Activator-Mediator binding regulates Mediator-cofactor interactions

Abstract

The 26-subunit, 1.2 MDa human Mediator complex is essential for expression of perhaps all protein-coding genes. Activator binding triggers major structural shifts within Mediator, suggesting a straightforward means to spatially and temporally regulate Mediator activity. By using mass spectrometry (MudPIT) and other techniques, we have compared the subunit composition of Mediator in three different structural states: bound to the activator SREBP-1a, VP16, or an activator-free state. As expected, consensus Mediator subunits were similarly represented in each sample. However, we identify a set of cofactors that interact specifically with activator-bound but not activator-free Mediator, suggesting activator binding triggers new Mediator-cofactor interactions. Furthermore, MudPIT combined with biochemical assays reveals a nonoverlapping set of coregulatory factors associated with SREBP-Mediator vs. VP16-Mediator. These data define an expanded role for activators in regulating gene expression in humans and suggest that distinct, activator-induced structural shifts regulate Mediator function in gene-specific ways.

Transcriptional regulation is driven in large part by transcription factors: DNA-binding proteins that target specific regulatory sites within the genome. Different transcription factors (or activators) recognize different sequence elements via their DNA-binding domains, whereas distinct activation domains within transcription factors interact with one or perhaps several components of the transcriptional machinery. One of the main activator targets within the transcriptional machinery is the Mediator complex (1, 2). Direct activator-Mediator interactions are thought to recruit and stabilize Mediator at the promoter; however, EM analyses of Mediator bound to different activation domains have indicated that activators may be playing other regulatory roles (3). In particular, activator binding induces significant structural shifts within Mediator (4), which imply an additional means to regulate the human Mediator complex. The sheer size (approximately 320 × 180 × 160 Å) and shape of Mediator provides an enormous surface area for protein–protein interactions, and the global structural shifts induced by activator binding likely expose distinct motifs within the Mediator complex. Potentially, such structural shifts may activate Mediator by triggering protein–protein interactions at the promoter.

To test this hypothesis, we purified Mediator in three different structural states. In one instance, Mediator was purified bound to the activator SREBP-1a; in another, Mediator was purified bound to the activator VP16; Mediator was also purified in an activator-free state (without an activator bound). Relative to the activator-free conformation, Mediator will adopt significantly different structural states when bound to SREBP-1a vs. VP16 (4). To assess Mediator subunit composition and potential associated factors in each structural state, we utilized the multidimensional protein identification technology (MudPIT) methodology, which enables a comprehensive, unbiased assessment of protein composition within even highly complex samples (5). Our results provide clear evidence that activator-induced structural shifts trigger Mediator-cofactor interactions. Moreover, a subset of factors interact specifically with SREBP-Mediator but not VP16-Mediator, suggesting distinct activator-dependent structural shifts within Mediator direct gene-specific regulatory functions. These data indicate that activator binding can dictate subsequent Mediator-cofactor interactions, providing a straightforward means by which Mediator activity (i.e., transcription) can be controlled in a spatial and temporal fashion.

Results

The MudPIT method uses a two-dimensional liquid chromatography separation method coupled to a tandem mass spectrometer for detection of tryptic peptides. A major advantage of this technique is that protein samples are digested in solution, which minimizes sample loss that is typical for standard in-gel protein sample preparation methods. This feature is especially important for analysis of Mediator, given the low-abundance of this macromolecular complex in human cells.

Activator-Induced Structural Shifts Do Not Dissociate any Mediator Subunits.

In past studies, we have purified human Mediator by using a combination of ion exchange and affinity chromatography steps, followed by glycerol gradient sedimentation. This rigorous purification protocol yields Mediator complexes that are devoid of additional associated factors. The purpose of this work was to use MudPIT to address whether additional cofactors might stably interact with Mediator and, if so, whether these interactions were dependent upon activator-induced Mediator structural shifts. Consequently, we adopted a simplified purification scheme such that potential Mediator-associated factors could be identified (Fig. S1). Importantly, the simplified purification protocol still included a series of high-salt washes to remove weak, nonspecific interactions. To purify Mediator in an activator-free state, we isolated complexes from HeLa nuclear extract by using antibodies against the Med1 or CDK8 subunits. Activator-bound Mediator complexes were purified with affinity chromatography resins containing the activation domain of VP16 (residues 411–490) or SREBP-1a (residues 1–50). After elution, the activator-bound Mediator samples were further purified over a glycerol gradient, and Mediator-containing fractions (> 1 MDa in size) were combined for MudPIT analysis.

After analysis of Mediator samples with the MudPIT protocol, identified peptides were subjected to a rigorous 1% false discovery rate threshold; furthermore, peptides that could be assigned to multiple different proteins were eliminated with an isoform resolver algorithm (6). As shown in Table 1, all consensus Mediator subunits (7) were identified in each sample (SREBP-Mediator, VP16-Mediator, CDK8 IP, and Med1 IP), with one exception: Med26 was not observed in the CDK8 IP sample. This was expected upon the basis of past work that demonstrated a mutually exclusive association of CDK8 and Med26 within Mediator (4). Total spectral counts, defined as an MS/MS event identifying a peptide corresponding to a specific Mediator protein, are shown for each of the four different Mediator samples in Table 1. With a few exceptions, spectral counts were similar—typically within a 2- to 3-fold range—for each of the 32 consensus Mediator subunits (26 core Mediator subunits, plus the CDK8 submodule). Moreover, when all spectral counts were summed for activator-bound Mediator and activator-free Mediator (IPs), there was less than a 2-fold difference in total Mediator spectral counts. Importantly, this indicates the total amount of Mediator was similar within each of the four MudPIT samples analyzed. These data also demonstrate that, whereas activator binding alters the conformational state of Mediator, activator-induced structural shifts—at least those induced by SREBP or VP16—do not cause dissociation of any Mediator subunit.

Table 1.

Spectral counts for consensus Mediator subunits

Shown are the total number of ms/ms assignments to peptides within each Mediator subunit from each Mediator purification. Med1: Med1 IP sample; CDK8: CDK8 IP sample; SREBP-Mediator; VP16-Mediator. Note that Med2, Med3, and Med5 are not present in human Mediator. CDK8 submodule components are shown separately at the bottom of the table.

Activator Binding Triggers Mediator-Cofactor Interactions.

The data outlined in Table 1 represent expected results from MudPIT analysis of Mediator and are consistent with a previous study by the Conaway lab, in which activator-free Mediator samples (i.e., purified with an antibody) were examined by MudPIT (7). The motivation for this study, however, was to determine whether additional cofactors might preferentially associate with activator-bound Mediator complexes, which will adopt distinct structural states compared with activator-free Mediator. As shown in Table 2, the MudPIT data provide support for the hypothesis that activator-dependent structural shifts trigger Mediator-cofactor interactions. Spectral counts for identified cofactors were summed for activator-bound Mediator complexes (VP16 or SREBP) and for activator-free Mediator (CDK8 IP or Med1 IP). Cofactors identified that were greater than 4-fold enriched in activator-bound Mediator samples are shown in Table 2. Note that for most cofactors, the fold enrichment represents a minimum value because most proteins listed in Table 2 had no detectable peptides in the activator-free Mediator samples (CDK8 IP or Med1 IP). To ensure that potential Mediator-cofactor interactions were not occurring indirectly via any nucleic acid tether, we completed control experiments in which Mediator samples were treated with benzonase, a promiscuous endonuclease that cleaves both single- and double-stranded RNA or DNA. Benzonase treatment did not affect the presence of Mediator-associated factors, as determined by quantitative Western blot analysis. Furthermore, we performed SREBP-Mediator purifications by using a different affinity tag (MBP instead of GST) to confirm that the cofactors identified in Table 2 did not result from potential interactions with GST.

Table 2.

MudPIT identifies a subset of factors that associate with activator-bound Mediator

Spectral counts are given for proteins identified in SREBP and VP16-activation domain purifications and also for CDK8 and MED1 IP experiments. Fold enrichment represents the combined spectral counts for SREBP and VP16 purifications divided by the combined spectral counts for the CDK8 and MED1 IP experiments. In cases where no spectra were identified in the IP samples, the fold enrichment represents a minimum value.

It was possible that enrichment of some factors listed in Table 2, such as CBP, might result from a direct interaction with SREBP (8) and not Mediator itself. Our purification protocol (Fig. S1) includes a glycerol gradient sedimentation step that will separate complexes upon the basis of size; complexes of approximately 1.0 MDa and greater were selected for MudPIT analysis. Thus, a 350 kDa SREBP-CBP binary complex (for example) would be separated during the glycerol gradient step. However, as an additional means to ensure the factors identified in Table 2 associate with SREBP-Mediator and not SREBP itself, we completed an orthogonal purification scheme, outlined in Fig. 1A. This protocol first involved isolation of SREBP-Mediator complexes with an SREBP affinity resin, as before. Following elution from the resin, complexes were passed over an anti-Med1 or an anti-CDK8 antibody column to ensure that only Mediator complexes would be retained, whereas potential SREBP-CBP complexes (for example) would flow through. After a series of high-salt washes, proteins were eluted from the Mediator antibody resin (Fig. 1B). The presence or absence of additional, Mediator-associated factors was then examined by Western blot analysis.

Fig. 1.

Activator binding triggers new Mediator-cofactor interactions. (A) Orthogonal purification scheme for confirmation of Mediator-associated cofactors. (B) Silver-stained acrylamide gels representing various stages in the orthogonal purification: IP input, A/G-beads only negative control, and CDK8 or MED1 IP elutions. Subunit identities are listed at the right. (C) Western blots for various cofactors identified from the MudPIT analysis. Note most, but not all, cofactors were confirmed as Mediator-associated in this assay. LRP130, HADHA, SKIV2L2, and SnoN did not track with Mediator through the orthogonal purification, suggesting these factors likely interact directly with the SREBP activation domain and not Mediator.

As shown in Fig. 1C, the results from this orthogonal purification protocol confirms that many cofactors identified in Table 2 are in fact Mediator-associated, as they were retained by the second, Mediator-specific antibody affinity resin. Note, however, that the orthogonal purification procedure did identify a few cofactors (LRP130, HADHA, SKIV2L2, SnoN) that do not appear to interact directly with Mediator and likely associate with SREBP itself. These results highlight the effectiveness of the orthogonal purification strategy in confirming whether cofactors identified in Table 2 are truly Mediator-associated. The 12 factors probed following the orthogonal purification (Fig. 1C) represent a good cross-section of the cofactors identified in Table 2. That is, cofactors representing different activities (e.g. mRNA processing, acetyltransferase, H2A.Z exchange, etc.) were examined. As expected, alternate orthogonal purification protocols further confirmed the results shown in Fig. 1C. For example, orthogonal purification of SREBP-Mediator with an anti-ADA3L antibody resin similarly supported an SREBP-Mediator-SAGA interaction (Fig. S2). Although past reports have suggested a direct interaction between human Mediator and SAGA (9), these data implicate activator-induced structural shifts in promoting and/or stabilizing these interactions. Not every SAGA subunit was identified in Table 2, which likely results from its association with only a subset of SREBP-Mediator complexes (i.e., SAGA is substoichiometric relative to Mediator itself). Experiments in which Mediator was immunodepleted from extracts yielded data that also supported the orthogonal purification results; however, such experiments are limited by the fact that Mediator cannot be effectively removed in this way, even following six immunodepletion steps (SI Text).

Because 8 of 12 factors tested positively through the orthogonal purification protocol (Fig. 1C), it is evident that the majority of factors identified in Table 2 are likely Mediator-associated and are not observed because of potential direct interactions with the activation domain itself (VP16 or SREBP-1a). It remains possible that in some cases a tripartite interaction might occur, in which the cofactor might interact simultaneously with the activation domain and a surface exposed within activator-bound Mediator. In any case, the Mediator-cofactor interactions identified in Table 2 and further validated in Fig. 1C appear to be triggered by activator binding, and, because activator binding causes major structural shifts within Mediator, this observation suggests activator-induced structural shifts regulate subsequent Mediator-cofactor interactions.

Distinct Cofactors Associate with SREBP-Mediator vs. VP16-Mediator.

The sweeping structural shifts induced by activator-Mediator binding likely expose motifs for protein–protein interactions; in agreement with this, additional cofactors were observed to stably associate with Mediator upon activator binding. Because SREBP-Mediator adopts a distinct conformational state relative to VP16-Mediator, it was hypothesized that distinct cofactors might associate with SREBP-Mediator vs. VP16-Mediator. The data in Table 2 support this hypothesis, because there are substantial differences between factors associated with SREBP-Mediator vs. VP16-Mediator. To further probe potential activator-selective Mediator-cofactor interactions, we completed a series of quantitative immunoblotting experiments with SREBP- and VP16-Mediator samples (Fig. 2A). These samples were purified by using the same protocol used for MudPIT analysis (Fig. 2B) and the presence/absence of various polypeptides was examined by Western blot analysis. As shown in Fig. 2C, the data correlate precisely with the MudPIT results shown in Table 2. For example, TRRAP, GCN5L, and reptin were identified in both SREBP and VP16-Mediator samples by MudPIT (Table 2). Each protein was also detected in the SREBP-Mediator and VP16-Mediator samples by Western blot analysis, as shown in Fig. 2C. Similarly, the MudPIT data revealed no peptides corresponding to ATM, GCN1L1, ADA2B, or ADA3L in the VP16-Mediator sample, whereas these proteins were well represented in the SREBP-Mediator sample. In agreement with these data, quantitative Western blotting confirmed a significant enrichment of these cofactors in SREBP-Mediator fractions (Fig. 2C), whereas ATM, GCN1L1, ADA2B, and ADA3L were nearly undetectable in the VP16-Mediator sample. Note that equivalent amounts of Mediator were examined in each experiment, as shown by the Med15 immunoblotting experiments (Fig. 2C). Combined with the MudPIT analysis summarized in Table 2, the data in Fig. 2 provide strong evidence that Mediator-cofactor association can be activator-selective and that this selectivity is conferred by distinct activator-bound Mediator structural states.

Fig. 2.

Distinct cofactors associate with SREBP-Mediator vs. VP16-Mediator. (A) Silver-stained acrylamide gels showing glycerol gradient fractions from the purification protocols shown in B. Mediator-containing fractions are denoted by the red boxes. (C) Quantitative Western blots confirm MudPIT data. Mediator-associated factors probed in immunoblotting experiments are shown at left. Factors shown in black font were observed to be present in both the SREBP-Mediator and VP16-Mediator samples, whereas factors shown in blue font were observed only in the SREBP-Mediator sample by MudPIT.

MudPIT Analysis of CDK8-Mediator.

Mediator exists in at least two major forms in human cells: core Mediator and CDK8-Mediator. Core Mediator is devoid of the CDK8 submodule and contains the Med26 subunit, whereas CDK8-Mediator contains the CDK8 submodule (CDK8, Cyclin C, Med12, and Med13) but lacks Med26 (3). Thus, Med26 is specific to core Mediator, whereas the CDK8 submodule is specific for CDK8-Mediator. Note, however, that the CDK8 submodule can also exist as a stable entity on its own (10). Med1 represents a Mediator subunit that is shared between core Mediator and CDK8-Mediator; thus, Med1 IP samples will represent a mix of CDK8-Mediator and core Mediator. As expected, Med26 was not detected in the CDK8 IP sample, whereas every other consensus Mediator subunit was identified (Table 1). Additionally, no pol II subunits were detected in the CDK8 IP sample, whereas spectral counts for pol II subunits were abundant in each of the other Mediator samples (Table S1). To verify these results, Med1 and CDK8 IP samples were probed for Med26 and the pol II subunit Rpb1, again by using Med15 to normalize the two samples (Fig. S3A). Quantitative Western blotting showed greater than 8-fold enrichment of Med26 and greater than 16-fold enrichment of Rpb1 in the Med1 IP sample, with no detectable Rpb1, nor Med26, in the CDK8 IP sample (Fig. S3B). These results are consistent with past reports that indicated mutually exclusive CDK8/pol II association with Mediator (11, 12).

Interestingly, MudPIT analysis revealed a set of factors that associate specifically with CDK8-Mediator (Table 3); these factors were well-represented in the CDK8 IP sample but not detected in the Med1 IP or activator-bound Mediator samples.

Table 3.

Spectral counts for cofactors observed exclusively in CDK8 IP samples

Discussion

The results described here have broad implications for how gene expression is regulated in human cells. First, it is evident that activators serve roles in controlling gene expression that extend beyond simple recruitment of factors (e.g., Mediator) to the promoter. In fact, activators appear to regulate Mediator function by altering its conformational state, thereby controlling subsequent interactions with other regulatory cofactors. We anticipate that in some cases, cofactor association with Mediator will, in turn, modulate the activity of these factors, upon the basis of previous work that demonstrated the GCN5L acetyltransferase and the CDK8 kinase alter their activity toward chromatin substrates upon association with Mediator (10, 14). Second, our results define a mechanism by which the general transcription machinery, in particular the Mediator complex, might actually adopt different functions in distinct promoter contexts. Different activators help regulate different sets of genes and, intriguingly, different activators induce distinct structural shifts within the human Mediator complex (4, 15, 16). Significantly, we observe these structural differences translate into differences in factors associated with SREBP-Mediator or VP16-Mediator, providing a means for Mediator to adopt gene-specific functions (Fig. 3). Thus, these data suggest a role for Mediator in orchestrating the recruitment and/or exchange of coregulatory factors at gene promoters and enhancers. Third, the MudPIT data forecast expanded roles for Mediator in the control of human gene expression (see below).

Fig. 3.

A model that summarizes the results and implications of this study. Mediator-cofactor interactions do not occur in the activator-free state; rather, activator binding signals a shift in Mediator structure only when engaged at the promoter, providing a straightforward means by which Mediator activity can be controlled in a spatial and temporal fashion. Note that activator binding not only enables Mediator-cofactor interactions, but different activators, which induce different structural shifts within Mediator, trigger interaction with distinct sets of coregulatory factors, providing a mechanism by which Mediator can adopt activator-specific functionality.

Importantly, numerous Mediator-cofactor interactions identified in the MudPIT experiments have been functionally validated with in vitro or cell-based experiments. For example, TRRAP and GCN5L associate with activator-bound but not activator-free Mediator complexes. Our laboratory previously determined that TRRAP and GCN5L stably assemble within CDK8-Mediator and that this “T/G-Mediator” complex phosphorylates S10 and acetylates K14 within histone H3, a mark associated with some active genes. In fact, CDK8 and GCN5L function synergistically in this context, providing a mechanistic basis for GCN5L association with CDK8-Mediator (14). The MudPIT data also identify CBP/p300 and components within the SAGA complex as factors that associate with activator-bound Mediator. A functional cooperativity in gene activation between Mediator-SAGA or Mediator-CBP/p300 has been described in past studies (9, 17). As another example, the MudPIT data indicate an enrichment in P-TEFb (Cyclin T and CDK9) and AFF4 in the CDK8-Mediator IP sample but not in the Med1 IP or the SREBP- or VP16-Mediator samples. This result implies a functional cooperativity between the CDK8 submodule and P-TEFb. In support of this, recent studies suggest CDK8-Mediator coordinates P-TEFb loading and activity during the activation of serum-response genes in human cells (13). These examples define clear functional roles for cofactors identified by MudPIT analysis of Mediator. Investigating the functional significance of other MudPIT-identified factors will be the subject of future experiments.

The MudPIT methodology has been described as a “hypothesis-generating engine” because of its ability to comprehensively identify polypeptides in an unbiased manner (18). Accordingly, the MudPIT data suggest that Mediator might regulate gene expression in ways not previously appreciated. For example, factors that control mRNA processing (GCN1L1), histone H2A.Z exchange (pontin, reptin, p400), and chromatin architecture (Smc1A, Smc3, IQGAP1, nesprin-2, ZW10, HMMR) are strongly represented in the MudPIT data. Numerous enzymatic activities also associate with Mediator (e.g., GCN5L, CBP/p300, DNA-PK, ATM, RECQL5). These additional Mediator-associated factors are observed almost exclusively with activator-bound Mediator (VP16 or SREBP), whereas activator-free Mediator samples contained only consensus Mediator subunits without associated coregulatory proteins (although see CDK8 IP data, Table 3). That additional factors were observed to associate with Mediator upon activator binding provides strong evidence that activator-induced structural shifts trigger Mediator-cofactor interactions. The biological rationale for this is clear: A requirement for the activator in directing subsequent Mediator-cofactor interactions ensures these interactions are controlled in a spatial and temporal fashion. Indeed, such a strategy prevents Mediator-cofactor interactions from occurring when such interactions might be unproductive, such as when Mediator is not stably bound to the promoter.

Most factors enriched in activator-bound Mediator samples are in fact specific to SREBP-Mediator; beyond the consensus Mediator subunits, few additional factors are observed with VP16-Mediator. This may reflect the fact that VP16 is a viral activator; as such, VP16-Mediator likely evades common regulatory strategies by avoiding association with a host of coregulatory proteins. The factors identified that associate with SREBP-Mediator are almost certainly substoichiometric relative to Mediator itself; that is, these factors likely associate with a fraction of SREBP-bound Mediator complexes and should not be considered consensus Mediator subunits. Human SREBP-1a is important for expression of several dozen genes (19). Factors that associate with SREBP-Mediator (but not VP16-Mediator) might play specialized roles in regulating a subset of SREBP target genes, whereas other Mediator-associated factors (e.g., CBP/p300) clearly serve more general roles at a wide variety of genes. It will be important in future work to further define potential gene-specific functions for cofactors that associate with SREBP-Mediator. These experiments will require comparative analyses at many SREBP target genes that correlate factor recruitment and gene expression with activation and nuclear localization of SREBP-1a. It will also be informative to compare and contrast MudPIT data from Mediator bound to an array of different transcription factors, such as nuclear receptors or p53, that induce structural shifts within Mediator distinct from SREBP or VP16 (15, 16). Potentially, a subset of Mediator-associated factors will be unique to p53-Mediator (for example) and the identity of these factors might provide insight regarding activator-specific regulatory mechanisms.

As anticipated, MudPIT analysis of CDK8-Mediator yielded no spectral counts for Med26, pol II, or the pol II-associated factor Gdown1; however, the MudPIT data also revealed intriguing differences among pol II and Mediator subunits. Several pol II subunits (Rpb2, Rpb3, Rpb5, Rpb8) were vastly overrepresented in activator-bound versus activator-free Mediator samples. A distinct role for Med7 in activated transcription was also implicated by the MudPIT data, because Med7 was enriched 37-fold in activator-bound Mediator samples. These results might reflect changes in Mediator-pol II interactions that occur upon activator-Mediator binding; these subunits may also play key roles in activator-dependent transcription. Additional structural and mechanistic studies will be required to confirm this, yet it is interesting to note that activator-induced structural shifts within Mediator have been linked to activation of promoter-bound, stalled pol II complexes (16).

Methods

Mediator Purification.

Activator-bound Mediator was purified from HeLa nuclear extract by using GST-SREBP-1a (residues 1–50) or GST-VP16 (residues 411–490) immobilized to Glutathione-Sepharose beads (GE Life Sciences), as outlined in Fig. 2B. Activator-free Mediator was purified from HeLa NE by using an antibody affinity resin (Med1 or CDK8). See SI Text for additional information.

MudPIT Analysis of Mediator.

Mediator subunits and associated proteins were identified by using a modified MudPIT procedure (20) used by the Conaway laboratory to identify the consensus Mediator subunits (7). See SI Text for further information. A summary of all MudPIT data is shown in Table S2.