Purification of Oct4-Interacting Proteins from ESCs
We have previously described a mouse ESC line in which, under self-renewing conditions, all the Oct4 protein in the cell has an N-terminal triple FLAG-tag (F-Oct4) (van den Berg et al., 2008
). Both F-Oct4 and the parental ZHBTc4 cells have a normal ESC morphology (Niwa et al., 2000; van den Berg et al., 2008
) and express normal levels of ESC markers Sox2, Sall4 (Figure S1
A available online), Klf4, Dax1, Zfp42, and Eras (Figure S1
B). This indicates that the F-Oct4 protein present in the F-Oct4 cells maintains their ESC identity. We prepared nuclear extracts from F-Oct4 cells and ZHBTc4 cells, which do not express F-Oct4 and serve as a control. FLAG-affinity purifications were performed from 1.5 ml of nuclear extract (equivalent to ~4 × 108
cells) with an improved protocol in which near-physiological salt conditions, low detergent concentrations, and low-adherence tubes were employed (see Experimental Procedures
for details). Benzonase nuclease was added to the extract to remove the remaining DNA (Figure S1
C), thereby eliminating protein interactions mediated indirectly by DNA bridging. Virtually all F-Oct4 in the extract was bound to the FLAG-antibody beads and subsequently eluted by FLAG peptide competition (Figure S1
D). An SDS polyacrylamide gel of the eluted fractions, stained with a sensitive Colloidal Coomassie protocol, showed Oct4 as the predominant band in the F-Oct4 sample (A). The control sample showed only one prominent band, which was also present in the F-Oct4 sample but was otherwise devoid of major contaminants. This indicates that our FLAG-mediated purification of Oct4 has a very good signal to background ratio. The presence of multiple bands of lower intensity in the F-Oct4 lane suggests that Oct4 interacts with a variety of proteins at substoichiometric levels. The majority of Oct4 runs at approximately its own molecular weight on a gel filtration column (Figure S1
E), unlike a stable complex such as NuRD. Therefore, most Oct4 interactions are likely to be weak and do not survive the 4 hr gel filtration procedure, in which dissociation causes an irreversible loss of the interaction. To independently verify candidate F-Oct4-interacting proteins, we also immunoprecipitated endogenous Oct4 from nuclear extracts of a different ESC line, 46C (Ying et al., 2003
), with an antibody that captured all Oct4 from the extract (Figure S1
F). Although we used the same buffer conditions and low-adherence tubes, this procedure gives higher background compared to the FLAG-affinity purification (B), because proteins that bind nonspecifically to the antibody beads or the tubes cannot be excluded from the eluate by FLAG-peptide elution, as they can in the FLAG purification strategy.
Purification of Oct4 and Its Interacting Proteins
We analyzed three independent F-Oct4 purifications and the endogenous Oct4 immunoprecipitation by mass spectrometry (). A representation of the identified proteins by a more quantitative measure, emPAI score (Ishihama et al., 2005
), is shown in Table S1
. Our list of more than 50 putative Oct4-associated proteins () contains 22 transcription factors of which half have a role in maintaining pluripotency (). These include Sall4, Klf5, Zfp143, Esrrb, and Sox2, the best-characterized Oct4 partner for which 3D structures of the Oct4-Sox2-DNA ternary complex have been reported (Reményi et al., 2003; Williams et al., 2004
). We also identified a number of chromatin-modifying complexes (CMCs). All of the subunits of the transcriptional repressor NuRD were specifically present, except for Rbbp4 (high background prevented inclusion of Rbbp4 in ). We detected subunits from the chromatin-remodeling complexes SWI/SNF and Trrap/p400, the Lsd1 histone demethylase complex, and components of the polycomb repression complex 1 (PRC1).
Oct4-Interacting Proteins as Identified by Mass Spectrometry Analysis of Purified Oct4 Samples
Transcriptional Network and Phenotype of Oct4-Interacting Proteins
Next we examined the presence of some of the identified interactors in Oct4 immunoprecipitates by immunoblotting. Indeed, we find that NuRD subunit Mta2 (C), spalt-like protein Sall4, histone-demethylase Lsd1 (D), Sall1, and Wdr5 (Figure S1
G) coprecipitate with Oct4, whereas immunoprecipitates of Mta2 (E) and Wdr5 (Figure S1
H) contain Oct4. Recently, it was suggested that a subset of the NuRD subunits (Mta1 and 2, Gatad2a and Gatad2b, Hdac1 and 2) forms an Oct4/Nanog-associated complex called NODE (Nanog- and Oct4-associated deacetylase; Liang et al., 2008
). We found that Oct4 binds the classical NuRD complex, as it was originally defined (Zhang et al., 1999
), including catalytic subunit Mi2β and Mbd3 and Rbbp7 (). Immunoblotting confirmed this; the proportionate amount of antigen detected for Mi2β, Mbd3, Mta1, and Mta2 was the same in FLAG-Oct4 and Mta2 IP samples (F). This suggests that Oct4-bound NuRD is similar or identical to classical NuRD in its composition and argues against the existence of Oct4-bound NuRD subcomplexes, such as NODE.
Oct4-Interacting Proteins Correlate with Gene Regulation by Oct4 and ESC Self-Renewal
Proteins that interact with Oct4 may be expected to be Oct4 cofactors in gene regulation and have DNA binding profiles that overlap with Oct4. Recently, two studies reported the genome-wide binding sites of different sets of ESC transcription factors (Chen et al., 2008b; Kim et al., 2008
). Five of the Oct4-interacting transcription factors identified here (Sox2, Nac1, Tcfcp2l1, Esrrb, Dax1) were investigated in those studies and were found to colocalize frequently with Oct4 (), including at the promoters of important pluripotency genes such as Nanog
(Chen et al., 2008b; Kim et al., 2008; Levasseur et al., 2008
Phenotypes are documented for ~60% of the identified Oct4-interacting proteins (). Of these, ~65% (21/32) of the tested factors () affect the ability of ESCs to remain undifferentiated. This includes most of the aforementioned transcription factors and subunits of all the identified Oct4-associated chromatin-modifying complexes (), except for the Lsd1 complex.
We then investigated whether genes encoding Oct4-interacting proteins are bound and regulated by Oct4. Gene expression profiling data from ZHBTc4 ESCs, which express Oct4 from a doxycycline-repressible transgene (Sharov et al., 2008
), was combined with two different sets of Oct4 ChIP data (Chen et al., 2008b; Kim et al., 2008
). We find that 14 factors (26%) are encoded by genes bound by Oct4 that are downregulated after 48 hr of doxycycline treatment (). This correlation of Oct4 binding and transcriptional regulation by Oct4 increases the interdependence of the associated proteins with Oct4, as previously observed (Wang et al., 2006
Purification of Interaction Partners of Sall4, Esrrb, Dax1, and Tcfcp2l1
Having established that our FLAG-affinity purification protocol identifies novel interactions that are independently verifiable and biologically relevant, an expanded network of Oct4 interactions was sought. Sall4, Esrrb, Dax1, and Tcfcp2l1 were selected for purification because of their consistent presence in all Oct4 purifications (). The spalt-like transcription factor Sall4 is important for stabilizing ESC self-renewal (Yuri et al., 2009; Zhang et al., 2006
). Orphan receptor Esrrb is important for ESC self-renewal (Ivanova et al., 2006; Loh et al., 2006
). Esrrb positively regulates the expression of key pluripotency gene Nanog
(van den Berg et al., 2008
), and overexpression of Esrrb allows short-term ESC maintenance without the addition of exogenous LIF (Zhang et al., 2008
). Esrrb is also capable of replacing KLF4 in somatic cell reprogramming (Feng et al., 2009
). Dax1 is an orphan receptor that is important for ESC self-renewal (Niakan et al., 2006
). Tcfcp2l1 colocalizes with Oct4 on many ESC promoters and may be important for optimal ESC proliferation (Chen et al., 2008b; Ivanova et al., 2006
). FLAG-tagged cDNAs were stably introduced into ZHBTc4 ESCs and clones selected that express the encoded proteins at levels similar to the endogenous proteins (Figure S2
A). These clones had comparable morphology and growth rate to the parental line (data not shown). Proteins were purified by our FLAG-affinity protocol, and coomassie-stained gels of the purified fractions from F-Sall4, F-Esrrb, and F-Tcfcp2l1 purifications showed prominent bands of the expected molecular weight (A) that reacted with the FLAG antibody (Figure S2
B). The presence of additional bands in the transcription factor purifications suggests the efficient copurification of associated proteins. F-Dax1 was not visible by coomassie blue staining (A), although it was almost completely depleted from the nuclear extract by the purification (Figure S2
B). Together with the weaker anti-FLAG western signals of F-Dax1 extracts and purified Dax1 fractions, compared to the other FLAG proteins (not shown), this suggests a relatively low expression level of F-Dax1 (and therefore of endogenous Dax1) in ESCs. B–2E provide summaries of the interacting proteins of Sall4, Dax1, Tcfcp2l1, and Esrrb (complete lists of identifications and information on Mascot scores, number of identified unique peptides, and emPAI scores are shown in Tables S2–S9
). To examine the Oct4 dependence of the interaction partner associations, we also performed the purifications 16 hr after doxycycline-mediated repression of Oct4, which removes essentially all Oct4 protein from ZHBTc4-derived cells (Niwa et al., 2000; van den Berg et al., 2008
). Purified fractions from two FLAG purifications of cells with or without doxycycline addition were analyzed by mass spectrometry. Doxycycline addition had no consistent effect on the vast majority of the identified interactions (Tables S2–S9
). Of the proteins affected by Oct4 modulation, only Esrrb was ever identified as an Oct4 interactor (Table 1
). The interaction between Esrrb and Sall4 appears to be sensitive to removal of Oct4 in the F-Sall4 purifications (Tables S2 and S6
). However, the mascot scores here are close to threshold, whereas in F-Esrrb purifications where Sall4 has a high Mascot and emPAI score, removal of Oct4 had no effect (Tables S5 and S9
). Taken together, this suggests that the identified interactions are unlikely to be bridged by Oct4, although many of the identified proteins also interact with Oct4.
Purification of F-Sall4, F-Dax1, F-Tcfcp2l1, F-Esrrb, and Their Interacting Proteins
We independently verified a number of the putative interactors of F-Sall4, F-Dax1, F-Tcfcp2l1, and F-Esrrb. Immunoprecipitation of Sall4 coprecipitated Sall1 and MTA2 (Figure S3
A), V5-tagged Zfp143 (Figure S3
B), and F-Nac1 (Figure S3
C), whereas Sall4 is present in immunoprecipitates of MTA2 (Figure S3
D) and F-Nac1 (Figure S3
E). GST-Dax1 pull-downs precipitated Sall4, Sall1, Oct4, Wdr5, and Esrrb (Figure S3
F). V5-Tcfcp2l1 immunoprecipitation brought down Esrrb and MTA2 (Figure S3
G), whereas GST-Esrrb pull-down coprecipitated MTA2, Sall4, Ep400 (Figure S3
H), V5-Dax1 (Figure S3
I), and F-Tcfcp2l1 (Figure S3
J). MTA2 immunoprecipitation coprecipitated Esrrb (Figure S3
An Oct4-Centered Interaction Network
We assembled the identified interactions of Oct4, Tcfcp2l1, Dax1, Sall4, and Esrrb into an interaction network containing 166 proteins (). This allows the visualization of the interactions between the purified tagged transcription factors and their interaction with multiple chromatin-modifying complexes (CMCs). The NuRD complex was associated with every tagged factor purified, except for Dax1 (, B–2E). The smaller set of interactors identified for Dax1 (C), compared to the other purified proteins, may be due to the purification of relatively small amounts of F-Dax1 protein (A). The Mascot and emPAI scores of NuRD are highest in the F-Sall4 purifications (B; Tables S2 and S6
). Sall4 also interacts with Sall1, Sall2, and Sall3 and associates with all the other tagged factors (B–2E). Binding of Sall4 to NuRD and Sall1 was previously observed (Yuri et al., 2009
). Our data suggest that spalt proteins form a unit with NuRD, which then can associate with other transcription factors. Sall4 interactors Nac1 and Bend3 (B) could also be part of this unit, as indicated by the fact that they were observed together in individual purifications of Tcfcp2l1 and Esrrb (data not shown). The SWI/SNF complex also associates with most tagged transcription factors (, B, 2D, and 2E). The Trrap/p400 complex is present with relatively high mascot and emPAI scores in Esrrb and Tcfcp2l1 purifications, with many subunits detected (D and 2E; Tables S4, S5, S8, and S9
). The PRC1/Mblr complex associates, besides Oct4, also with Tcfcp2l1 (D).
Protein Interaction Network of Oct4 and Its Associated Proteins Sall4, Dax1, Tcfcp2l1, and Esrrb
We find that the purified factors often bind efficiently to evolutionary related proteins. In addition to spalt proteins, we observed interactions between Tcfcp2l1, Tcfcp2, Ubp1, and Grhl2 (D), all of which are related to the Drosophila
Grainyhead transcription factors (Wilanowski et al., 2002
), whereas Esrrb binds the related protein Esrra (E). This suggests that despite diversification, these proteins can still act together in transcription regulation.
Some of the purified factors harbor extensive sets of unique interacting proteins that may mediate their specific function in ESCs. For example, Tcfcp2l1 interacts with many proteins involved in DNA metabolic processes (D) such as DNA replication (Polb, Asf1a, Rpa1) and DNA repair (Xrcc1, 5, 6, Msh2, 6, lig3, EMSY, Prkdc, pnkp) and related pathways such as cell cycle progression or cell proliferation (Hells, Msh2, Mybl2, EMSY).
Orphan receptor Esrrb, which is related to the estrogen receptor, was found to associate with Ncoa3 and Nrip and the TRX/Mll chromatin-modifying complex (E). Intriguingly, Esrrb also interacts with the Mediator complex, RNA polymerase II subunits (RNApol2), and TBP plus Tafs (TFIID complex; E; Tables S5 and S9
), which are all components of the basal transcription machinery (Sikorski and Buratowski, 2009
). The association of Esrrb with Mediator and RNApol2 is DNA independent as shown by the fact that it was not affected by benzonase treatment of the extract (F). Moreover, recombinant GST-Esrrb also interacted efficiently with Mediator and RNA pol2 (G).
The network provides links with protein modification and signaling pathways. For example, Oct4 associates with Rbpj, a transcription factor that acts as the nuclear effector of the Notch signaling pathway (Bray, 2006
), suggesting a connection between Notch-regulated and Oct4-regulated gene expression. Sall4 shows an interaction with Usp9x (B), an essential component of the TGF-β/BMP signaling pathway, which activates Smad4 by removing a monoubiquitin group (Dupont et al., 2009
). Another Sall4-associated factor, Cxxc5 (B), is regulated by TGF-β signaling in neural stem cells, binds Wnt-signaling mediator Dvl, and inhibits Wnt signaling (Andersson et al., 2009
). By interacting with both Usp9x and Cxxc5, spalt proteins may provide a physical link between the TGF-β and Wnt signaling pathways. Oct4, Esrrb, Tcfcp2l1, and Dax1 bind the glycosyl transferase Ogt (O-GlcNAc Transferase; , B–2E), an enzyme that adds N-acetylglucosamine groups (O-GlcNAc) to proteins.
The network contains a number of transcription factors with a high level of interconnectivity, characteristic of network hubs. Examples of such hubs are Zfp143 and Klf5. Zfp143 interacted with Oct4, Sall4, and Tcfcp2l1 (, B and 2D) and was present in one Esrrb purification (not shown). Klf5 was present in Oct4, Sall4, and Tcfcp2l1 purifications (, B and 2D). The purified factors Esrrb, Tcfcp2l1, Dax1, and Sall4 were selected on their interaction with Oct4, but they also have an Oct4-independent interaction with one another. All these highly connected factors affect ESC self-renewal when depleted (), suggesting that physical interaction may play a role in regulating this process. A possible rationale for this correlation, codependent recruitment to DNA, will be tested experimentally below.
Oct4-Dependent Recruitment of Dax1, Tcfcp2l1, and Esrrb
Our purifications showed the physical interaction of Oct4 with Dax1, Tcfcp2l1, and Esrrb. To investigate the relevance of these interactions for the ESC transcriptional network, we tested the effect of acute Oct4 depletion by 12 hr doxycycline treatment, on the recruitment of Dax1, Tcfcp2l1, and Esrrb to a number of genomic binding sites to which Oct4 also binds (Chen et al., 2008b; Kim et al., 2008
). Indeed, depletion of Oct4 reduced recruitment of F-Dax1, F-Tcfcp2l1, and F-Esrrb to several of their targets (A–4C). For example, Dax1 recruitment to the Rest
promoters, which are both also occupied by many other ESC transcription factors (Chen et al., 2008b; Kim et al., 2008
), is dependent on Oct4. Our data suggest that Oct4 can provide an anchor on the DNA for the recruitment of several of its associated factors.
Oct4-Dependent Genome Targeting by Dax1, Tcfcp2l1, and Esrrb