Activation mechanism of the β2-adrenergic receptorRon O. Drora,1,2, Daniel H. Arlowa,2, Paul Maragakisa,2, Thomas J. Mildorfa, Albert C. Pana, Huafeng Xua,David W. Borhania, and David E. Shawa,b,1

aD. E. Shaw Research, New York, NY 10036; and bCenter for Computational Biology and Bioinformatics, Columbia University, New York, NY 10032

Edited by Michael Levitt, Stanford University School of Medicine, Stanford, CA, and approved September 6, 2011 (received for review June 29, 2011)

A third of marketed drugs act by binding to a G-protein-coupledreceptor (GPCR) and either triggering or preventing receptor acti-vation. Although recent crystal structures have provided snapshotsof both active and inactive functional states of GPCRs, these struc-tures do not reveal the mechanism by which GPCRs transitionbetween these states. Here we propose an activation mechanismfor the β2-adrenergic receptor, a prototypical GPCR, based on atom-ic-level simulations in which an agonist-bound receptor transitionsspontaneously from the active to the inactive crystallographicallyobserved conformation. A loosely coupled allosteric network, com-prising three regions that can each switch individually betweenmultiple distinct conformations, links small perturbations at theextracellular drug-binding site to large conformational changes atthe intracellular G-protein-binding site. Our simulations also exhi-bit an intermediate that may represent a receptor conformation towhich a G protein binds during activation, and suggest that thefirst structural changes during receptor activation often take placeon the intracellular side of the receptor, far from the drug-bindingsite. By capturing this fundamental signaling process in atomic de-tail, our results may provide a foundation for the design of drugsthat control receptor signaling more precisely by stabilizing specificreceptor conformations.

molecular dynamics ∣ ligand ∣ biased signaling ∣ functional selectivity ∣beta-adrenergic

G-protein-coupled receptors (GPCRs) represent the largestclass of drug targets, and most drugs that act on GPCRs

do so by either triggering or preventing receptor “activation,”which involves a transition from an inactive, nonsignaling recep-tor conformation to an active conformation that induces G-pro-tein-mediated signaling. This conformational change allows theGPCR to transmit signals into the cell in response to the bindingof drugs or endogenous ligands to the cell’s exterior. Crystallo-graphy has now revealed structures of two GPCRs, the β2-adre-nergic receptor (β2AR) and rhodopsin, in both an inactive state(bound to an inverse agonist) and an active state (bound to bothan agonist and a G protein or G-protein-mimetic intracellularbinding partner) (1–5).

Despite these fundamental advances, the mechanism by whichGPCRs transition between inactive and active states remainsunclear. What intermediate conformations does a receptor adopten route to its active state? Do different regions of the receptormove in a concerted or an independent manner during activa-tion? How do subtle agonist-induced conformational changesin the ligand-binding site lead to much larger conformationalchanges at the distant G-protein-binding site?

These questions are particularly important in light of pharma-cological data suggesting that GPCRs can adopt multiple, distinctsignaling states in response to the binding of different ligands(6–8). Although the number of functionally distinct states andtheir conformations remain unknown, some are thought to be in-termediates along the activation pathway (7, 9). The rational de-sign of drugs that would bring about a particular, desired mode ofGPCR signaling requires a better understanding of the differentstates accessible to each GPCR and of the mechanisms by whichbound ligands can differentially induce or stabilize those states.

Here, we use molecular dynamics (MD) simulations to pro-pose an atomically detailed activation mechanism for β2AR, anarchetypal GPCR. Although MD and related techniques havebeen utilized extensively to study conformational dynamics ofGPCRs (e.g., refs. 10–16), unbiased simulation of the transitionfrom one crystallographically validated GPCR functional stateto another has historically been infeasible, because the timescalesof such transitions exceed those that have been accessible toMD. We recently demonstrated the feasibility of such simulations(17), using specialized computer hardware that dramatically ac-celerates MD computations (18); that study included two simula-tions in which agonist-bound β2AR transitioned from an active toan inactive crystallographically observed conformation. In thatpaper, we used those simulations only to confirm that the ener-getically favored conformation of agonist-bound β2AR in the ab-sence of an intracellular binding partner matched the (inactive)conformation determined crystallographically (17). In the currentpaper, we use many such simulations, performed under severaldifferent conditions, to build a detailed picture of the molecularmachinery, intermediate states, and conformational dynamicsunderlying β2AR activation.

ResultsTwo crystal structures of agonist-bound β2AR were availablewhen we performed this study. In one, the receptor was stabilizedin an active conformation by the binding of a G-protein-mimeticnanobody in addition to a potent agonist; this structure [the“active structure,” Protein Data Bank (PDB) entry 3P0G; ref. 2]closely resembles the crystal structure of meta II (active) rhodopsin(5). In the other agonist-bound β2AR structure (PDB entry 3PDS;ref. 17), the receptor lacks an intracellular binding partner; thisstructure is almost identical to that of inactive, inverse-agonist-bound β2AR (the “inactive structure,” PDB entry 2RH1; ref. 1).

These crystal structures suggest that, even though agonistbinding likely increases the fraction of active receptors, most ofthe receptor population remains inactive in the absence of a Gprotein or G-protein-mimetic binding partner. In other words,the lifetime of the active state is shorter than that of the inactivestate in the absence of the G protein. Accordingly, unbiased MDsimulations starting from the active structure, with the cocrystal-lized agonist BI-167107 but without the nanobody, transitionedfrom the active to the inactive receptor conformation on time-scales of several microseconds (17); the reverse transition hasnever been reported in unbiased simulations and has been shownexperimentally to occur on substantially longer timescales (19).

Author contributions: R.O.D., D.H.A., P.M., A.C.P., D.W.B., and D.E.S. designed research;R.O.D., D.H.A., P.M., T.J.M., and H.X. performed research; R.O.D., D.H.A., P.M., T.J.M.,and A.C.P. analyzed data; and R.O.D., D.H.A., P.M., T.J.M., A.C.P., D.W.B., and D.E.S. wrotethe paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.

Freely available online through the PNAS open access option.1To whom correspondence may be addressed. E-mail: Ron.Dror@DEShawResearch.com orDavid.Shaw@DEShawResearch.com.

2R.O.D., D.H.A., and P.M. contributed equally to this work.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1110499108/-/DCSupplemental.

18684–18689 ∣ PNAS ∣ November 15, 2011 ∣ vol. 108 ∣ no. 46 www.pnas.org/cgi/doi/10.1073/pnas.1110499108

In the present study, we performed 76 unbiased, all-atom MDsimulations of β2AR starting from the active structure, with thecocrystallized agonist but without the nanobody (Table S1); thereceptor was embedded in a hydrated lipid bilayer, and the simu-lations ranged in duration from 2 to 50 μs. In 36 of these simula-tions, the receptor transitioned spontaneously from the activeconformation to a conformation matching the inactive structure(see SI Text and Table S2). In each of these cases, the receptor,once settled into the inactive conformation, remained there forthe remainder of the simulation. These simulations did not incor-porate any prior knowledge of the crystallographically observedinactive conformation, so the fact that they reached it sponta-neously provides an indication of their predictive power; none-theless, the pathways traversed between these two states maybe affected by inaccuracies in the underlying force field models.In nine additional simulations of the crystallized complex, includ-ing the nanobody, the receptor never adopted an inactive confor-mation (see SI Text and Table S1).

While this paper was under review, a crystal structure of activeβ2AR in complex with a G protein was reported (3); the confor-mation of β2AR in this structure is similar to that in the nano-body-bound active structure. Five additional simulations of β2ARstarting from its conformation in the β2AR–G-protein complexstructure produced results similar to those reported in the re-mainder of this paper (see SI Text and Tables S1 and S2).

By analyzing the transitions of the agonist-bound receptorfrom the active conformation to the inactive conformation—aprocess we call deactivation—we can infer properties of themechanism of the reverse transition, receptor activation. Eachdeactivation pathway represents a possible activation pathway inreverse. More importantly, these simulations allow us to watch inaction the machinery that links the conformation of the ligand-binding site to that of the G-protein-binding site.

Simulations Display an Intermediate G-Protein-Binding Site Confor-mation. During the deactivation process, the receptor’s G-pro-tein-binding site typically assumed three major, discrete confor-

mations: active, inactive, and a previously unobserved intermedi-ate (Fig. 1). For the initial part of each simulation, usually severalhundred nanoseconds, the intracellular side of the receptorremained in a conformation matching that observed crystallogra-phically in complex with the nanobody, even though the nano-body was absent from the simulation. This active conformationdiffers from that of the inactive receptor primarily throughshifts and rotations of helices 5, 6, and 7; additionally, the intra-cellular end of helix 7 (including the conserved NPxxY motif,Asn3227.49–Tyr3267.53) maintains a distorted helical configura-tion in the active conformation, with the side chain of Tyr3267.53separating helix 6 from helix 3. (Superscripts refer to Ballesteros–Weinstein residue numbering; ref. 20). Crystal structures ofmeta II (active) rhodopsin and of opsin (both with and withoutthe Gα C-terminal helix bound) display a similar arrangement ofhelices 5, 6, and 7 (5, 21, 22), except that the intracellular end ofhelix 6 is shifted approximately 5 Å farther from the helical bun-dle in the active structure of β2AR (and in this initial part of theβ2AR simulations).

Next, the G-protein-binding site usually adopted an intermedi-ate conformation substantially different from either crystallo-graphic conformation. Upon transition to this intermediate, helix7 assumes its inactive conformation, with its intracellular endadopting an undistorted α-helical conformation and moving awayfrom helices 3 and 6, such that Tyr3267.53 loses contact with helix6 (Fig. 1). This helix 7 motion opens a space between the intra-cellular ends of helices 2, 3, 6, and 7, accommodating approxi-mately five additional water molecules (Fig. S1). Helices 3, 5,and 6 maintain a more active-like conformation; in particular,helix 6 maintains its separation from helix 3 (Fig. 1). Helices 5and 6 display substantial mobility within the intermediate, how-ever, sometimes adopting a conformation where helix 6 movesapproximately 7 Å in the direction of helix 5, which in turn movesapproximately 5 Å toward helix 3 (Fig. S2).

After a typical residence time of several hundred nanosecondsto several microseconds in the intermediate conformation, thereceptor shifted into the inactive conformation. The intracellular

A B

Fig. 1. The β2AR G-protein-binding site adopts three major conformations: active, inactive, and a previously unobserved intermediate. (A) The distance be-tween Arg1313.50 and Leu2726.34 Cα atoms (a measure of helix 6 displacement) and the rmsd of Asn3227.49–Cys3277.54 backbone atoms [a measure of helix 7conformation; relative to the inactive structure (PDB entry 2RH1)] are plotted for a simulation in which an agonist-bound receptor, initially in the conformationof the active structure (PDB entry 3P0G) spontaneously transitioned to an inactive conformation. Circles represent simulation snapshots sampled every 6 ns.Diamonds correspond to the active (3P0G) and inactive (2RH1) crystal structures, and to the average value of these coordinates in a simulation of the inactivestructure with the cocrystallized inverse agonist bound (2RH1-sim; Table S1, condition R). (B) Representative structures of the three major conformationalclusters, each superimposed on the inactive structure (light blue). Data are from simulation 11 (Table S2).

Dror et al. PNAS ∣ November 15, 2011 ∣ vol. 108 ∣ no. 46 ∣ 18685

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

end of helix 6 moved inward, rotating such that the side chains offour residues (Met2796.41, Leu2756.37, Leu2726.34, and Glu2686.30)previously separated from helix 3 by helix 5 now formed contactswith helix 3. The side chain of Tyr2195.58 moved from betweenhelices 3 and 6 to the lipid-exposed side of helix 6, adoptingthe dominant conformation seen in simulations of the inactivestructure (14) and allowing the Arg1313.50–Glu2686.30 ionic locksalt bridge to form (Fig. S3). Despite the bound agonist, the con-formations and dynamics of the G-protein-binding site followingdeactivation were nearly identical to those seen in inverse ago-nist-bound β2AR simulations (14); the receptor interconvertedbetween two distinct inactive conformations—a predominant onewith the ionic lock formed and aminor one with it broken (Fig. S4).

The β2AR contains several titratable residues, includingAsp792.50, Asp1303.49, and His1724.64, for which protonation statesare uncertain or may change during activation (16, 24). We per-formed simulations with each of these residues, and combinationsthereof, in different protonation states (Table S1). Although wefound that simulations with Asp1303.49 charged typically reachedthe inactive state nearly twice as quickly as those with it neutral (SIText)—in accord with experimental observations indicating thatAsp3.49 protonates upon activation of rhodopsin (23) and likelyof β2AR as well (24)—the deactivation pathway was substantivelyaffected only by the protonation state of Asp792.50. With Asp792.50charged, deactivation consistently proceeded via the intermediatedescribed above.With Asp792.50 neutral, the receptor followed thiscanonical pathway in three of four simulations that deactivated,but took an alternative pathway in the remaining one, suggestingthat the two pathways may coexist in a certain pH range. Althoughbroadly similar to the canonical pathway, the alternative pathwaydiffers in that the initial motion of helix 7 is not to its inactive con-formation; rather, it adopts an undistorted helical configuration in

place, without moving away from helix 6 (Fig. S5). As the deacti-vation process proceeds, helix 7 eventually relaxes to its inactiveconformation, but only after helix 6 has transitioned to its owninactive conformation. The conformation of helix 7 in the inter-mediates on this pathway is similar to that observed in recentcrystal structures of the agonist-bound A2A adenosine receptor(25, 26). Several other simulations followed the initial steps ofthe alternative deactivation pathway but did not reach the inactiveconformation (Fig. S5). Although we focus on simulations withAsp792.50 charged in the following sections, the key results alsohold for the alternative pathway.

The Ligand-Binding and G-Protein-Binding Sites Appear to Be LooselyCoupled.Although the differences in the ligand-binding site of theactive and inactive structures are much more subtle than those inthe G-protein-binding site, these small changes led Rasmussenet al. (2) to suggest an allosteric network connecting the two sites(Fig. 2A). The primary change in the ligand-binding site uponactivation is a motion of helix 5 toward helix 6, with Ser2075.46in particular moving approximately 1.7 Å to form a hydrogenbond with the agonist’s p-hydroxyl group. As a result, Ser2075.46and Pro2115.50 move into the space occupied by Ile1213.40 in theinactive structure. Ile1213.40 consequently shifts toward helix 6into the space previously occupied by Phe2826.44, which in turnmoves away from helix 3, tilting the intracellular end of helix6 away from helix 3.

Our simulations not only capture this allosteric network in ac-tion, but also indicate that it involves three distinct regions that caneach switch individually between multiple conformational states.The residues in and just below the ligand-binding site transitionfrom their active to their inactive conformation as the G-protein-binding site deactivates, but they do not all move in concert with

Fig. 2. The allosteric network underlying β2AR activation involves three regions that each switch individually between discrete conformations. (A) Activestructure of β2AR, with three key regions highlighted. (B) Representative structures of the dominant conformations for each region in simulation. (C) Quantitiesindicative of each region’s conformation: the vectorial displacement of the Ser2075.46 Cα atom away from helix 7 (see SI Text) and the distance between theSer2075.46 side-chain oxygen and agonist p-phenol oxygen atoms (ligand-binding site); the rmsd from the active and inactive structures of the nonsymmetric,non-hydrogen atoms in Ile1213.40 and Phe2826.44 (connector); and the two quantities plotted in Fig. 1A (G-protein-binding site). The time series are smoothedwith a 9.9-ns running average. These quantities and related ones are used to classify the conformation of each region (see SI Text). The shaded bars belowthe plots indicate which conformation is adopted by each region as a function of time (active/inactive for the ligand-binding site and connector; active/intermediate/inactive for the G-protein-binding site), with color coding matching that of B. Data are from simulation 11 (Table S2).

18686 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1110499108 Dror et al.

one another or with theG-protein-binding site. Instead, they divideinto two spatial clusters, with the residues in each tending to movetogether: one cluster consists of helix 5 residues in the ligand-bind-ing site, near Ser2075.46, whereas the other, which we term the con-nector region (or simply connector) given its location between theligand-binding and G-protein-binding sites, comprises primarilyIle1213.40 and Phe2826.44 (Fig. 2B). Both regions adopt discreteconformational states which resemble either their active or inactivestructures, although the connector acts more as a binary switchwhereas the ligand-binding site displays more conformational di-versity (Fig. 2 B and C, and Fig. S6).

Although the ligand-binding site couples to the connector, andthe connector to the G-protein-binding site, both couplings areloose (Fig. 3, Figs. S7 and S8, and Table S3). The ligand-bindingsite fluctuates between active and inactive conformations re-gardless of the conformations of the connector and the G-pro-tein-binding site. The relative population of ligand-binding siteconformations does depend on the connector conformation,however, with the ligand-binding site adopting an active confor-mation 78% and 11% of the time when the connector is in itsactive or inactive conformations, respectively, in our simulations(an important caveat is that these simulations do not representequilibrium sampling). The connector can also switch betweenactive and inactive conformations when the G-protein-bindingsite is in the active or intermediate conformations; in our simula-tions, the connector was in its active conformation 86%, 11%,and 0% of the time when the G-protein-binding site was active,intermediate, or inactive, respectively. In other words, the partsof the receptor appear to behave not like interlocked gears thatmove in synchrony, but rather like a set of loosely coupledswitches, each of which influences the probability that its neigh-bor(s) will adopt one state or another.

Activation May Begin near the G-Protein-Binding Site. Despite theloose coupling between the connector and the G-protein-bindingsite, our simulations suggest that the inactive G-protein-bindingsite conformation precludes an active connector conformation.When the G-protein-binding site adopts active or intermediateconformations—both characterized by an outward shift of theintracellular end of helix 6—the connector can assume bothactive and inactive conformations, and often switches betweenthem (Fig. 3 and Fig. S6). By contrast, the connector consistentlyadopts an inactive conformation before the G-protein-bindingsite does, and does not assume an active conformation once theG-protein-binding site has taken on the inactive conformation.

This behavior suggests that, during the reverse process of recep-tor activation, the G-protein-binding site likely leaves its inactiveconformation before the connector assumes an active confor-mation. This order of events runs counter to a simple sequentialmodel of activation in which agonist-induced structural changes inthe ligand-binding site lead to changes in the connector, whichonly then lead to changes in the G-protein-binding site. Instead,our simulations suggest that the activation process typically beginsat the G-protein-binding site, with the rest of the receptor onlysubsequently adopting an active conformation. Our simulationsare not sufficiently long to capture a full transition from the inac-tive to the active receptor state, however, and activation pathwaysbeginning at the ligand-binding site may also be possible.

If activation may begin on the intracellular side of the receptor,what role does agonist binding play in the activation process?Experimentally, agonists bind more tightly to the active state thanthe inactive state (27), and agonist binding thus shifts the ligand-binding site equilibrium between active and inactive conforma-tions; free-energy-perturbation calculations confirm that this is thecase in our simulations as well (see SI Text). Once the G-protein-binding site has adopted an intermediate conformation, the agonistthus stabilizes the active conformation of the connector, helping

Fig. 3. The ligand- and G-protein-binding sites are loosely coupled, but an inactive G-protein-binding site restricts the connector to its inactive conformation.Each set of three horizontal bars indicates which conformation is adopted by the ligand-binding site, the connector, and the G-protein-binding site during thatsimulation. The upper inset shows which conformations are adopted in each major receptor state in our agonist-bound simulations; large text indicates apredominant conformation. Simulation numbers match those in Table S2. The final simulation was initiated from the inactive structure, with the inverseagonist carazolol bound (Table S1, condition R).

Dror et al. PNAS ∣ November 15, 2011 ∣ vol. 108 ∣ no. 46 ∣ 18687

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY

to maintain the G-protein-binding site in its intermediate or activeconformations. Thus, the initial step of activation appears todepend on a preexisting conformational equilibrium in the G-pro-tein-binding site, which the agonist helps bias toward active states,much along the lines proposed by Schwartz et al. (28).

DiscussionOur simulations of β2AR transitioning from an active state toan inactive state allow us to build a picture of the machinery un-derlying the reverse process, receptor activation. The simulationsshow how agonist-induced changes in the conformation of theligand-binding site can favor consolidation of activating confor-mational changes at the distant G-protein-binding site. Threefunctionally relevant regions of the receptor—the ligand-bindingsite, the G-protein-binding site, and a connector between them—

appear to be only loosely coupled to one another. Each can takeon multiple distinct conformations, and each can typically changeconformation without a corresponding conformational change inthe other two regions.

This rather soft coupling mechanism has several implicationsfor signaling by drugs and endogenous ligands. It suggests thatpartial agonists need not induce distinct conformations of theligand-binding site to achieve distinct levels of efficacy; instead,it suffices for them to differentially shift the equilibrium betweenactive and inactive conformations of the ligand-binding site.Loose coupling also helps explain the difficulty of stabilizingβ2AR in a homogeneous active state with an agonist (29): Evenan agonist that locks the ligand-binding site into an active con-formation may not lock the entire receptor in its active state.

In our simulations, β2AR also adopts a set of intermediateconformations on the pathway between the active and inactivestates. Indeed, the ability of the ligand-binding site, the G-pro-tein-binding site, and the connector to individually change con-formation implies that many possible combinations—each ofwhich represents a distinct receptor conformation—may be vis-ited during activation or deactivation. The intermediate confor-mations of the G-protein-binding site (shown in Fig. 1 for thecanonical pathway and in Fig. S3 for the alternative pathway)deserve special attention because of their potential implicationsfor intracellular signaling. An extensive body of pharmacologicalevidence indicates that individual GPCRs are capable of adoptingmultiple conformations with distinct intracellular signaling pro-files in response to binding of different ligands (6–8). Fluores-cence spectroscopy and ligand-binding affinity measurementshave suggested that activation by full agonists involves at least

one intermediate conformation (30, 31). The major intermediateconformations of the G-protein-binding site adopted in oursimulations may well have a different signaling profile from thecrystallographically observed active conformation and may bedifferentially stabilized by different ligands.

We speculate that, during the typical activation process, theG protein may first bind to the receptor when the receptor’sG-protein-binding site is in an intermediate conformation. Themajor intermediate of Fig. 1 has an even larger cleft betweenthe intracellular ends of helices 2, 3, 5, 6, and 7 than the activeconformation, and it could thus also accommodate the C-term-inal α-helix of a G protein (Fig. S9). Binding of a G protein (ora G-protein mimetic) to this intermediate may promote its tran-sition to the active conformation: In the presence of an appro-priately positioned binding partner, the inward motion of theNPxxY region would allow burial of otherwise exposed hydro-phobic surface area in helix 7. Indeed, spectroscopic studies ofrhodopsin have suggested a two-step interaction mechanism inwhich GDP-bound transducin first forms a complex with anintermediate conformation of rhodopsin, and subsequently thetwo undergo a conformational change resulting in the fully activereceptor conformation and GDP release from the G protein (32).An appropriate nanobody might stabilize the G-protein-bindingsite in the intermediate rather than active conformation, enablingcrystallographic determination of this intermediate structure.

Our simulations suggest that receptor activation may occur asshown in Fig. 4. Once the G-protein-binding site has adoptedan intermediate conformation—presumably due to fluctuationwithin a preexisting conformational equilibrium—the connectorcan adopt an active conformation. The agonist then serves to sta-bilize the ligand-binding site, and thus the connector, in an activeconformation. Finally, perhaps as a result of G-protein binding,helix 7 changes configuration such that the G-protein-binding siteadopts its active conformation.

Several caveats are worth noting. First, molecular dynamicssimulations are subject to errors from several sources, so theactual activation mechanism might differ from that suggested byour simulations. Second, we did not simulate spontaneous disso-ciation of the nanobody, an event that should take place only onfar longer timescales. The receptor might follow a different path-way between inactive and active states in the presence of ananobody or G protein, particularly if such a binding partner isnecessary to induce early conformational changes on the activa-tion pathway (although fluorescence data suggests that activation-like conformational fluctuations in helix 6 of the agonist-bound

Fig. 4. A typical β2AR activation pathway as inferred from our simulations. InA, an agonist has bound to an inactive receptor. (B) The activation process beginson the receptor’s intracellular side with an outward motion of helix 6, past Tyr2195.58, bringing the receptor to an intermediate state. Here, the connector andthe ligand-binding site are in equilibrium between inactive and active conformations; the bound agonist stabilizes the active conformations. (C) A G proteinmay bind to the intermediate state, favoring the final step on the activation pathway: conformational change in the NPxxY motif (helix 7) that positions theside chain of Tyr3267.53 close to that of Tyr2195.58. Other activation pathways are also possible.

18688 ∣ www.pnas.org/cgi/doi/10.1073/pnas.1110499108 Dror et al.

receptor take place even in the absence of an intracellular bindingpartner; ref. 2). Finally, intracellular loop 3 (ICL3), which hasbeen either absent from or unresolved in each of the availableβ2AR structures, was also absent from our simulations. Removalof the bulk of ICL3 by partial tryptic digest does not appreciablyalter receptor function (33), but ICL3 likely does interact with theG protein and might affect the activation pathway.

Rhodopsin activation has proved more amenable to experimen-tal characterization than activation of other GPCRs. Althoughthese experiments have not provided the same level of spatial andtemporal detail as our simulations, rhodopsin activation involvessimilar conformational changes (5, 34) and also takes place via aset of intermediate states (35). In agreement with our results onβ2AR, FTIR spectroscopy on azido-labeled rhodopsin suggeststhat the first intracellular conformational changes involve helices5 and 6 (36), and recent NMR data indicate that Tyr7.53 andparticularly Tyr5.58 play a critical role in rhodopsin activation (37).On the other hand, rhodopsin has several early activation inter-mediates that appear to involve only small conformational changesnear the binding pocket (35). That these intermediates are popu-lated in an obligate, sequential order is suggestive of an “inducedfit” activation process in which retinal isomerization causes confor-mational changes in the ligand-binding site that then propagatetoward the intracellular side. It also suggests tighter couplingbetween receptor regions, consistent with the extremely low basalactivity and extremely high photon detection efficiency of rhodop-sin (38). These differences suggest some variability in signal trans-duction mechanisms among GPCRs, despite the involvement ofhighly conserved residues (39).

A promising frontier in drug development is the design ofligands that differentially modulate signaling pathways governedby a single GPCR, stimulating beneficial pathways while blockingdeleterious ones (6–8). Rational design of such “biased” or “func-tionally selective” ligands demands a structural understandingof the different conformations a GPCR can assume and of howeach conformation influences various downstream G-protein andarrestin signaling pathways. Identifying the complete set of func-tionally relevant GPCR conformations and their implicationsfor downstream signaling will require substantial additional work:These conformations may differ from one receptor to another,and some may lie off the activation pathway. By capturing insimulation an activation mechanism of an archetypal GPCR,however—both the major conformational intermediates and themachinery responsible for coupling the ligand- and G-protein-binding sites—our results may serve as a step toward the goal ofcreating drugs that provide finer control of GPCR signaling.

MethodsWe performed all-atom molecular dynamics simulations of β2AR, with lipidsand water represented explicitly, using the CHARMM (Chemistry at HarvardMolecular Mechanics) force field (40) on Anton (18), a special-purposemachine that accelerates such simulations by orders of magnitude. No arti-ficial (biasing) forces were applied. Further details are provided in SI Text.

ACKNOWLEDGMENTS. We thank Brian Kobilka, Morten Jensen, and WillyWriggers for helpful discussions, Kim Palmo for assistance in evaluating forcefields, and Rebecca Kastleman and Mollie Kirk for editorial assistance.

1. Cherezov V, et al. (2007) High-resolution crystal structure of an engineered humanβ2-adrenergic G protein-coupled receptor. Science 318:1258–1265.

2. Rasmussen SGF, et al. (2011) Structure of a nanobody-stabilized active state of the β2adrenoceptor. Nature 469:175–180.

3. Rasmussen SGF, et al. (2011) Crystal structure of the β2 adrenergic receptor-Gs proteincomplex. Nature, 10.1038/nature10361.

4. Palczewski K, et al. (2000) Crystal structure of rhodopsin: A G protein-coupled recep-tor. Science 289:739–745.

5. Choe HW, et al. (2011) Crystal structure of metarhodopsin II. Nature 471:651–655.6. Galandrin S, Oligny-Longpre G, Bouvier M (2007) The evasive nature of drug efficacy:

Implications for drug discovery. Trends Pharmacol Sci 28:423–430.7. Urban JD, et al. (2007) Functional selectivity and classical concepts of quantitative

pharmacology. J Pharmacol Exp Ther 320:1–13.8. Rajagopal S, Rajagopal K, Lefkowitz RJ (2010) Teaching old receptors new tricks:

Biasing seven-transmembrane receptors. Nat Rev Drug Discov 9:373–386.9. Kobilka BK, Deupi X (2007) Conformational complexity of G-protein-coupled recep-

tors. Trends Pharmacol Sci 28:397–406.10. Saam J, Tajkhorshid E, Hayashi S, Schulten K (2002) Molecular dynamics investiga-

tion of primary photoinduced events in the activation of rhodopsin. Biophys J83:3097–3112.

11. Shi L, et al. (2002) β2 Adrenergic receptor activation: Modulation of the proline kink intransmembrane 6 by a rotamer toggle switch. J Biol Chem 277:40989–40996.

12. Grossfield A, Pitman MC, Feller SE, Soubias O, Gawrisch K (2008) Internal hydrationincreases during activation of the G-protein-coupled receptor rhodopsin. J Mol Biol381:478–486.

13. Tikhonova IG, et al. (2008) Atomistic insights into rhodopsin activation from a dynamicmodel. J Am Chem Soc 130:10141–10149.

14. Dror RO, et al. (2009) Identification of two distinct inactive conformations of theβ2-adrenergic receptor reconciles structural and biochemical observations. Proc NatlAcad Sci USA 106:4689–4694.

15. Provasi D, Filizola M (2010) Putative active states of a prototypic G-protein coupledreceptor from biased molecular dynamics. Biophys J 98:2347–2355.

16. Vanni S, Neri M, Tavernelli I, Rothlisberger U (2010) A conserved protonation-inducedswitch can trigger “ionic-lock” formation in adrenergic receptors. J Mol Biol397:1339–1349.

17. Rosenbaum DM, et al. (2011) Structure and function of an irreversible agonist-β2adrenoceptor complex. Nature 469:236–240.

18. Shaw DE, et al. (2009) Millisecond-scale molecular dynamics simulations on Anton.Proceedings of the Conference on High Performance Computing, Networking,Storage and Analysis (SC09) (Assoc Computing Machinery, New York).

19. Vilardaga JP, Bünemann M, Krasel C, Castro M, Lohse MJ (2003) Measurement ofthe millisecond activation switch of G protein-coupled receptors in living cells. NatBiotechnol 21:807–812.

20. Ballesteros JA, Weinstein H (1995) Integrated methods for the construction of threedimensional models and computational probing of structure function relations inG protein-coupled receptors. Methods Neurosci 25:366–428.

21. Scheerer P, et al. (2008) Crystal structure of opsin in its G-protein-interacting confor-mation. Nature 455:497–502.

22. Park JH, Scheerer P, Hofmann KP, Choe HW, Ernst OP (2008) Crystal structure of theligand-free G-protein-coupled receptor opsin. Nature 454:183–187.

23. Vogel R, et al. (2008) Functional role of the “ionic lock”—an interhelical hydrogen-bond network in family A heptahelical receptors. J Mol Biol 380:648–655.

24. Ghanouni P, et al. (2000) The effect of pH on β2 adrenoceptor function: Evidence forprotonation-dependent activation. J Biol Chem 275:3121–3127.

25. Xu F, et al. (2011) Structure of an agonist-bound human A2A adenosine receptor.Science 332:322–327.

26. Lebon G, et al. (2011) Agonist-bound adenosine A2A receptor structures revealcommon features of GPCR activation. Nature 474:521–525.

27. Kent RS, De Lean A, Lefkowitz RJ (1980) A quantitative analysis of beta-adrenergicreceptor interactions: Resolution of high and low affinity states of the receptor bycomputer modeling of ligand binding data. Mol Pharmacol 17:14–23.

28. Schwartz TW, Frimurer TM, Holst B, Rosenkilde MM, Elling CE (2006) Molecularmechanism of 7TM receptor activation—a global toggle switch model. Annu RevPharmacol Toxicol 46:481–519.

29. Ghanouni P, et al. (2001) Functionally different agonists induce distinct conformationsin the G protein coupling domain of the β2 adrenergic receptor. J Biol Chem276:24433–24436.

30. Swaminath G, et al. (2004) Sequential binding of agonists to the β2 adrenoceptor:Kinetic evidence for intermediate conformational states. J Biol Chem 279:686–691.

31. Liapakis G, Chan WC, Papadokostaki M, Javitch JA (2004) Synergistic contributions ofthe functional groups of epinephrine to its affinity and efficacy at the β2 adrenergicreceptor. Mol Pharmacol 65:1181–1190.

32. Shichida Y, Morizumi T (2007) Mechanism of G-protein activation by rhodopsin.Photochem Photobiol 83:70–77.

33. Rubenstein RC, Wong SK, Ross EM (1987) The hydrophobic tryptic core of theβ-adrenergic receptor retains Gs regulatory activity in response to agonists and thiols.J Biol Chem 262:16655–16662.

34. Altenbach C, Kusnetzow AK, Ernst OP, Hofmann KP, Hubbell WL (2008) High-resolution distance mapping in rhodopsin reveals the pattern of helix movementdue to activation. Proc Natl Acad Sci USA 105:7439–7444.

35. Park PSH, Lodowski DT, Palczewski K (2008) Activation of G protein-coupled receptors:Beyond two-state models and tertiary conformational changes. Annu Rev PharmacolToxicol 48:107–141.

36. Ye S, et al. (2010) Tracking G-protein-coupled receptor activation using geneticallyencoded infrared probes. Nature 464:1386–1389.

37. Goncalves JA, et al. (2010) Highly conserved tyrosine stabilizes the active state ofrhodopsin. Proc Natl Acad Sci USA 107:19861–19866.

38. Smith SO (2010) Structure and activation of the visual pigment rhodopsin. Annu RevBiophys 39:309–328.

39. Nygaard R, Frimurer TM, Holst B, Rosenkilde MM, Schwartz TW (2009) Ligand bindingand micro-switches in 7TM receptor structures. Trends Pharmacol Sci 30:249–259.

40. MacKerell AD, Jr, et al. (1998) All-atom empirical potential for molecular modeling anddynamics studies of proteins. J Phys Chem B 102:3586–3616.

Dror et al. PNAS ∣ November 15, 2011 ∣ vol. 108 ∣ no. 46 ∣ 18689

BIOPH

YSICSAND

COMPU

TATIONALBIOLO

GY