Multivalent and Bidirectional Binding of Transcriptional Transactivation Domains to the MED25 Coactivator

Abstract

The human mediator subunit MED25 acts as a coactivator that binds the transcriptional activation domains (TADs) present in various cellular and viral gene-specific transcription factors. Previous studies, including on NMR measurements and site-directed mutagenesis, have only yielded low-resolution models that are difficult to refine further by experimental means. Here, we apply computational molecular dynamics simulations to study the interactions of two different TADs from the human transcription factor ETV5 (ERM) and herpes virus VP16-H1 with MED25. Like other well-studied coactivator-TAD complexes, the interactions of these intrinsically disordered domains with the coactivator surface are temporary and highly dynamic (‘fuzzy’). Due to the fact that the MED25 TAD-binding region is organized as an elongated cleft, we specifically asked whether these TADs are capable of binding in either orientation and how this could be achieved structurally and energetically. The binding of both the ETV5 and VP16-TADs in either orientation appears to be possible but occurs in a conformationally distinct manner and utilizes different sets of hydrophobic residues present in the TADs to drive the interactions. We propose that MED25 and at least a subset of human TADs specifically evolved a redundant set of molecular interaction patterns to allow binding to particular coactivators without major prior spatial constraints.

Assessment of machine‐learning predictions for the Mediator complex subunit MED25 ACID domain interactions with transactivation domains

“…In particular, it has been shown recently [54] that the percentage of the predicted disordered region in the sequences of unassigned domains had an inverse correlation coefficient with model quality for AlphaFold and RoseTTAFold models, indicating that model quality is lower because the sequences might be disordered. Third, as previously observed with the yeast acidic transcription activator Gcn4‐MED15 complex [49,55,56] and human MED25‐PEA3s [3,13,14,27], TADs likely bind MED25 ACID in multiple conformations and orientations that prevent accurate prediction of their correct binding mode. It has been recently shown by different groups that AlphaFold can be actually tuned either by subsampling [42,57] or clustering [58] the multiple sequence alignments to provide multiple conformations of the same protein.…”

“…As defined by NMR chemical shift perturbation (CSP) and mutagenesis studies of ACID domain in complex with VP16 and ERM TADs [2,13,21,22], the H1 binding site is delineated by strands β1‐β3‐β5, while the H2 binding site involves helix α1 and strands β6‐β7. These ACID surfaces appear adequately designed to accommodate the specific patterns of bulky hydrophobic and negatively charged residues found in different acidic transactivation domains [2,3,10,12,13,19,21,22,25–27] (Fig. 1).…”

Assessment of machine-learning predictions for MED25 ACID domain interactions with transactivation domains

“…In particular, it has been shown recently (52) that the percentage of the predicted disordered region in the sequences of unassigned domains had an inverse correlation coefficient with model quality for AlphaFold and RoseTTAFold models, indicating that model quality are lower because the sequences might be disordered. Third, as previously observed with the yeast acidic transcription activator Gcn4-MED15 complex (46, 53, 54) and human MED25-PEA3s (3, 13, 14, 26), TADs likely bind MED25 ACID in multiple conformations and orientations that prevents accurate prediction of their correct binding mode. It is also possible that AlphaFold training procedure may have been biased by overrepresented experimental structures in the PDB.…”

“…As defined by NMR chemical shift perturbation (CSP) and mutagenesis studies of ACID domain in complex with VP16 and ERM TADs (2,13,21,22), the H1 binding site is delineated by strands β1-β3-β5, while the H2 binding site involves helix α1 and strands β6-β7. These ACID surfaces appear adequately designed to accommodate the specific patterns of bulky hydrophobic and negatively charged residues found in different acidic transactivation domains (2,3,10,12,13,19,21,22,(24)(25)(26) (Figure 1). The H1 binding site is formed mainly by strands β1-β3-β5 as defined by NMR chemical shift perturbation (21,22) and is shown in gray.…”

E26 transformation-specific transcription variant 5 in development and cancer: modification, regulation and function

“…Mediator complex subunit 25 physically associates with N-terminal TAD of ETV5 as a coactivator and recruits Mediator to the promoter of target gene matrix metallopeptidase (MMP) 1 to initiate transcription by RNA polymerase II (Fig. 1 B) [ 27 , 29 ]. NRD, mapped from residues 73 to 298 AA, inhibits the transcriptional activity of N-terminal TAD, which is dependent on the its small ubiquitin-related modifier (SUMO) sites (K89, K263 and K293) (Fig.…”