Myc interacts with the NuA4 HAT complex in ES cells

We did not observe overlap of proteins existing between the core protein interaction network (Wang et al., 2006) and the Myc-centered protein interaction network (Figure S1C). Although this may be due to the stringency of our conditions for recovery of protein complexes, within each network we observed a high degree of interactions, strongly suggesting that these two networks, and their protein complexes, are physically separate. Interestingly, we observed that Myc interacts with many proteins in a recognized conserved protein complex known as NuA4 HAT (or the Tip60-Ep400 complex) (Doyon and Cote, 2004) as shown in Figure 1A (pink cells) and Figure 1B (proteins in a pink circle). Myc, Max, Dmap1, Tip60, Trrap, and Ep400 are tightly interconnected within the network; however, Gcn5 and E2F4 show a lower degree of association, suggesting their weak or indirect interaction with Myc/NuA4. It has been suggested that transcription factors, such as Myc, p53, and E2Fs, require the NuA4 complex to activate downstream targets in non-ES cell contexts (Ard et al., 2002; McMahon et al., 1998). Our data (Figure 1 and Table S1) strongly support the view that Myc interacts with an intact NuA4 HAT complex in ES cells, also implying that histone 3 and 4 acetylation (AcH3, and AcH4, respectively) signatures may also be generated in part by the Myc/NuA4 complex via Tip60 in ES cells. Previous RNAi based phenotypic analyses in ES cells revealed that factors in the NuA4 HAT complex, including Ep400, Dmap1, Tip60, Trrap, Ruvb1 and Ruvb2, are critical to ES cell identity (Fazzio et al., 2008) (also our observation, Figure S1D and S1E). These findings imply a crucial role for the Myc/NuA4 complex in ES cells.

Construction of a Myc-centered protein-DNA interaction network in ES cells

To identify genomic targets of Myc and its associated factors tested in Figure 1, we performed bioChIP-chip (Kim et al., 2008). Since Tip60 and Gcn5 generate AcH3 and AcH4 histone modification signatures, we also performed ChIP reactions using native antibodies against AcH3 and AcH4. We found that the six factors we tested (Myc, Max, Dmap1, Tip60, E2F4, and Gcn5) co-occupy many target promoters in close proximity (Figure 2A).

Open in a separate window

Figure 2

Myc-centered protein-DNA interaction network in ES cells

(A) Representative view of Myc, Max, Dmap1, Tip60, E2F4, and Gcn5 occupancy at the target loci.

(B) Number of target promoters bound by each factor or associated with each histone modification signature. Blue represents factors or histone signatures involved in PrC complexes. Red represents factors involved in ES cell core factors, and Green represents Myc and Myc-related factors or histone signatures (D–E). See also Figure S2 and Table S2.

(C) Relative position of chromosomal target loci of each factor in the Myc cluster (upper panel) and Core cluster (bottom panel) shown in (B) and (C) to the TSS.

(D) Target correlation map: The degree of target co-occupancy of each pair of factors (either transcription factor or histone modification signature) is shown. Yellow indicates more frequent colocalization of each pair of factors.

(E) Median distance map: Median distances between the loci co-occupied by two tested factors (except PrC complex proteins) shown in (D). Yellow indicates closer colocalization.

To obtain a global view of individual and multiple transcription factor occupancy, we combined this new data set with previously published ChIP-chip or ChIP-sequencing data sets (Boyer et al., 2006; Chen et al., 2008; Ding et al., 2009; Hu et al., 2009; Kim et al., 2008; Shen et al., 2008), and tested the factor occupancy or histone modification signatures (see Supplemental Data). The numbers of genes that are occupied by a tested factor or marked by a tested histone modification signature are summarized in Figure 2B and Table S2 with a hierarchical clustering image based on target co-occupancy. We then calculated the degree of target co-occupancy of each pair of factors. As shown in the target correlation map in Figure 2D, we observed three major clusters. Factors in Polycomb complexes are associated with the H3K27me3 signature to form a distinct cluster (PrC cluster, blue-colored box in Figure 2D and blue letters in Figure 2B and ​and2D).2D). Core pluripotency factors, including Nanog, Sox2 and Oct4 and others, form an independent cluster (Core cluster, red-colored box and red letters). Myc forms a cluster with other factors and AcH3, AcH4, and H3K4me3 signatures (Myc cluster, Green-colored box and green letters).

We calculated the median distances between binding peaks of each pair of factors using the same cluster information shown in Figure 2D (except for the PrC cluster due to availability of the processed data). The target distance map demonstrates that the factors within the Core or Myc clusters regulate their common targets in close proximity, whereas the factors belonging to a different cluster regulate their common targets in a relatively remote manner (Figure 2E).

Previously, we observed that Myc occupies more target genes than the ES cell core factors (Kim et al., 2008). Similarly, we observed that the factors in the Myc cluster, such as Max, nMyc, E2F4, and Dmap1, tend to occupy more targets than factors in the Core or PrC clusters (Figure 2B), suggesting more global roles in their target gene regulation. The majority of binding peaks generated by the factors in the Myc cluster are more centered at the transcription start site (TSS) compared to the target binding peaks of the factors in the Core cluster (Figure 2C). The factors in the Myc cluster may interact with basal transcription machinery, whereas core factors have both promoter and upstream enhancer targets as described (Chen et al., 2008; Loh et al., 2006). In summary, our data suggest that the factors belonging to each of the distinct clusters (Core, PrC, and Myc) regulate their own rather similar downstream targets in close proximity and may be functionally separated in regulating aspects of ES cell identity.

Target co-occupancy of factors within the Myc cluster has a positive correlation with histone H3 and H4 acetylation signatures

Our prior work revealed that Myc target promoters correlate positively with an active H3K4me3 signature, and negatively with a repressive H3K27me3 signature (Kim et al., 2008). Since Myc is associated with histone acetylation (Frank et al., 2001), we tested the correlation between target occupancy of each factor in the Myc cluster and the histone modification status of their target promoters. As shown in Figure 3A, the majority of the factors in the Myc cluster harbor significantly higher levels of H3K4me3, AcH3, and AcH4 signatures on their target promoters over background (at least >150%). On the contrary, the H3K27me3 signature is significantly underrepresented on the target promoters of approximately half of the factors in the Myc cluster. Interestingly, Cnot3 and Trim28 target promoters show bivalent modifications (both H3K4me3 and H3K27me3 positive), suggesting that, although these factors share many common targets with Myc, they may have different functions compared to the other factors in the cluster.

Open in a separate window

Figure 3

Histone modification signatures on the target promoters of Myc cluster proteins

(A) Histone marks on the target promoters of each factor in the Myc cluster. “All” represents all promoters.

(B) Histone marks and target co-occupancy of 7 factors in Myc cluster (Myc, Max, nMyc, Dmap1, E2F1, E2F4, and Zfx).

Additionally, we tested the relationship between the factor co-occupancy (7 factors in the Myc cluster shown in Figure 2D and Figure 2E, including Myc, Max, nMyc, Dmap1, E2F1, E2F4, and Zfx) and histone modification signatures. As shown in Figure 3B, target promoters co-occupied by multiple factors in the Myc cluster show a higher level of histone acetylation than the common targets of fewer factors. Targets occupied by 7 factors show approximately 400%, and 220% of AcH4, and AcH3 signatures, respectively, over the background level. Upon the decrease of co-occupancy, the level of these signatures decreased on their common targets. We failed to observe correlation between co-occupancy and the H3K4me3 signature, presumably due to the abundance of H3K4me3 marks across many promoters (>60 % of all promoters)(Kim et al., 2008). The repressive signature H3K27me3 displays a reverse correlation with the Myc cluster factor co-occupancy (Figure 3B).

Modules defined by transcriptional regulatory subnetworks in ES cells

Since we observed a strong positive correlation between target co-occupancy of the factors in the Myc cluster and histone acetylation signatures, we examined the correlation between target co-occupancy and gene expression. As shown in Figure 4A, targets co-occupied by 7 or 6 factors in the Myc cluster are more active than the common targets of five or fewer factors in ES cells (red line), and repressed upon differentiation (blue line). To test if the information generated from the factor co-occupancy in the Myc cluster is functionally relevant, we compared KEGG pathways (Dennis et al., 2003; Ogata et al., 1999) enriched in the genes that are common targets of at least 6 factors among 7 factors in the Myc cluster (Myc, Max, nMyc, Dmap1, E2F1, E2F4, and Zfx; black bar in Figure 4A), and the global target genes of Myc. Many cancer-related pathways (red letters in Figure 4B and Table S3) are enriched in the genes co-occupied by the factors in the Myc cluster. In contrast, these cancer-related pathways are not enriched within the global set of genes occupied by Myc. This observation strongly suggests that factor co-occupancy in the Myc cluster does not represent a random subset of Myc targets, and may provide additional information in understanding the combinatorial function of factors in the Myc cluster in ES cells and in cancer cells (Figure 4B).

Open in a separate window

Figure 4

ES cell modules

(A) Gene expression profiles (log₂, left y-axis) upon J1 ES cell differentiation (wild type ES cells: ES, differentiated ES cells for 14 days: dES day 14) are shown as moving window averaged lines (ES; red line, dES day 14; blue line, bin size 100 and step size 1). Randomized genes are sorted (x-axis) by the target co-occupancy of 7 factors in the Myc cluster (right y-axis). Black bar represents target genes co-occupied by at least 6 factors among the 7 factors in the Myc cluster.

(B) Enrichment of KEGG pathways. All Myc target genes (gray bars, total 3733 genes) and genes co-occupied by at least 6 factors among the 7 factors tested marked by black bar in (A) (black bars, total 1756 genes). See also Figure S3A, S3B and Table S3.

(C) ES cell modules: Three ES cell modules are defined based on the target co-occupancy within each cluster shown in Figure 2D. See also Table S3.

(D) GSEA analyses show the gene activity of the three ES cell modules (Core, PrC and Myc modules) as well as the previously defined ESC-like module (Wong et al. 2008a) upon ES cell differentiation (wild type ES cells; ES vs. 14 days differentiated ES cells; dES).

(E) Average gene expression values (log₂) of each module (C) are tested upon ES cell differentiation (ES day0, dES day2, dES day7, and dES day14, respectively). Data are represented as mean ± SEM. See also Figure S2B and Figure S3C.

We previously demonstrated that common targets of multiple factors in the core pluripotency network are significantly active in ES cells. However, when these factors occupy targets alone or with few factors they are not associated with activation of target genes (Kim et al., 2008). Since the targets co-occupied by 7 factors in the Myc cluster show the strongest gene activity (Figure 3B and Figure 4A), we classified common target gene ‘modules’ in ES cells based on the target co-occupancy within the clusters shown in Figure 2; the PrC module, the Core module, and the Myc module (Figure 4C and listed in Table S3). Definition of each module is as follows; the Core module is comprised of genes co-occupied by at least 7 factors among 9 factors shown in the Core cluster (Smad1, Stat3, Klf4, Oct4, Nanog, Sox2, Nac1, Zfp281, and Dax1), depicted in the red box in Figure 2D. The PrC module genes are the common targets of PrC cluster proteins, Suz12, Eed, Phc1, and Rnf2 (blue box in Figure 2D). The Myc module is comprised of genes that are common targets of 7 factors (Myc, Max, nMyc, Dmap1, E2F1, E2F4, and Zfx) in the Myc cluster (green box in Figure 2D). For construction of the Myc module, we excluded Tip60, Gcn5, and Rex1 due to their relatively small number of targets (Figure 2B), and Trim28, and Cnot3 due to the bivalent signature on their target promoters (Figure 3A), as well as the discrepancy of their target similarity within the Myc cluster (Figure S2A). Additional gene sets co-occupied by different combinations of factors in Myc cluster were also tested but showed no significant difference, because the majority of target genes among the tested sets are shared (see below and Figure S2B). Lists of gene sets tested are summarized in Table S3. Indeed, the Core module includes previously known factors in core regulatory circuitry, such as Nanog, Oct4, Rest, Sox2, Tcf3, and Rex1. The PrC module includes genes generally repressed in ES cells, including Hox cluster genes. As shown in Figure 4C, the overlap between each module is minimal and they are involved in different pathways (Figure S3A and S3B).

We then tested activity of each module (hereafter ‘module activity’, averaged expression of all genes in each module in a given expression data set) in ES cells as compared to module activity in differentiated cells. Gene activity of a previously identified ‘Core ESC-like gene module’(hereafter ‘ESC-like module’) (Wong et al., 2008a) was also tested. Gene set enrichment analysis (GSEA) (Subramanian et al., 2005) revealed that the Core, Myc, and previously identified ESC-like modules are highly active in ES cells. As anticipated, the PrC module is repressed in ES cells (Figure 4D). We additionally tested the activity of each module during a time-course of ES cell differentiation. As shown in Figure 4E and Figure S3C, in ES cells the Core module is most active, yet the Myc and ESC-like modules show some activity; these modules become repressed with time during differentiation, whereas the PrC module shows an opposite pattern.

Functional separation of Core and Myc modules in partial iPS cells

Although we observed that both the Core and Myc modules are active in ES cells, the genes that comprise the Core module are distinct from those of the Myc module (Figure 4C). To test if the modules can be functionally separable, we tested the module activity of our three ES cell modules, along with the ESC-like module (Wong et al., 2008a) in other cell types, including iPS cells, partial iPS (piPS) cells, and mouse embryonic fibroblasts (MEFs). Global gene expression profiles of ES and iPS cells are highly similar (Takahashi and Yamanaka, 2006). Relying on a publicly available data set (Sridharan et al., 2009), we tested if the module activity is similar between ES and iPS cells. Similar to the data shown in Figure 4E, the Core and Myc modules are highly active in both ES and iPS cells (Figure 5A and Figure S3D). The PrC module is inactive in both cell types, as expected. In MEFs, the module activity pattern is similar to the module activity shown in differentiated ES cells shown in Figure 4E, suggesting that strongly active Core and Myc modules, as well as an inactive PrC module, may characterize the pluripotent state of cells, such as ES and iPS cells.

Open in a separate window

Figure 5

Module activity in various cells

(A) Average gene expression values (log₂) (Sridharan et al., 2009) of ES cell modules (Core, Prc and Myc) and previously defined ESC-like module are tested in ES cells (ES), iPS cells (iPS), MEFs (MEF), and partial iPS cells (piPS). See also Figure S3D.

(B) Average gene expression values (log₂) (Bild et al., 2006) of each module tested in (A) upon induction of Myc in human epithelial cells (Myc induction) and in control cells (WT). Human orthologues of genes in three ES cell modules are tested (listed in Table S3) and data are represented as mean ± SEM (A–B).

Previous work has shown that piPS cells exist at an intermediate stage in the reprogramming process (Maherali et al., 2007). The endogeneous ES cell core regulators Oct4 and Nanog are not re-activated in piPS cells, whereas they are reactivated in fully reprogrammed iPS cells. To test if the ES cell modules we have defined are functionally separable in piPS cells, we analyzed ES module activity using gene expression data from piPS cells (Figure 5A and Figure S3D) (Sridharan et al., 2009). We found that the activity of the Myc module in piPS cells is comparable to that in ES cells and iPS cells, but the Core module is not re-activated in piPS cells. These data demonstrate that the regulatory modules defined in ES cells may be considered functionally separable units, not arbitrary subdivisions of the overall ES cell signature. Of particular note, the ESC-like module (Wong et al., 2008a) shows similar module activity to our Myc module, but not to the Core module in piPS cells.

ES cell module activity in cancer

Others have described ESC-like gene modules (Wong et al., 2008a) or ES-cell like gene expression signatures (Ben-Porath et al., 2008) that have been widely used in assessment of cancer gene signatures. With the three ES cell modules we defined as new analytical tools, we readdress the relatedness of ES cell and cancer gene signatures as a series of case studies. For analyses of human data, human orthologues of mouse genes were used (Table S3).

Myc induction does not activate the Core module in human epithelial cells

We tested ESC-like modules (both Core ESC-like gene and mouse ESC-like gene modules) (Wong et al., 2008a) and found that they behave similarly to our Myc module in various settings (Figure 4E, Figure 5A and data not shown). Since we observed that our defined Core and Myc modules can be functionally separated in piPS cells (Figure 5A), we examined whether the induction of Myc may activate the Core module in a different cellular context. It has been reported previously that the induction of Myc activates the ESC-like module in adult human epithelial cells (Wong et al., 2008a). As shown in Figure 5B, upon reanalysis of this dataset (Bild et al., 2006), we find that the Core module is not activated following Myc induction, whereas the Myc module is strongly represented. In addition, core factors in ES cells, such as Nanog and Oct4, are also not activated by Myc induction (Figure S4A). These observations confirm that the Myc and Core modules are functionally separable, and also support the view that the overall ES cell expression signature can be subdivided into functionally distinct units. Our refined analysis argues against the prior conclusion that Myc induction leads to activation of an ESC-like gene module in human epithelial cells (Wong et al., 2008a).

ES cell modules in mouse MLL myeloid leukemia models

We have assessed the relevance of our ES cell modules within a mouse model of acute myeloid leukemia (AML). Expression of MLL alleles leading to expression of fusion products, such as MLL-AF9, MLL-ENL, MLL-AF10, MLL-AF1p, and MLL-GAS7, initiates leukemia. MLL associated leukemia models in mice have served as platforms for purifying and examining the gene expression profiles of leukemia stem cells (LSCs; also called LIC, leukemia initiating cells) (Krivtsov et al., 2006). It has also been suggested that LSCs are present at a higher frequency in leukemic mice in which AML was initiated by MLL-ENL or MLL-AF9 as compared with MLL-AF10, MLL-AF1p, and MLL-GAS7 (Somervaille et al., 2009). We tested the activity of our defined modules in these leukemias. We first observed that the Core module is not active in any of the AMLs as compared to the Core module activity of a control group (Figure 6A). Moreover, we failed to detect an active Core module in AMLs demonstrated to have high LSC frequency (MLL-ENL and MLL-AF9) (Figure 6A). In contrast, we observed active Myc module expression in high frequency LSC AMLs (MLL-ENL and MLL-AF9), but not in low frequency LSC AMLs (MLL-AF10, MLL-AF1p, and MLL-GAS7) or control.

Open in a separate window

Figure 6

ES cell modules in mouse MLL myeloid leukemia models

(A) Average gene expression values (log₂) of ES cell modules and the previously defined ESC-like module are tested in various mouse models of acute myeloid leukemia (AML) initiated by MLL-AF9, MLL-ENL, MLL-AF10, MLL-AF1p, and MLL-GAS7 (Somervaille et al., 2009). (B) Average gene expression values (log₂) of each module are tested in a c-kit high MLL-AF10 leukemia cell population (MLL-AF10 c-kit high) and a c-kit low MLL-AF10 leukemia cell population (MLL-AF10 c-kit low) (Somervaille et al., 2009). Data are represented as mean ± SEM (A–B). See also Figure S4.

It has been reported that the previously defined ESC-like gene module (Wong et al., 2008a) is prominent in a MLL-AF10 leukemia cell population enriched for LSCs (c-kit high) as compared to c-kit low cells (Somervaille et al., 2009). As shown in Figure 6B, we observed a stronger Myc module activity in the LSC-enriched population. However, this cell population shows relatively inactive Core module activity. In both of the tests shown in Figure 6A and Figure 6B, we observed that the activity of the previously defined ESC-like gene module (Wong et al., 2008a) is similar to the activity of the Myc module rather than the Core module. If the gene expression findings are functionally relevant to self-renewal of LSCs, our findings undermine the notion that reactivation of an ESC-like pattern is critical for LSCs in this setting. In contrast, Myc module activity alone appears to correlate with LSC frequency in mouse AML models. Core module activity does not appear to be a major determinant of LSC frequency in AML.

Protein complex pulldown and mass-spectrometry

One-step affinity purification and protein complex identification using nuclear extracts from ES cell lines expressing BirA only (reference) or both BirA and biotin tagged proteins (sample) with streptavidin-agarose were performed as described previously (Kim et al., 2009; Wang et al., 2006). Further details are documented in Supplemental Data.

ChIP-chip

At least three biological replicates of ChIP and bioChIP reactions were performed for each factor as described previously (Kim et al., 2009; Kim et al., 2008). Detailed procedure and a list of antibodies used for native antibody ChIP reactions are available in Supplemental Data.

We thank Jennifer Trowbridge for critical reading of the manuscript, the Taplin Biological Mass Spectrometry Facility at Harvard Medical School for mass-spectrometry and peptide identification, and the Microarray Core Facility at the Dana Farber Cancer Institute for ChIP sample processing. The project described is partially supported by Award Number K99GM088384 to J.K. from the NIH/NIGMS. S.H.O. is an investigator of the Howard Hughes Medical Institute.