- Top of page
- Results and Discussion
- Supporting Information
To address the problem of specificity in G-protein coupled receptor (GPCR) drug discovery, there has been tremendous recent interest in allosteric drugs that bind at sites topographically distinct from the orthosteric site. Unfortunately, structure-based drug design of allosteric GPCR ligands has been frustrated by the paucity of structural data for allosteric binding sites, making a strong case for predictive computational methods. In this work, we map the surfaces of the β1 (β1AR) and β2 (β2AR) adrenergic receptor structures to detect a series of five potentially druggable allosteric sites. We employ the FTMAP algorithm to identify ‘hot spots’ with affinity for a variety of organic probe molecules corresponding to drug fragments. Our work is distinguished by an ensemble-based approach, whereby we map diverse receptor conformations taken from molecular dynamics (MD) simulations totaling approximately 0.5 μs. Our results reveal distinct pockets formed at both solvent-exposed and lipid-exposed cavities, which we interpret in light of experimental data and which may constitute novel targets for GPCR drug discovery. This mapping data can now serve to drive a combination of fragment-based and virtual screening approaches for the discovery of small molecules that bind at these sites and which may offer highly selective therapies.
G-protein coupled receptors (GPCRs) represent a large and diverse superfamily of integral membrane proteins, triggering a wide range of signal transduction pathways in response to such stimuli as neurotransmitters, hormones and photons (1,2). Their ubiquity in the human genome and engagement in key physiological processes makes them a top class of pharmaceutical target (3). Testament to this, GPCRs are associated with approximately 30% of current drugs (4) and are linked to such diseases as cancer (5) and those of the cardiovascular and central nervous systems (6,7).
Despite encompassing over 1000 members, spread over six subfamilies (8), GPCRs share a common architecture consisting of seven transmembrane α-helices (TM1–TM7) connected by three intracellular loops (ICL1-ICL3) and three extracellular loops (ECL1-ECL3). Recent experimental evidence has demonstrated that GPCRs are highly dynamic structures, embracing a spectrum of conformational states from fully inactive through fully active (9). GPCR activity is manifested through their stimulation of G-proteins, which interact with the intracellular face of GPCRs and, in turn, engage with effector proteins to regulate levels of second-messenger molecules (10). Most GPCRs appear to possess an intrinsic, basal level of activity in the absence of any ligand. The binding of ligands (from the extracellular medium) then shifts the conformational equilibrium toward the fully active state in the case of agonists and toward the fully inactive state in the case of inverse agonists (11). To further emphasize the conformational heterogeneity of GPCRs, it has been established that ligands can regulate different downstream signalling cascades through the same GPCR. This phenomenon is known as ‘functional selectivity’ and is thought to reflect the ability of different ligands to stabilize highly specific receptor conformations (12).
G-protein coupled receptor structural biology has recently enjoyed a boom (13), with high-resolution crystallographic structures now available for the avian β1 adrenergic receptor (β1AR) (14), human β2 adrenergic receptor (β2AR) (15–17) and human adenosine A2A receptor (18), as well as the proposed active conformation of bovine opsin in complex with a G-protein fragment (19). These structures have yielded great insights into the structure–function relationship between ligand binding and G-protein activation, whereby ligands are thought to stabilize/induce a variety of conformational rearrangements, which are characteristic of different states and can have diverse downstream effects (20–22). From a drug design perspective, the structures have afforded a detailed picture of the ligand-binding pocket, formed in an extracellular cleft leading to the transmembrane core, and have opened new drug discovery avenues (23). For example, new structure-based virtual screening efforts have emerged, in a traditionally ligand-based field, reporting the successful discovery of several new active compounds (24,25).
References to the GPCR ligand-binding site have, hitherto, alluded to the ‘orthosteric’ site, which is defined as the pocket bound by the endogenous activating ligand (26). GPCR drug discovery to date has predominantly been concerned with this site, yielding an impressive repertoire of drugs that compete with the endogenous ligand and generate a variety of efficacies. Such orthosteric ligands include well-known examples such as the anti-hypertension drug atenolol and the anti-asthma drug salbutamol. However, there has been formidable recent interest in compounds that modulate GPCR activity through an ‘allosteric’ mechanism [as reviewed in (26–29)]. Such allosteric ligands bind to a site that is topographically distinguished from the orthosteric site (OS) and thus do not compete with orthosteric ligands (30). Allosteric ligands may (i) modify the binding and/or efficacy properties of an orthosteric ligand (termed ‘allosteric modulators’) or (ii) affect the activation state of the GPCR by themselves (termed ‘allosteric agonists’) (31). The binding of an allosteric ligand may therefore be considered to stabilize/induce GPCR conformations that either increase/decrease the affinity of orthosteric ligand binding and/or affect G-protein stimulation. The main allure of allosteric ligands is their potential for greater receptor selectivity, by binding to specific GPCR subtypes (26). The strong evolutionary conservation of the OS (across closely related GPCRs) has caused problems with cross-reactivity of orthosteric ligands, which can lead to undesirable therapeutic side-effects. For example, orthosteric drugs that act on the β1AR for the treatment of heart disease may cross-react with the β2AR and cause unwanted pulmonary effects (and vice-versa for anti-asthma drugs that can cause cardiac effects) (32). Similarly, the development of orthosteric antipsychotic drugs, acting at the muscarinic acetylcholine receptor, has been plagued with a lack of subtype specificity, leading to side-effects and motivating the discovery of more selective allosteric drugs (33). The amino acid sequences of allosteric binding sites are more likely to have diverged among members of a receptor subclass than the OS, therefore conferring specificity and reducing the potential for off-target activity. A key advantage of allosteric modulators is that they exert no effect by themselves, serving only to tune the effect of endogenous ligands and thus causing less disruption to the normal physiological profile of the GPCR. This activity also has implications for toxicity, whereby, because the effect of allosteric modulators is saturable, overdose-associated risks are reduced. Finally, it has been noted that allosteric ligands identified to date display greater structural diversity than their orthosteric counterparts, thus opening up their chemical space and making them more amenable to modifications to solve ADMET problems (34). In summary, allosteric GPCR ligands have generated considerable excitement in the field of GPCR drug discovery and offer various modes of receptor modulation, with a degree of specificity unattainable at the OS.
Allosteric GPCR ligands have now been described for a wide range of GPCRs, including members from all three human classes (26,27,29). Two allosteric drugs have already been FDA-approved: cinacalcet, which binds the calcium-sensing receptor in hyperparathyroidism, and maraviroc, which binds the CCR5 chemokine receptor in HIV infection, as well as several candidates undergoing clinical studies (35). Perhaps the two best studied subclasses are those of the muscarinic receptors (Class A) and the metabotropic glutamate receptors (Class C), which bind allosteric ligands with diverse pharmacological activities and play important roles in the diseases of the central nervous system (6). Despite encouraging progress in the discovery and functional characterization of allosteric GPCR ligands, the structural biology of their binding sites is still poorly understood. An outstanding question is where do they bind? Site-directed mutagenesis has been employed to give the approximate locations and identify interacting residues for several allosteric ligands and receptors (31). Such experimental mapping studies have been invaluable in providing the first details of interaction sites and exposing their diversity, occurring in both solvent-exposed and bilayer locations (6). However, given the time and labor-intensive nature of such methods, as well as the need for an existing allosteric ligand, there is a strong case for the implementation of faster, more predictive approaches with which to identify putative allosteric binding sites. Such tools are particularly germane in the case of GPCRs, which have proven especially difficult to crystallize and therefore do not enjoy the same degree of structural data which has supported allosteric drug design in soluble proteins, such as kinases (34). Establishing the topographical location of allosteric sites in GPCRs would drive the structure-based, de novo discovery of allosteric ligands and help elucidate the structural basis of their function.
The recent milestone of high-resolution structural data for ligand-activated GPCRs provides a role for computational methods in allosteric drug discovery. The motivation to uncover new drug binding sites is not a new one and has been fueled by the characterization of several recent sites which have therapeutic potential (36). Given the three-dimensional structure of a target protein, a number of algorithms have been developed to scan the entire protein surface for cavities, which are capable of binding small molecules and are potentially druggable [as reviewed in (37,38)]. Such methods aim to detect and score such pockets based on various concepts of molecular recognition. These range from a purely geometric treatment of the binding pocket [e.g. PocketPicker (39)] to more rigorous energy-based calculations that typically attempt to dock a series of probe molecules to candidate pockets and estimate the strength of their interaction [e.g. GRID (40)]. FTMAP (41) is one of the most recent energy-based mapping algorithms and was originally conceived as a faster, computational equivalent of an experimental technique known as the multiple solvent crystal structures (MSCS) method (42). With the MSCS approach, the target protein is co-crystallized in the presence of diverse organic solvent probe molecules, and it has been demonstrated that the probes tend to cluster at functionally important sites. Similarly, FTMAP docks a panel of 16 probe molecules (representing a variety of functional groups and drug fragments) to the protein surface and uses an empirical scoring function to determine low-energy poses. FTMAP is distinguished from other methods by a combination of its clustering scheme, which differentiates between consensus sites (CSs) (which represent putative binding sites) and isolated, non-specific binding events and an efficient sampling method (41). FTMAP (and its predecessor CS-Map) have been validated against a range of pharmaceutical targets (including renin aspartic protease, elastase and glucocerebrosidase), showing excellent agreement with binding sites identified by X-ray crystallography, for both organic solvents and drug molecules (41,43,44). These encouraging correlations with existing structural data suggest that the FTMAP method also has the potential to work in a prescriptive fashion, in the identification of novel druggable allosteric binding sites.
The aforementioned mapping algorithms typically depend on the availability of one, or occasionally a few, experimentally determined atomic structures of the target protein. Considering the structural flexibility of proteins, this static representation of the target can be extremely restrictive and is a recognized flaw in many protein–ligand docking efforts (45,46). Consequently, a variety of schemes have been proposed that allow target flexibility to be taken into account and range from modeling simple sidechain changes to full backbone and sidechain mobility (47,48). In the context of binding site identification, the incorporation of protein flexibility is appealing as allosteric pockets may only form transiently and relatively infrequently in the dynamics of the protein and may therefore be missed in experimental structures. Also, the topography of allosteric sites may change, exposing different protein residues and altering their physicochemical properties. The role of flexibility is even more pronounced in the case of GPCRs, renowned for their strong intrinsic conformational plasticity that has hampered crystallization efforts (49). Molecular dynamics (MD) simulation is a popular method for the modeling of protein motions and generation of ensembles of protein conformations, which evolve from an experimental starting structure (50). Several MD simulation approaches have now been reported, examining the conformational dynamics of GPCRs, often with a view to capturing the structural rearrangements that accompany receptor activation [e.g. (51,52)]. MD simulations have also demonstrated their value in drug discovery applications by exposing dramatic binding site changes. For example, a relatively short (2 ns) MD simulation of HIV-1 integrase revealed a new inhibitor binding site that led to the discovery of the first integrase inhibitor, raltegravir (53). More subtle binding site dynamics have been used in virtual screening applications, whereby putative compounds are docked to a range of conformers, as opposed to a single, experimental structure, to improve rank ordering (54). It is therefore appealing to couple a computational mapping analysis with an MD-based ensemble of GPCR conformations, so as to potentially discover novel allosteric sites, as well as further characterizing any existing sites.
In this work, we report the computational mapping of potential allosteric sites on the surface of the human β1 (β1AR) and β2 (β2AR) adrenergic receptors (Figure 1), which may be exploited in the structure-based design of allosteric ligands. The recent milestone of high-resolution crystal structures for these pharmaceutically important GPCRs makes them ideally suited to such a mapping analysis. β1AR and β2AR are members of the β-adrenoceptor subfamily of Class A GPCRs and play key roles in heart muscle contraction and bronchial smooth muscle relaxation, respectively. They have therefore been targeted in the treatment of heart disease and pulmonary problems, through a range of orthosteric drugs. Unfortunately, these receptors typify a limitation endemic throughout GPCRs, and indeed many other drug targets, in the extremely high sequence conservation of their OSs (94% identity over 16 residues). Consequently, subtype selectivity is a key issue in their pharmacological control, and there is a pressing need to identify drugs with a greater capacity to discriminate between related receptors. Allosteric drugs pose one such solution, and we have therefore employed the FTMAP algorithm to search for druggable hot spots in non-orthosteric regions of the receptors. To address the role of conformational flexibility, we have applied FTMAP to both the experimental structure and an ensemble of 15 representative MD simulation structures, for each receptor. Here, we used the multicopy MD approach, whereby a series of six independent 40-ns trajectories was generated for each receptor in a phospholipid bilayer, yielding a combined total simulation time of approximately 0.5 μs. This study was inspired by previous work that used the CS-Map algorithm to expose novel binding sites in MD snapshots of a soluble viral target – the H5N1 influenza neuraminidase (55). The current work is distinguished by use of the improved FTMAP code and its application to a pair of human membrane protein targets, as well as a global search of the entire protein surface. Our results define a set of five key non-orthosteric regions that act as consensus binding sites for organic probes and which may represent targets for allosteric ligands. Interestingly, the sites are distributed across both solvent-exposed and lipid-exposed surfaces, and it appears that one site may be exclusive to one receptor subtype, while all others are generally shared. Furthermore, some of the sites are not apparent in the experimental structures, only being revealed in the MD-generated conformers. We characterize each site in the context of experimental data and propose they will serve to guide fragment-driven and virtual screening studies for the identification of allosteric compounds, which may be shared with related GPCRs.
Figure 1. Snapshots of the bilayer-embedded β1AR (A) and β2AR (B) MD simulation systems. Receptors are shown in cartoon representation (left), with the seven TM helices colored blue and other structures colored orange. Residues comprising the orthosteric site (OS) are shown in green stick representation. The co-crystallized orthosteric ligand is superimposed and shown in black molecular surface representation. Receptors are shown in molecular surface representation (center, right) to illustrate the regions being mapped by probes. Solvent-accessible surfaces are colored green, while other regions are colored gray. Extracellular and intracellular views show the solvent-exposed regions in more detail, from above and below the bilayer. Palmitoyl-oleoyl phosphatidylcholine phospholipid molecules are shown in space-filling representation and are colored by atom type.
Download figure to PowerPoint
- Top of page
- Results and Discussion
- Supporting Information
The allosteric modulation of GPCR activity is the focus of a growing branch of drug discovery, searching for novel therapeutic agents to control the numerous pathologies they play a role in. Among a range of advantages over their orthosteric counterparts, allosteric ligands offer the prospect of highly specific targeting, by binding to less conserved regions of the receptor surface. Despite the development of experimental screening methods for allosteric GPCR lead discovery, in silico structure-based approaches are lacking, largely because of the renowned problems associated with GPCR crystallization. In the absence of experimental structures of GPCRs complexed with allosteric ligands, there is a clear window of opportunity for predictive computational approaches using recent unbound structures. In this work, we report the application of a fragment-based algorithm, FTMAP, to map the surface of the human β1AR and β2AR GPCR structures for druggable sites distinct from the OS. To incorporate the flexibility of the receptors, we have mapped a series of 15 diverse structures taken from a series of MD simulations of each receptor in a phospholipid bilayer.
By focusing on the key interacting protein residues, we have defined a set of five putative allosteric binding sites, four of which are shared between receptor types and one of which is unique to β2AR, which we have interpreted with corroborating experimental evidence. Sites 1 and 4 are found in the solvent-exposed extracellular and intracellular mouths, with Site 1 representing a well-known region of allosteric ligand-binding activity in related GPCRs. Key gating motions from MD simulations suggest that allosteric ligands binding at the extracellular mouth may block the entrance or exit of orthosteric ligands by bridging opposing structures at the entrance. The possible function of allosteric ligands at Site 4 is less clear, as this is an interaction site for G-proteins; however, we speculate they may be capable of stabilizing the open form of the cavity and influencing the conformation of the ionic lock. Sites 2, 3 and 5 represent pockets formed at the protein–lipid interface, in the hydrophobic core of the lipid bilayer. Occupying the junctions of TM helices, it is likely that compounds filling these locations would increase inter-helical packing interactions and thus restrict conformational flexibility. This effect is supported by experimental evidence at Sites 3 and 5, where occlusion of the pockets has been shown to increase stability of the receptors and may stabilize distinct states with desirable therapeutic effects. Site 5 is of particular interest as significant probe-binding events are only seen in β2AR, suggesting it may be an excellent target for β2AR-selective therapies. While structural evidence of protein–drug interactions in the hydrophobic core of the membrane is lacking, a number of lipophilic drugs have demonstrated strong membrane partitioning coefficients and have been proposed to access their membrane-associated receptors from the lipid phase rather than the aqueous phase (88).
From a methodological point of view, the use of MD simulations to model flexibility of the receptors allowed FTMAP to detect some sites not apparent in the static experimental structures and for us to observe the transient nature of some pockets. Sampling different receptor conformations is especially appealing for such flexible proteins as GPCRs, as even subtle rearrangements may expose or conceal key ‘cryptic’ binding sites. Given the broad distribution of the representative MD structures used in this study (see Supporting Information), we conclude that a multi-copy simulation approach of this timescale is successful in generating enhanced conformational diversity. However, despite recent advances in the evolution of computer hardware, conventional MD simulations still suffer from incomplete conformation sampling for systems of this size. It seems reasonable to predict that a more extensive exploration of the conformational landscape could lead to the identification of further druggable binding sites. It is therefore tempting to apply new computational methods in the generation of more diverse structural ensembles. Such methods include accelerated MD (89), replica exchange (90) and conformational flooding (91) and may expose sites that are formed in regions of the energy landscape distant from the experimentally captured conformation.
Having identified a series of potential allosteric binding sites, this work will serve as a springboard for structure and fragment-based lead identification methods. An obvious starting point is in the virtual screening of existing drug-like compound libraries for potential high-affinity ligands at each pocket, which can then be assayed for binding and allosteric activity. An alternative approach involves the design and synthesis of novel compounds using the poses of docked probe molecules from our analysis with fragment-based techniques (92). Probes can be grown into high-affinity small molecules that interact with further protein residues (fragment evolution). Also, when multiple probes are bound simultaneously, they can be fused to form a single molecule (fragment linking). This approach could also be used to form proposed bitopic compounds in the extracellular mouth of the receptors, by joining orthosteric compounds to putative allosteric probes (82). While the GPCRs featured in this study are logical targets to begin such screening studies and are of considerable therapeutic interest in their own right, it is likely that additional druggable sites are present on related GPCRs, which may be amenable to homology modeling approaches for a similar analysis.