Using two alternative two-hybrid approaches, we have obtained a detailed protein–protein interaction network for 20 out of the 25 subunits of the yeast Mediator complex (Figure ). This map encompasses and extends the one that has been established previously using a pull-down approach (24
). We have found novel interactions that link the three modules of the core Mediator complex. In several cases, we have delineated interacting domains between subunits and provide evidence that the interactions are direct. In addition, two-hybrid interactions, co-immunoprecipitation and gel filtration experiments indicated that Med31 (Soh1) is associated with the Mediator complex in yeast. Investigating the putative Drosophila Mediator middle module counterpart, we showed here that several predicted metazoan homologues of yeast subunits are engaged in conserved interactions suggesting that, even though the primary sequences of Mediator subunits have extensively diverged, the overall structure of eukaryotic Mediator has been conserved during evolution.
Figure 8 Integrated interaction map of the yeast Mediator. (A) Connection map of Mediator subunits. The direct links between the different Mediator subunits found in this work, by Kang et al. (24–26), Ito et al. (28) and Uetz et al. (27) are combined (more ...)
In this work, we draw an interaction map of budding yeast Mediator, connecting most of its subunits (Figure ). Our results are in agreement with a general topology of Mediator composed of a head, middle and tail module (18
). New interactions were found to connect these head, middle and tail modules. Med14 (Rgr1) was previously shown to connect the tail and middle modules (19
). In this study, we found that Med14 (Rgr1) interacts with Med1 and Med10, thereby identifying its middle module partners. Our mapping data also revealed that Med5 (Nut1) interacts with both Med16 (Sin4) and Med1, thereby determining the position of Med5 (Nut1) within the Mediator complex. Moreover, we showed an interaction between Med21 (Srb7) and Med3 connecting the middle and tail module. Similarly, we identified a link between the middle and head modules through the Med21 (Srb7)–Med17 (Srb4) interaction. In addition to the contacts identified between the Cdk8 and middle module (24
), we found that the Cdk8 module is connected with the head module via an interaction between Med13 (Srb9) and Med17 (Srb4).
Since it has been reported that interactions with transcriptional regulators or the CTD of Rpb1 can change the conformation of the Mediator complex and alter the interaction between mediator subunits (38
), it should be noted that the interactions reported in this study may not occur all at the same time, but may be dependent on the transcriptional state of the Mediator complex. Such conformational changes have been proposed to play a pivotal role in regulating Mediator function (39
). In this light, it is interesting that most of the interactions we find between the individual submodules were only found by screening the genomic library, which was designed to contain fragments, rather than by directly testing the interactions between full-length proteins. Such submodule interactions are of interest for understanding the two different global conformations of Mediator and the observation that most of these are not picked up when examining full-length proteins indicates the possibility of a regulatory mechanism that involves masking certain domains. Whatever the mechanism, it is clear that the submodule interactions which are demonstrated here are of highest interest for further investigating how Mediator function is regulated through the conformational switch.
Med31 (Soh1) is a very well-conserved protein that co-purifies with metazoan Mediator (11
). In yeast, Med31 (Soh1) was genetically associated with transcription but was not found in highly purified Mediator complex (41
). Here, Med31 (Soh1) fragments were selected in two-hybrid screens with Med10 and Med21 (Srb7), in keeping with Ito et al
. data (28
) who used Med31 (Soh1) as bait and selected Med10 and Med21 as major interactants among others. To buttress the conclusion that Med31 (Soh1) is a bona fide yeast Mediator subunit, we showed that Med31 (Soh1) co-immunopurified with Med17 (Srb4) and vice versa
. Additionally, Med31 (Soh1) and Med17 (Srb4) co-eluted in a single high molecular weight fraction on a gel filtration column. Several possibilities could explain why Med31 (Soh1) was not previously identified as a yeast Mediator subunit. Med31 (Soh1) might be a labile subunit that is lost during purification. Alternatively, Med31 (Soh1) may have escaped mass spectrometric identification since tryptic digestion of the protein is expected to produce peptides under 500 Da or over 2000 Da, except for one, outside the usual window of detection of the technique.
It was previously proposed that the majority of the yeast Mediator proteins is conserved in eukaryotic organisms (10
). Significantly, some of the homologue assignations have been confirmed by the purification of Schizosaccharomyces pombe
). Similarly, a large human Mediator-like complex has been reported to contain homologues of yeast Med11, Med18 (Srb5), Med19 (Rox3) and Med22 (Srb6) (43
). Nevertheless, since the primary sequences of most Mediator subunits have widely diverged across the eukaryotic kingdom, their functional conservation could be questioned. To address these issues, we reasoned that if the yeast and Drosophila Mediator subunits interacted similarly in a two-hybrid assay, one could argue (i) that the homologue assignments were correctly predicted, (ii) that the interactions are likely direct since no bridging yeast proteins might interact in a stable manner with so-widely divergent proteins and (iii) that the structural organization of the subunits is conserved. The conservation of several interactions between the Drosophila and yeast middle module subunits thus strongly suggested that at least part of the Mediator middle module is structurally conserved across evolution. In addition, the detected interaction between the Med4 (TRAP36) and Med10 (Nut2) Drosophila proteins confirmed a direct contact between the yeast Med4 and Med10 proteins. Along the same line, in Caenorhabditis elegans
, a proteome-wide two-hybrid analysis indicated a conserved contact between the worm Med22 (MDT-22/ZK970.3) and Med11 (MDT-11/R144.9) homologues (44
We were unable to find interactions with Drosophila Med9 (CG5134), possibly due to the insertion of a hydrophobic linker between the Gal4 domains and the subunit. Nevertheless, a recent Drosophila proteome-wide screen detected a contact between the predicted fly Med9 and Med4 (45
) strongly suggesting that their respective assignation was correct. Our inability to detect a protein–protein interaction of the fly Med1 homologue (i.e. Trap220) probably does not stem from a false assignation. Indeed, the five conserved segments found in Med1 (10
) have been systematically found in the same order within putative ORFs from a larger spectrum of eukaryotic species (Henri-Marc Bourbon, unpublished observations). In addition, the formal possibility that the tested Drosophila Med1 is non-functional could be ruled out since the full-length cDNA used in the two-hybrid assays could fully rescue Trap220
loss-of-function mutants (Muriel Boube, unpublished data). At this stage, we rather favour the simple hypothesis that the Gal4–Med1 protein fusions were not appropriately folded, modified and/or expressed in yeast for interaction with its partners.
Only 9 out of 28 interactions were obtained both by the mating assay and by the library screening, indicating that each two-hybrid approach has its own bias. A number (10
) of subunit interactions were found in the mating assays but not in the library screens. In some cases, the FRYL2 library may not have been exhaustively screened because of low transformation efficiencies due to the presence of a toxic bait or because the relevant interaction was weak in comparison with that of other preys. Another possible caveat is that clones encompassing the interaction domains may be absent from the tested library. Conversely, some of the interactions observed by screening were not found in the corresponding mating assay. It is possible that a Mediator subunit inhibitory domain, be it for protein folding, stability, or contact, has to be removed to observe the interaction since part of the protein is lacking in the selected fragments when a discrepancy between mating and screening occurs.
Five Mediator subunits out of 25 did not interact with any other partner. Two of these, subunits, Med2 and Med15 (Gal11), behaved as activators in the two-hybrid system and hence could only be tested as GAD
fusions against the other subunits fused to the GBD
and were not screened with the library, decreasing the probability that we could detect a protein contact. Interestingly, the three remaining subunits, Cdk8, CycC and Med12 (Srb8), all belong to the Cdk8 module which has an inhibitory role, raising the possibility that their binding to the reporter gene promoter through the GAD
may have prevented its activation and thus the detection of an interaction. The lack of interaction of three of the four Cdk8 module subunits might also stem from poor representation of interacting fragments in the FRYL2 library, even though it has already been carefully prepared and successfully used in many other studies in addition to this one (32
). We found previously that fusions of entire high molecular weight proteins, like the two large subunits of Pol I, II or III, may be unable to select interacting protein fragments in screens or to interact when tested directly with their partners (33
) (Michel Werner, unpublished data). This observation might explain why we were unable to identify partners for Med15 (Gal11; 120 kDa) and Med12 (Srb8; 168 kDa). Nevertheless, we could find partners for 80% of Mediator subunits, a success rate that compares favourably with other approaches.
For the GST pull-down experiments, each possible pair within the head or middle module has been co-expressed in a baculovirus system (24
). Checking the interactions that were found by these biochemical techniques against two-hybrid approaches could clearly (i) confirm interactions, (ii) reveal potential biases (iii) lead to the discovery of new interesting interactions, and (iv) delineate interacting domains. Strikingly, 11 out of the 20 two-hybrid interactions concerning intra-head or intra-middle module proteins were supported by co-immunoprecipitation data. Only the co-immunoprecipitations of Med17 (Srb4) with Med19 (Rox3) or Med20 (Srb2) were not confirmed by our two-hybrid assays.
When using the two-hybrid system to look for yeast protein interactions, there is the formal possibility that a third partner might bridge the two proteins under investigation leading to activation of the reporter gene. Thus, a two-hybrid interaction cannot be taken as an evidence of a direct contact. In our protein–protein interaction map, 11 of the 20 intra-module connections confirmed the Kang et al
) co-immunoprecipitation data and are thus direct. To investigate the nature of the contacts between the protein pairs that have not been found using the GST pull-down assay, we devised a method taking advantage of the availability of families of truncated fragments. These were tested against their Mediator subunit partners, based on the reasoning that if a single protein–protein contact was lost at a time, we could infer that it was direct. Using this test, we could show that the Med4–Med9, Med4–Med21 (Srb7), Med7–Med9 and Med8–Med20 (Srb2) contacts were direct. In retrospect, it is interesting to note that all the interactions between RNA polymerase subunits, predicted using the two-hybrid system (33
), were indeed demonstrated to be direct in the crystallographic structure of Pol II (47
). It is thus likely that most of the Mediator contacts reported here behave in the same way.
A two-hybrid screen with a library of genomic fragments has a number of advantages even though it is not exhaustive. This approach allows the identification of short interaction domains that are able to fold independently. It is notorious that the Mediator subunits are difficult to produce at high levels, hampering their crystallographic analyses. The definition of the interaction domains might help in that respect. In addition, coupled with phylogenetic analyses, these domains may also be used as targets for directed mutagenesis experiments to decipher the functional significance of intra- or extra-module contacts, both in yeast and metazoans.