Stringing along the estrogen receptor to engage with DNA

Estrogen receptor alpha (ERα), a ligand-regulated transcription factor and member of the large nuclear receptor superfamily of transcription factors, plays major roles in health and disease. Together with the endogenous hormone, estradiol, ERα regulates the development and maintenance of female reproductive tissues and modulates the normal function of many other nonreproductive organs, such as bone, brain, vasculature, liver and pancreas. It is also the principal driver of many breast cancers and is the target of endocrine therapies based on antiestrogens (such as tamoxifen or fulvestrant) that block its activity, or aromatase inhibitors (such as letrozole) that prevent the production of estradiol. Understandably, there has been a keen and longstanding interest in elucidating the structure of the estrogen receptor and understanding how it functions at a mechanistic level as it displays its spectrum of activities. Using advanced computational simulations, the Wolynes group at Rice University has recently made a major contribution (1) that illuminates the dynamic motions that ERα undergoes when—triggered by the binding of activating ligands—it interacts with DNA, the fundamental first step in its primary role as a transcription factor. The intriguing mechanism that they propose is likely to be relevant to the approximately 50 other members of the nuclear receptor superfamily (2).

Many of the basic structural features of ERα, as well as those of other members of the nuclear receptor superfamily, have been elucidated over the past half century through progressive advances in biochemical and molecular methods and NMR, X-ray crystallographic, and cryoelectron microscopic (CryoEM) structure determination (2). The basic structural features of ERα (Fig. 1 A and B) comprise two well-folded domains, a central DNA-binding domain (DBD, the C-domain) and a ligand-binding domain (LBD, the E-domain), which are connected and flanked by three unstructured sequences, the A/B domain at the N terminus, the linker or hinge D domain between the DBD and LBD, and a short C-terminal F domain. ERα generally functions as a dimer with the principal dimer interaction site centered in the LBD (Fig. 1C). After activating ligands bind to ERα, coregulating factors then assemble on the DNA-bound ERα dimer to functionally link it to the basal complex of proteins that control gene transcription (3). Beyond this simple lexicon of ligand binding and then DNA binding followed by activation or repression of transcription, little is known about the dynamic features of these interactions and how they proceed on the molecular scale. The roles, if any, played by the unstructured domains have remained particularly mysterious (4).

Fig. 1.

Schematic representation of the marionette mechanism of ERα binding to DNA. (A) Segmentation of the ERα linear sequence into distinct, color-coded domains. (B) Pictorial representation of ERα-structured and -unstructured domains. (C) Initial pose of ERα domains C–E assembled as a dimer connected through the LBD. (D) Mapping domains C–E onto a marionette. (E) Pictorial representation a pose of the ERα dimer bound to DNA. (F) A structural view of this pose presented in the highlighted publication (1). LBD (ligand-binding domain)—light and dark green; Hinge—orange; DBD (DNA-binding domain)—light and dark red.

In their recent publication (1), Chen et al. use advanced computational and simulation methodologies to probe the dynamics of interactions that take place among the ERα-structured domains, both with one another and with DNA. They chart a series of movements of the ERα dimer in which the unstructured hinge sequences between the DBDs and LBDs take a leading role in bringing the whole ligand-triggered ERα dimer into interactions with DNA (Fig. 1 E and F). The manner in which the motions of the structured domains were controlled through the initial DNA interactions of the unstructured hinge domains (Fig. 1 C and E) reminded the Wolynes group of how a puppeteer controls the movement of a marionette puppet through flexible strings (Fig. 1D). Hence, they denoted this process as a “marionette mechanism.”

Chen et al. use advanced computational and simulation methodologies to probe the dynamics of interactions that take place among the ERα structured domains, both with one another and with DNA.

It is important to appreciate the magnitude of the task of simulating the motions of a system as large as complexes of ERα with DNA. To undertake this work, the investigators built upon a computational strategy they had developed for simulating smaller biological systems, adapting it to the larger realm of dynamic features of ERα binding to DNA. They used a specialized, coarse-grained, and computationally efficient force field for the prediction of protein structures, which they supplemented with frustration analysis to highlight snags (local energy minima) that might be impeding the progressive release of energy in a folding or interaction process.

As a starting point, they constructed a DNA-bound ERα dimer based on carefully selected X-ray crystallographic structures of the ERα ligand- and DNA-binding domains, linked together with the flexible hinge domains. Extensive dynamic simulation and energy landscape analyses led to a minimized dimeric ERα–DNA complex that showed different interdomain and DNA interactions for each monomer, an asymmetric aspect consistent with that seen for ERα and other nuclear receptors by CryoEM analyses (5, 6), as well with other experimental results (7). In these and further simulations, the two unstructured N-terminal A/B and the C-terminal F domains were omitted, but the unstructured hinge domains linking the DBD and LDBs were retained and were found to play important roles in their further studies of the ERα–DNA-binding process. The flexibility of these hinge domains also made analysis of the subsequent dynamic simulations particularly challenging.

From this structural starting point, the investigators initiated multiple conformational simulation trajectories, doing it separately for four different states of ERα ligand binding, a complex with an antagonist (4-hydroxytamoxifen), an agonist (diethylstilbestrol), both the agonist and a fragment of a coactivator, and with no ligands (apo-ERα). Because of hinge domain flexibility, each simulation produced a large ensemble of conformations that principal component analyses displayed as rather shallow energy landscapes having the shape of a broad, irregular basin. Each state of ligand binding gave characteristic surface depressions in the basin that represented distinctly different interactions of the ligand- and DNA-binding domains with one another and with the hinge sequences and DNA. Thus, the state of ligand binding had a pronounced effect on how the different structured domains and the flexible hinges communicated with one another as the receptor dimer moved to settle into comfortable final DNA-bound poses. Most interesting was their finding that the hinge domains played a major role in the initial phases of bringing the agonist-liganded receptor to DNA. Simulations in which the charge and other changes were made in the hinge domains resulted in pronounced alterations in pathways of receptor–DNA interaction, supporting specificity in the role of the hinges in guiding ERα–DNA binding.

The overall picture that has emerged is illustrated in Fig. 1. In the initial state, ERα is a dimer secured through interactions between the two ligand-binding domains, with the DNA-binding domains simply tethered to these structured domains via the unstructured hinge regions (Fig. 1 A–C). The interaction of this loose assemblage with DNA appears to involve first the unstructured hinge domains, with this hinge engagement with DNA shortening the LBD–DBD distance and enabling the LBDs to begin to control the motion of the DBDs. What then follows is a progressive shift of the site of ERα–DNA interaction from the hinges to the DNA-binding domains. It was the role that the flexible hinge domains play in guiding these movements and communications between the two structured domains during the sequence of events for ERα–DNA binding that connoted the marionette mechanism to the authors (Fig. 1 D–F).

The advanced simulation and structural analyses displayed in this publication are computationally demanding, and given the lack of experimental information about the dynamic features of ERα interaction with DNA, the marionette mechanism they propose provides an intriguing model for further consideration in the broader context of ligand-regulated nuclear receptor activity. Their model might also be refined by further computational work and eventually become better benchmarked with current and evolving experimental findings.

A central challenge in analyzing simulations of ERα conformations stems from the flexibility of the hinge regions, because this resulted in a huge ensemble of energetically similar structures giving broad and shallow energy landscapes. Through their use of principal component analyses, however, the authors were able to extract shallow minima that corresponded to distinctive structural poses from which a pathway of progressive reorganization could be extracted. They note that despite its flexibility, the hinge domain plays an important role because amino acid sequence changes in the hinge affect the ERα–DNA-binding pathway. Historically, flexible loop structures in proteins have been considered mostly passive connectors between elements of secondary and tertiary structures, but it is apparent that they can also have important functional roles. Although it is shorter than the hinge domains, a thematically related case in ERα is the role of the loop linking helix 11 and 12 in the ERα LBD. This loop is a hot spot for activating mutations found in advanced breast cancers that have become resistant to endocrine therapies (8). Simulations have shown that in its native sequence, this loop provides an energetic, springlike force that keeps the receptor “off” in the absence of agonist ligands, with this spring force being eliminated by the activating mutations, which enable the LBD to fold into an active state without agonist binding (9, 10).

There are a few high-resolution structures of the complete sequence of smaller nuclear receptors, the peroxisome proliferator activated receptor gamma (PPARγ)/retinoid X receptor alpha (RXRα) and the liver X receptor (LXR)/retinoid X receptor (RXR) heterodimers complexed with DNA; these structures also reveal extensive and asymmetric patterns of interactions among the domains (11, 12). Benchmarking these experimentally determined structures with some of the predicted ERα conformations obtained by simulation in this study could reveal to what degree the results from the marionette mechanism might be generalized across the large nuclear receptor superfamily. In addition, CryoEM is providing increasingly high-resolution structures of nuclear receptors interacting with DNA and coregulators (5, 6). The authors clearly recognize that these models could be used as comparators for their predicted conformations, and they propose that their simulations might be useful in improving the resolution of these CryoEM structures.

Finally, as the authors recognize, they are of necessity working with a simplified system that lacks the N- and C-terminal unstructured domains of ERα, regions that are known to have major effects on the range of ERα activities (4). They are also considering only direct ERα interactions with DNA, whereas much of the transcriptional activity of ERα involves its interaction at different sites by a tethered binding to other proteins bound to other DNA sites, and there are a multitude of coactivator and corepressor proteins through which ERα mediates it diverse transcriptional effects (3). Even more distant from the scope of their modeling are the actions of estrogen receptor outside of the nucleus, where the receptor is tethered to membranes after lipid modifications and acts by initiating kinase cascades without entry into the nucleus (13, 14).

While there is much further work remaining, this elegant study has brought to the fore an intriguing pathway, the marionette mechanism, as a model for the dynamics of ERα interaction with DNA, a stepping stone along the way to further computational and experimental work in the study of ERα actions.