A molecular model of mammalian Mediator

Endogenous tagging of Mediator subunits in mouse CH12 cells¹⁴ allowed us to use an immunoaffinity purification approach to obtain Mediator fractions suitable for cryo-EM analysis of the complex (Supplementary Fig. 1 and Supplementary Table 1). We used state-of-the-art cryo-EM analysis to build on our published low-resolution (~6 Å) structural analysis of the mammalian Mediator complex¹⁴ and were able to obtain a cryo-EM map with an overall resolution of 4.0 Å (Fig. 1a, Supplementary Figs. 2 and 3, and Supplementary Table 2). Portions of the Head and Tail modules and MED14 reached a maximum resolution of ~3.5 Å (Fig. 1b). The Middle’s hook and the Head’s neck, known to interact with the RNA polymerase II (RNAPII) carboxy-terminal domain (CTD)^9,15 showed the highest conformational variability. Secondary structure elements were clearly resolved throughout the cryo-EM map and densities for bulky amino acid side chains were apparent throughout MED14, the lower portion of the Head module, and the Tail module. Following on our previous naming convention, we assigned all non-core/non-CKM subunits to the Tail module (Supplementary Table 3).

The 4.0 Å resolution mMED cryo-EM map and information from published structures of yeast Mediator^9,10,11 allowed us to determine the structures of 25 out of 26 non-kinase module mMED subunits, including 16 core subunit, 7 Tail subunits (MED15, MED16, MED24, MED27, MED28, MED29, and MED30) whose structure and precise location and interactions were previously unknown, and 2 Tail subunits (MED23 and MED25) whose structures were partially (MED25¹³) or completely (MED23¹²) known, but only as individual proteins outside the context of the entire Mediator. Focused refinement of densities at either end of the Middle module provided additional insight into the structure of the hook (MED14 N-terminus, MED10, and MED19) and the N-terminal portion of MED1. Combining all of this information resulted in a molecular model including 24 mMED subunits, and a partial MED1 model (Fig. 1c, Supplementary Movie 1 and Supplementary Table 3). This left MED26 as the only mMED subunit for which only general interaction information, but no detailed structure is available. Information about the structure of the four subunits in the Cdk8 kinase module (CKM) has come from studies in yeast¹⁶ and interaction of the CKM with mMED has been characterized by EM^16,17.

Several features of the mMED model are worth noting. First, the structure and organization of core mMED subunits is remarkably conserved from yeast to mammals, but yeast and mammalian core Mediators differ considerably in conformation. Second, a centrally positioned mammalian MED14 enables inter-module interactions, as observed in yeast, but mammalian MED14 has a much larger C-terminal portion that is heavily involved in Tail interactions. Third, core subunit domains (MED14 N-terminus, MED17 N-terminus, MED6 C-terminus) that cross-module boundaries to make inter-module connections that link the Middle’s knob and Head’s neck domains are conserved between yeast and mammals. Fourth, the structures of MED14 and each module show particular characteristics. The Head shows a highly integrated structure, with considerable intertwining of conserved subunits with metazoan-specific ones, and extensive interfaces with MED14 and the Tail. The Middle shows an extended structure, with component subunits that, with the exception of MED1, are almost exclusively helical, resulting in overall rigidity and the potential for conveying long-range structural rearrangements. The Middle’s interaction with other modules is limited to focused contacts. The Tail is divided into upper and lower portions¹⁴. Subunits in the lower Tail are large and mostly self-contained. In contrast, smaller subunits in the upper Tail have extended structures and large interfaces amongst themselves and with MED14 and the Head.

Structure and conformation of the mammalian Head module

The structure and subunit organization of the mammalian Head module are very similar to those of the yeast Head, with MED17 acting as the central scaffold for Head assembly (Fig. 2a). Mammalian MED17 is larger than its yeast counterpart due to the presence of an insertion (aa 384–419) and a ~110 residue C-terminal extension (aa 540–650) (Fig. 2b). Both of these mammalian-specific MED17 domains are involved in interactions with other subunits, with the insertion forming an extended interface with the Head’s MED18–MED20 subcomplex (Fig. 2c) and the longer C-terminus adopting an extended structure involved in interaction with MED14 and the Tail. While some Head subunits like MED18–MED20 are nearly identical in structure and conformation between yeast and mouse (Fig. 2d, top), others like MED8 adopt the same structure but show a different conformation (Fig. 2d, bottom). Mammalian Head subunits MED11 and MED22 also differ from their yeast counterparts in conformation and by having longer, well-ordered C-terminal helices (Fig. 2e). Reflecting the similarity and differences between corresponding subunits, the yeast and mammalian Head modules have the same overall structure but adopt a different conformation. This difference in conformation results from a change in the relative positions of the top (neck) and bottom (jaws) portions of the Head (Fig. 2f). explained by flexing at the connection between the two, which is solely composed of extended loop domains that allow all of the subunits that cross the interface to flex (Fig. 2f, inset).

a Overall mammalian Head module structure. b Comparing the secondary structure of yeast (S. pombe) and mammalian (M musculus) MED17 proteins highlights their similarity and reveals the presence of an insertion and a C-terminal extension in the mammalian protein. c Like its yeast counterpart, MED17 functions as the central scaffold subunit for the Head module and shows a similar overall structure. A mammalian-specific insertion interacts with MED18/MED20 (bottom, right), while a C-terminal extension, also mammalian-specific, adopts an extended, well-ordered structure (bottom, left). d The structure of mammalian MED18–MED20 is remarkably similar to that of their yeast counterparts (yeast in gray). The MED8 structure is also conserved, but the conformation of the yeast and mammalian proteins differ (yeast in gray). e The structures of mammalian MED11 and MED22 are mostly conserved (yeast in gray), but mammalian MED11 and MED22 have longer, well-ordered C-terminal α-helices that are involved in interaction with metazoan-specific upper Tail subunits. f Comparing the mammalian (red) and yeast (gray) Head modules shows a change in overall conformation due to flexing at the neck–jaws interface, which is formed by flexible loops in all subunits that cross the interface (inset).

Structure and conformation of the mammalian Middle module and MED14

The mammalian Middle is similar to the yeast Middle (Fig. 3a), with a notable difference being that the N-terminal portion of MED1 is much better ordered (likely due to strong contacts with the Tail) and partially resolved in the mMED cryo-EM map. The central portion of the mammalian Middle can be divided into top and bottom sections formed by long helical bundles formed, as seen in yeast, by MED7–MED21 (top) and MED4–MED9 (bottom) (Fig. 3b). MED31, the Mediator subunit showing the highest sequence conservation across eukaryotes, also shows high structural similarity between yeast and mammals (Fig. 3b, bottom panel). As observed for the Head, individual corresponding subunits are very conserved structurally, yet the overall conformation of the yeast and mammalian Middle modules differ. This is again due to flexing, in this case at the MED7–MED21/MED4–MED9 interface (Fig. 3c, see inset for details). Some flexibility of the Middle module structure was suggested by X-ray analysis of S. pombe Mediator¹¹.

a Overall mammalian Middle module structure, including a partial MED1 model. b The structure of MED7–MED21 (top), MED4–MED9 (middle) and MED31 (bottom) are mostly conserved between yeast and mammalian Mediators (yeast subunits in gray). c The conformation of the mammalian Middle (light blue) differs considerably from that of the yeast Middle (gray) due to flexing at the MED7–MED21/MED4–MED9 interface (see inset for details). d The structure of mammalian MED14’s N-terminal portion structure is remarkably similar to the corresponding portion of yeast MED14, but the mammalian subunit is considerably larger due to an extended C-terminus (highlighted by red rectangle). Elucidation of the MED14 structure was facilitated by considerable detail in the cryo-EM map (inset). e The mammalian MED14 C-terminus interacts extensively with the Tail module (left). The position of MED14’s extreme C-terminal portion (highlighted by red ellipse on the left) was confirmed by difference mapping of a MED14 Δ1206–1458 mMED mutant. A MED14 Δ1206–1458 mMED class average and the difference map obtained after subtracting an aligned mMED class average are shown in the top right. Difference map densities colored by standard deviation (as indicated in the color scale) and superimposed on a mMED class average are shown in the bottom right.

The distal portions of the Middle module are only partially ordered (6–10 Å) and were further analyzed through focused refinement. As anticipated based on localization by subunit deletion^10,14, MED1 forms the lower end of the Middle module and in mMED has extended contacts with MED24 in the Tail (Supplementary Fig. 4a). The very N-terminus of MED1 wraps around MED4–MED9 and the subunit is composed of alternating α-helical and β-sheet domains (Supplementary Figs. 4b and 5). The tertiary organization of MED1 is reminiscent of MED14 (see below) and might result in flexibility that would explain why the MED1 portion of the cryo-EM map is poorly resolved (~6–8 Å) in comparison to other regions. Nonetheless, the partial MED1 model is consistent with recently reported cross-linking mass spectrometry data¹⁸ and provides insight into the structure of this important nuclear receptor target in the context of the larger Mediator structure. Focused refinement of the knob/hook resulted in comparatively improved density for the domains (Supplementary Fig. 6), but they remained poorly resolved and molecular models for the corresponding subunits are informed by consideration of secondary structure similarity to their yeast counterparts and the published X-ray structure of the yeast Mediator knob and hook¹¹ (Supplementary Fig. 7).

The theme of subunit structure conservation continues with MED14. Sequence analysis predicted similarities between corresponding N-terminal portions of mammalian and yeast MED14¹⁴ and the two are in fact remarkably similar in both secondary structure and tertiary organization (Fig. 3d). The alternating pattern of α-helical and β-sheet domains seen in yeast Med14 is also present in mammalian MED14 and, in fact, that pattern continues into the much larger (mouse MED14 1454 aa, S. cerevisiae MED14 1082 aa, S pombe MED14 879 aa) C-terminal portion of the mammalian MED14 (Fig. 3d, highlighted by red rectangle), which is heavily involved in interactions with the large (~790 kDa) mammalian Tail module. Modeling of the extended and intricate structure of mammalian MED14 was made possible by the presence of considerable detail in the cryo-EM map (Fig. 3d, inset). Mammalian MED14 extends nearly all the way across mMED, with its N-terminus forming part of the hook at the top of the Middle module and its C-terminus located over >350 Å away at the interface between the upper and lower portions of the Tail (Fig. 3e).

Although sequence homology between yeast and mammalian Mediator subunits is generally low (<25% on average), the structures of individual subunits and modules are highly conserved. Flexing of the mammalian Head and Middle modules at their hinge regions results in a remarkable correspondence between the structures of the yeast (S pombe) and mammalian (M musculus) core Mediators.

Structure of the mammalian Tail module

Information about the structure of the Tail has been limited to approximate localization of Tail subunits in the context of low (16 Å S cerevisiae Mediator¹⁰) or intermediate (6 Å M musculus Mediator¹⁴) cryo-EM maps, a proposed S cerevisiae Tail molecular organization map based on integrative modeling¹⁹, tentative localization of part of the yeast Med27 C-terminus (in a 4 Å S pombe Mediator cryo-EM map⁹), and a recently published X-ray structure of human MED23¹². Our mMED molecular model (Fig. 1c) provides a detailed view of the entire Tail.

The upper Tail forms an extended connection between core Mediator (specifically the Head and C-terminal portion of MED14) and the lower Tail (Fig. 4a, left). The four subunits forming the upper Tail (MED27–30) are similar in size and structure. MED28–29–30 are all 180–200 aa long, and so is MED27 if a globular C-terminal domain (~100aa) is considered separately. Despite their structural similarity, upper Tail subunits could be distinguished from one another and identified based on differences in the length of specific helices and loops and the presence of bulky side-chain densities (Supplementary Figs. 8 and 9). The upper Tail subunits are organized in pairs (MED27–MED29 and MED28–MED30) to form a double bracket-like structure (Fig. 4a, right). As observed in S pombe⁹, the MED27 C-terminal globular domain sits between the jaws of the Head module, adding to the complexity of the Tail-core interface. In all four proteins, an N-terminal 2-helix coiled coil is followed by a short loop and a third α-helix arranged roughly perpendicular to the N-terminal coiled coil (Fig. 4b). The N-terminal coiled coils of MED27–29 contact the C-terminal portion of MED16 in the lower Tail and then wrap around the backside of the MED14 C-terminus towards the Mediator core. The α-helices that follow the N-terminal coiled coil in both MED27 and MED29 interdigitate with the MED28–30 coiled coils to form a 6-helix bundle. Finally, the C-terminal helices in MED28–30 move toward the Head jaws where they form a 4-helix bundle with C-terminal helices from MED11 and MED22. The helical bundles in the upper Tail are stabilized by extensive hydrophobic interactions (Fig. 4c).

As observed in S cerevisiae Mediator¹⁰, the mammalian lower Tail is organized around MED16, which establishes contacts with all other lower Tail subunits (Fig. 5a). MED16 has a bipartite structure (Fig. 5b), with its N-terminal portion forming a large β-propeller, as predicted¹⁹ (Fig. 5b, bottom right), and its C-terminal portion being entirely α-helical (Fig. 5b, left). The massive distal end of the lower Tail, abutting the MED16 β-propeller, is formed by MED23 and MED24, which have roughly homologous structure and organization (Fig. 5c). Nestled between MED23, MED24, and MED16 is the folded N-terminal von Willebrand domain of MED25²⁰ (aa 15–216, Fig. 5a, b). No density corresponding to the C-terminal portion of MED25 (accounting for over two-thirds of the protein) was detected in the mMED cryo-EM map, which is explained by the presence of a disordered loop (aa ~198–392) connecting the folded N-terminal and C-terminal domains (Fig. 5d, bottom).

a Overall lower Tail structure (upper Tail in light yellow). b MED16 structure with β-propeller N-terminal and α-helical C-terminal domains colored from blue to red (top left) and map-to-model comparisons (top right), and details of the N-terminal β-propeller domain (bottom). c Models for MED23 and MED24 (left) show remarkable similarity in overall organization of the subunits (subunit color changes from blue (N-terminus) to red (C-terminus)) and correspondence between the models and the cryo-EM map (right). d Structure of the N-terminal von Willebrand MED25 domain (top) that interacts with other lower Tail subunits, and map-to-model comparison (middle left). The MED25 C-terminal activator interaction domain (ACID, PDB 2L23) is connected to the von Willebrand domain by a long flexible loop (middle left and bottom) and was not detected in the cryo-EM map. e Secondary structure diagram for MED15 (top) showing that its N-terminal portion is expected to be mostly disordered. The C-terminal portion evident in the mMED cryo-EM map (middle) is highlighted by the dashed rectangle and starts with two short anti-parallel α-helices (bottom, left) connected to a folded domain at the very C-terminus by an extended loop. f Localization of the MED15 C-terminus by MBP labeling and difference mapping. A MED15-MBP mMED class average (left) and the difference map obtained after subtraction of an aligned mMED class average from it (right) are shown in the top row. The middle row shows difference map densities colored by standard deviation (as indicated in the color scale) and superimposed on a mMED class average. The bottom row is a diagram showing the position of the MED15 C-terminus in the overall mMED structure.

The last subunit in the mMED Tail, MED15, shows perhaps the most peculiar structure and arrangement. With the exception of a small N-terminal folded domain, the first ~530 MED15 residues are expected to be disordered (Fig. 5e, top) and were not detected in the mMED cryo-EM map. However, the C-terminal portion of MED15 was well resolved (Fig. 5e, middle). The first portion of MED15 visible in the cryo-EM map (residues 620–652) forms two short, roughly anti-parallel alpha helices (Fig. 5e, middle and bottom left). These helices are followed by a long loop (residues 654–674) that travels across the lower Tail module along the MED14 C-terminus to connect to a folded domain at the very C-terminus of MED15 (residues 678–787, Fig. 5e, middle and bottom right). This unusual and very extended organization of MED15 was confirmed by maltose-binding protein (MBP) labeling of the MED15 C-terminus (Fig. 5f). The folded domain at the MED15 C-terminus sits between the N-terminal portions of MED23 and MED24 (Supplementary Fig. 10a). Other subunit interactions between lower Tail subunits happen through a combination of charged, hydrophobic and cationic-π interactions (Supplementary Fig. 10b). Interestingly, peripheral subunits in the lower Tail are rather self-contained. For example, both MED23 and MED25 can be absent without compromising Tail integrity or its interaction with core mMED¹⁴. This could reflect relatively recent incorporation of these subunits into Mediator.

Disease-associated mMED mutations

The mMED model provides information about the location of known disease-associated mutations in Mediator subunits (Supplementary Fig. 11), which can provide insight into their effect. For example, the human L371P MED17 mutation is associated with postnatal onset microcephaly²¹. The corresponding mouse mutation, MED17 L369P, would sit near the end of a helix that interfaces with the helical bundle formed by MED11–MED22 (Head) and MED28–MED30 (upper Tail). The leucine to proline mutation would very likely alter the MED17 helix and could affect the conformational dynamics of the Head. Consistent with this hypothesis, the corresponding mutation in Sc Mediator (ScMED17 M504P) did not destabilize the Head module, but impaired Mediator–RNAPII interaction²². Similarly, decreased MED25 association with Mediator resulting from a MED25 Y39C mutation in the folded N-terminal MED25 domain that is linked to a severe genetic syndrome²³ is explained by our observation that the MED25 N-terminal domain interacts directly with other subunits in the lower Tail.

Tail-core interfaces in mammalian Mediator

The mammalian Tail has an extensive and intricate interface with core Mediator, wrapping around the entire bottom of the core (Fig. 6a). The upper Tail has a convoluted interface with Head subunits, with the C-terminal α-helices of metazoan-specific subunits MED28 and MED30 forming a helical bundle with the extended C-terminal helices in MED11 and MED22 (Fig. 6b). Additional complexity in the upper Tail–Head interface comes from the globular C-terminal domain of MED27, which interacts with MED18–MED20 and is close to the MED17 insertion domain (Fig. 6c). The upper Tail also interacts with MED14, as MED27–MED29 wrap around it to reach down towards MED16 and interdigitate with the two short, antiparallel helices formed by MED15 residues 620–652, which end up positioned between the MED27 and MED29 coiled coils (Fig. 6d). This is consistent with well-established interaction of MED15 with MED27 (Med3) and MED29 (Med2)^10,19,24,25. The well-ordered conformation of the extended MED15 loop is explained by extensive and specific interactions with MED14, with charged residues in both subunits aligned opposite to one another along the entire length of the loop (Fig. 6e, top right). In addition to extensive contacts with MED23 and MED24, the C-terminal folded MED15 domain also has an extended interface with MED14 (Fig. 6e, bottom right), and is contacted by the mammalian-specific C-terminal extension in MED17 (Fig. 6f). This very extended MED15 structure results in the subunit running all the way across the Tail from the MED23–24 N-terminal domains to the MED27–28–29–30 assembly on the opposite side, putting MED15 at the center of the core–Tail interface. Finally, crosslinking-mass spectrometry analysis of yeast Mediator pointed to interaction between MED1 and MED24 (yeast Med5)¹⁹ and the mMED structure provides additional insight into the MED1–MED24 interaction. Although we cannot generate a detailed model of the folded N-terminal portion of MED1, interpretation of the MED1 focused refinement map suggests that β-strands rich in hydrophobic residues formed by aa ~437–481 in the folded MED1 N-terminal domain interact with a hydrophobic patch on the surface of the MED24 N-terminus (Fig. 6g).

Tail effect on core mMED conformation

The mammalian Mediator structure shows high interconnectivity of component subunits, which stabilizes the conformation of most of the complex. Compared to yeast Mediator, the presence of metazoan-specific subunits that form the upper Tail, along with extensive subunit interactions across the entire mammalian Mediator, result in a more conformationally stable complex. This is evidenced by local resolution analysis of the mMED cryo-EM map (Fig. 1b), which points to a generally stable structure, with the notable exception of the Middle’s knob/hook and Head’s neck domains involved in interaction with RNAPII and the CKM (Fig. 1d). Interestingly, although individual yeast and mammalian core Mediator subunits have very similar structures, overall core conformation differs between the yeast and mammalian Mediators, primarily due to rearrangements in the Head and Middle modules facilitated by their intrinsic pliability.

For yeast Mediator there is strong evidence that changes in core conformation are necessary for formation of a Mediator-RNAPII holoenzyme⁹. We previously reported¹⁴ that release of the Middle–Tail interaction by deletion of MED1 (the only direct interaction between the Middle and Tail modules is the contact between Middle subunit MED1 and Tail subunit MED24) leaves the Tail unchanged but allows core mammalian Mediator to adopt a different conformation in which the CTD-binding gap between the knob and the neck (Fig. 1d) narrows. Importantly, this is accompanied by a considerable (~3-fold) increase in mMED interaction with RNAPII¹⁴. In mammalian Mediator, rigidity of the Tail structure and the nature of Tail–core interactions, both explained by the mMED molecular model, seem to enable the Tail to influence core mMED conformation. A rigid upper Tail that includes metazoan-specific subunits and displays a large interface with the Head and an extended MED14 C-terminus, keeps the lower Tail in a fixed position. In turn, attachment of the Middle to the Tail through the MED1–MED24 contact keeps the mammalian core Mediator in a specific conformation. Considering the effect of MED1 deletion on mMED core Mediator conformation and polymerase interaction suggests that a primary effect of the mammalian Mediator Tail could be to modulate Mediator function by biasing core Mediator conformation towards a state that limits interaction with RNAPII. Although we only obtained EM data for MED1-related effects, the mMED molecular model indicates that any change in Tail structure that decreased Tail rigidity or its interaction with the core, could potentially lead to changes in core conformation.