and di-methylate histone H3 lysine 9 (H3K9me2). Results: Widely interspaced zinc finger (WIZ) regulates H3K9me2 levels through a mechanism that involves retention of G9a on chromatin. Conclusion: The G9a-GLP-WIZ complex has unique functions when bound to chromatin that are independent of the H3K9me2 mark. Significance: Combining pharmacologic and genetic manipulations is essential to any translational hypotheses related to G9a function. G9a and GLP lysine methyltransferases form a heterodimeric complex that is responsible for the majority of histone H3 lysine 9 mono- and di-methylation (H3K9me1/me2). Widely interspaced zinc finger (WIZ) associates with the G9a-GLP protein complex, but its role in mediating lysine methylation is poorly defined. Here, we show that WIZ regulates global H3K9me2 levels by facilitating the interaction of G9a with chromatin. Disrupting the association of G9a-GLP with chromatin by depleting WIZ resulted in altered gene expression and protein-protein interactions that were distinguishable from that of small molecule-based inhibition of G9a/GLP, supporting discrete functions of the G9a-GLP-WIZ chromatin complex in addition to H3K9me2 methylation.

G9a and GLP lysine methyltransferases form a heterodimeric complex that is responsible for the majority of histone H3 lysine 9 mono-and di-methylation (H3K9me1/me2). Widely interspaced zinc finger (WIZ) associates with the G9a-GLP protein complex, but its role in mediating lysine methylation is poorly defined. Here, we show that WIZ regulates global H3K9me2 levels by facilitating the interaction of G9a with chromatin. Disrupting the association of G9a-GLP with chromatin by depleting WIZ resulted in altered gene expression and protein-protein interactions that were distinguishable from that of small molecule-based inhibition of G9a/GLP, supporting discrete functions of the G9a-GLP-WIZ chromatin complex in addition to H3K9me2 methylation.
Transcriptional silencing and the H3K9 mono-and di-methylation functions of G9a and GLP are dependent on heterodimerization through their respective SET domains (25). Studies in mouse G9a and Glp knock-out cells demonstrate that loss of either G9a or GLP results in degradation of the remaining protein partner (13,25). Reconstitution of these cells with a heteromeric complex consisting of either a G9a or a GLP catalytic mutant suggests that the catalytic activity of GLP, but not G9a, is dispensable for H3K9 methylation (13).
Although the interaction between G9a and GLP has been studied extensively, the interaction between these proteins and a third complex member, widely interspaced zinc finger (WIZ) protein, is less well understood. The mouse Wiz gene products were identified as two alternatively spliced isoforms, WizS and WizL, which were abundantly expressed in the brain compared with other tissue types (26). WIZ has C 2 H 2 -type zinc finger motifs, originally characterized in the Xenopus TFIIIA (27) and Drosophila Krüppel transcription factors (28). Typical C 2 H 2 -type zinc finger motifs are separated by seven amino acids. The WIZ zinc fingers are widely spaced, being separated by distances ranging from 16 to 258 amino acids in mouse (26) and from 16 to 263 amino acids in the longest human splice variant. Mutational analysis has demonstrated that WIZ interacts with the C-terminal SET domain of G9a or GLP through its C-terminal zinc finger (29). WIZ has been shown to bridge the interaction between G9a-GLP and the transcriptional co-repressors C-terminal binding protein 1 (CtBP1) and CtBP2, possibly to help recruit G9a-GLP to specific genomic loci (29), and is a non-histone methylation target for G9a (18).
WIZ has also been implicated in G9a and GLP protein stability (29). We found that WIZ knockdown leads to an H3K9me2 loss that is not attributable to the degradation of G9a or GLP protein; rather, WIZ is important for the retention of G9a on chromatin. Using pharmacological inhibition of G9a and genesilencing methods to regulate WIZ, we compared G9a localization with evidence of enzymatic activity. Although the loss of G9a chromatin localization mediated by WIZ and small molecule inhibition of G9a and GLP activity (30) both resulted in a loss of H3K9me2, these treatments variably affected gene expression and G9a protein-interacting partners.

Experimental Procedures
Cell Culture-HEK293T cells were grown in DMEM supplemented with 10% FBS. FLAG-HA-G9a HeLa cells (9) were grown in MEM supplemented with 5% FBS and 5% horse serum. Stable pools of WIZ shRNA knockdown and control shRNA knockdown cells were created as follows. FLAG-HA-G9a HeLa cells (9) were transduced with lentivirus containing WIZ shRNA with a target sequence identical to the WIZ siRNA used in this study (Sigma Mission WIZ, SHCLNV; clone ID TRCN0000253784) or a control shRNA that contains an shRNA insert that does not target any known genes from any species (Sigma Mission pLKO.1-puro non-target shRNA, SHC016V). Transduced cells were selected with 5 g/ml puromycin, and the resulting stable pools were grown in MEM supplemented with 5% FBS, 5% horse serum, and 5 g/ml puromycin.
siRNA Knockdown and Compound Treatment-Cells were treated with the indicated media supplemented with 1 M UNC0638 (30) or the volume equivalent of DMSO for 48 or 96 h, as indicated. All siRNAs were purchased from GE Healthcare Dharmacon, and transfections were performed as per the manufacturer's instructions with DharmaFECT 4 transfection reagent. siRNAs used in this study were as follows: control (scrambled) siRNA (Dharmacon non-targeting 001810-04), G9a siRNA (Dharmacon 006937-05), and WIZ siRNA custom (ON-TARGETplus, Dharmacon) with sense sequence 5Ј-CAGCAGAGGUCAAGGCCAAUU-3Ј and antisense sequence 5Ј-UUGGCCUUUGACCUCUGCUGUU-3Ј.
Quantitative RT-PCR-RNA was extracted using the RNeasy Plus mini kit (Qiagen), and RNA was quantified on a NanoDrop 2000 spectrophotometer (Thermo Fisher Scientific). The reverse transcription reaction was performed with 500 ng of RNA using Superscript III First Strand Synthesis Supermix (Invitrogen 1172-050) according to the manufacturer instructions. The resulting cDNA was diluted 1:10, and 2 l of the mixture was used for each PCR reaction. Quantitative PCR was performed using Roche FastStart Universal SYBR Green Master Mix (Roche Applied Science) in a 384-well plate format on an ABI 2900HT instrument. Oligo sequences are available upon request.
Immunoprecipitation-Whole cell protein extract was prepared by lysing cells in Cytobuster TM protein extraction reagent (EMD Millipore 71009) supplemented with 25 units of Benzonase nuclease (EMD Millipore 101654). Subcellular fractionation was performed using the Pierce Subcellular Fractionation Kit (Thermo Fisher Scientific 78840) according to the manufacturer's instructions. FLAG-HA-G9a was immunoprecipitated from the samples described with Anti-FLAG® M2 affinity gel (Sigma A2220) overnight at 4°C and washed three times in TBST (20 mM Tris-HCl, pH 8.0, 150 mM NaCl, and 0.1% Tween 20). Bound proteins were eluted by boiling in 1ϫ SDS-PAGE loading buffer and subjected to Western blot analysis.
G9a Protein Stability Assay-FLAG-HA-G9a cells were plated on 6-well plates at a density of 5 ϫ 10 5 cells/well. After 24 h, cells were transfected with siRNA as described previously or treated with 1 M UNC0638 (or volume equivalent of DMSO) and incubated for 48 h. Cells were then treated with 1 M MG132 (or volume equivalent of DMSO) for 12 h followed by the addition of 25 g/ml (Fig. 3A) or 50 g/ml (Fig. 3B) cycloheximide (or volume equivalent of ethanol) for 0, 4, 6, 9, or 13.5 h. Cells were lysed in Cytobuster TM protein extraction reagent (EMD Millipore 71009) supplemented with 25 units of Benzonase nuclease (EMD Millipore 101654). The protein extract concentration was determined with Bradford reagent (Bio-Rad 500-0006) and a standard curve generated by known quantities of bovine serum albumin. Ten g of lysate was run on SDS-PAGE and subjected to Western blotting.
Western Blotting and Densitometry-Western blotting was performed using a Typhoon imager (GE Healthcare) as described elsewhere (31) or using a LI-COR Odyssey as described elsewhere (32). Densitometry was performed using ImageJ software or LI-COR Image Studio software.
Chromatin Immunoprecipitation-Chromatin immunoprecipitation was performed using the ChIP-IT High Sensitivity kit (Active Motif 53040) according to the manufacturer's instructions, except that the input samples were purified by phenol: chloroform extraction. The following antibodies were used: anti-H3K9me2 (Abcam ab1220), anti-H3K9me3 (Abcam ab8898), anti-G9a (Abcam ab40542), and Rabbit IgG (Jackson ImmunoResearch 011-000-003). ChIP samples were quantitated using the Qubit 2.0 fluorometer (Invitrogen). A portion of each sample was confirmed by qPCR as described above. The fold change over background was calculated using the ⌬⌬Ct method (33), where immunoprecipitation (IP)/input for the ChIP antibody signal was compared with the IP/input value for IgG (background). After a confirmation of the ChIP signal over background, the remainder of the sample was submitted for high-throughput sequencing. Oligo sequences are available upon request.
High-throughput Sequencing and Data Analysis-Libraries were prepared for high-throughput using the Illumina TruSeq kit as per the manufacturer's recommendations. Single-end 50-bp reads were sequenced on an Illumina HiSeq. Reads were checked to be of sufficient quality, filtered to remove those with a significant adapter contribution using TagDust (34), and aligned to the reference human genome (hg19) using Bowtie (35). Regions of significant ChIP enrichment relative to input control ("peaks") were identified using MACS2, assuming fragments were 250 bp in length. Heat maps were generated in MATLAB by plotting the normalized ChIP-seq signal around the midpoints (Ϯ2 kb) of the union set of G9a peaks. Changes in the ChIP signal were illustrated by taking the arithmetic difference between treated and untreated samples. ChIP-seq data are available from NCBI Gene Expression Omnibus (accession number GSE67317). Genomic distributions were computed using CEAS (36). The CCCTC-binding factor (CTCF) and H3K4me3 ChIP-seq data from 293T cells were mined from BAM files provided by ENCODE (37). To find regions with significantly altered ChIP-seq enrichment, the number of overlapping aligned reads for all 500-bp non-overlapping windows across the genome was first tabulated for all samples. Only those windows with an average number of aligned reads exceeding 10 were retained. A binomial test was then applied to compare the G9a ChIP-seq signal between control siRNA-treated and WIZ siRNA-treated cells at each of these windows; p values were corrected for multiple testing using the q value. Windows with q Ͻ 1 ϫ 10 Ϫ5 were retained and used for subsequent analyses.
Gene Expression Microarrays-RNA was extracted using the RNeasy Plus mini kit (Qiagen) and quantified on the Qubit fluorometer (Invitrogen). RNA quality was verified on an Agilent bioanalyzer before hybridization to human 8 ϫ 60 K arrays (Agilent Technologies 28004). After scanning, microarray images were processed using Agilent Feature Extraction software according to the manufacturer-supplied protocols. The resulting background-subtracted and local regression (LOESS)normalized log ratio data were collected and filtered to remove data generated from poor quality or low intensity spots. Probes with Ͼ30% missing data were removed from further evaluation. Dye swap replicates were inverted to match the log ratio direction of non-dye swap replicates. The remaining missing data were then imputed using weighted K-nearest neighbors (KNNimpute) (38). Testing for differential expression was performed using the statistical analysis of microarray (SAM) (39) one-class option for a comparison with DMSO or scrambled siRNA control and the two-class unpaired option when comparing UNC0638 with WIZ siRNA knockdown. A conservative false discovery rate threshold of 0.5% was set to identify differentially expressed genes in both experiments. The differential expression results were visualized using hierarchical cluster analysis based on average linkage clustering and a Pearson correlation distance metric. All statistical analyses were carried out using the R package (v3.1.1). Expression data are available from NCBI Gene Expression Omnibus (accession number GSE70914).
G9a Purification for Proteomics-G9a was purified from the HeLa FLAG-HA-G9a cell line (9) grown in a volume of 5 liters (ϳ8 ml cell pellet), treated with either 1 M UNC0638 or volume equivalent of DMSO for 48 h. Cells were centrifuged, washed once in PBS, and then frozen at Ϫ80°C prior to purification. Purification was performed as follows. All buffers were supplemented with 1ϫ protease inhibitor mixture (Roche 05056489001). Cell pellets were thawed on ice, resuspended in 40 ml of nuclei prep buffer (50% glycerol, 0.5% Nonidet P-40, 1 mM PMSF, 10 mM Tris-HCl, pH 7.4, 10 mM NaCl, and 3 mM MgCl 2 ), and incubated on ice for 10 min or until the nuclei were visible with trypan blue stain. Nuclei were pelleted, washed once with reticulocyte standard buffer (RSB: 10 mM Tris-HCl, pH 7.4, 10 mM NaCl, and 3 mM MgCl 2 ), resuspended in 24 ml of nuclei lysis buffer (0.2% Triton X-100, 20 mM HEPES, pH 7.9, 1.5 mM MgCl 2 , 0.56 M NaCl, 10 mM KCl, and 0.5 mM DTT) supplemented with 250 units of Benzonase nuclease (EMD Millipore 101654), and incubated at 37°C for 10 min followed by 1 h at 4°C. The nuclear lysate was clarified by ultracentrifugation for 30 min at 40,000 rpm in a Ti70.1 rotor, and the salt concentration was lowered from ϳ0.42 to ϳ0.3 M with no salt nuclei lysis buffer (10 mM HEPES, pH 7.9, 1.5 mM MgCl 2 , 10 mM KCl, and 0.5 mM DTT) followed by ultracentrifugation as described above. Lysate was transferred to a chromatography column (Bio-Rad 731-1550), and anti-FLAG M2 affinity gel (Sigma A2220) equilibrated in 0.42 M NaCl nuclei lysis buffer (ϳ100 l of beads) was incubated with the clarified lysate over-night at 4°C. Beads were washed with 50 ml of wash buffer (0.2% Triton X-100, 10 mM HEPES, pH 7.9, 1.5 mM MgCl 2 , 0.3 M NaCl, and 10 mM KCl) and transferred to a 0.8-ml chromatography spin column (Pierce 89868). The protein was eluted by rotation at room temperature in 200 l of elution buffer (50 mM Tris-HCl, pH 7.5, 0.5% Nonidet P-40, 0.15 M NaCl, and 0.2 mg/ml 2ϫ FLAG peptide) three times for a total of ϳ600 l of eluate. Eluates were diluted to 1.5 ml final volume in wash buffer and transferred to a chromatography column (Bio-Rad 732-6008). Anti-HA-agarose (Sigma A2095) was equilibrated in wash buffer (ϳ100 l beads) and incubated with the eluates overnight at 4°C. Beads were washed with 30 ml of wash buffer and then transferred to a 0.8-ml chromatography spin column (Pierce 89868). Proteins were eluted in 200 l of elution buffer (50 mM Tris-HCl, pH 7.5, 0.5% Nonidet P-40, 0.15 M NaCl, and 1.0 mg/ml HA peptide) four times for a total of ϳ800 l of eluate. A portion of the eluates was visualized by silver staining, and the remaining sample was submitted for mass spectrometry analysis. Proteomic Analysis-Multidimensional protein identification technology (MudPIT) was carried out as described previously (40,41). In brief, proteins from each elution were precipitated with trichloroacetic acid, and the resulting pellets were washed with 100% cold acetone. The dried protein pellets were then resuspended in 8 M urea and reduced and alkylated with tris(2-carboxyethyl)phosphine and chloroacetamide as described previously (42). The proteins were then subjected to proteolytic digestion with LysC overnight at 37°C with shaking followed by digestion with Trypsin Gold (Promega) after dilut-ing the sample to 2 M urea. The reaction was quenched with formic acid and then pressure-loaded onto a microcapillary column pack with both strong cation exchange resin and reverse phase resin as described previously (42). A nine-step MudPIT run was performed for each sample on an LTQ-Velos Pro that was in-line with a Proxeon Easy nanoLC (41). The resulting RAW data files were processed by protein database matching using SEQUEST within Proteome Discoverer 1.4 (Thermo Fisher Scientific). Database matching was performed using a Homo sapiens FASTA database downloaded from UniProt on  03/16/2014. The database also contained potential sample contaminant proteins including human keratins, IgGs, and proteolytic enzymes. The database search was performed with ϩ57 daltons on cysteine as a static modification and the following modifications as dynamic: ϩ16 daltons on methionine (oxidation), ϩ14 daltons on lysine (mono-methylation), ϩ28 daltons on lysine (di-methylation), and ϩ42 daltons on lysine (tri-methylation). All peptide-spectral matches were required to have a false discovery rate of Յ1%. SAINT (significance analysis of interactome) probability scores were calculated as outlined in the CRAPome (contaminant repository for affinity purification) and various previous publications (41,(43)(44)(45). Mock purifications were performed in parallel in this study, and analysis was performed using the same methods described above for SAINT analysis.

Enzymatic Inhibition of G9a or Knockdown of WIZ Expression Leads to a Global Reduction in H3K9 Methylation Levels-
We initially compared H3K9me2 levels following chemical inhibition of G9a catalytic activity and reduction in G9a protein levels due to WIZ knockdown. Inhibition of G9a catalytic activity in HEK293T cells with a small molecule probe, UNC0638 (30), did not result in a change in G9a, GLP, or WIZ mRNA levels ( Fig. 1A) but did lead to a global reduction in H3K9me2 levels and an increase in G9a protein (Fig. 1, B and C). Treatment of HEK293T cells with a siRNA against WIZ reduced WIZ mRNA 5-fold (p Ͻ 0.0001) but did not affect G9a or GLP mRNA levels (Fig. 1D). H3K9me2 (p ϭ 0.03) and WIZ (p ϭ 0.001) protein levels decreased compared with the control siRNA (Fig. 1, E and F). There was also a slight (1.25-fold) decrease in G9a levels, which agrees with a previous study that linked WIZ with G9a protein stability (29). To determine whether WIZ knockdown affects long-term G9a protein stability, we stably expressed WIZ shRNA or control in a HeLa cell line that expresses a double tagged version of G9a (9). These cells demonstrated no change in G9a or GLP mRNA levels (p ϭ 0.1311), whereas WIZ mRNA levels were reduced 2-fold (p ϭ 0.035) (Fig. 1G). Even after multiple passages of this cell line, the overall change in G9a protein levels remained very small (1.25fold, p ϭ 0.02), whereas H3K9me2 levels were reduced 2.5-fold (p ϭ 0.01) and WIZ protein levels were 2.5-fold lower (input, p ϭ 0.02) (Fig. 1, H and I). These data suggest that reduction in H3K9me2 associated with WIZ silencing was not due to progressive loss of G9a protein.
WIZ Is Important for G9a Interaction with Chromatin-Because other zinc finger proteins have been shown to regulate lysine methyltransferase activity (46,47), we investigated the possibility that WIZ is important for G9a retention on chromatin. To assess G9a chromatin interaction, we performed subcellular fractionation of HEK293T cells that were treated with UNC0638, WIZ siRNA, or combinations of the two to determine whether these treatments had an additive effect. Neither treatment nor the combination of the two had an effect on G9a or GLP mRNA levels, whereas WIZ mRNA levels were reduced to a similar extent in both the siRNA knockdown and the combination treatment (Fig. 2A). H3K9me2 levels were visibly reduced after treatment with UNC0638 in combination with control siRNA (Fig. 2B, compare lanes 5 and 6, 4.8-fold) or WIZ siRNA (Fig. 2B, compare lanes 5 and 8, 8.6-fold) and were slightly reduced after treatment with DMSO combined with WIZ siRNA (Fig. 2B, compare lanes 5 and 7, 1.6-fold). Consistent with the changes in H3K9me2 levels, G9a and GLP association with the chromatin fraction was altered after treatment with UNC0638 or WIZ siRNA. G9a levels decreased slightly after the addition of UNC0638 (Fig. 2B, compare lanes 5 and 6,  1.9-fold), WIZ siRNA (Fig. 2B, compare lanes 5 and 7, 1.8-fold), or a combination of both treatments (Fig. 2B, compare lanes 5 and 8, 6.2-fold). Similar to G9a, GLP levels decreased in the chromatin fraction after the addition of UNC0638 (Fig. 2B, compare lanes 5 and 6, 1.9-fold), WIZ siRNA (Fig. 2B, compare lanes 5 and 7, 1.7-fold), or a combination of both treatments (Fig. 2B, compare lanes 5 and 8, 9.1-fold). The dramatic reduction in G9a and GLP protein levels in the chromatin fraction after the combination treatments (Fig. 2B, lane 8) indicates that enzymatic inhibition and reduction in WIZ protein levels have an additive effect, suggesting that WIZ may help stabilize G9a interaction with chromatin.
Because we observed increased G9a in the soluble fraction after UNC0638 treatment not attributable to an increase in G9a mRNA (Figs. 1A and 2A), we examined G9a protein stability in HeLa FLAG-HA-G9a cells. We inhibited de novo protein synthesis with cycloheximide (CHX) in combination with MG132 proteasome inhibitor, UNC0638, or WIZ siRNA knockdown (Fig. 3). Although treatment with UNC0638 for 48 h resulted in higher levels of G9a, none of the treatments demonstrated diminished G9a protein turnover including up to 13.5 h of CHX treatment (Fig. 3A). As a control, the levels of a control protein, COXIV, were reduced after 6 h of CHX treatment (Fig. 3B). Extending the experiment beyond 13.5 h was not possible (Fig.  3A) due to cell toxicity. Thus, we cannot rule out the possibility that UNC0638 increases long-term G9a protein stability.
To further examine G9a association with chromatin, we performed subcellular fractionation of HeLa cells expressing FLAG-HA-tagged G9a followed by FLAG IP of G9a. G9a association with GLP, WIZ, H3K9me3, or H3 did not change following UNC0638 treatment (Fig. 4, A and B), but H3K9me2 levels in three biological replicate experiments were lower in the input (p ϭ 0.030) and IP (p ϭ 0.001) samples compared with the H3 input signal (Fig. 4B). We repeated the experiment using a FLAG-HA-G9a HeLa stable pool expressing a control or WIZ shRNA knockdown (Fig. 4C). Although G9a continued to interact with GLP when WIZ was knocked down, its interaction with both K9 di-methyl (2.2-fold) and tri-methyl (2.3-fold) H3 and total H3 (2.2-fold) was noticeably reduced. These data indicate that a reduction in WIZ protein levels decreases the association of G9a with histones in the chromatin fraction (Fig. 4D). We then examined G9a and WIZ localization to either the nuclear or chromatin fractions (equivalent to Fig. 4, A and C, Input). Treatment with UNC0638 resulted in an increase in G9a protein levels in the nuclear compartment as noted earlier (Fig. 2B), but G9a levels remained unchanged in the chromatin fraction, and no alteration in WIZ protein levels was observed (Fig. 4E). In contrast, WIZ shRNA knockdown significantly reduced G9a levels in the chromatin fraction (Fig. 4F, 2-fold).
To confirm that this effect was WIZ-specific, we transfected a plasmid expressing the short human WIZ splice variant cDNA into the stable WIZ silenced HeLa cells (Fig. 1E) followed by subcellular fractionation (Fig. 5). The short isoform (WIZS) was selected for reconstitution for several reasons. First, it contains five zinc fingers that are highly conserved in mouse and human WIZ, whereas the additional zinc fingers present in the long isoform (WIZL) are less conserved (26,48). Second, in mice, WizL and WizS are expressed at different times during development, with WizS being highly expressed in the developing cerebellum and WizL expression restricted to the mature cerebellum (26). Despite the disruption of WizL in C57BL/6 mice due to the insertion of an ETn retrotransposon, there is no obvious phenotype in the cerebellum, suggesting a nonessential role for WizL (48). Third, G9a and GLP interact with the C-terminal zinc finger of WIZ, which is present in the protein product for both splice variants (29). Note that the WIZ siRNA recognizes the 3Ј-end of the WIZ transcript, and therefore both isoforms were knocked down after treatment (Fig. 5A). Because WIZ mRNA was reduced about 2-fold in the shRNA stable cell line (Fig. 1E), and the pWIZ vector was overexpressed com-pared with endogenous WIZ (data not shown), we did not mutate the region targeted by WIZ shRNA in the pWIZ vector.
In the WIZ knockdown pool, G9a protein levels were reduced in the chromatin fraction compared with the control knockdown pool (p ϭ 0.024, Fig. 5, B and C). Transfection of the pWIZ vector resulted in an increase in G9a protein levels in the chromatin fraction to levels comparable with those of the control knockdown pool (p ϭ 0.028, Fig. 5, B and C). Thus, the protein product of the WIZS isoform is sufficient to restore G9a and WIZ protein levels to chromatin.
To determine whether G9a chromatin occupancy was affected by enzymatic inhibition or reduced levels of WIZ, we performed chromatin immunoprecipitation followed by highthroughput sequencing (ChIP-seq). HEK293T cells were treated with 1 M UNC0638 or DMSO for 48 h (Fig. 6A) or 96 h  ( Fig. 6B) or control siRNA or WIZ siRNA for 72 h (Fig. 6C) followed by ChIP with antibodies against G9a or H3K9me2 (30) (Figs. 6 -8).
We first normalized the ChIP-seq enrichment computed over the union set of G9a binding sites to account for differences in the sequencing depth. We then assessed the difference in the G9a and H3K9me2 ChIP-seq signals between UNC0638treated (Fig. 6, A and B) or WIZ siRNA (Fig. 6C)-treated and untreated cells by subtracting the signal from control cells. A reduction in the H3K9me2 signal was observed for all treated samples compared with controls (Fig. 6, A-C and E). Following UNC0638 treatment, the G9a signal was elevated (Fig. 6, A, B,  and D), which might be explained by the increase in the G9a protein levels (Figs. 2B, 3, and 4A). This elevation in the G9a signal was general and subtle, with no one site in the -treated samples having a statistically significant increase over controls. The signal difference decreased and became more diffuse after 96 h. In contrast, WIZ siRNA treatment resulted in a loss of G9a signal (Fig. 6, C and D). Examples of the changes at specific regions are highlighted in Fig. 7. These data suggest that the decreased G9a chromatin binding associated with a reduction in WIZ is responsible for the loss of H3K9me2.
Assessment of the genomic distribution of regions with G9a occupancy revealed that the majority of G9a signal was localized to introns or distal intergenic regions (Fig. 8A). UNC0638 treatment resulted in a slight increase in G9a occupancy at distal intergenic regions at 48 h (2.8%) and 96 h (4%), with no other large changes in occupancy distribution noted. Treatment with WIZ siRNA had little effect on overall G9a distribution, with the remaining G9a signal increasing slightly at promoters (1.6%) and decreasing slightly at distal intergenic regions (2.7%). Because WIZ siRNA knockdown resulted in an overall loss of G9a signal, we examined the genomic distribution of regions with reduced G9a occupancy compared with control siRNA knockdown (Fig. 8B). We identified 14,116 regions where G9a occupancy was significantly reduced following WIZ knockdown (binomial test, q Ͻ 1 ϫ 10 Ϫ5 ). Many of these differential regions were distal from genes. ϳ83% of these sites were greater than 10 kilobases (kb) from a transcription start site (TSS), and the median distance to the nearest TSS was ϳ30 kb. There were, however, 2389 sites within 10 kb of a TSS. Despite the small proportion of sites (ϳ17%) in the vicinity of a TSS, we noted that the loss of G9a was slightly skewed toward promoters and coding exons following WIZ siRNA treatment relative to the genomic coverage of those features. By associating the sites of G9a signal loss with ENCODE ChIP data for HEK293 cells (37), we observed signal enrichment with H3K4me3 and the CTCF (Fig. 8C), both of which are associated with active transcription.
The Effects on Gene Expression by WIZ Knockdown Do Not Completely Overlap with Those by Enzymatic Inhibition of G9a-Because WIZ silencing and G9a inhibition both led to reduced H3K9me2 levels, we asked whether gene expression would be similarly affected. G9a can act as both a co-activator and co-repressor of transcription (49 -53). As the ability of G9a to affect transcription is not fully dependent on its catalytic activity, we predicted that treatment with UNC0638 would have a different effect on global gene expression compared with WIZ knockdown.
Microarray-based gene expression analysis was performed using cDNA from HEK293T cells treated with G9a siRNA, WIZ siRNA, or UNC0638 (Fig. 9) and compared with cDNA from control siRNA knockdown or vehicle (DMSO) treatment. We identified genes that demonstrated a concordant (Fig. 10A) or divergent (Fig. 10B) change in gene expression. Gene expression was median-centered to highlight the magnitude of differential expression.
For genes with a similar trend in RNA abundance after either WIZ knockdown or UNC0638 treatment (Fig. 10A), 72% (147 of 205 genes at Ͻ0.5% false discovery rate) exhibited lower expression, suggesting a significant effect of these treatments on the G9a co-activator function. In comparison, G9a knockdown resulted in a 50% overall gene repression (135 of 205 genes). Because the loss of G9a protein results in the degradation of GLP as well as loss of the proteins that would normally interact with the heterodimer (13,25), it is not surprising that G9a knockdown affects gene expression differently than either enzymatic inhibition (UNC0638 treatment) or loss of chromatin association (WIZ knockdown). Indeed, this observation is also true for genes with divergent expression changes following WIZ knockdown versus UNC0638 treatment (Fig. 10B).
For differentially expressed genes (Fig. 10B, 118 genes at Ͻ0.5% false discovery rate), enzymatic inhibition resulted in derepression of gene expression, whereas WIZ knockdown led to transcriptional repression. As noted earlier, only ϳ17% of the regions of G9a loss were within 10 kb of a TSS. Of these sites, 42% (or 1015) were near a gene that was represented on our microarray. Although only 32 of those regions were near a gene with a significant (q Ͻ 0.05) change in gene expression (Fig. 10C), all but one of those genes were repressed following WIZ knockdown. In contrast, UNC0638 treatment showed less overall change in the gene expression profiles in the same regions. The fact that loss of G9a ChIP signal after WIZ knockdown is associated with signals associated with active transcription, but are not generally seen directly at transcription start sites (Fig. 8, B and C), could indicate that G9a exerts co-transcriptional influence from a distance.
Taken together, these data suggest that binding of G9a to chromatin is important for its role as a co-activator, whereas enzymatic activity is associated with co-repressor function. These observations are consistent with the promoter and exonskewed loss of G9a following siRNA WIZ knockdown (Fig. 8, B and C) and with prior studies, indicating that G9a catalytic activity is dispensable for co-activator function (49,50,(52)(53)(54).

Enzymatic Inhibition with UNC0638 and WIZ Knockdown
Differentially Affect the Interaction of G9a with Its Proteinbinding Partners-Based on the observation that WIZ-mediated loss of G9a association with chromatin had different consequences for gene expression compared with enzymatic inhibition, we predicted that G9a protein-protein interactions would also be differentially affected. To determine how enzymatic inhibition of G9a compares with WIZ knockdown, FLAG-HA-tagged G9a (9) was tandem affinity-purified from HeLa cells treated with UNC0638 or DMSO (Fig. 11A), or cells from the stable control or WIZ knockdown pools (Fig. 11B), and subjected to mass spectrometry analysis. Significant G9aassociated proteins were determined using the SAINT (44) scoring method (Fig. 12A). Both the known and novel associations with prey protein were identified. Enzymatic inhibition of G9a had a greater effect on protein association compared with WIZ knockdown (Fig. 12B). Differences could be due to the fact that there was only about a 40% reduction in WIZ protein levels in the knockdown cells compared with the control, whereas UNC0638 treatment at 1 M results in Ͼ90% inhibition of catalytic function (30). Alternatively, in the absence of WIZ, G9a may retain catalytic activity toward non-histone substrates.

Discussion
We demonstrated that WIZ is important for G9a interaction with chromatin. Stable knockdown of WIZ reduced H3K9me2 levels and resulted in a very slight reduction in G9a protein levels, which remained stable over time (Fig. 1, G-I). This WIZmediated effect was specific, because reconstitution of exoge- FIGURE 10. Microarray data indicating that enzymatic inhibition and WIZ siRNA knockdown have both overlapping and distinct effects on gene expression. A, heat map of genes with a similar expression pattern relative to control. Red indicates genes that have increased expression compared with control, and green indicates genes that have decreased expression compared with control. B, heat map of differentially expressed genes between HEK293T cells with UNC0638 versus WIZ siRNA knockdown. Gene expression has been median-centered to highlight the magnitude of differential expression. Lane numbers indicate biological replicates. C, comparison of gene expression and G9a occupancy. Loss of G9a occupancy in WIZ siRNA knockdown HEK293T cells correlates with a reduction in gene expression of a subset of genes (listed as Nearest Gene). Gene expression is represented as -fold change in gene expression relative to control from Ϫ2.0-fold (blue) to 2.0-fold (red). G9a occupancy is represented as -fold change in G9a occupancy relative to control from Ϫ2.0-fold (blue) to 2.0-fold (red).
nously expressed WIZ into WIZ knockdown cells increased G9a chromatin-associated protein (Fig. 5, B and C), and genome-wide, as WIZ knockdown resulted in a dramatic loss of G9a occupancy on chromatin (Fig. 6).
The WIZ protein contains C 2 H 2 -type zinc finger motifs that have an unusually widely spaced configuration (26). In Petunia, the EPF family of proteins has two widely separated C 2 H 2 zinc fingers that recognize and bind to specific DNA sequences (55)(56)(57). Drosophila proteins with widely spaced zinc finger motifs include the chromatin-associated suppressor of variegation 3-7 (58 -60) and the homeotic protein Teashirt, which binds sequence-specific gene-regulatory sites in vitro (61) and behaves as a transcriptional repressor when transfected into human cells (62). The murine orthologues of Teashirt also act as transcriptional repressors and are involved in developmental roles similar to fly Teashirt (63,64). Thus, despite an unusual configuration, proteins with widely spaced zinc finger motifs are compatible with DNA binding and transcriptional regulation. This conclusion was further supported by findings pub-lished while our manuscript was being prepared for submission (65).
Co-transcriptional Activity of G9a-Although both loss of WIZ and small molecule inhibition of G9a reduced the H3K9me2 levels, they appeared to do so by distinct mechanisms. UNC0638 binds in the substrate-binding site of G9a and is noncompetitive with the cofactor S-adenosyl-L-methionine (30). Yet, although UNC0638 treatment effectively decreased cellular H3K9me2 levels, it did not generally displace G9a from chromatin. In contrast, silencing the G9a complex member WIZ resulted in the loss of both H3K9me2 and G9a occupancy on chromatin (Figs. 6 and 7).
The power of a combinatorial approach to the study of G9a signaling was highlighted by differential gene expression analysis, which suggested that G9a catalytic activity was associated with the co-repressor function, whereas G9a occupancy on chromatin was more closely related to its function as a co-activator (Fig. 10B). The G9a catalytic domain is sufficient for silencing a target reporter gene (66,67); however, G9a promotes DNA methylation through its ankyrin repeat domain independently of its lysine methylation activity (13, 68 -71). Furthermore, the catalytic activity of G9a is dispensable for transcriptional activation of several genes (49,50,(52)(53)(54), and G9a has a number of non-histone substrates, including itself, which would be affected by enzymatic inhibition but potentially not a knockdown of WIZ (17)(18)(19)(20)(21). Our results confirmed that catalytic activity was dispensable for the regulation of a subset of genes on our microarray (Fig. 10B). A comparison of either UNC0638 treatment or WIZ siRNA knockdown to a G9a siRNA knockdown would not have revealed this stark contrast, because the loss of G9a produced a gene expression pattern that overlapped with both treatment conditions (Fig 10B).
Our ChIP-seq data showed that G9a loss occurred mainly at distal intergenic and intronic regions of the genome (Fig. 8B); however, the signal was slightly skewed toward promoters and coding exons. The sites of G9a loss also overlapped with sites of H3K4me3 and CTCF binding (Fig. 8C). These data were consistent with our gene expression analysis, which associated G9a occupancy on chromatin with its co-activator activity at a subset of genes (Fig. 10B).
Despite the strong association between G9a occupancy and transcriptional activity, we noted that the overall median distance of sites of G9a loss following WIZ siRNA knockdown to the nearest TSS was ϳ30 kb (Fig. 8B). In addition, the majority of sites occupied by G9a were located in distal intergenic regions (Fig. 8A). These findings suggested that G9a co-transcriptional activity in HEK293T cells was exerted mostly at a significant distance from the site of transcription initiation. The H3K9me2 mark is unique compared with some other histone modifications because of its broad regional deposition (72). It has been proposed that G9a target specificity is determined by multiple interactions with chromatin or sequence-specific DNA-binding molecules (6,8). It would be interesting to determine whether G9a binding mediates chromatin looping to allow its interaction with multiple sites in the genome from a localized domain.
G9a Protein-interacting Partners-The loss of H3K9me2 by two distinct mechanisms allowed for comparison of the FIGURE 11. Silver-stained gels showing purified samples submitted for mass spectrometry analysis. Tandem affinity purification of G9a was performed with FLAG followed by HA. HA elution samples were submitted for mass spectrometry analysis. The FLAG elution (FLAG Elute) and HA purification unbound (HA UB) samples are shown for reference. A, purification from HeLa FLAG-HA-G9a cells treated with DMSO or 1 M UNC0638 for 48 h. B, purification from HeLa FLAG-HA-G9a cells with either control shRNA or WIZ shRNA stable knockdown.
We noted that enzymatic inhibition of G9a had more effect on its protein-interacting partners compared with WIZ knockdown (Fig. 12B). These data could be the result of an inefficient WIZ knockdown (WIZ protein levels were reduced by ϳ40%), or they could indicate that G9a displaced from chromatin maintains catalytic activity toward non-histone substrates. In support of the latter interpretation, proteins with the greatest abundance of changes in UNC0638-treated cells compared with controls were non-histone substrates or closely associated with non-histone substrates of G9a. Automethylation of G9a has been shown to be important for its interaction with CBX3 (17, 19, 67). Other non-histone substrate protein interactions that are reduced following treatment include CDYL1, HDAC1, GLP, and WIZ (18,21). Knockdown of WIZ did not result in a change in association of G9a with CDYL1 or CDYL2, again implying that G9a enzymatic activity is important for this interaction. Other associations that were affected by enzymatic inhibition, but not WIZ knockdown, include CBX1 and ZNF644. These data suggest that methylation of non-histone substrates by G9a could be important for maintaining interactions with various protein complexes.
Overall, our findings underscore the potential of combining small molecule probes with gene silencing to dissect molecular pathways. Understanding the mechanistic basis for differences between pharmacologic and genetic manipulations is essential to any translational hypotheses related to G9a function (83)(84)(85).