A Conserved Mediator Hinge Revealed in the Structure of the MED7 (cid:1) MED21 (Med7 (cid:1) Srb7) Heterodimer*

The Mediator of transcriptional regulation is the central coactivator that enables a response of RNA polymerase II (Pol II) to activators and repressors. We present the 3.0-Å crystal structure of a highly conserved part of the Mediator, the MED7 (cid:1) MED21 (Med7 (cid:1) Srb7) heterodimer. The structure is very extended, spanning one-third of the Mediator length and almost the diameter of Pol II. It shows a four-helix bundle domain and a coiled-coil protrusion connected by a flexible hinge. Four putative protein binding sites on the surface allow for assembly of the Mediator middle module and for binding of the conserved subunit MED6, which is shown to bridge to the Mediator head module. A flexible MED6 bridge and the MED7 (cid:1) MED21 hinge could account for changes in overall Mediator structure upon binding to Pol II or activators. Our results support the idea that transcription regulation involves conformational changes within the general machinery.

Regulation of eukaryotic mRNA transcription requires multiprotein coactivators, which transmit signals from gene-specific transcription factors to Pol II 1 (1). Over the last decade, evidence from many laboratories converged on the Mediator complex as a central Pol II coactivator (2)(3)(4). Mediator was discovered in the yeast Saccharomyces cerevisiae by its ability to enable activated transcription in an in vitro system containing Pol II and the general transcription factors (5)(6)(7). Genome-wide studies showed that Mediator is required for regulated transcription of the majority of yeast genes (8). Mediator also stimulates basal transcription (9,10). The yeast Mediator comprises 25 polypeptide subunits, of which 11 are essential and 22 are at least partially conserved in sequence throughout eukaryotes (11,12). Nine of the Mediator subunits are products of the srb genes, which were discovered in a genetic screen for suppressors of truncations of the C-terminal domain of the largest Pol II subunit (13,14). Four of the Srb proteins reside in an independent Cdk-cyclin subcomplex that is only found in a subpopulation of Mediator complexes.
Mediator promotes initiation complex assembly through activator-Mediator, Mediator-Pol II, and Mediator-general transcription factor contacts (15). Mediator also stimulates the kinase activity of the general transcription factor TFIIH, which phosphorylates the Pol II C-terminal domain, apparently triggering the transition from transcription initiation to elongation (6). Mediator is recruited to active genes ahead of Pol II (16) and is thought to remain near the promoter to facilitate transcription reinitiation (17,18). Electron microscopy shows that Mediator undergoes strong structural changes upon interaction with Pol II (19 -21) and with transcription activators (22) that are obvious even at 35 Å resolution (21). The causes and consequences of these changes are, however, poorly understood, and the molecular mechanism of Mediator remains enigmatic mainly because of a complete lack of detailed structural information.
The Mediator subunit architecture was inferred from biochemical, genetic, and electron microscopic studies. Biochemical studies defined three submodules of Mediator, the MED17 (Srb4), the MED9/10, and the MED15 (Gal11) submodules (23), which were tentatively correlated with three density lobes in electron microscopic images, termed the head, middle, and tail module, respectively (24). The head and middle modules form a core Mediator that can be isolated from yeast (17). Core Mediator enables 4-fold activation of transcription in nuclear extracts compared with 18-fold activation for the complete Mediator. The middle module is the most conserved part of Mediator and comprises subunits MED7, MED21, MED10, MED1, MED4, MED9, and possibly MED31. With the exception of MED31 (Soh1) (25), MED7 and MED21 show the highest degree of sequence homology of all core Mediator subunits (40 and 45% between yeast and human, respectively). The high conservation of MED7 and MED21 is reflected in their essential function in yeast (26,27) and in a requirement for MED21 in mouse development (28).
To explore the general Mediator mechanism, we have determined the structure of the highly conserved heterodimeric MED7⅐MED21 subcomplex in two different conformational states and have analyzed the multiple direct protein interactions of this subcomplex in vitro. We show that highly conserved elements confer flexibility to the Mediator complex that may be important for function.

EXPERIMENTAL PROCEDURES
Cloning-The genes of MED10His and MED4His were cloned into pET21b vector (Novagen) using the restriction sites NdeI and NotI so that a C-terminal hexahistidine tag (His) was introduced. For bicistronic expression of MED7 with MED21His, MED4 with MED21His, and MED10 with MED21His, the genes of MED7, MED4, and MED10 were cloned into the pET21b vector using the restriction sites NheI and EcoRI. The gene of MED21 was then inserted together with a second ribosomal binding site at the 5Ј region of the MED21 gene as described (29). The sequence of the primer was GGACGCGTCGACAATAATTTT-* This work is supported by the Deutsche Forschungsgemeinschaft, the Fonds der Chemischen Industrie, and the EMBO Young Investigator Programme. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The GTTTAACTTTAAGAAGGAGATATACATATGACAGATAGAGATTAA-CACAATTGC (the start of the MED21 open reading frame is underlined). The restriction sites used were SalI and NotI. The genes for variants MED7⌬N⌬C⅐MED21His (comprised of residues 102-205 of MED7) and MED7⌬N⌬C⅐MED21 were cloned in the same way in the pET24b vector, which contains a kanamycin resistance cassette. For MED7⌬N⌬C⅐MED21 the restriction site XhoI was used instead of NotI. MED7⌬N⌬C⅐MED21⌬C (comprising residues 1-132 of MED21) was cloned in pet21b and contains methionine mutations (see x-ray structure determination). Glutathione S-transferase (GST)-MED17⌬N, comprising residues 241-688 of yeast MED17 and an N-terminal GST tag, was cloned into the BamHI site of pGEX-3x (Amersham Biosciences). The gene of MED6 and an additional ribosomal binding site at its 5Ј-region were inserted into the Xma site, fused to a C-terminal His tag. Genomic DNA was used as a template for all constructs except for MED17, where a MED17-containing plasmid was used (30). MED6His was obtained from N. Lehming (31).
Protein Expression and Purification-All proteins were expressed for 16 h at 20°C in Escherichia coli BL21 DE3 (Stratagene). Heterodimeric complexes were coexpressed from a single bicistronic plasmid. For coexpression of a third protein, a second plasmid was cotransformed that contains the appropriate gene and a different antibiotic resistance (ampicillin). For protein purification, cell pellets were thawed and lysed by sonication in buffer A (150 mM NaCl, 50 mM Tris, pH 8.0, 10 mM ␤-mercaptoethanol). After centrifugation, the supernatant was loaded onto a Ni-NTA column (Qiagen) equilibrated with buffer A. After washing with 30 ml of buffer A, bound proteins were eluted with buffer A containing 200 mM imidazole. All complexes except MED6His⅐ MED7⌬N⌬C⅐MED21 and GSTMED17⌬N⅐MED6His were further purified by anion exchange chromatography (Mono Q, Amersham Biosciences). The column was equilibrated with buffer B (50 mM NaCl, 20 mM Tris, pH 8.0, 5 mM dithiothreitol), and the proteins were eluted with a linear gradient of 20 column volumes from 50 mM to 1 M NaCl in buffer B. After concentration, the sample was applied to a Superose-6 high resolution gel filtration column (Amersham Biosciences) equilibrated with buffer C (150 mM NaCl, 20 mM Bicine, pH 8.5, 5 mM dithiothreitol). Purification of the triple methionine mutant was as for MED7⌬N⌬C⅐ MED21His, except for the Ni-NTA purification step. Instead, the mutant was precipitated with 35% saturated ammonium sulfate solution. For crystallization, the MED7⌬N⌬C⅐MED21His peak fraction was concentrated to 16 mg/ml. MED6His⅐MED7⌬N⌬C⅐MED21 was further purified by ammonium sulfate precipitation as above followed by gel filtration as above. GSTMED17⌬N⅐MED6His was purified by sequential Ni-NTA and GST affinity chromatography according to the manufacturer (CL4B, Amersham Biosciences). Bands of copurified proteins were generally verified by Edman sequencing.
X-ray Structure Determination-All diffraction data were collected at the Swiss Light Source, Villigen, Switzerland. For MAD phasing, three additional methionines were introduced at positions of conserved hydrophobic residues in MED21⌬C (L5M/L119M/L125M). Selenomethionine was incorporated as described (32). MAD experiments were performed on selenomethionine-labeled crystals, and diffraction data were processed with DENZO and SCALEPACK (33) (see Table I). Four selenium sites were correctly identified with program SOLVE (34) and refined using SHARP (35). Three additional selenium peaks were detected in residual electron density maps with SHARP. Phasing with SHARP using seven consistent peaks led to an electron density map that revealed distinct ␣-helices. A total of 10 methionine residues are present in the amino acid sequences and were ordered in the electron density map, except the two N-terminal methionines and Met-42 of MED21. Using the selenium sites as sequence markers, most of MED7 and MED21 could be built into the electron density map at 3.6 Å of resolution with program O (36). This initial model was repositioned in the unit cell of the native crystal by rigid body refinement with CNS (37). Model-phased maps at the final resolution of 3.0 Å allowed us to complete and refine the structure with CNS to a free R-factor of 28.9% (see Table I). The second crystal form, belonging to space group C222 1 , was phased using the 4-helix bundle of the refined MED7⅐MED21 structure. A difference electron density map phased with the bundle domain showed locations of the coiled-coil protrusions in the two heterodimers of the asymmetric unit that deviated substantially from that observed in the original structure. The coiled-coil region was adjusted. An extended loop between MED21 helices ␣1 and ␣2 that was disordered in the P4 3 22 crystal form was added as polyalanine in one of two heterodimers in the asymmetric unit, and the resulting model was refined to a free R-factor of 31.4% (see Table I). The MED21 loop ␣1-␣2 is better ordered in the C222 1 crystal form since it forms a crystal contact that is not observed in the P4 3 22 crystal form. In both refined structures, none of the residues fall in disallowed regions of the Ramachandran plot (see Table I).

Subcomplex Mapping-Individually expressed recombinant
Mediator subunits are generally insoluble, explaining the current lack of Mediator subunit structures. Insolubility apparently results from a loss of structural integrity when subunits are outside their natural multiprotein context. To overcome this obstacle, we coexpressed the highly conserved and essential S. cerevisiae Mediator subunits MED7 and MED21 in E. coli with the use of a bicistronic vector (compare "Experimental Procedures"). A stable MED7⅐MED21 complex was purified and subjected to partial proteolysis to probe for flexible regions that may interfere with crystallization. Chymotrypsin treatment resulted in the removal of 101 poorly conserved N-terminal residues of MED7 (Fig. 1). A corresponding variant MED7⌬N (MED7 residues 102-222) still formed a stable complex with MED21 after coexpression. Proteinase K treatment of the purified MED7⌬N⅐MED21 complex resulted in cleavage of a short C-terminal portion of MED7. A subsequently prepared variant MED7⌬N⌬C (residues 102-205) still bound MED21 strongly. This variant comprises only the highly conserved region of MED7, which shows 59% sequence homology between the yeast and human proteins. Except for a short C-terminal truncation, MED21 remained stable in all proteolysis experiments. In contrast, individually expressed MED21 was readily cleaved before residues 38 and 76 (Fig. 1B). Therefore, MED21 is protected from degradation upon MED7 binding. Taken together, iterative proteolysis and truncation of coexpressed and copurified subunits allowed us to map a stable subcomplex, and this approach may be used to obtain potentially crystallizable portions of other multiprotein complexes.
Structure Determination-The preparations of the MED7⌬N⌬C⅐MED21 subcomplex were monodisperse and homogeneous according to dynamic light scattering and size exclusion chromatography. The subcomplex formed crystals that grew to a maximum size of 0.6 ϫ 0.3 ϫ 0.3 mm and comprise two heterodimers per asymmetric unit (space group C222 1 ). Despite the large size of the crystals, diffraction extended only to 3.3 Å in favorable cases, most likely because of the very high solvent content of 78%. To increase the resolution limit of the crystals, we truncated eight non-conserved amino acid residues from the MED21 C terminus (variant MED21⌬C, residues 1-132, Fig. 1B). The resulting complex MED7⌬N⌬C⅐MED21⌬C formed crystals that adopted space group P4 3 22. These crystals were slightly more stable, and could be used for structure determination. The structure was solved with MAD data from a selenomethionine-substituted crystal and was refined to a free R-factor of 28.9% against native diffraction data extending to 3.0 Å of resolution ("Experimental Procedures," Table I). We used the refined structure of the MED7⌬N⌬C⅐MED21⌬C heterodimer to solve the structure in the initial crystal form (space group C222 1 ) by molecular replacement. We refined this structure to a free R-factor of 31.4% with data extending to 3.3 Å of resolution (Table I).
Elongated Two-domain Structure-MED7 and MED21 form a very elongated heterodimer of purely helical structure (Fig. 1C). The heterodimer extends over 110 Å, corresponding to 1 ⁄3 of the Mediator length (21) and amounting almost to the diameter of Pol II (39) (Fig. 1C). This elongated shape is consistent with an unusually short retention in size exclusion chromatography (Fig. 2). The retention of the MED7⅐MED21 species lies between that of the standard proteins aldolase (157 kDa) and catalase (230 kDa) and may be explained by formation of tetramers (dimers of heterodimers). Two different types of tetramers are observed in the crystal (Fig. 3,  C-D), which show a computed (40) Stokes radius of 42-44 Å. This is in good agreement with the Stokes radii for aldolase and catalase (47 and 49 Å, respectively), whereas the Stokes radius of the MED7⅐MED21 heterodimer is only 31 Å. Thus, the complex apparently forms stable tetramers in solution. Dynamic light scattering also indicates a molecular weight well above that of the heterodimer, but the data quality did not allow for quantification.
MED7 and MED21 show no sequence similarity, but they have the same structural organization, comprising three extended helices. MED7 and MED21 tightly pack against each other, forming a heterodimer with two domains. A 4-helix bundle domain is formed by the two N-terminal ␣-helices of each subunit, and the C-terminal helices of the two subunits form a long coiled-coil protrusion (Fig. 1C). A "hinge" region connects the bundle domain to the coiled-coil protrusion.
A Conserved Flexible Hinge-Comparison of the MED7⅐MED21 structure in the two crystal forms reveals that the coiled-coil protrusion can undergo a hinge movement with respect to the bundle domain, giving rise to a 10-Å displacement of the C-terminal end of the protrusion (Fig. 1D). This repositioning of the protrusion is accommodated by a slight conformational adjustment in the hinge region. Thus, the MED7⅐MED21 heterodimer has an intrinsic flexibility that allows for a relative repositioning of the two domains. This flexibility is not due to the short C-terminal truncation of MED21 in the C222 1 crystal form since the two chemically identical heterodimers in the asymmetric unit also show two different conformations indicating the same hinge motion and since the truncated residues are not involved in packing interactions in the C222 1 crystal form. The high sequence conservation of amino acid residues in the hinge region of MED7 and MED21 (Fig. 1B) strongly suggests that the observed flexibility is functionally significant. A hinge movement is also predicted by molecular dynamics simulation with the Dynamite server (dynamite.biop.ox.ac.uk/dynamite).
MED7-MED21 Interaction-MED7 and MED21 form a continuous hydrophobic interface, which explains why subunits that are individually expressed in E. coli are structurally unstable and either poorly soluble or easily degraded. The hydrophobic residues in the MED7-MED21 interface along the coiled-coil protrusion are mostly conserved throughout eukaryotes and show a spacing typical for coiled coils (MED7 residues Leu-175, Leu-178, Leu-182, Ile-189, Ile-192; MED21 residues Ile-94, Leu-97, Leu-101, Val-104, Ala-111) (Figs. 1B and 2). In MED7 the regular spacing of the hydrophobic residues is discontinued only at residue Lys-185. The regular pattern in MED21 is similar, except for residues Gln-90, Lys-K108, and Lys-115, which, however, also participate in foldstabilizing interactions. The conserved MED21 residue Gln-90 forms hydrogen bonds with Glu-172 and Arg-171 of MED7 and may contribute to the specificity of the heterodimeric interac-

FIG. 2. Size exclusion chromatography.
The MED7⅐MED21 complex shows an unusually short retention in size exclusion chromatography. The elution profile of MED7⅐MED21 on a Superose-6 column (Amersham Biosciences) is overlaid to those of standard proteins (thyroglobulin, ferritin, catalase, and aldolase). Protein elution was monitored by UV absorption at 280 nm. For details see "Experimental Procedures." mAU, milliabsorbance units. tion. Given the high degree of sequence conservation of the crystallized regions, the MED7⅐MED21 structure must essentially be the same in all species. This conservation in structure explains why a chimeric MED21 protein consisting of the two human N-terminal helices (corresponding to yeast residues 1-81) and the yeast C-terminal helix (residues 82-140) is functional in vivo (41) and why truncation of the MED21interacting region of MED7 is lethal in Caenorhabditis elegans (42).
A Unique Leucine Zipper-The coiled-coil protrusion resembles a canonical leucine zipper as it occurs in transcription factors like c-Jun (43) or Gcn4 (44,45). The two helices of the coiled-coil separate at their C-terminal end. The inner side of this open end of the zipper is conserved and hydrophobic (MED21 residues Leu-118, Val-122, Ile-126, and Phe-129; MED7 residues Leu-203 and Val-199) and, thus, chemically different from the corresponding region in leucine zipper-containing transcription factors, which show exposed polar or charged residues for DNA binding in this region. Instead of binding DNA, the open end of the MED7⅐MED21 coiled-coil may interact with other Mediator subunits. In the crystal two heterodimers pack against each other via their coiled-coil ends (Fig. 3C), indicating that the open end of the coiled-coil may allow for protein interactions.
Conserved Interactions within the Middle Module-To map direct protein-protein interactions of the MED7⅐MED21 complex with other subunits in the Mediator middle module, we tested for copurification of subunits after their coexpression in E. coli (Fig. 4A). Such copurification can successfully map strong and specific direct protein-protein interactions, as demonstrated in our structure determination of the MED7⅐MED21 complex. The copurification assay is very stringent, because many different nonspecific competitor proteins are present in the E. coli lysate, because the stoichiometry of the complexes can be estimated with Coomassie-stained gels, and because the protein-protein complexes must persist over several copurification steps even when elevated salt concentrations of 600 mM NaCl are used. E. coli was cotransformed with two plasmids, a bicistronic plasmid expressing the MED7⅐MED21 heterodimer and a second plasmid with a different antibiotic resistance, expressing a third subunit. We found that the MED7⅐MED21 heterodimer strongly binds MED10 and MED4, which show 39 and 33% sequence homology between yeast and human, respectively, and are the most conserved core Mediator subunits aside from MED7, MED21, MED31 (Soh1), and MED6 (Fig. 4A). In contrast, the other two subunits of the middle module, MED1 or MED9, do not copurify with the MED7⅐MED21 complex. To test if MED21 alone is sufficient for the interactions with MED10 or MED4, we constructed bicistronic vectors for coexpression. These experiments revealed that MED21 alone is capable of binding MED10 or MED4 (Fig. 4B).
MED6 Bridges the Two Mediator Core Modules-We could additionally demonstrate that the MED7⅐MED21 heterodimer binds directly to MED6, which shows 34% sequence homology between yeast and human (Fig. 4C). Although MED7, MED21, MED4, and MED10 are all subunits of the middle module, MED6 is an integral part of the head module (46), suggesting that MED6 bridges between these two modules. To test if MED6 binds directly to MED17 (Srb4), the architectural subunit of the head module (47), we tagged MED6 with a Cterminal hexahistidine tag (His), fused MED17 to a N-terminal GST tag, and coexpressed the two subunits from a bicistronic vector. We could copurify the two subunits in two subsequent affinity chromatography steps using a Ni-NTA and a glutathione column (Fig. 4C). Successful purification of the complex was independent of the order of the affinity columns. The weakly conserved N-terminal part of MED17 and the nonconserved C-terminal part of MED6 are not required for binding since truncated variants of MED17 (residues 241-688) and MED6 (residues 1-214) were sufficient for the interaction. Our results are consistent with a functional interaction between MED17 and MED6 observed previously (46). The results are further consistent with a very recent study of Mediator subunit interactions by yeast two-hybrid analysis (48). In conclusion, MED6 physically bridges between the two Mediator core modules, interacting with MED17 in the head module and with the MED7⅐MED21 heterodimer in the middle module.

DISCUSSION
Conserved Mediator Core Architecture-Our data show that the MED7⅐MED21 heterodimer tightly binds to other highly conserved Mediator subunits via its evolutionarily conserved regions and, thus, plays a central architectural role within Mediator. Our findings are consistent with published data on Mediator subunit interactions, which stem from coexpression of subunits in insect cells (23), coimmunoprecipitation (46), the split ubiquitin assay (49), and from yeast two-hybrid analysis (50). These studies additionally showed that the remaining subunits of the middle module, MED1 and MED9, are both bound by MED7, and MED1 is also bound by MED4. Because we did not detect MED1-MED7 or MED9-MED7 interaction with the N-terminally truncated MED7 variant, this indicates that MED1 and MED9 bind to the MED7 N-terminal region. Taken together, these results establish the MED7⅐MED21 complex as the conserved assembly scaffold of the middle module (Fig. 5). Sequence-based predictions of secondary structure (51) and coiled-coils (52) strongly suggest that subunits MED10 and MED4 are largely helical and form coiled-coils that may mediate the strong subunit interactions observed here and may span large distances on the Pol II surface.
Analysis of the molecular surface of the MED7⅐MED21 heterodimer reveals that the majority of the surface is hydrophobic, consistent with the observed extended interactions with other subunits. Four conserved surface patches (Fig. 3B) may serve as protein interaction sites and befit the architectural role of the MED7⅐MED21 heterodimer. Patch 1 and 2 are near the hinge region. Patch 2 includes the highly conserved N-terminal helix ␣1 of MED21 (residues Arg-4, Gln-7, Leu-8) and a part of MED21 helix ␣2 (residues Leu-76, Ser-79, Leu-80, Fig. 3). Because the highly conserved 7 Nterminal MED21 residues are required for MED6 binding (49), we propose that patch 2 constitutes the MED6 binding site. Patches 1, 3, and 4 may bind to subunits of the middle module but may also be involved in interactions with Pol II. Patches 3 and 4 are involved in crystal contacts. Patch 3 corresponds to the open end of the coiled-coil that stacks against a neighboring coiled-coil in the crystal (Fig. 3C), and patch 4 mediates dimerization of bundle domains of neighboring MED7⅐MED21 heterodimers in the crystal (Fig. 3D). The resulting two types of tetramers in the crystal do not show additional conserved surface patches (Fig. 3, E-F). Taken together, the high conservation of the MED7⅐MED21 heterodimer, its many interactions with conserved subunits, and its extended conserved hydrophobic surface patches all indicate that the structural architecture of the central region of the core Mediator is the same in all species.
Conserved Hinges and Mediator Function-Electron microscopy of free yeast Mediator and Mediator bound to Pol II revealed a dramatic structural rearrangement (20,21). Whereas free Mediator is relatively compact, it becomes extended upon Pol II binding and is wrapped around the polymerase surface in the Pol II-Mediator complex (Fig. 5A). This transition apparently involves a large change in the relative position of the Mediator middle and head modules (21). Our data suggest that the MED6 subunit plays a central role in this process because it bridges between the two modules of the core Mediator. Secondary structure prediction for MED6 reveals extended loop regions and helices with low probability (51), indicating a strong intrinsic flexibility for MED6. MED6 may form a conserved flexible connection between the head and middle modules. The bridging role of MED6 is relevant for Mediator function in vivo. A point mutation in MED17 suppresses temperature-sensitive mutations in the N-terminal region of MED6 that weaken the interaction with MED17 (54). The interaction of MED21 with MED6 is apparently essential in vivo, as a deletion mutant of yeast lacking the 15 N-terminal residues of MED21 is not viable (31).
Strong structural changes have also been observed in the mammalian Mediator coactivators CRSP and ARC upon binding of activator proteins (22,38). In addition to the MED6 hinge, the intrinsic flexibility of the MED7⅐MED21 heterodimer may account for these conformational changes. The hinge region between the two domains of the heterodimer may allow for flexibility within the middle module, because the relative locations of subunits bound to surface patch 3 on the coiled-coil and patches 1, 2, and 4 on the bundle domain can change. Binding of MED6 to the hinge region of the MED7⅐MED21 heterodimer may restrict or coordinate hinge motions. The repositioning of different parts of the core Mediator, enabled by conserved hinges as suggested here and triggered by the interaction with various partners, may be crucial for Mediator function.  5. Mediator architecture. A, outline of Mediator bound to Pol II according to electron microscopy (21). The suggested head, middle, and tail modules are indicated. B, Mediator subunit architecture. All depicted subunits are at least partially conserved throughout eukaryotes, except MED2 and MED3. The percentage of sequence homology between yeast and human subunits is indicated for the well conserved Mediator subunits. The four-subunit kinase module and the putative Mediator subunit MED5 have been omitted. For size comparison, the structure of the MED7⅐MED21 complex and the Pol II-Mediator complex in A are drawn to scale. Black rectangles indicate direct protein-protein interactions detected in this study.