Wnt3a-dependent and -independent Protein Interaction Networks of Chromatin-bound β-catenin in Mouse Embryonic Stem Cells

Canonical Wnt signaling is repeatedly used during development to control cell fate, and it is often implicated in human cancer. β-catenin, the effector of Wnt signaling, has a dual function in the cell and is involved in both cell adhesion and transcription. Nuclear β-catenin controls transcription through association with transcription factors of the TCF family and the recruitment of epigenetic modifiers. In this study, we used a strategy combining the genetic manipulation of mouse embryonic stem cells with affinity purification and quantitative mass spectroscopy utilizing stable isotope labeling with amino acids in cell culture to study the interactome of chromatin-bound β-catenin with and without Wnt3a stimulation. We uncovered previously unknown interactions of β-catenin with transcription factors and chromatin-modifying complexes. Our proof-of-principle experiments show that β-catenin can recruit the H3K4me2/1 demethylase LSD1 to regulate the expression of the tumor suppressor Lefty1 in mouse embryonic stem cells. The mRNA levels of LSD1 and β-catenin are inversely correlated with the levels of Lefty1 in pancreas and breast tumors, implying that this mechanism is common to mouse embryonic stem cells and cancer cells.

The canonical Wnt signaling pathway plays multiple roles in development and is often dysregulated in human cancers (1).
The key event in the canonical Wnt signaling is the regulation of ␤-catenin. Under normal conditions, the levels of ␤-catenin are low as a result of its phosphorylation by the destruction complex and subsequent ubiquitination and degradation by the proteasome (1). The destruction complex contains the scaffold proteins APC and Axin1, protein phosphatase 2a, and the kinases glycogen synthase kinase (GSK)-3␤ and casein kinase I alpha. Upon binding of Wnt ligands to the receptors Frizzled and LRP5/6, the destruction complex is turned off via translocation to the plasma membrane, where it binds Dishevelled. Thus, ␤-catenin is stabilized, translocates to the nucleus, and associates with TCF factors and with various proteins that are implicated in chromatin structure and RNA polymerase II regulation (1,2).
The interactions of ␤-catenin with chromatin effector proteins are concentrated at the C-terminal domain of ␤-catenin, as shown for the histone acetyltransferases CBP and p300, the histone methyltransferase MLL, the nucleosome repositioning SWI/SNF family member protein Brg1, the Mediator component MED12, and the BCL-9/Pygo complex, which has the ability to bind methylated H3K4 (2). It is not probable that all these bulky multiprotein assemblies simultaneously occupy the C-terminal domain of ␤-catenin. Other transcription factors use the same repertoire of co-activators via sequential or cycling recruitment (3). It is likely that ␤-catenin serves as a platform which orchestrates the sequential recruitment of chromatin remodeling factors in a stimulus-dependent manner.
In mES 1 cells, the epigenetic regulation of gene expression has been shown to take place at the levels of DNA methylation, histone modification, nucleosome packaging and rearrangement, and higher order chromatin organization (4). Unique to key developmental genes in mES cells are bivalent chromatin domains with both activating and repressing marks, which contribute to the pluripotency potential of mES cells (5). Although Wnt/␤-catenin signaling has been shown to play a role in the self-renewal and pluripotency of mES cells (6), the nuclear interaction partners of ␤-catenin in mES cells have not been extensively studied. Because mES cells have low intrinsic Wnt signaling activity, which can be further stimulated by the addition of exogenous Wnt3a (7), they are an excellent model for studying the dynamics of ␤-catenin complex formation upon Wnt3a stimulation.
In this study we used affinity purification and mass spectroscopy combined with stable isotope labeling with amino acids in cell culture (SILAC) to identify the Wnt3a-dependent and -independent chromatin interactomes of ␤-catenin in mES. We show the dynamics of complex formation upon Wnt3a treatment and identify among the ␤-catenin interaction partners both repressing and activating chromatin effector proteins. We found that ␤-catenin can associate with the H3K4me1/2 histone demethylase LSD1 to repress transcription. LSD1 is recruited to the proximal enhancer of Lefty1 in a ␤-catenin-dependent manner, and Lefty1 is up-regulated in ␤-catenin Ϫ/Ϫ cells, which correlates with increased H3K4me2/3 levels at the Lefty1 proximal enhancer element.

EXPERIMENTAL PROCEDURES
Cell Culture and Labeling-SR1-Cre-ERT2 cells were routinely cultured on feeder cells in mES cell medium (Dulbecco's modified Eagle's medium supplemented with 15% fetal calf serum, 1:1000 inhouse produced LIF, 1 mM penicillin/streptomycin, 2 mM L-glutamine, 1 mM sodium pyruvate, 1x non-essential amino acids, and 0.1 mM 2-mercaptoethanol) at 37°C in a humidified atmosphere and 8% CO 2 . For labeling, the cells were cultured in ES cell medium on gelatinized plates without feeders. One cell population was supplemented with normal isotope containing L-lysine and L-arginine (Sigma-Aldrich Corporation, St. Louis, MO), and another with heavy isotope labeled 13 C 8 -lysine and 13 C 6 15 N 4 -arginine (Sigma-Aldrich Corporation, St. Louis, MO), generating mass shifts of ϩ8 and ϩ10 Da, respectively. mES cells were grown for 6 days in labeling medium before further treatment.
Cell Treatment-Wnt3a (R&D Systems, Minneapolis, MN) was used at a concentration of 150 ng/ml. The GSK3 inhibitor SB-216763 (Sigma-Aldrich Corporation, St. Louis, MO) was used at a concentration of 10 M. LiCl was used at a concentration of 40 mM. (Z)-4hydroxytamoxifen (Sigma-Aldrich Corporation, St. Louis, MO) was used at a concentration of 100 ng/ml. (Z)-4-hydroxytamoxifen was added to the medium on day 1, and 24 h later the cell medium was replaced with medium without (Z)-4-hydroxytamoxifen. 72 h after treatment, the cells were harvested.
Protein Extract Preparation-The cells were harvested by being scraped off the plates in Dulbecco's PBS without Ca and Mg (PAA Laboratories, Pasching, Austria) and lysed in Buffer A (1 mM Hepes pH 7.6, 0.5% Nonidet P-40, 0.5 mM MgCl 2 , 1 mM DTT with protease inhibitors (EDTA-free protease inhibitor mixture, Roche, Mannheim, Germany), and phosphatase inhibitors (50 mM sodium fluoride, 1 mM sodium orthovanadate)). The cells were incubated for 20 min at 4°C and then vortexed for 30 s. The chromatin was collected via 10 min of centrifugation at 2500 ϫ g, and the supernatant was transferred to a new tube (soluble fraction). The buffer was adjusted to 10 mM Hepes pH 7.6, 100 mM NaCl, 1 mM MgCl 2 . The chromatin fraction was washed twice with Buffer B (10 mM Hepes pH 7.6, 1 mM MgCl 2 , 100 mM NaCl, 1 mM DTT with protease inhibitors (EDTA-free protease inhibitor mixture, Roche, Mannheim, Germany), and phosphatase inhibitors (50 mM sodium fluoride, 1 mM sodium orthovanadate)), and the DNA and RNA were digested via incubation in Buffer B with 1 kU benzonase (Sigma-Aldrich Corporation, St. Louis, MO) for 15 min at 25°C (the chromatin-bound fraction). The soluble fraction and the chromatin-bound fraction were centrifuged for 15 min at 20,000 ϫ g to remove residual cell debris and membranes.
Co-immunoprecipitation-The protein concentration of each extract was determined via Bradford assay (Thermo Fisher Scientific, Rochester, NY). For each reaction, 1.5 mg of total protein extract and 5 g of anti-␤-catenin antibody (Cell Signaling Technology, Danvers, MA) were used. The protein extract was incubated with the antibody for 2 h at 4°C with rotation. The antibody-bound material was collected via the addition of 30 l pre-washed magnetic Dynabeads ProteinG (Invitrogen, Carlsbad, CA). The beads were washed three times with 1 ml of Buffer B. Beads from parallel co-immunoprecipitations using light and heavy labeled extracts were mixed (8), and bound proteins were eluted in LDS sample buffer (Invitrogen, Carlsbad, CA).
LC-MS/MS-The fractionated cell extracts or the eluted proteins from the co-immunoprecipitations were separated on 4%-12% gradient SDS-PAGE gels (Invitrogen, Carlsbad, CA) and stained with colloidal Coomassie Blue. Each gel lane was cut into 12 equal gel slices, and the proteins from each gel slice were in-gel digested with trypsin (Promega, Madison, WI) as described elsewhere (9). Tryptic peptides from each gel slice were analyzed via nanoflow HPLC (Agilent, Boeblingen, Germany) coupled to a nanoelectrospray LTQ-Orbitrap XL mass spectrometer (Thermo Fisher Scientific, Waltham, MA) operating in positive ion mode. Peptides were desalted and subsequently separated on an analytical column (15 cm length and 75 m inner diameter, packed into a picotip emitter with an 8 m tip opening) at flow rate of 250 nl/min and a linear gradient of 2%-60% buffer B (80% acetonitrile and 0.5% acetic acid) for 50 min. Data-dependent acquisition of eluting peptides was applied and consisted of one survey scan in the Orbitrap with the resolution set at 60,000 at m/z 400 and the automatic gain control target at 10 6 , followed by MS/MS of the five most intense precursors in the LTQ using collision-induced decay fragmentation with previously fragmented ions dynamically excluded for 90 s. The general mass spectrometric conditions were as follows: spray voltage, 2.3 kV; no sheath and auxiliary gas flow; ion transfer tube temperature, 160°C; collision gas pressure, 1.3 mTorr; normalized collision energy using wide-band activation mode; 35% for MS 2 . Ion selection thresholds were 500 or 1000 counts for MS 2 , depending on the experiment. An activation q ϭ 0.25 and activation time of 30 ms were applied.
Data Analysis-The raw files from the LTQ-Orbitrap XL were analyzed using MaxQuant software (version 1.0.13.13) (10). Peak lists generated by the first module of MaxQuant (Quant.exe) were queried by the Mascot search engine (Mascot Daemon 2.2, Matrix Science, London, UK) against the mouse International Protein Index database (version 3.87) supplemented with a list of frequent contaminants and concatenated with the reverse sequences of all entries. The total number of sequences in the database was 110,425. Enzyme specificity was set to trypsin with up to three missed cleavages and allowing cleavages N-terminal to proline and C-terminal to aspartate. Modifications were cysteine carbamidomethylation (fixed), protein N-terminal acetylation, asparagine and glutamine deamidation, and methionine oxidation (variable). The maximum mass deviations for monoisotopic precursor ions and MS/MS peaks were restricted to 5 ppm and 0.5 Da, respectively. The peptides were filtered for a maximum false discovery rate of 1% in MaxQuant. Only unique and razor peptides with a posterior error probability of less than 0.05 and proteins with a ratio count of at least three were accepted and used for quantification (10). All proteins are listed in supplemental Tables S1-S4 with accession numbers, numbers of unique peptides, percent sequence coverage, quantification significance, and variability as recorded by MaxQuant.

␤-catenin Chromatin Interactome in Mouse Embryonic Stem Cells
Gene Ontology Analysis-Gene Ontology (GO) term enrichment analysis was performed with AmiGO version 1.8 (11) using the GO database (released September 29, 2012). The database filter was set to MGI, and the default settings were used. The enriched GO terms were plotted with REViGO (12) with the "allowed similarity" set to large.
Graphical Representation of Proteomics Data-For the graphical representation, R version 2.14.1 (13) was used. The scripts are available in the supplemental "Experimental Procedures" section.
Chromatin Immunoprecipitation-For ChIP, five confluent 15 cm dishes were used per experiment. The cells were fixed with 1% formaldehyde for 15 min at room temperature. After three washes with PBS, the cells were harvested with silicon scrapers and put on ice. Cells were lysed in lysis buffer (50 mM Tris pH 8.1, 150 mM NaCl, 10 mM EDTA, 1% SDS) and sonicated to shear the crosslinked DNA. For sonication we used the Covaris S2 system using the following parameters: 12 cycles, 15% duty cycle, intensity ϭ 10, 500 cycle burst, 60 s duration, and frequency sweeping mode. 5 g of each antibody were incubated overnight with cell extract containing 250 g of DNA. Subsequently, Dynal Protein G magnetic beads were added for 1 h. The beads were washed four times for 4 min with low-salt buffer (20 mM Tris pH 8.1, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS) and once with high-salt buffer (20 mM Tris pH 8.1, 500 mM NaCl, 2 mM EDTA, 1% Triton X-100, 0.1% SDS). The immune complexes were eluted in 100 l of elution buffer (1% SDS, 100 mM NaHCO 3 ) with RNase A and protease K to 500 g/ml at 37°C for 1 h. Crosslinking was reversed by overnight incubation at 65°C. Whole cell extract DNA (reserved from the sonication step) was also subjected to crosslink reversal. The following antibodies were used: anti-LSD1 raised in rabbit (Abcam, Cambridge, UK), anti-␤-catenin raised in mouse (1:5, in-house produced), anti-H3K4me2/3 raised in mouse (Abcam, Cambridge, UK), rabbit-IgG (Diagenode, Liè ge, Belgium), mouse-IgG (Santa Cruz Biotechnology, Santa Cruz, CA), and rabbit-IgG (Santa Cruz Biotechnology, Santa Cruz, CA). The qPCR analysis was performed as described elsewhere (14). The primers encompassing the Lefty1 enhancer have been described elsewhere (14) and are 5Ј-GTAGCCAGCAGACAGGACAA-3Ј and 5Ј-ATC-CCCAATCCACATTCACT-3Ј. All data were first normalized to the percentage of input, including the IgG control. Because the IgG control represents the noise, the noise/input was subtracted from the signal/input. The average of two biological experiments was plotted with error bars showing the standard error of the mean.
Real-time PCR-Total RNA was isolated using an RNeasy Mini Kit (Qiagen, Hilden, Germany). cDNA was prepared using a RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher Scientific, Rochester, NY). qPCR was performed using Absolute SYBR Green ROX mix (Thermo Fisher Scientific, Rochester, NY). The primer sequences for Lefty1 were 5Ј-ACTCAGTATGTGGCCCTGCTA-3Ј and 5Ј-AACCT-GCCTGCCACCTCT-3Ј. The primer sequences for Tbp were 5Ј-CG-GTCGCGTCATTTTCTC-3Ј and 5Ј-GGGTTATCTTCACACACCATGA-3Ј. The results represent the average of three experiments performed in triplicate, and the error bars show the standard error from the mean.
In Silico Binding Protein Analysis-Known and predicted protein associations were analyzed and visualized with STRING version 9.0 (15). The Active Prediction Methods used were "Experiments," "Databases," and "Textmining," with "Required Confidence" set to "low confidence (0.150)." Statistical Analysis-The statistical analysis was done in Excel. The p values for the data represented in Fig. 6 were determined via one-tailed Student's t test. Pearson correlation was done in Excel. In the pancreas cancer dataset (PubMed ID 19260470), the relative fold change between the tumor and the normal tissue was used. In premalignant breast cancer samples, the measured values were taken.

RESULTS
Experimental Setup-In order to study the interactome of ␤-catenin via co-immunoprecipitation from metabolically labeled cells using SILAC, we needed a cell system in which we could identify the ␤-catenin binding proteins before large changes in the proteome had taken place. For that, we took advantage of the SR1-Cre-ERT2 system. 2 It has been previously shown that ␤-catenin flox/Ϫ mES cells are phenotypically wild type (16). SR1-Cre-ERT2 cells are ␤-catenin flox/Ϫ cells that have been stably transfected with a vector carrying tamoxifen-inducible Cre recombinase. The addition of tamoxifen to the medium causes the removal of the ␤-catenin sequence between the two loxP sites via recombination, thereby generating ␤-catenin Ϫ/Ϫ mES cells (Fig. 1A) (17). Three days after the induction of recombination, ␤-catenin was not detectable via Western blot (Fig. 1B).
In order to study the interactome of chromatin-bound ␤-catenin, we set out to establish a protocol that would allow us to separate the chromatin-bound from the soluble ␤-catenin. As a criterion for successful separation, we used a Western blot for E-cadherin, which binds ␤-catenin at the cell membrane (18). The separation of the cytoplasmic and the nuclear ␤-catenin with commercially available kits was not appropriate, because we could detect E-cadherin in the nuclear fraction (Fig. 1C). To solve this problem, we developed a one-step protocol wherein the Nonidet P-40 detergent concentration was essential. After fractionation, E-cadherin was detected in the soluble, but not in the chromatin, fraction (Fig.  1D). The transcription factors OCT-3/4 and SOX2 were found exclusively in the chromatin fraction (19 -21), along with Lamin B, a major component of the nuclear lamina (22) (Fig. 1D). Thus, our fractionation procedure could successfully separate the chromatin-bound ␤-catenin from the soluble ␤-catenin.
To prepare the labeled protein extracts, we grew SR1-Cre-ERT2 cells under SILAC conditions. We used medium containing normal isotope L-arginine and L-lysine for preparing the "light" extracts and medium containing heavy isotope labeled 13 C 6 15 N 2 -lysine (Lysϩ8) and 13 C 6 15 N 4 -arginine (Argϩ10) for preparing the "heavy" extracts. Both the soluble and the chromatin fractions from the tamoxifen-and mocktreated cells were tested to ensure the high similarity of the heavy and light material (i.e. the extracts contained mostly the same proteins in the same amounts, as judged by their ratios).
In the forward experiments, the mock-treated cells were incubated with the heavy medium and the tamoxifen-treated cells were incubated with the light medium. In the reverse experiments, the labeling was swapped. The extracts were analyzed at 3 days after the induction of recombination (supplemental Figs. S1A and S1B, supplemental Tables S1 and S2). As expected, ␤-catenin was efficiently down-regulated, as judged by the heavy-to-light normalized ratios of 4.9 in the chromatin fraction (supplemental Table S1) and 12.6 in the soluble fraction (supplemental Table S2). In the soluble fraction we identified 3643 proteins, of which 3533 proteins showed quantitative ratios (supplemental Table S2). Accordingly, from 2089 identified proteins in the chromatin fraction, we obtained quantitative ratios for 2026 proteins (supplemental Table S1). In both fractions, the quantitative ratios were grouped around 1:1, showing similar proteomes for the ␤-catenin flox/Ϫ and ␤-catenin Ϫ/Ϫ cells (supplemental Figs. S1A and S1B).
Wnt3a-independent Interactome of ␤-catenin in the Chromatin of mES Cells-To study the interactome of ␤-catenin in mES cells, we prepared chromatin extracts from SILAC labeled SR1-Cre-ERT2 cells that had been either mock-or tamoxifen-treated for 3 days. Two biological replicas, a forward and a reverse experiment, were performed. Equal amounts of heavy or light protein extracts were incubated with anti-␤-catenin antibodies immobilized on magnetic beads (Fig. 1E). The beads were mixed before the last wash. The proteins that were retained by the antibodies were recovered and run on SDS-PAGE gels, and the separated proteins were trypsinized. The tryptic peptides were analyzed via electrospray ionization mass spectroscopy, and a heavy-to-light ratio for each SILAC pair was determined (Fig. 1E).
We identified 1034 proteins in the two biological replicas, of which 962 had quantitative ratios ( Fig. 2A) Table S3). ␤-catenin was specifically enriched in both forward and reverse experiments, with normalized ratios of 8.8 and 4.6, respectively (supplemental Table S3). To determine an empirical threshold for the specific enrichment of interaction partners in our experiment, we searched the list of identified SILAC pairs for known ␤-catenin interacting proteins. We found Pontin (RUVBL1) and Reptin (RUVBL2), two helicases that have been previously identified as nuclear ␤-catenin interaction partners (23,24). They were enriched in both experiments, with normalized ratios of 2.0 and 1.6 for Pontin and 1.9 and 2.0 for Reptin in the forward and the reverse experiment, respectively (supplemental Table S3). Thus, we set as an enrichment threshold a normalized ratio of more than 1.5-fold in both the forward and the reverse experiments.

FIG. 1. Experimental setup for the affinity purification of chromatin-bound ␤-catenin interacting partners.
A, scheme of the ␤-catenin flox allele. Upon treatment with tamoxifen, the genomic sequence from exon two to exon six is recombined to delete the remaining ␤-catenin. B, Western blot of total cell extracts from SR1-Cre-ERT2 after 3 days of mock or tamoxifen treatment. Treatment of these cells with tamoxifen leads to a loss of ␤-catenin. C, D, optimization of the biochemical procedure for the separation of the chromatin-bound ␤-catenin from the cytoplasmic ␤-catenin. C, Western blot with anti-E-cadherin and anti-␤-catenin antibodies after a standard fractionation shows that the nuclear/chromatin fraction was still contaminated with membrane proteins. D, Western blot with antibodies against membrane (E-cadherin) and chromatin (OCT4, SOX2, LaminB) proteins. An optimized protocol yields a clear separation between the membrane/cytoplasmic and chromatin fractions of ␤-catenin. E, schematic representation of the affinity purification and of the protein identification procedures.

␤-catenin Chromatin Interactome in Mouse Embryonic Stem Cells
When these criteria were applied to our dataset, 76 proteins were found to bind ␤-catenin specifically in the chromatin of mES cells (Wnt3a-independent interactome) ( Table I). 20% of these proteins (16 of 76) have been previously shown to associate directly or indirectly with ␤-catenin (Table I). Thus, 80% of the Wnt3a-independent interactome comprises novel interaction partners of chromatin-bound ␤-catenin. A STRING (15) analysis of the Wnt3a-independent ␤-catenin interactome revealed a complex network of known or predicted functional interactions among the enriched proteins (Fig. 2B).
Wnt3a-dependent ␤-catenin Interactome in the Chromatin of mES Cells-In order to be able to directly compare the interactomes of ␤-catenin before and after Wnt3a treatment, we decided to study the same SR1-Cre-ERT2 cell line after Wnt3a stimulation. ␤-catenin flox/Ϫ cells respond to Wnt3a stimulation with the up-regulation of the canonical Wnt signaling targets Axin2, Cdx1, and brachyury (T) within 4 h after Wnt3a supplementation (25), and SR1-Cre-ERT2 are similarly Wnt3a responsive. 2 Interestingly, we could not observe a significant increase in the protein levels of chromatin-bound ␤-catenin in mES cells after Wnt3a, LiCl, or GSK3a inhibitor treatments (supplemental Fig. S2A). To study the interactome of ␤-catenin in mES cells after Wnt3a stimulation, we immunoprecipitated ␤-catenin from the chromatin of SILAC labeled SR1-Cre-ERT2 cells that had been mock or Wnt3a treated for 4 h. In the two biological replicas, the forward and the reverse experiments, we identified 1132 proteins in total, of which 1043 had quantitative ratios (Fig. 3A) (supplemental Table S4). The normalized ratios for ␤-catenin were 1.0 and 1.3, respectively. As a cutoff for enrichment, we used, as before, a normalized ratio of 1.5 in both the forward and the reverse experiments. We identified 69 proteins that matched our criteria (Wnt-3a-dependent interactome) (Fig. 3B) (Table II). Only 10% of those had been previously shown to associate with ␤-catenin (Table II), whereas 90% represented novel interaction partners. Interesting is the comparison of the ␤-catenin Wnt3a-independent and Wnt3a-dependent interactomes. 64 proteins of the Wnt3a-dependent interactome were found  (15). Only the proteins with known and predicted interactions are shown in the schematic drawing. The color code represents k-means clustering of the interaction likelihood among the different proteins. The thicker non-dashed lines between proteins represent high-confidence interactions, whereas thinner dashed lines between proteins represent low-confidence interactions.

. Wnt3a-dependent interactome of the chromatin-bound ␤-catenin in mES cells 4 h after Wnt3a stimulation.
A (upper panels), scatter plot of normalized heavy-to-light ratios of identified proteins and total peptide intensities. Proteins with quantitative ratios in both the forward and the reverse experiment, or in either of them, are represented on the graph. A (lower panels), scatter plot of normalized ratios of proteins with quantitative ratios in both the forward and the reverse experiment. Proteins with fold change ratios above 1.5 in both experiments are colored red. B, ␤-catenin interacting proteins of the Wnt-dependent proteome with fold change ratios above 1.5 in the forward and the reverse experiments were analyzed using the STRING protein interaction database (15). Only the proteins with known and predicted interactions are shown in the schematic drawing. The color code represents k-means clustering of the interaction likelihood among the different proteins. The thicker non-dashed lines between proteins represent high-confidence interactions, whereas thinner dashed lines between proteins represent low-confidence interactions. c Proteins that were found with ratios of more than 1.5 in the biological replicas in both the Wnt3a-independent and the Wnt3a-dependent proteomes.
d Proteins found to bind ␤-catenin with ratios of less than 1.5 after Wnt3a stimulation.  Fig. S2B). These proteins represent a subset of binding partners that interact with ␤-catenin in the chromatin of mES cells, and this interaction is potentiated upon exogenous activation of canonical Wnt signaling. GO Term Enrichment Analysis-To determine in which processes the identified ␤-catenin interaction partners are involved, we performed molecular function, biological process, and cellular component GO term enrichment analysis of the ␤-catenin Wnt3a-independent and -dependent interactomes (supplemental Fig. S3, supplemental Tables S5 and S6). In both interactomes there were gene functions associated with protein binding (GO:000551), nucleic acid binding (GO: 0003676), and small molecule binding (GO:0036094) (supplemental Figs. S3B and S3D), as well as various GO functions associated with metabolism and gene expression (supplemental Figs. S3A and S3C). Interestingly, however, in the Wnt3a-independent proteome there was overrepresentation of gene functions associated with DNA binding (GO:0003677) and nucleoside binding (GO:0001882), whereas in the Wnt3adependent proteome there was enrichment of gene functions associated with RNA binding (GO:0003723), translation factor activity, nucleic acid binding (GO:0008135), and translational elongation factor activity (GO:0003746) (supplemental Figs. S3B and S3D). Furthermore, in the Wnt3a-independent proteome there were clusters of gene functions associated with chromosome organization (GO:0051276), chromatin remodeling complex (GO:0016585), and chromatin organization (GO:0006325), whereas in the Wnt3a-dependent proteome there was enrichment in gene functions associated with RNA processing (GO:0006396), nuclear export (GO:0051168), transport (GO:0006810), and nucleocytoplasmic transport (GO:0006913) (supplemental Figs. S3A and S3C). These find-ings imply overlapping, but also functions of ␤-catenin in mES cells with or without Wnt3a stimulation.
Screen Validation-To assess the quality of the screen, we decided to test some of the identified proteins from both the Wnt3a-dependent and the Wnt3a-independent ␤-catenin interactomes that had not been previously characterized for their ability to bind ␤-catenin in different biochemical and cellular assays. We used specific anti-␤-catenin antibodies to immunoprecipitate the endogenous ␤-catenin from W4 wildtype mES, and we used Western blot to test for the copurification of SALL4 and LSD1, two proteins that were found in the Wnt3a-independent ␤-catenin interactome. SALL4 is a transcription factor essential for the self-renewal of mES cells, and it exerts its function by directly binding two other key transcription factors in mES cells, Nanog and Oct4 (26 -30). LSD1 is a histone demethylase that participates both in silencing complexes by catalyzing the demethylation of both diand monomethylated lysine 4 (H3K4me2/1) and in activating complexes, where it demethylates lysine 9 of histone H3 (H3K9me2/1) (31,32). mES cells deficient for LSD1 show a precautious expression of developmental markers and a loss in global levels of DNA methylation (33,34).
To confirm that the LSD1/␤-catenin and SALL4/␤-catenin interactions indeed take place within the chromatin of wildtype mES cells, we used a cell-based proximity ligation assay (35,36). This method is suitable for visualizing protein interactions in situ and provides information about the localization of protein complexes. It relies on the use of protein-specific primary antibodies and secondary antibodies with attached DNA (cytosine-5-)-methyltransferase activity C-5 cytosine-specific DNA methylase IPI00317740: Gnb2l1 WD domain, G-beta repeat IPI00170008: Snrpa1 mRNA processing IPI00134621: Ran b,c Nucleocytoplasmic transport Ras family IPI00130280: Atp5a1 Purine nucleotide biosynthetic process ATP synthase alpha/beta family IPI00474430: Sfrs2 RNA splicing RNA recognition motif IPI00307837: Eef1a1 Translation factor activity Elongation factor Tu GTP binding domain a According to Gene Ontology. b Proteins previously shown to associate directly or indirectly with ␤-catenin according to literature search. c Proteins that were found with ratios of more than 1.5 in the biological replicas in both the Wnt3a-independent and Wnt3a-dependent proteomes.

␤-catenin Chromatin Interactome in Mouse Embryonic Stem Cells
DNA strands that participate in ligation, replication, and sequence decoding reactions (35,37). As a control, we used 16.1 cells, which are ␤-catenin deficient (38). By using specific anti-␤-catenin and anti-SALL4 antibodies, we detected the SALL4/␤-catenin interaction in the chromatin of wild-type W4 mES cells, but not in the chromatin of control 16.1 mES cells (Fig. 5A). Similarly, we were able to detect the interaction of ␤-catenin and LSD1 in the nuclei of wild-type W4 mES cells, but not in the nuclei of control 16.1 mES cells (Fig. 5B). Thus, these results validated the correct subcellular localization of the novel ␤-catenin complexes that we uncovered with our SILAC-based screen.
From the Wnt3a-dependent interactome, we set out to validate the interaction of ␤-catenin with Hmg2. This protein has already been shown to form functional complexes with ␤-catenin in the superficial zone of articular cartilage that are essential for the maintenance of tissue homeostasis (39). This interaction is indirect via Lef-1 and is controlled by the canonical Wnt signaling (39). Interestingly, however, in our mES cell model, ␤-catenin was found to bind Hmg2 in the Wnt3aindependent interactome, but the complex formation was further enhanced by the addition of exogenous Wnt3a (Tables  I and II). We used the proximity ligation assay with anti-␤catenin and anti-Hmg2 antibodies and compared the complex formation in W4 wild-type mES cells that were either treated with Wnt3a for 4 h or mock-treated. We found that the ␤-catenin-Hmg2 complexes were readily observed in mocktreated W4 cells, but the signal was increased after treatment with Wnt3a (Fig. 5C). Hmg2 is a nonspecific DNA-binding and -bending protein that facilitates the formation of complex nucleoprotein complexes involved in the regulation of gene expression (40 -43). Therefore, it is possible that the enhancement of the ␤-catenin-Hmg2 interaction upon Wnt3a treatment reflects the active ␤-catenin-dependent restructuring of the chromatin.
Functional Characterization of the ␤-catenin-LSD1 Interaction-To validate the functional relevance of our screen, we decided to study the significance of the ␤-catenin-LSD1 interaction in the regulation of gene expression. It was recently shown that LSD1 binds to a proximal enhancer of Lefty1 and regulates Lefty1 expression in mES cells by controlling the methylation status of H3K4 (14,44). Lefty1 encodes a transforming growth factor beta (TGF␤) superfamily protein (45) that antagonizes the functions of Activin and Nodal and is essential in the establishment of left-right asymmetry of the embryo (46,47). It has been suggested that the function of Lefty1 in stem cells is to protect them from the action of developmental inducers (48). We found that Lefty1 is one of the early genes that become up-regulated in SR1-Cre-ERT2 mES cells after the addition of tamoxifen (Fig. 6D). This suggests that Lefty1 is a direct target of ␤-catenin and that ␤-catenin participates in a repressive complex that controls Lefty1 expression, possibly with LSD1. We hypothesized that ␤-catenin and LSD1 are recruited to the very same enhancer. To test this hypothesis, we carried out chromatin immunoprecipitations with anti-LSD1 and anti-␤-catenin antibodies in SR1-Cre-ERT2 cells mock-treated or treated with tamoxifen, and we amplified the genomic sequences corresponding to the previously characterized enhancer of Lefty1 (14). Interestingly, we found that both ␤-catenin and LSD1 were bound to the Lefty1 enhancer in SR1-Cre-ERT2 cells, but this binding was greatly diminished upon the removal of functional ␤-catenin (Figs. 6A, 6B). This suggests that the formation of the repressive complex on the Lefty1 enhancer is dependent on the presence of functional ␤-catenin. To show that the formation of the ␤-catenin-LSD1 complex is functionally relevant, we tested the levels of H3K4me2/3 at the Lefty1 enhancer. We found that the H3K4me2/3 levels at the Lefty1 enhancer, which mark active genes, increased upon the removal of functional ␤-catenin (Fig. 6C). Our findings suggest that a repressive complex containing ␤-catenin and LSD1 controls the levels of Lefty1 in mES cells. The recruitment of LSD1 to the Lefty1 enhancer, and thereby the demethylation of H3K4me2/3, is dependent on the presence of ␤-catenin at this enhancer.

␤-catenin and LSD1 Expression Levels Are Inversely Correlated with Lefty1 Expression Levels in Pancreas Cancer Cells and in the Normal Epithelium of Breast Cancer Patients-
Lefty1 has been recently identified as a tumor suppressor in human pancreas cancer cells (49). We used publically available data (PubMed ID 19260470) to compare the mRNA expression levels of ␤-catenin and LSD1 with those of Lefty1 in pancreas cancer tumors (50). Interestingly, the mRNA expression levels of ␤-catenin and LSD1 were inversely correlated with those of Lefty1 (r ϭ Ϫ0.39 and r ϭ Ϫ0.49, respectively), which is in agreement with our data from the mES cell model (Figs. 7A, 7C). Importantly, this correlation is not unique to pancreas cancer and can be observed in pre-malignant tissues. We compared the mRNA expression levels of ␤-catenin and LSD1 with those of Lefty1 first in histologically normal breast epithelium of non-cancer controls, and second in the

␤-catenin Chromatin Interactome in Mouse Embryonic Stem Cells
Lefty1 (r ϭ Ϫ0.44), as well as between the expression levels of LSD1 and those of Lefty1 (r ϭ Ϫ0.55) (Figs. 7B, 7D). Taken together with our data, this implies that the down-regulation of the tumor suppressor Lefty1 by ␤-catenin and LSD1 might be an early event in malignant transformation. DISCUSSION We used affinity purifications and SILAC-MS to study the interactome of chromatin-bound ␤-catenin in mES cells under standard culturing conditions and after treatment with Wnt3a. We identified 76 proteins from the Wnt3a-independent inter-actome and 69 proteins from the Wnt3a-dependent interactome of ␤-catenin. We showed that ␤-catenin uses the demethylase activity of one of its newly identified interaction partners, LSD1, to regulate the expression of Lefty1 in mES cells. Both LSD1 and ␤-catenin bind to the proximal enhancer of Lefty1, and the recruitment of LSD1 to the Lefty1 enhancer and the H3K4me2/3 levels at this enhancer are dependent on ␤-catenin. LSD1 and ␤-catenin mRNA levels are inversely correlated with the levels of Lefty1 in human tumors.
Our work identified an extensive network of protein interactions of ␤-catenin in mES cells. 80% of the newly identified

␤-catenin Chromatin Interactome in Mouse Embryonic Stem Cells
proteins in the Wnt3a-independent proteome and 90% of those in the Wnt3a-dependent proteome are novel interaction partners. Interestingly, most of the proteins that were identified in the Wnt3a-independent interactome of ␤-catenin showed a normalized ratio of less than 1.5 in the Wnt3a-dependent interactome. This implies that these interactions are rather stable and are not influenced by additional Wnt3a stimulation. Seven proteins of the Wnt3a-independent interactome were also specifically enriched in the Wnt3a-dependent interactome. Their interaction with ␤-catenin appeared to get stronger with increasing levels of ␤-catenin. One explanation for that would be that these proteins participate in the formation of new ␤-catenin complexes. For example, Hmg2 is a nonspecific DNA-binding and DNA-bending protein (43), and the interaction with ␤-catenin could be one of the early events in the formation of ␤-catenin-containing complexes.
Recently, it has been shown that ␤-catenin can be recruited to target genes in mES cells not only by the canonical Tcf transcription factors (52), but also by OCT-3/4 (53, 54) and KLF4 (38). Interestingly, we found that in the Wnt3a-independent interactome, ␤-catenin associates with SALL4, another transcription factor that is part of the core pluripotency network (27,28,30). One implication of that fact would be that some yet-to-be-identified genes might be regulated through the ␤-catenin-SALL4 interaction and, probably, through associated chromatin-modifying activity. One candidate for such activity would be the Mi2-nucleosome remodeling and deacetylase (NuRD) complex. SALL4 was shown to repress the expression of the stem cell factors SALL1 and PTEN through the recruitment of the Mi2-NuRD complex in human embryonic stem cells (55). The Mi2-NuRD complex utilizes the histone deacetylase activity of Hdac1 or Hdac2 to silence its targets, which in mES cells include the pluripotency genes Zfp42, Tbx3, Klf4, and Klf5 (55). Indeed, we found two components of the Mi2/NuRD complex in our screen: Hdac1 and Gatad2a (56). Gatad2a (p66) has already been genetically identified as a component of canonical Wnt signaling in the fly (57), implicating the Mi2-NuRD complex in the repression of Wnt target genes. Our studies suggest that in mES cells, Wnt signaling and the Mi2-NuRD complex meet at the level of ␤-catenin. Further studies are required in order to test whether SALL4 -Mi2-NuRD-␤-catenin complexes can form on target promoters. To identify putative target genes, it would be essential to compare the genome-wide binding sites of SALL4 in mES cells with those of ␤-catenin and of Mi2-NuRD.
Besides its histone deacetylase activity, the Mi2-NuRD complex has been shown recently to associate with the H3K4me2/1 demethylase LSD1 in cancer cells (58). In embryonic stem cells, LSD1 co-occupies nine of every ten enhancers with the NuRD complex, suggesting a significant functional overlap (14). Interestingly, we identified LSD1 as a novel interaction partner of ␤-catenin in the Wnt3a-independent interactome. In human breast cancer cells, the Mi2-NuRD-LSD1 complex suppresses the expression of members of the TGF␤1 signaling pathway (58). At least one member of the TGF␤ signaling pathway, Lefty1, has been shown to be a direct transcriptional target of LSD1 in embryonic stem cells (14). Lefty1 is a nodal antagonist that has recently been shown to act as a tumor suppressor in pancreatic cancer (49). We found that Lefty1 is up-regulated in ␤-catenin Ϫ/Ϫ mES cells and that ␤-catenin is recruited to the proximal enhancer of Lefty1. The active enhancers in embryonic stem cells, including the Lefty1 enhancer, are marked by H3K4me1 and H3K4me2, but not by H3K4me3 (14,59,60). By using an anti-H3K4me2/3 antibody, we showed that the demethylase activity of LSD1 at the Lefty1 enhancer is dependent on ␤-catenin, implying that ␤-catenin recruits LSD1 to the Lefty1 enhancer. Recently, it was also found that SALL4 binds the proximal enhancer of Lefty1 (14). It is tempting to speculate that, at this particular enhancer, a tertiary complex of SALL4 -␤-catenin-LSD1 might form. The regulation of Lefty1 expression by LSD1 and ␤-catenin might not be restricted to the mES cell model. Indeed, we found that the levels of Lefty1 are inversely correlated with those of ␤-catenin and LSD1 in pancreatic cancer and in pre-malignant breast cancer epithelium. Thus, the mechanism of repression of the tumor suppressor Lefty1 by ␤-catenin and LSD1 might be common to mES cells and cancer cells, and this mechanism could provide novel therapeutic targets for human disease.