Subunit redundancy within the NuRD complex ensures fidelity of ES cell lineage commitment

Multiprotein chromatin remodelling complexes show remarkable conservation of function amongst metazoans, even though components present in invertebrates are often present as multiple paralogous proteins in vertebrate complexes. In some cases these paralogues specify distinct biochemical and/or functional activities in vertebrate cells. Here we set out to define the biochemical and functional diversity encoded by one such group of proteins within the mammalian Nucleosome Remodelling and Deacetylation (NuRD) complex: Mta1, Mta2 and Mta3. We find that, in contrast to what has been described in somatic cells, MTA proteins are not mutually exclusive within ES cell NuRD and, despite subtle differences in chromatin binding and biochemical interactions, serve largely redundant functions. Nevertheless, ES cells lacking all three MTA proteins represent a complete NuRD null and are viable, allowing us to identify a previously undetected function for NuRD in maintaining differentiation trajectory during early stages of lineage commitment.


Introduction
Mammalian cells contain a number of proteins capable of using ATP hydrolysis to shift nucleosomes relative to the DNA sequence, thereby facilitating chromatin remodelling. In mammals, these ATP-dependent chromatin remodelling proteins usually exist within multiprotein complexes and play essential roles in the control of gene expression, DNA replication and repair (Hargreaves and Crabtree 2011;Narlikar et al. 2013;Hota and Bruneau 2016).
NuRD (Nucleosome remodelling and deacetylation) is one such multiprotein complex which is unique in that it contains both chromatin remodelling and protein deacetylase activity. NuRD is highly conserved amongst metazoans and has been shown to play important roles in cell fate decisions in a wide array of systems (Denslow and Wade 2007;Signolet and Hendrich 2015). For example, in embryonic stem (ES) cells NuRD controls nucleosome positioning at regulatory sequences to finely tune gene expression (Reynolds et al. 2012;Bornelöv et al. 2018) and in somatic lineages NuRD activity has been shown to prevent inappropriate expression of lineage-specific genes to ensure fidelity of somatic lineage decisions (Denner and Rauchman 2013;Knock et al. 2015;Gomez-Del Arco et al. 2016;Loughran et al. 2017). It was recently demonstrated that this is achieved in both ES cells and B-cell progenitors by restricting access of transcription factors to regulatory sequences (Liang et al. 2017;Loughran et al. 2017;Bornelöv et al. 2018). Additionally, aberrations in expression levels of NuRD component proteins are increasingly being linked to cancer progression (Lai and Wade 2011;Mohd-Sarip et al. 2017).
NuRD is comprised of two enzymatically and biochemically distinct subcomplexes: a chromatin remodelling and a deacetylase subcomplex. The chromatin remodelling subcomplex contains a nucleosome remodelling ATPase protein (Chd3/4/5) along with one of the zinc finger proteins Gatad2a/b and the Doc1/Cdk2ap1 protein, while the deacetylase subcomplex contains class I histone deacetylase proteins Hdac1/2, the histone chaperones Rbbp4/7, the Metastasis Tumour Antigen family of proteins, Mta1, Mta2 and Mta3 and, in pluripotent cells, the zinc finger proteins Sall1/4 (Lauberth and Rauchman 2006;Allen et al. 2013;Kloet et al. 2015;Bode et al. 2016;Low et al. 2016;Miller et al. 2016;Spruijt et al. 2016;Zhang et al. 2016). These two subcomplexes are bridged by Mbd2/3, creating intact NuRD. While HDAC and RBBP proteins are also associated with other chromatin modifying complexes, the CHD, MBD, GATAD2 and MTA proteins are obligate NuRD components. Changes in subunit composition in large multiprotein, chromatin modifying complexes such as PRC1 and BAF has been shown to correlate with distinct changes in function to sites of action in the chromatin in a cell-type specific manner (Ho and Crabtree 2010;Morey et al. 2012). The NuRD complex might therefore be expected to show similar diversity in both composition and function and in fact, diversification of NuRD function has been described through differential incorporation of different isoforms of NuRD component proteins (Bowen et al. 2004). For example, Mbd2 and Mbd3 are mutually exclusive within NuRD (Le Guezennec et al. 2006). While Mbd2 is not required for mammalian development, Mbd3 is essential for early postimplantation mouse development (Hendrich et al. 2001).
Mbd2/NuRD is a methyl-CpG binding co-repressor complex which is dispensable for early development but Mbd3/NuRD, a transcriptional modulator found at sites of active transcription, has been shown to play important roles in regulation of cell fate decisions in multiple developmental systems (Feng and Zhang 2001;Reynolds et al. 2012;Gunther et al. 2013;Reynolds et al. 2013;Shimbo et al. 2013;Menafra et al. 2014). In brain development NuRD complexes containing either CHD3, CHD4 or CHD5 play distinct roles during cortical development (Nitarska et al. 2016).
Further functional and biochemical diversification occurs through alternate use of the three MTA proteins within NuRD. MTA proteins function as a scaffold around which the deacetylase subcomplex is formed, comprising a 2:2:4 stoichiometry of MTAs:HDACs:RBBPs (Millard et al. 2013;Smits et al. 2013;Millard et al. 2016;Zhang et al. 2016). The three MTA proteins are highly conserved, differing from each other predominantly at their C-termini.
The MTA1 protein was originally identified because of its elevated expression in metastatic cell lines (Toh et al. 1994), and subsequently all three MTA proteins have been shown to be up-regulated in a range of different cancer types (Covington and Fuqua 2014;Sen et al. 2014;Ma et al. 2016). The MTA1 and 3 proteins were shown to form distinct NuRD complexes in breast cancer cells and in B-cells and were recruited by different transcription factors to regulate gene expression (Mazumdar et al. 2001;Fujita et al. 2003;Fujita et al. 2004;Si et al. 2015). These studies did not detect biochemical interactions between MTA3 and the other MTA proteins, leading to the conclusion that MTA proteins are mutually exclusive within NuRD. In contrast, Mta1 was shown to interact with Mta2 in MEL cells, possibly indicating that mutual exclusivity may be cell type-specific (Hong et al. 2005). While all three Mta genes are expressed in ES cells, detailed biochemical analysis of interactions of MTA proteins with one another or with the various NuRD components in ES cells has not previously been described.
Functional evidence does not support a strict lack of redundancy amongst MTA proteins during mammalian development. While zygotic deletion of Chd4 or Mbd3 results in pre-or peri-implantation developmental failure respectively (Kaji et al. 2007;O'Shaughnessy-Kirwan et al. 2015), mice deficient in any one of the three MTA proteins show minimal phenotypes. Mice lacking either Mta1 or Mta3 are viable and fertile (Manavathi et al. 2007)(Mouse Genome Informatics), while mice lacking Mta2 show incompletely penetrant embryonic lethality and immune system defects (Lu et al. 2008). In the current study we took a systematic approach to dissecting MTA protein biochemical and functional diversity. We find that, in contrast to what has been described in somatic cells, MTA proteins are not mutually exclusive within NuRD in ES cells and serve largely redundant functions. Furthermore, ES cells lacking all three MTA proteins are viable and represent a complete NuRD null, allowing us to identify a previously undetected function for NuRD in early stages of lineage commitment.

Mta proteins are not mutually exclusive within the NuRD complex in ES cells
The absence of a detected interaction between the MTA2 and MTA3 proteins in human cells (Fujita et al. 2003;Si et al. 2015), and the observation that different MTA proteins can show different protein-protein interactions in B-cells ) has led to the conclusion that the MTA proteins are mutually exclusive within NuRD, and could hence confer functional diversity to the NuRD complex (Lai and Wade 2011). To investigate the biochemical specificity of the MTA proteins in an unbiased manner, we used gene targeting of endogenous loci to produce three different mouse ES cell lines in which an epitope tag was fused to the C-terminus of each MTA protein (Fig. 1A). Although MTA genes show different expression patterns in preimplantation mouse development, all three are expressed in peri-implantation and early postimplantation epiblast, the tissue most similar   to the naïve ES cell state (Fig. S1). We therefore considered ES cells to be a good system in which to investigate the function of MTA proteins.
Each tagged protein was expressed at levels comparable to those of wild type proteins and was found to interact with other NuRD component proteins by immunoprecipitation (Fig. 1B). Each MTA protein was also able to immunoprecipitate both of the other MTA proteins in addition to unmodified forms of itself (Fig. 1B). Each NuRD complex contains two copies of an MTA protein (Millard et al. 2013;Smits et al. 2013;Zhang et al. 2016), so the identification of an interaction between MTA proteins means that individual NuRD complexes in ES cells could contain either homodimers or heterodimers consisting of any combination of the three MTA proteins.
To investigate the potential biochemical diversification conferred by different MTA proteins in pluripotent cells we used our tagged cell lines to identify proteins interacting with each of the MTA proteins by label-free quantitative mass spectrometry. Each protein robustly co-purified with all known NuRD component proteins, including each of the MTA proteins, confirming that MTA proteins are not mutually exclusive within NuRD in ES cells ( Fig. 1C). In addition to NuRD components, each of the MTA proteins co-purified with Wdr5 as well as a number of zinc finger proteins, most of which had previously been identified as NuRD-interacting proteins (Bode et al. 2016;Spruijt et al. 2016;Ee et al. 2017;Matsuura et al. 2017).
As NuRD is assembled from a deacetylase subcomplex and a remodeller subcomplex joined through the Mbd3 protein ( Fig. 1A), loss of Mbd3 is expected to result in dissociation of these two subcomplexes. Endogenous tagging for each Mta protein was performed in an ES cell line harbouring a floxed Mbd3 allele, and IP/Mass spectrometry was repeated in ES cells after Mbd3 deletion in order to enrich for interactions specific for the deacetylase subcomplex. The majority of interactions with non-NuRD components was lost after Mbd3 deletion, indicating that most of these proteins do not directly associate with either the MTA proteins or with the deacetylase subcomplex ( Fig. 1C, Table S1). Exceptions to this were Zfp296, which was identified as interacting with all three MTA proteins in an Mbd3independent manner, and Pwwp2a, which co-purified with Mta1 in both wild type and Mbd3-null cells. Notably, Zfp219 and Zfp512b both showed Mbd3-independent interactions specifically with Mta2 ( Fig. 1C, Table S1).
By quantitating the abundance of peptides sequenced in each experiment we found that the interactomes for both Mta1 and Mta2 showed a depletion of peptides associated with the remodeller subcomplex (i.e. Chd4, Gatad2a/b and Cdk2ap1) in Mbd3-null cells, whereas proteins associated with the histone deacetylase subcomplex (Mta proteins, Hdac1/2, Sall proteins and Rbbp4/7) remained present at similar levels ( Fig 1D). In contrast the Mta3 interactome showed an increased interaction with Mbd2 and no relative loss of either sub-complex in the absence of Mbd3. Thus both Mta1 and Mta2 can form part of a stable deacetylase subcomplex in the absence of intact NuRD, but Mta3 is preferentially found in intact NuRD complexes. All three MTA proteins associated with Mbd2 in the absence of Mbd3, indicating that they can all can contribute to Mbd2/NuRD. Together these data show that the MTA proteins are found exclusively within the NuRD complex in ES cells.

MTA proteins show subtle differences in genome-wide binding patterns
To test whether MTA proteins confer differential chromatin binding to NuRD complexes we subjected our ES cell lines expressing epitope-tagged MTA proteins to chromatin immunoprecipitation followed by high throughput sequencing (ChIP-Seq). The binding profiles of the three proteins were largely, but not completely overlapping ( Fig. 2A).
Mta3 peaks were almost entirely associated with Mta1 and/or Mta2-binding, while 31% of Mta1 peaks and 22% of Mta2 peaks were not associated with any other MTA protein ( Fig.   2A). The vast majority of MTA peaks overlapped with Chd4 peaks, indicating that they represent NuRD-bound regions (Figs. 2B, C). Mta3 peaks were almost exclusively associated with Chd4 binding, consistent with Mta3 being preferentially in intact NuRD (Fig. S2A). In contrast, 15% of Mta1 peaks and 10% of Mta2 peaks did not overlap with the Chd4 dataset.
Sites found associated with all three MTA proteins were highly enriched for Chd4 and Mbd3 binding, consistent with these being core NuRD binding regions (Fig. 2C, top panels).
These sites were also enriched for marks of active promoters (H3K4Me3, H3K27Ac) and active enhancers (H3K4Me1, H3K27Ac, P300; Fig. 2C, D), both of which are hallmarks of NuRD-associated regions (Miller et al. 2016;Bornelöv et al. 2018). The same was true for sites bound by any two of the three MTA proteins (Fig. S2B). The majority of sites occupied by one MTA protein, but not the other two, were also occupied by Chd4 but to a lesser extent than is seen at core NuRD sites ( H3K27ac Ep300 H3K4me3 log2(ChIP/Input) ChIP-seq datasets, whereas "Mta1 Only", "Mta2 Only" or "Mta3 Only" refer to peaks only called for that protein. D. Average enrichment of density plots in C are plotted for different subsets of Mta ChIP-seq peaks. enrichment for H3K4Me3, indicating that Mta1 is not found at promoters either without one of the other Mta proteins or within the context of complete Mbd3/NuRD (Fig. 2C, D, S2C). In contrast, sites bound by Mta2 and Mta3, but not Mta1 showed particular enrichment for H3K4Me3 (Fig. S2B). None of the MTA-bound sequences showed any enrichment for methylated DNA (Fig. S2D). In all cases MTA-only sites lacking Chd4, which could represent sites of binding by the histone deacetylase subcomplex only, showed characteristics of inactive enhancers, in that they were moderately enriched for H3K4Me1 and P300, but not for H3K4Me3, H3K27Ac or H3K36Me3 (Fig S2C). Genes associated with binding by only one or two MTA proteins, with or without Chd4 are associated with a similar distribution of GO terms as core NuRD sites ( Fig S2E) which is not consistent with the idea that the MTA proteins are directing NuRD activity to specific gene subsets.

Mta1/Mta2/Mta3 triple knockout is a total NuRD null
To identify specific functions for the different MTA proteins, we obtained gene trap  Mta123∆ ES cells at levels comparable to wild type expression resulted in restoration of Mbd3 protein levels, demonstrating that contact with at least one of the MTA proteins is sufficient for Mbd3 protein stability (Fig. 3E). In contrast to Mbd3-null ES cells which display a significant depletion, but not a complete loss of NuRD due to partial compensation by Mbd2 (Kaji et al. 2006), Mta123∆ ES cells were completely devoid of any detectable intact NuRD (Fig. 3C, F). The Mta123∆ ES cells are thus a total NuRD null ES cell line, which allowed us to examine, for the first time, the consequences of a complete loss of the NuRD complex in a viable mammalian cell system.

The NuRD complex safeguards cellular identity during differentiation
Twice as many genes showed an increase rather than a decrease in expression levels in Mta123∆ ES cells compared to control ES cells by RNA-seq (Fig. 4A). This is consistent with NuRD acting predominantly, but not exclusively, as a transcriptional repressor. The ratio of increased to decreased gene expression in Mta123∆ cells was very similar for genes bound by all three Mta proteins, Mta1+2, Mta1 only or Mta2 only (Fig. 4B), indicating that NuRD's representing sites bound by the deacetylase subcomplex, were not associated with specific gene misexpression events in the Mta123∆ ES cells (Fig. S4A, B).
Comparing global gene expression profiles using principal component analysis (PCA) showed that Mta123∆ and Mbd3∆ ES cells in self-renewal conditions were more similar to each other than either was to wild type cells (Fig. 4C). Mta123∆ ES cells were most distinct from wild type cells, consistent with them representing a more complete NuRD knockout ( Fig. 4C, S4C). Consistent with this, genes driving the NuRD-specific separation (Principal Component 2; PC2) were generally misexpressed to a greater degree in Mta123∆ ES cells than in Mbd3∆ ES cells (Fig 4D, S4C). As has been shown for Mbd3∆ ES cells, both NuRD mutant lines moderately misexpressed pluripotency-associated genes in ES cells (Reynolds et al. 2012) (Fig. 4D). Yet in 2iL conditions pluripotency-associated genes were mis-expressed to a much lesser extent than genes associated with differentiation (Fig. 4D). This demonstrates that, in addition to maintaining pluripotency gene expression levels, NuRD also functions in ES cells to prevent inappropriate activation of lineage-specific gene expression.
Despite showing inappropriate expression of a large number of genes normally associated with differentiated cells, Mta123∆ ES cells could be maintained as morphologically undifferentiated ES cells in the very restrictive 2iL conditions (Fig. 3B). We therefore next asked what impact complete loss NuRD activity had upon the differentiation capacity of Mta123∆ ES cells. Upon removal of the two inhibitors and LIF from the culture media wild type cells began to adopt the flatter morphology of neurectoderm ( Fig. 4E) (Ying et al. 2003). This was accompanied by downregulation of pluripotency-associated genes and the activation of a neural gene expression programme (Fig. 4F, S4C). Mbd3∆ ES cells are able to respond to the absence of self-renewal factors but have a very low probability of adopting a differentiated fate when induced to differentiate in N2B27 conditions (Kaji et al. 2006;Reynolds et al. 2012). Consistent with these findings, after 5 days in differentiation conditions Mbd3∆ ES cells showed some signs of having responded to differentiation conditions, but still retained pockets of morphologically undifferentiated cells (Fig. 4E). In contrast the completely NuRD-null Mta123∆ ES cells appeared to have all exited the selfrenewal programme and adopted a flat, monolayer morphology (Fig. 4E). The ability of
Since the differentiation process may be considered as a combination of exit from selfrenewal (downregulation of pluripotency-associated genes) and acquisition of lineage specific gene expression programs, we focussed on changes to these two classes of genes by RT-qPCR over a differentiation time course. Pluripotency-associated genes such as Esrrb, Klf4, Nanog and Pou5f1 were downregulated during differentiation of both wild type and Mta123∆ ES cells (Fig. 4F). While this downregulation was not absolutely dependent on the presence of functional NuRD, the magnitude and kinetics of the response varied in a genedependent manner. Genes associated with acquisition of neural fate were activated in the presence of wild type NuRD, a response which was reduced in the NuRD null line (Fig. 4F).
Globally, Mta123∆ ES cells misexpressed considerably more genes associated with development and differentiation than did Mbd3∆ ES cells after 48 hours of differentiation ( Fig. S4E). Exogenous expression of either Mta1, Mta2 or Mta3 in Mta123∆ ES cells was able to rescue the ability of cells to activate neural gene expression to different extents during differentiation (Fig. S5B), consistent with their ability to rescue the morphological phenotypes. Mta123∆ ES cells induced to differentiate towards a mesoderm fate similarly failed to appropriately activate differentiation markers, and exogenous expression of individual MTA proteins was again able to rescue the defect to varying degrees (Fig. S5C).
Together these data show that while NuRD activity is not required for ES cells to exit the naïve state, it is required for activation of lineage-appropriate gene expression as well as to prevent expression of genes associated with other cell types.
To better understand how NuRD-null cells respond when induced to differentiate we compared their gene expression profiles to a transcription landscape made from single cells leftwards along PC1; Fig 5A, S5D), they did not travel as far along this trajectory as wild type cells, instead occupying a space between the E4.5 and E5.5 epiblast states. Mta123D cells travelled even less far along PC1 than Mbd3∆ ES cells, and rather than maintain the appropriate differentiation trajectory they also travelled along PC2, occupying a space between E4.5 epiblast and E4.5 primitive endoderm. This further demonstrates that not only is NuRD important for cells to be able to adopt the appropriate gene expression programme for a given differentiation event, but it is also important for cells to maintain an appropriate differentiation trajectory.
If Mta123∆ ES cells were undergoing a specific trans-differentiation event towards trophectoderm during ES cell differentiation then this could become more pronounced if exposed to normal differentiation conditions in a chimaeric embryo. If, in contrast, they are simply unable to differentiate properly, they would not be expected to contribute to early embryos. To distinguish between these possibilities, we assessed the ability of Mta123∆ ES cells to differentiate in chimaeric embryos. Equal numbers of control or Mta123∆ ES cells expressing a fluorescent marker were aggregated with wild type morulae and allowed to develop for 48 hours. While wild type cells contributed to the ICM of host embryos with 100% efficiency, Mta123∆ ES cells showed significantly reduced contribution and increased levels of apoptosis ( Fig. 5B-D). Those Mta123∆ cells that did survive in blastocysts were predominantly, but not always found in the inner cell mass and expressed Sox2 but not Cdx2, indicating that they had not undergone inappropriate differentiation towards a trophectoderm fate. This is most consistent with Mta123∆ cells being unable to properly adjust to an ICM environment and enter a normal differentiation path. The ability to contribute to the ICMs of chimaeric embryos was rescued by constitutive expression of Mta2 in Mta123∆ cells (Fig. 5C). We therefore propose that NuRD functions not only to establish the correct lineage identity of cells during the differentiation process, but also prevents inappropriate gene expression to maintain an appropriate differentiation trajectory (Fig. 6).

Figure 6. Model of NuRD function during differentiation of pluripotent cells.
NuRD facilitates lineage commitment of ES cells after exit from pluripotency (red arrows), allowing cells to form differentiated cell types (Top). In the absence of Mbd3, residual NuRD activity ensures cells retain the appropriate differentiation trajectory, but the cells are unable to reach a differentiated cell fate (dotted red arrows; Middle). In the absence of all three MTA proteins there is no residual NuRD activity and ES cells are unable to either achieve appropriate lineage commitment, or to maintain the proper differentiation trajectory (dotted arrows, bottom panel).

Discussion
Here we provide a biochemical and genetic dissection of the core NuRD component Protein subunit diversity is often found in chromatin remodelling complexes which specifies functional diversity (Hargreaves and Crabtree 2011;Morey et al. 2012;Hota and Bruneau 2016). This is also the case for the NuRD complex, where alternate usage of Mbd2/3, Chd3/4/5, and Mta1/3 has been found to result in alternate functions for NuRD complexes (Feng and Zhang 2001;Fujita et al. 2003;Nitarska et al. 2016). Yet our findings that Mta1 and Mta3-null mice are viable and fertile, and that we detect no major differences in the abilities of different MTA proteins to rescue the Mta123∆ ES cell phenotypes indicates that MTA proteins exhibit considerable functional redundancy.
Different MTA proteins are capable of interacting with each other in ES cells, so how the mutual exclusivity reported in other cell types might be achieved is not clear. One possibility is that the variable inclusion in NuRD of zinc finger proteins, such as the Sall proteins in ES cells, could influence the MTA makeup of NuRD complexes. This class of variable NuRD interactors, which include Sall1/2/3/4, Zfp423 (Ebfaz), Zfpm1/2 (Fog1/2), and Bcl11b, interact with RBBP and/or MTA proteins via a short N-terminal motif (Hong et al. 2005;Lauberth and Rauchman 2006;Lejon et al. 2011). Of this class of proteins Sall1 and Sall4 are the most highly expressed in ES cells, and Sall4 can associate with all three MTA proteins ( Fig. 1C) (Miller et al. 2016). In contrast the Zfpm1 (Fog1) protein was shown to preferentially associate with Mta1 and Mta2, but not Mta3, in a somatic cell line (Hong et al. 2005). Hence it is possible that different proteins using this N-terminal motif to interact with NuRD in different cell types could act to skew the proportion of different MTA proteins included in the NuRD complex.
Mta1 and Mta2 both contain two distinct RBBP interaction domains, while Mta3 lacks the C-terminal most RBBP interaction domain (Millard et al. 2016). The Mta1 and Mta2 proteins show minimal loss of stability in the absence of Mbd3, but Mta3 requires an interaction with Mbd3 to be completely stable (Bornelöv et al. 2018). It is possible that the additional interaction with an Rbbp protein confers stability to Mta1 and Mta2 in the absence of Mbd3. This would be consistent with our interpretation that Mta3 preferentially exists within an intact NuRD complex (Figs 1, 2). Rbbp4/7 confer histone H3 binding to the NuRD complex, so Mta3-containing NuRD may bind chromatin less tightly than Mta1and/or Mta2-containing NuRD. Consistent with this possibility, we identified >2x more Mta1-and Mta2-assoicated ChIP-seq peaks than Mta3-peaks ( Fig. 2A).
Pluripotent cells lacking Mbd3 have been used extensively to show that NuRD plays important roles in control of gene expression during early stages of exit from pluripotency in vitro and in vivo (Kaji et al. 2006;Kaji et al. 2007;Latos et al. 2012;Reynolds et al. 2012). conditions, and they show a more pronounced differentiation defect. We propose that the increased transcription of a wide array of lineage inappropriate genes in Mta123∆ ES cells renders them incapable of not only achieving the correct differentiation state, but also of maintaining the proper differentiation trajectory (Fig. 6). In this model the primary function of NuRD in ES cells is to silence inappropriate gene expression, which ensures fidelity of differentiation. A further, more specialised function contained within the core function and dependent upon Mbd3 (but not Mbd2), is to respond properly to context-appropriate differentiation signals to achieve specific differentiation states.

Mouse embryonic stem cells
Mouse embryonic stem cells (ESCs) were grown on gelatin-coated plates in 2i/LIF conditions as described (Ying et al. 2008). The following "Knockout First" alleles were obtained from EUCOMM as heterozygous ES cell lines (Illustrated in Fig. S1A (Nichols et al. 2009).
The epitope tagged ESC lines Mta1-Avi-3×FLAG and Mta2-GFP were generated by traditional gene targeting, while the Mta3-Avi-3×FLAG line was generated using a CRISPR/Cas9 genome editing approach. All Mta epitope tagged lines were made in an Mbd3 Flox/background (Kaji et al. 2006). Transient expression of Dre recombinase was then used to remove the selectable marker (Anastassiadis et al. 2009).
ES cells were induced to differentiate towards a neuroectoderm fate by removal of two inhibitors and LIF and culturing in N2B27 media as described (Ying et al. 2008).
Mesendoderm differentiation was performed as follows: ES cells were plated at 10 4 cells/cm 2 in N2B27 on fibronectin-treated 6-well plates and cultured for 48 hours. Medium was then replaced with 10 ng/ml activin A and 3 µM CHIR99021 in N2B27 and cultured further.

Mice
All animal experiments were approved by the Animal Welfare and Ethical  (Hayashi et al. 2002) to create null alleles.
Chimaeric embryos were made by morula aggregation with 8-10 ES cells per embryo as described (Hogan et al. 1994) and cultured for 24-48 hours prior to fixation and immunostaining. ES cells used stably expressed a PiggyBac Kusabira Orange transgene.

Label-free pulldowns and label-free quantitation (LFQ) LC-MS/MS analysis
Label-free pulldowns were performed in triplicate as previously described Miller et al. 2016). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD009855.

Chromatin immunoprecipitation (ChIP), sequencing, and analysis
Chromatin immunoprecipitations were carried out as previously outlined (Reynolds et al. 2012). For sequencing of ChIP DNA, samples from six (Mta1-FLAG and Mta2-GFP) and four (Mta3-FLAG) individual ChIP experiments were used. Antibodies used are listed in Table   S1. ChIP-seq libraries were prepared using the NEXTflex Rapid DNA-seq kit (Illumina) and sequenced at the CRUK Cambridge Institute Genomics Core facility (Cambridge, UK) on the Illumina platform. Reads were aligned using bowtie version 0.12.8 and the arguments -m 1.
Peaks were called using macs2 version 2.1.0 with the default Q-value of < 0.05. The results were merged to get 7 different sets of peaks defined by the combination of bound Mta proteins.
DNA Methylation data was obtained from (Shirane et al. 2016). Output methylation files were filtered for base pairs with a coverage of over 3. The methylation percentage was averaged using a sliding window of 250 bp to give a smoothed track for the heatmaps/profile.
High throughput sequence datasets used in this manuscript are listed in Supplemental Table 2.   B. Features of peaks containing two of the three MTA proteins, as in Figure 3C. C. Average enrichment of indicated features is shown for peaks for each Mta protein which do or do not also contain Chd4. The number of peaks in each set is indicated as n. D. Average enrichment of DNA methylation across MTA peaks with or without Chd4, as in Panel C. E. Significance of GO-term enrichment for genes associated with peaks for different combinations of Mta proteins and Chd4.   Figure S4. Control of gene expression by the MTA proteins (Related to Figure 4).