Ligand- and mutation-induced conformationalselection in the CCR5 chemokine Gprotein-coupled receptorRavinder Abrola,1,2,3, Bartosz Trzaskowskia,1,4, William A. Goddard IIIa,1,2, Alexandre Nesterovb, Ivan Olaveb,and Christopher Ironsb

aMaterials and Process Simulation Center, California Institute of Technology, Pasadena, CA 91125; and bPharmSelex Inc., Seattle, WA 98102

Contributed by William A. Goddard III, July 27, 2014 (sent for review January 25, 2014)

We predicted the structural basis for pleiotropic signaling of the C-Cchemokine type 5 (CCR5) G protein-coupled receptor (GPCR) bypredicting the binding of several ligands to the lower-energyconformations of the CCR5 receptor and 11 mutants. For each case,we predicted the ∼20 most stable conformations for the receptoralong with the binding sites for four anti-HIV ligands. We found thatnone of the ligands bind to the lowest-energy apo-receptor confor-mation. The three ligands with a similar pharmacophore (Maraviroc,PF-232798, and Aplaviroc) bind to a specific higher-energy receptorconformation whereas TAK-779 (with a different pharmacophore)binds to a different high-energy conformation. This result is inagreement with the very different binding-site profiles for theseligands obtained by us and others. The predicted Maraviroc bind-ing site agrees with the recent structure of CCR5 receptor cocrystal-lized with Maraviroc. We performed 11 site-directed mutagenesisexperiments to validate the predicted binding sites. Here, we inde-pendently predicted the lowest 10 mutant protein conformationsfor each of the 11 mutants and then docked the ligands to theselowest conformations. We found the predicted binding energies tobe in excellent agreementwith our mutagenesis experiments. Theseresults show that, for GPCRs, each ligand can stabilize a differentprotein conformation, complicating the use of cocrystallized struc-tures for ligand screening. Moreover, these results show that a sin-gle-point mutation in a GPCR can dramatically alter the availablelow-energy conformations, which in turn alters the binding site,potentially altering downstream signaling events. These studies val-idate the conformational selection paradigm for the pleiotropicfunction and structural plasticity of GPCRs.

protein structure prediction | ligand–protein binding prediction |conformational ensemble | functional selectivity | GEnSeMBLE

Since the finding that individuals lacking the C-C chemokinereceptor type 5 (CCR5) gene are resistant to HIV, the CCR5

receptor has been established as a coreceptor for macrophage-tropic viruses, including HIV, to enter host cells (1). Severaldrugs aimed at disrupting the HIV–CCR5 coupling have beendeveloped. To understand the structural mechanism of ligandbinding to G protein-coupled receptors (GPCRs) and how theseCCR5 targeting drugs work, we predicted the low-energy struc-tures of the apo CCR5 receptor, which in turn were used topredict the binding sites for various ligands. The focus was onligands already known to inhibit CCR5: Maraviroc (MVC) (2),PF-232798 (PF) (3), Aplaviroc (APL) (4), and TAK-779 (TAK) (5)because there are abundant experimental data on their binding tonumerous CCR5 mutants, to test the predicted structures. The li-gand structures are shown in Scheme S1. For three ligands (MVC,PF, and APL), we predict a similar binding site pharmacophorecentered on a protonated nitrogen, whereas the other ligand (TAK)containing a quaternary nitrogen leads to a very different bindingsite pharmacophore. To validate our predictions, we identified 11singly mutated CCR5 receptor forms selected to test the predictedbinding region for the ligands. The predicted structures of ligand-

bound CCR5 complexes presented here have already been used intwo recent studies to provide a structural basis for (i) biophysicalmeasurements (6) mapping the binding mode of MVC to CCR5using genetically encoded photo–cross-linkers and (ii) viral entryexperiments (7) probing the basis for G protein-coupled and-uncoupled CCR5 receptors by MVC-resistant and -sensitive HIV-1 viruses.While this work was being finalized for submission, the CCR5

receptor was crystallized with MVC (8), allowing a direct struc-tural validation of the predicted structures as discussed inPrediction of Ligand–CCR5 Structures and Comparison with theCrystal Structure.

ResultsPrediction of the CCR5 Structure. To predict the ensemble oflow-energy structures (conformations) for CCR5, we used theGEnSeMBLE method (9), which performs a complete conforma-tional sampling (∼11 billion) of transmembrane (TM) helix confor-mations accessible in the membrane. This method starts withtemplates for positioning the seven TM helices based onavailable experimental/predicted structures and then samplesa large number of helix orientation angles as defined in Fig. 1.During conformational sampling, we first sampled only the

helix rotation angles (η) (on a 15° grid), and the BiHelix energy

Significance

The C-C chemokine receptor type 5 (CCR5) G protein-coupled re-ceptor (GPCR) is a prime target for preventing HIV invasion. Amajor difficulty in developing effective therapeutics is that theCCR5 exhibits an ensemble of ∼10–20 distinct low-energy con-formations, each of which might favor binding to differentligands and/or lead to different downstream functions. X-raystructures generally provide only one of these conformations. Weapplied the GEnSeMBLE methodology to predict this ensemble,and we designed and carried out 11 experiments to validate theability of this ensemble to predict binding of an HIV therapeuticto CCR5. We found that each of the mutations changes thebinding site. The predicted effects of mutations on binding are inexcellent agreement with experiments, providing CCR5 structuresfor designing new ligands.

Author contributions: R.A. and W.A.G. designed research; R.A., B.T., W.A.G., A.N., I.O.,and C.I. performed research; R.A., B.T., W.A.G., A.N., I.O., and C.I. analyzed data; and R.A.,B.T., and W.A.G. wrote the paper.

The authors declare no conflict of interest.1R.A., B.T., and W.A.G. contributed equally to this work.2To whom correspondence may be addressed. Email: wag@wag.caltech.edu or abrolr@csmc.edu.

3Present address: Departments of Medicine and Biomedical Sciences, Cedars-Sinai MedicalCenter, Los Angeles, CA 90048.

4Present address: Centre of New Technologies, University of Warsaw, 02-089, Warsaw,Poland.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1413216111/-/DCSupplemental.

13040–13045 | PNAS | September 9, 2014 | vol. 111 | no. 36 www.pnas.org/cgi/doi/10.1073/pnas.1413216111

(10) was used to evaluate the energy of all (24)7 ∼ 4.6 billion TMhelix bundle conformations. For the human CCR5 receptorstructure prediction, this sampling was done starting with each ofthe following four X-ray–derived templates: bovine rhodopsin(PDB ID code 1u19) (11), human β2 adrenergic receptor (PDBID code 2rh1) (12), human adenosine A2A receptor (PDB IDcode 3eml) (13), and turkey β1 adrenergic receptor (PDB IDcode 2vt4) (14). The full seven-helix bundle was built for the top1,000 conformations from each template, and the side chainswere optimized using the SideChain Rotamer Energy AnalysisMethod (SCREAM) side-chain placement protocol (15), fol-lowed by minimization (10 steps using Dreiding force field) (16).The β2 receptor template-based conformational ensemble led tothe lowest energies (most stable conformations). We then se-lected the most stable 16 CCR5 conformations, for each of whichwe simultaneously sampled all three helix orientation angles (θ, ϕ,and η) allowing −10°, 0°, and +10° for the θ tilt angle and the −15°,0° , and +15° range for both ϕ azimuthal and η rotation angles.This procedure led to a total of (27)7 ∼11 billion TM bundleconformations, for each of which we evaluated the energy rapidlyusing the SuperBiHelix method (17). Then the lowest-energy2,000 conformations were built into seven-helix bundles and op-timized, from which we selected the 20 lowest-energy con-formations (labeled WT1 to WT20) as shown in Table S1. Fromthese conformations, we selected eight structurally diverse seven-helix structures (highlighted rows in Table S1) for further analysisand ligand docking. The lowest-energy conformation WT1 cor-responds to the predicted apo conformation of the receptor. Noneof the experimentally obtained GPCR structures (with the ex-ception of opsin) has been ligand-free so it remains to be con-firmed whether WT1 resembles the apo conformation of theCCR5 receptor.These best eight diverse structures were then used to predict

the binding of CCR5 ligands. The next section shows that MVCbinds most strongly to the WT7 conformation rather than thelowest-energy WT1 conformation.

Prediction of Ligand–CCR5 Structures and Comparison with theCrystal Structure. The four ligands (MVC, PF, APL, and TAK)were minimized using the B3LYP flavor of density functionaltheory (DFT) (with the 6–311G** basis set) using the Jaguarsoftware package (Jaguar, version 7.8; Schrödinger, LLC). Aconformational search was performed over the rotatable bondsfor each ligand, and ∼20–30 conformations were selected (basedon energy and diversity) (SI Materials and Methods) for each li-gand. Then, each of these conformations was docked to thepreviously selected eight CCR5 conformations using theDarwinDock/GenDock method (18). This method for predictingligand–protein structures samples a complete set (∼50,000) ofligand poses in the potential binding regions, which are thenclustered based on root-mean-squared (rms) distances to obtain

∼2,000 family heads whose energies are then evaluated to select∼200 families, for each of which the energies of all members areevaluated. The top 100 poses from this process are optimized tofinally select the best protein–ligand structure. These ligand–CCR5 structure predictions were performed independently foreach of the eight CCR5 structures using the DREIDING forcefield (16).The final predicted most stable structure for the MVC/CCR5

complex is shown in Fig. 2, overlaid with the complex structurefrom the recent crystal structure (8) and showing the most im-portant ligand–residue interactions. Binding of MVC stabilizedthe WT7 structure of CCR5, which was also stabilized by PF andAPL (Fig. 3 A and B) whereas TAK selected the WT10 confor-mation (Fig. 3C). Similar results have been obtained earlier for theadenosine A3AR receptor, where all four selective agonists sta-bilizedWT15 whereas all four selective antagonists stabilized WT2or WT3 (19) conformations; for none of these eight ligands wasthe complex with WT1 of lowest energy. These results emphasizethe importance of having the ensemble of low-energy structures,and they suggest that ligands can control the final conformationand perhaps downstream function of CCR5 as well as GPCRsin general.The biggest changes in the CCR5 structure upon binding

MVC (WT1 to WT7) were in the η rotation angles of helices 2and 7 (compare the binding site residues in a 2D representationin Fig. S1 A and D). The 20-degree rotation of helix 2 completelychanges the interhelical hydrogen bond network of CCR5. In thewild-type protein, the structure of CCR5 is stabilized by severalinteractions, including a conserved N48 (TM1)–D76 (TM2) hy-drogen bond, S63 (TM2)–R126 (TM3) hydrogen bond, and fi-nally a 1-2-7 hydrogen bond network Y37 (TM1)–W86 (TM2)–T284 (TM7). The 1-2-7 network and the W86–T284 hydrogenbond, in particular, result in a compact structure for the proteinthat is not suitable to develop strong interactions with largeligands having a tertiary nitrogen atom, such as MVC. This isbecause the interaction with T284 forces W86 to be close toE283, the crucial residue for interacting with the basic part of theligand. To overcome this problem, the W86-T284 hydrogen bondmust break, which leads to a slight destabilization of the struc-ture, which forces helices 2 and 7 to rotate to find favorableinteractions with other residues from other helices, such as π–πstacking with Y108. The existence of a similar receptor activationnetwork was proposed recently (20), but it did not include T284;

Fig. 1. Coordinate frame for the TM helix orientation in a membrane(reprinted with permission from ref. 28). The x-y plane defined as the middleplane of the membrane bilayer. The helical axis, defined as the least momentof inertia axis, intersects this plane at position x,y with residue “h.” Thehelical axis is tilted by an angle θ with respect to the z axis that is perpen-dicular to the membrane. The azimuthal angle ϕ of this tilt in the x-y plane andthe rotation angle of the helix around this axis is the angle η.

Fig. 2. Comparison of the predicted MVC binding site (MVC in yellow andbinding site residues in cyan) with the experimental X-ray MVC binding site(MVC in purple and binding site residues in green).

Abrol et al. PNAS | September 9, 2014 | vol. 111 | no. 36 | 13041

BIOCH

EMISTR

Y

instead, the network consisting of Y37, W86, Y108, and E283residues was proposed.Thus, binding of MVC causes helix 2 to rotate by 20 degrees,

which opens up the binding pocket formed by helices 1, 2, and 7.This rotation also allows the W86 residue to find an interactionwith Y108 on helix 3, while interacting with the ligand, whereasT284 still forms a hydrogen bond with Y37 on TM1. Addition-ally these rotations aid the hydrogen bonding between N71(TM2)–N293 (TM7) due to a more favorable geometry of these tworesidues. Moreover, the 15-degree rotation of H7 allows the E283residue to find a better orientation, optimizing the interaction be-tween the carboxylic group and the basic nitrogen from the tropanering of MVC, as well as the hydroxyl group of Y251 on helix 6.The binding pocket of the predicted complex model is in very

good agreement with the crystal structure of the MVC:CCR5complex published just recently (8). Fig. 2 also shows the MVC–CCR5 interactions observed in the crystal structure of thecomplex for comparison with the predicted structure. Fig. S1 Aand B shows these interactions using a 2D representation. Allimportant interactions between the MVC and CCR5 residueswere predicted, including (i) the salt bridge between the tropanenitrogen atom and Glu283, (ii) the hydrogen bond between thecarboxyamide nitrogen and Tyr251, and (iii) the hydrophobicinteractions of the phenyl group with Tyr108, Phe109, andPhe112. The predicted structure is not perfect and missed someinteractions: (i) the hydrogen bond between the amine moiety ofthe triazole group and Tyr37 and hydrogen bonds between oneof the fluorines on the cyclohexane ring with Thr195 and Thr259.The accuracy of the predicted CCR5 structure (obtained using

the human β2 adrenergic receptor template) shows the strengthof the BiHelix and SuperBihelix methodology. This approachallows sampling of a complete set of TM rotations and local tilts,going well beyond simple homology modeling. This method alsoallows the residues crucial for ligand binding to be correctlypositioned in the binding pocket, leading to high-quality proteinstructures starting from relatively low-homology templates (25%sequence identity between β2AR and CCR5 in the TM region).The MVC ligand does not select the lowest-energy receptor

conformation (WT1) for binding but a higher-energy one (WT7).The helix rotation η angles change from [0, −120, 0, 0, 15, 0, 0]for the WT1 conformation to [15, 0, 0, 105, 15, 0, 0] for the WT7conformation, which is very close to the [15, 0, 10, 105, −15, 0, 0]angles calculated for the crystal conformation of CCR5 bound toMVC (8). The predicted TM domains are also in good agree-ment with the crystal structure (crystal TMs/predicted TMs):TM1, 26–57/26–57; TM2, 64–92/64–93; TM3, 97–132/98–131;TM4, 142–166/142–166; TM5, 187–222/192–224; TM6, 229–261/229–259; and TM7, 269–297/276–299.The predicted PF binding mode (Fig. 3A) is similar to that of

MVC, based also on the same protein configuration WT7. Thisresult is reasonable because the ligand structures are rather similar.Thus, the mechanism of PF binding to CCR5 protein is similar tothat of MVC binding, involving also the rotation of helices 2 and 7,which opens up the 1-2-7 binding pocket to allow the ligand to bindto E283. Fig. S1C shows the PF binding site in a 2D representation.Although APL has a slightly different molecular scaffold,

it shares with MVC and PF a strongly basic nitrogen atom

positioned in the center of the molecule (see central N atom inthe ligand structures shown in Scheme S1). Indeed, it interactsstrongly with the E283 anchor point (Fig. 3B), similar to theMVC and PF cases. Thus, the preferred CCR5 configuration isWT7, just as for MVC and PF. However, the benzoic acid groupof APL provides an additional anchoring point for the ligand, farfrom helix 7 and E283. It interacts with the K191 residue locatedon the verge of helix 5 and extracellular loop 2, leading to a largereduction of binding upon K191A mutation. This result is con-sistent with the experimental >100-fold decrease in binding af-finity of APL upon K191A mutation (21, 22).The mechanism of binding of APL also involves breaking or

weakening the 1-2-7 hydrogen bond network (Y37-W86-T284),with E283 providing the anchor for the middle part of the ligand.Interestingly, the benzoic acid group also interacts with R168 onextracellular loop 2, also shown to be important for ligandbinding in experimental assays (23).The TAK antagonist is based on a completely different scaf-

fold containing a quaternary amine group in place of the pro-tonated nitrogen for the other three ligands, and giving thewhole ligand a formal charge of +1. We predict that TAK prefersto bind to the WT10 conformation (Fig. 3C). The overallmechanism of opening the binding site via disruption of theW86-T284 hydrogen bond is preserved. As in the previous cases,the 1–2-7 hydrogen bond network is broken or weakened due tothe 20-degree rotation of helix 2 and formation of a π-π stackinginteraction between W86 and Y108. The T284-Y37 hydrogenbond stabilizes the ligand-bound structure, in agreement with theexperimental data that show lower binding affinity of TAK uponthe Y37A mutation (21). Additionally, the 15-degree rotation ofhelix 5 allows the T195 residue to rotate away from the toluylgroup of the ligand, making more space available in the 3-5-6pocket. On the other hand, the relatively large rotation of weaklyinteracting H4 does not seem to have any impact on the structureof the protein or ligand binding.A detailed analysis of the TAK binding site shows that W86 is

the strongest interacting residue with this ligand, with a −3.7 kcal/molcontribution to the overall interaction energy, which agreeswell with the experimental data. This favorable interaction comesfrom an imperfect π–π stacking between the tryptophan residueand the benzo[7]annulene part of the ligand. We predict thatthe contribution of E283 residue is the second strongest one(−3.3 kcal/mol).On the other hand, the interaction of the ligandwithY108 is very weak so that the relatively large effect of the Y108Amutation seen experimentally results because WT22 becomes themost stable TAK bound conformation in the Y108A receptormutant. Thus, the large effect of the mutation is indirect throughstabilizing a different protein conformation rather than a changein direct interaction with the ligand. All other residues in-vestigated experimentally show very weak interactions with TAK,consistent with the experimental mutation data.

Mutation Experiments. Based on the binding structure for MVCto WT CCR5, we selected 15 single mutants to validate thepredicted structure: F109Y, F112A, Q194A, Y251F, D276A,Q277A, W86A, A90H, T105A, Y108A, F109A, I198A, Y251A,Q280A, and E283A. For 11 of these mutants, binding studies of

Fig. 3. The predicted binding sites for (A) PF-232798 (PF), (B) Aplaviroc (APL), and (C) TAK-779 (TAK).

13042 | www.pnas.org/cgi/doi/10.1073/pnas.1413216111 Abrol et al.

MVC and PF were carried out. The predicted results weredramatically inconsistent with experiments for the A90H andF109A mutants. Also, the experiments suggested a biggerchange in binding energy for PF vs. MVC in the A90H mutant.A fundamental assumption in our selection of mutations wasthat the structure of the binding site would not change signifi-cantly (the established paradigm for assessing the effect ofmutations). These experimental results suggested that this as-sumption might be invalid for CCR5 and potentially otherGPCRs. This result is also consistent with our earlier studies onbinding of the ligand BX471 to CCR1 receptor, where we foundthat a single mutation of human CCR1 to mouse and rat CCR1dramatically changed the binding (24).Rather than starting all over to predict the hierarchy of CCR5

structures for each of the 15 mutant receptors, we simplified theprocedure as follows. We went back to the 100 lowest-energystructures from the conformational hierarchy predicted for WTCCR5; we mutated the single residue in question for each ofthose structures, and then we reassigned side chains, usingSCREAM (15), and minimized the structures. The net result isoften a dramatic reordering in the ranking of these 100 receptorstructures. We then selected the 20 lowest-energy structures fordocking the four ligands.This approach leads to dramatically different hierarchies, as

shown in Table S2. For example, the W86A mutant makes theWT30 conformation fourth best. This outcome is easy to un-derstand from the structures. WT30 conformation has the W86residue up against TM7 (Fig. S2), but replacing this residue byAla in the W86A mutant removes this steric clash, dramaticallyimproving the rank of the WT30 conformation to fourth.Next, we docked the 20–30 conformations for each of the four

ligands to each of the ∼10 receptor conformations for each of themutated proteins using the DarwinDock/GenDock methods(18). The results are shown in Table 1. For the top-scoring poses,a water displacement penalty was applied if a polar residue wasmutated to a nonpolar residue and could have accommodated awater molecule in the apo protein conformation. This penaltyaffected only the Q194A mutant.The results in Table 1 show the pronounced effect of muta-

tions in the active site. Thus, the WT7 configuration stabilized byMVC binding to WT CCR5 is not the best binding configurationfor any of the 11 mutants, all of which were aimed at modifying thebinding! Instead, WT5 conformation appears seven times, WT2

four times, WT3 three times, and WT10 once. These mutants weregenerated experimentally (as described in SI Materials and Methods,Experimental Generation of CCR5 Mutants) and used to measure Kivalues for the binding of different ligands to CCR5WT and mutantreceptors (SI Materials and Methods, Ligand-Binding Experiments).Comparing relative Ki values to relative predicted binding energiesfor each case led to the results in Fig. 4, which show excellentagreement with experimentally obtained Ki values for MVC bindingto the mutants relative to the WT receptors, with the bindingconstants varying over a range larger than 3 logs. The trend for theother three ligands is also correct, but the correlation is weaker thanthat for MVC, due in part to the fact that the experimental changesin binding upon mutations span less than three orders of magnitude(less than 3 logs) for these ligands, which is only slightly larger thanthe best expected resolution from the calculations of ∼2 logs [Fig.S3 (PF), Fig. S4 (APL), and Fig. S5 (TAK)]. These results overallconfirm the prediction that different conformations can be stabi-lized under different conditions (mutants and/or ligands).

DiscussionAnalysis of the Ligand–CCR5 Binding Site in the Context of Mutationsand Ligand Structure–Activity Relationship Series. The experimentalassays confirmed the predictions that the binding modes of MVCand PF232798 are rather similar, but with one exception. TheW86A mutation gives a 629-fold decrease in binding of PF anda 29.8-fold decrease in binding for MVC. [Kondru et al. reporteda 10-fold decrease in MVC binding upon W86A mutation (22) inslightly different conditions.] Indeed, the calculations find thatfor W86A, both MVC and PF bind to WT7, but with predictedenergy changes of 3.1 kcal/mol and 3.5 kcal/mol.The difference in impact of the W86A mutation on MVC/PF

binding comes in part from structural differences between thosetwo ligands. Although they share a similar scaffold, they havea different stereochemistry for the five-membered nitrogen-richring that forces the boat conformation of the tropane ring for PFwhereas MVC retains the favorable chair conformation. Alsothere is an additional cyclohexane ring in the terminal part ofMVC that affects the pKa value of the N3 atom, making it morebasic. The quantum mechanical calculations predict the pKavalue of the nitrogen in the five-membered ring to be 5.0–6.0 forMVC, but 6.0–7.0 for PF, suggesting that MVC may exist ina doubly protonated state in the experimental environment (thetropane N atom is very basic for both MVC and PF, with a pre-dicted pKa of ∼9). Because, for WT, the residue W86 is close tothe five-membered ring of both MVC and PF, this change in theprotonation state could have been responsible for the predicteddifferent interaction with tryptophan through a strong cation–πinteraction (25).

Table 1. Predicted lowest-energy protein structures for wild-type and mutant receptors in the apo and ligand-bound forms

Mutations

Lowest-energy CCR5 conformations

Apo MVC PF APL TAK

Wild-type wt1 wt7 wt7 wt7 wt10F109Y wt3 wt5 wt5 wt5 wt3F112A wt1 wt5 wt5 wt5 wt3Q194A wt1 wt5 wt5 wt5 wt4Y251F wt1 wt2 wt22 wt4 wt3D276A wt1 wt2 wt7 wt3 wt2Q277A wt1 wt5 wt5 wt5 wt3W86A wt1 wt7 wt7 wt7 wt3A90H wt3 wt10 wt5 wt10 wt3T105A wt1 wt2 wt6 wt6 wt3Y108A wt3 wt2 wt22 wt5 wt22F109A wt1 wt5 wt7 wt7 wt3I198A wt1 wt5 wt5 wt5 wt25Y251A wt2 wt3 wt5 wt5 wt3Q280A wt3 wt5 wt7 wt7 wt7E283A wt3 wt3 wt22 wt22 wt3

Ligands used are Maraviroc (MVC), PF-232798 (PF), Aplaviroc (APL), andTak-779 (TAK). This shows that a single mutation can dramatically affect thebinding site.

Fig. 4. Comparison of the predicted effect of Maraviroc binding to mutantsvs. experimental data for the 11 cases studied experimentally. The red line isthe fit (R2 = 0.89) whereas the green lines indicate a deviation of 1 kcal/mol.

Abrol et al. PNAS | September 9, 2014 | vol. 111 | no. 36 | 13043

BIOCH

EMISTR

Y

However, our predicted structure makes this hypothesis un-likely because W86 is oriented in a way not allowing for theπ–cation interaction in the PF. Despite a similar orientation ofW86 in the MVC-bound and PF-bound models, we predicta difference in the interaction energy between those two ligandsand the tryptophan residue. The binding site energy analysisallows the ligand–protein interaction to be partitioned intocontributions from each residue. For the MVC binding site, wefound a ligand interaction energy of −2.0 kcal/mol with W86(which is less than the E283 contribution of −3.2 kcal/mol) whereasfor PF it is −4.2 kcal/mol (almost identical to −4.3 kcal/molcontribution of E283). This difference can be understood interms of the geometries of the ligands (the different stereo-chemistry of the tropane ring-five membered ring connections)and the different position of the phenyl ring with respect to thetropane ring for MVC and PF, leading to different interactionenergies.A second important difference in the experimental mutation

data for these two ligands occurs for the A90H mutation, whichchanges the PF binding significantly, but has almost no effect onMVC binding. PF prefers the WT5 conformation for the A90Hmutant whereas MVC prefers WT10. Even though both MVCand PF are anchored by E283 and occupy a similar position inthe protein, they are structurally different in the part thatinteracts with helix 2. The bulkier pyridyl group of PF positionsthe ligand closer to the A90 residue, which does not affect thebinding unless this residue is mutated to a larger moiety: i.e., thehistidine residue. Indeed, MVC is relatively far from the A90residue (which is on top of helix 2); as a result, replacing A90 bythe bulky histidine does not affect its binding. However, the li-gand PF is positioned closer to A90 so that the mutation tohistidine causes a steric clash, decreasing binding. This differencearises from the difference in the receptor conformation for theA90H mutant preferred by MVC (WT10) versus PF (WT5).On the other hand, all 11 mutations investigated force the

TAK-bound CCR5 system to switch to a different conformation.This conformational change has, however, little impact on thecomputational mutation results. For all six mutations studiedexperimentally for TAK, the results from both structures were ingood agreement with experimental data.Unfortunately, the structure–activity relationship (SAR) series

for the PF compound showed (3) little structural diversity. Theexperimental data presented in ref. 3 agrees well with thePF232798 binding pose predicted here. All 3-Substituted 1,4,6,7-Tetrahydro-imidazo[4,5-c]pyridines that were tested experimentally(compounds 39 and 41a–h from ref. 3) show a very high affinity, onthe level of the PF232798 system. In the predicted docking pose,this part of the ligand interacts with the flexible N terminus of theprotein and there is enough space to incorporate any substituent

suggested in the original work. The modification of the ligandyielding 3-substituted derivatives (compounds 20a–h and 29–32from ref. 3) positions the terminal part of the ligand closer toTM2, but avoiding steric clashes with any protein residues. Thisshort SAR series further supports the predicted PF232798binding pose.APL (Aplaviroc) has been developed from a series of spi-

rodiketopiperazine derivatives showing a relatively high CCR5activity (5). In that study (5), four different spirodiketopiperazinederivatives were evaluated for anti–HIV-1 activity, and moleculeE913 was selected as the lead compound for a newer round ofoptimizations. Our predicted docking pose of APL is in agree-ment with the experimental data for this series of compounds.The main difference among the earlier spirodiketopiperazinederivatives E910, E913, E916, and E917 is the length of themolecule’s terminal part targeting the TM3-TM4-TM5 region ofCCR5. According to the predicted pose, there is enough space toincorporate even the longest, heptylbenzene tail present in E910.This region is also highly hydrophobic, with a number of residues(F109, F112, F168, and I198) able to make relatively stronginteractions with the phenyl ring present at the end of the hy-drophobic tail of all these ligands. The difference responsible forthe strong binding of APL is the addition of the benzoic acidgroup to the end of this tail, which is the most important dif-ference between APL and E913. In APL, the phenoxy linker isable to position the new carboxylic group to promote a stronginteraction with K191, known to be important in APL bindingfrom the experimental mutation data.For TAK-779, the original paper (4) does not disclose any

SAR series. Other Takeda compounds with known CCR5 ac-tivity (like TAK-220) are completely different.

Implications for the Conformational Ensemble View of GPCRs. Fig. 5summarizes the results for all ligands and mutants. Fig. 5A showsthat MVC, PF, and APL prefer to bind to the WT7 receptorconformation, not to the lowest-energy WT1 conformation ofthe apo-protein. TAK prefers to bind to the WT10 conforma-tion. This difference between TAK and other ligands arises be-cause TAK has a quaternary nitrogen group in place of thetertiary nitrogen of the other ligands. Fig. 5B shows how differ-ent mutants of the receptor stabilize distinct conformations andindicates which conformation is preferred by ligands for differentmutants. These changes in conformation of the mutated apo-protein explain the differential mutational data for W86A andA90H mutants on binding of MVC vs. PF compounds. This re-sult has direct implications for the functional selectivity ofGPCRs because it shows that different receptor conformationsare stabilized under different conditions (ligand binding and/or

Fig. 5. (A) Conformational shuffling upon binding of different ligands. (B) CCR5 conformations preferred by different ligands and mutants.

13044 | www.pnas.org/cgi/doi/10.1073/pnas.1413216111 Abrol et al.

mutations), which potentially can have distinct downstreameffects on signaling.These results show that CCR5 is a relatively plastic protein

with distinctly different protein conformations stabilized by dif-ferent ligands. A single mutation can considerably alter the en-ergetics of the low-lying conformations to alter the preferredbinding site. This outcome reinforces the conformational selec-tion basis of the GPCR pleiotropic function. These results sug-gest that the paradigm of designing a ligand for a particularprotein structure will not be sufficient for developing new ther-apeutics to CCR5 (and likely other GPCRs). One must considerall of the low-energy protein conformations in the design and indetermining structure–activity relationships. Moreover, one mustbe cautious about interpreting mutation data. Thus, we considerthat methods presented in this work for predicting the effect ofmutations on GPCR structure will be essential to developinghighly selective and active ligands targeting these receptors.

ConclusionsWe developed an ensemble-based method that samples billionsof conformations to eventually select an ensemble of ∼20 low-energy protein conformations for the CCR5 receptor andshowed how this conformational ensemble changes upon bindingof four ligands and how it changes because of mutations. Toassess the validity of the binding sites, 15 mutations were pre-dicted in CCR5 that should significantly modify the binding.Experimental binding studies were carried out for 11 of thesemutants, and several cases were found that disagreed signifi-cantly from the expectations based on the predicted pose forbinding of the ligand MVC to the WT protein. This result mo-tivated us to predict the ensemble of low-energy protein struc-tures for each of the 15 mutants, some of which showed majorchanges in the conformations available for ligand binding. Wepredicted the optimum ligand pose for the ensemble of low-en-ergy CCR5 conformations for each of 11 mutants and comparedwith our experimental binding results. We found that ligandbinding to the predicting mutated receptor conformations agreeswell with the experimental mutation data for all four ligands.This work leads to several important general conclusions. (i) Dif-ferent ligands can stabilize dramatically different seven-helix bundle

conformations, complicating the use of experimental structures fordesigning new strongly interacting ligands. (ii) For CCR5 (andlikely other GPCRs), even a single mutation can dramaticallychange the binding site, complicating experimental validation ofligand-binding structures. This result invalidates the standard par-adigm of using directly the GPCR structure bound to the ligand toexplain mutation-binding data. (iii) The changes in protein con-formation stabilized by different ligands and/or mutants might leadto changes in functionality (a structural basis for GPCR pleiotropy).Indeed functional differences between agonists and inverseagonists (or antagonists) to GPCRs are well established. For ex-ample, recent studies show that single, double, and triple mutantforms of the CB1 receptor exhibit dramatic changes in function,switching from full inactivity to high constitutive activity and backfor sequential mutations (26, 27). (iv) The GEnSeMBLE approachfor predicting the ensemble of thermally accessible protein–ligandcomplexes provides a tool for accurate GPCR–ligand predictionsconsistent with the GPCR pleiotropy paradigm. The validation byboth the mutation data and the subsequently published crystalstructure for the MVC:CCR5 complex indicates that this approachmay be trusted for new GPCRs for which there are little or no data.

Materials and MethodsThe computational methods used in this study for the structure prediction ofCCR5 receptor in the WT and mutant forms as well as ligand docking weredescribed briefly in Results. Key aspects of the method for predicting re-ceptor structure are that we do a very complete sampling of helix tilts androtations (∼11 billion conformations) and use energy as the sole criterion toselect a conformational ensemble for docking the ligands. Key aspects of themethod for predicting the binding site are that for each of the ligands’multiple torsional conformations, we docked ∼50,000 poses to the receptor’sfull conformational ensemble, from which we eventually select the best posesolely on the basis of energy. SI Materials and Methods contains detaileddescriptions of those computational methods along with the experimentalmethods used in this study for the receptor mutant generation and ligand-binding assays.

ACKNOWLEDGMENTS. We thank Soo-Kyung Kim and Andrea Kirkpatrickfor useful discussions. A portion of this research was funded by a giftfrom Accelerator/PharmSelex. The balance was from gifts to the Materialsand Process Simulation Center.

1. Dragic T, et al. (1996) HIV-1 entry into CD4+ cells is mediated by the chemokine re-ceptor CC-CKR-5. Nature 381(6584):667–673.

2. Dorr P, et al. (2005) Maraviroc (UK-427,857), a potent, orally bioavailable, and selectivesmall-molecule inhibitor of chemokine receptor CCR5 with broad-spectrum anti-humanimmunodeficiency virus type 1 activity.Antimicrob Agents Chemother 49(11):4721–4732.

3. Stupple PA, et al. (2011) An imidazopiperidine series of CCR5 antagonists for thetreatment of HIV: The discovery of N-(1S)-1-(3-fluorophenyl)-3-[(3-endo)-3-(5-iso-butyryl-2-methyl-4,5,6,7-tetrahydro-1H-imidazo[4,5-c]pyridin-1-yl)-8-azabicyclo[3.2.1]oct-8-yl]propylacetamide (PF-232798). J Med Chem 54(1):67–77.

4. Baba M, et al. (1999) A small-molecule, nonpeptide CCR5 antagonist with highlypotent and selective anti-HIV-1 activity. Proc Natl Acad Sci USA 96(10):5698–5703.

5. Maeda K, et al. (2001) Novel low molecular weight spirodiketopiperazine derivativespotently inhibit R5 HIV-1 infection through their antagonistic effects on CCR5. J BiolChem 276(37):35194–35200.

6. Grunbeck A, et al. (2012) Genetically encoded photo-cross-linkers map the binding siteof an allosteric drug on a G protein-coupled receptor. ACS Chem Biol 7(6):967–972.

7. Berro R, et al. (2013) Use of G-protein-coupled and -uncoupled CCR5 receptors byCCR5 inhibitor-resistant and -sensitive human immunodeficiency virus type 1 variants.J Virol 87(12):6569–6581.

8. Tan Q, et al. (2013) Structure of the CCR5 chemokine receptor-HIV entry inhibitormaraviroc complex. Science 341(6152):1387–1390.

9. Abrol R, Griffith AR, Bray JK, Goddard WA, III (2012) Structure prediction of G protein-coupled receptors and their ensemble of functionally important conformations. Mem-brane Protein Structure and Dynamics: Methods and Protocols, Methods in MolecularBiology, eds Vaidehi N, Klein-Seetharaman J (Humana, New York), Vol 914, pp 237–254.

10. Abrol R, Bray JK, Goddard WA, III (2011) Bihelix: Towards de novo structure predictionof an ensemble of G-protein coupled receptor conformations. Proteins 80(2):505–518.

11. Okada T, et al. (2004) The retinal conformation and its environment in rhodopsin inlight of a new 2.2 A crystal structure. J Mol Biol 342(2):571–583.

12. Cherezov V, et al. (2007) High-resolution crystal structure of an engineered humanbeta2-adrenergic G protein-coupled receptor. Science 318(5854):1258–1265.

13. Jaakola VP, et al. (2008) The 2.6 angstrom crystal structure of a human A2A adenosinereceptor bound to an antagonist. Science 322(5905):1211–1217.

14. Warne T, et al. (2008) Structure of a beta1-adrenergic G-protein-coupled receptor.Nature 454(7203):486–491.

15. Kam VWT, Goddard WA, III (2008) Flat-bottom strategy for improved accuracy inprotein side-chain placements. J Chem Theory Comput 4(12):2160–2169.

16. Mayo SL, Olafson BD, Goddard WA, III (1990) Dreiding: A generic force-field formolecular simulations. J Phys Chem-Us 94(26):8897–8909.

17. Bray JK, Abrol R, GoddardWA, III, Trzaskowski B, Scott CE (2014) SuperBiHelixmethod forpredicting the pleiotropic ensemble of G-protein–coupled receptor conformations. ProcNatl Acad Sci USA 111(1):E72–E78.

18. Goddard WA, III, et al. (2010) Predicted 3D structures for adenosine receptors boundto ligands: comparison to the crystal structure. J Struct Biol 170(1):10–20.

19. KimS-K, Riley L, Abrol R, JacobsonKA,GoddardWA, III (2011) Predicted structures of agonistand antagonist bound complexes of adenosine A3 receptor. Proteins 79(6):1878–1897.

20. Hall SE, et al. (2009) Elucidation of binding sites of dual antagonists in the humanchemokine receptors CCR2 and CCR5. Mol Pharmacol 75(6):1325–1336.

21. Maeda K, et al. (2006) Structural and molecular interactions of CCR5 inhibitors withCCR5. J Biol Chem 281(18):12688–12698.

22. Kondru R, et al. (2008) Molecular interactions of CCR5 with major classes of small-molecule anti-HIV CCR5 antagonists. Mol Pharmacol 73(3):789–800.

23. Berro R, et al. (2011) Multiple CCR5 conformations on the cell surface are used dif-ferentially by human immunodeficiency viruses resistant or sensitive to CCR5 in-hibitors. J Virol 85(16):8227–8240.

24. Vaidehi N, et al. (2006) Predictions of CCR1 chemokine receptor structure and BX 471antagonist binding followed by experimental validation. J Biol Chem 281(37):27613–27620.

25. Dougherty DA (1996) Cation-pi interactions in chemistry and biology: A new view ofbenzene, Phe, Tyr, and Trp. Science 271(5246):163–168.

26. Ahn KH, Scott CE, Abrol R, Goddard WA, III, Kendall DA (2013) Computationally-predicted CB1 cannabinoid receptor mutants show distinct patterns of salt-bridgesthat correlate with their level of constitutive activity reflected in G protein couplinglevels, thermal stability, and ligand binding. Proteins 81(8):1304–1317.

27. Scott CE, Abrol R, Ahn KH, Kendall DA, Goddard WA, III (2013) Molecular basis fordramatic changes in cannabinoid CB1 G protein-coupled receptor activation uponsingle and double point mutations. Protein Sci 22(1):101–113.

28. Abrol R, Kim S-K, Bray JK, Griffith AR, Goddard WA, III (2011) Characterizing andpredicting the functional and conformational diversity of seven-transmembraneproteins. Methods 55(4):405–414.

Abrol et al. PNAS | September 9, 2014 | vol. 111 | no. 36 | 13045

BIOCH

EMISTR

Y

Supporting InformationAbrol et al. 10.1073/pnas.1413216111SI Materials and MethodsComputational Methodology. We manually aligned the C-C che-mokine receptor type 5 (CCR5) transmembrane (TM) sequenceto the sequences of four available G protein-coupled receptors(GPCRs): bovine rhodopsin (PDB ID code 1u19), human β2adrenergic receptor (PDB ID code 2rh1), human adenosine A2Areceptor (PDB ID code 3eml), and turkey β1 adrenergic receptor(PDB ID code 2vt4) using the conserved residues found in eachTM. We predicted the lengths of the CCR5 TMs based on thePredicTM approach that predicts the hydrophobic helical re-gions and combines them with the output of three predictionservers—Porter (1), APSSP2 (2) and PsiPred (3)—to predict anyhelical extensions beyond the membrane. The starting structureof the TM bundle for each template was obtained by mutatingthe template residues to the CCR5 residues (as in homologymodeling) but using the SideChain Rotamer Energy AnalysisMethod (SCREAM) side-chain optimization protocol and a 100-step minimization of TMs (with the Dreiding force field) per-formed independently for each helix (i.e., without the presenceof other helices) (4, 5). Afterward, the mutated and optimizedhelices were put back together to form a seven-helix bundle. Thisapproach yielded four CCR5 seven-transmembrane helical do-mains, each based on a different experimental GPCR structure.In the next step, the BiHelix step (6) was performed, followed bythe SuperBiHelix step (7), as described in Prediction of the CCR5Structure in the main text.Because some loop residues (e.g., K191) were suggested to be

important for ligand binding, we modeled the extracellular loopsof CCR5 before ligand docking, Thus, we built the extracellularloops and N terminus using homology approaches and the se-quence similarity to human β2 adrenergic receptor loops (ex-tracellular loop 1, 12 first residues of the extracellular loop 2)and human adenosine A2A receptor loops (last 14 residues ofthe extracellular loop 2, extracellular loop 3). We built the Nterminus using the part of the bovine rhodopsin N terminusfused with the experimental NMR structure of the middle part ofthe CCR5 N terminus (PDB ID code 2RLL). After mutating theresidues to the respective CCR5 residues using the SCREAMprotocol, the loops were connected to the proper helices, and theconnected regions were minimized for 1,000 steps using a Dreidingforce field.The intracellular loops and helix 8 were modeled after docking,

using an approach similar to that used for the extracellular loops,with homology modeling and sequence similarity to human β2adrenergic receptor loops.Before the docking of each CCR5 ligand described in this work

[Maraviroc (MVC), PF-232798 (PF), Aplaviroc (APL), and TAK-779 (TAK)], we carried out a search over conformational space.Here, torsionally diverse conformations for each ligand were gen-erated by the Maestro MacroModel program using a 0–10 kcal/molwindow and requesting 10,000 poses. These conformations wereclustered at 2.0 Å heavy-atom rmsd into a number of diversefamilies whose family heads were then docked to the selectedCCR5 models. This procedure yielded 17 different conformationsfor Aplaviroc, 31 conformations for Maraviroc, 35 conformationsfor PF-232798, and 22 conformations for TAK-779. Each of theseconformations was docked to each of the ∼10 ensembles of proteinconformations using the DarwinDock protocol.The DarwinDock protocol used in this study relied on sampling

a complete set of ligand poses (∼50,000) in the potential bindingregions for each ligand conformation. In the first step, each of theCCR5 model binding sites had the bulky, hydrophobic residues

(ILE, LEU, PHE, TRP, TYR, and VAL) mutated to alanineresidues to allow flexibility in the binding site. Next, ∼300 poseswere selected for each CCR5 model–ligand pair using the lowest-energy criterion a Dreiding force field and three different scoringcriteria: total energy, polar energy (defined as the sum of cou-lombic and H-bonding energy contributions to the total energy),and hydrophobic energy (defined as Van der Waals energy con-tribution to the total energy). Afterward, the bulky hydrophobicresidues were restored to the original ones, and the entire bindinghad the side chains optimized using the SCREAM protocol. Thisapproach allows a different side-chain conformation for each ofthe 300 poses. In the next step, a 10-step optimization of thebinding site was performed using a Dreiding force field fromwhich we selected 50% of the lowest-energy poses for furtherrefinement. The computational charge neutralization of the wholesystem was then performed by mutating the charged residues ofthe CCR5 (ASP, ARG, GLU, and LYS) to their charge-neutralforms as well as detaching the most basic hydrogen atom fromAplaviroc, Maraviroc, or PF-232798 to change the formal chargeof these ligands from +1 to 0. These neutral protein-ligand sys-tems were subjected to a 10-step minimization run, and half ofthe lowest-energy poses were selected for the final step. Here,a 100-step minimization of the entire ligand–protein complexwas performed and the best systems were selected based on thelowest total energy of the neutral system.The procedure of obtaining models of CCR5 mutants was the

following for each mutation listed in Table 1. For each of the top100 CCR5 models obtained in the SuperBihelix step, the specificresidue was mutated, the respective side chain was optimized usingthe SCREAM protocol, and the entire protein was minimized for100 steps. Then, the CCR5 mutant models were rescored using anaverage of total and interhelical energies with the Dreiding forcefield. This procedure allowed for the evaluation of the structuralchanges in the apo form of the protein upon mutations.The top 10 mutant receptor structures were selected and used

for ligand docking. In this step, a simpler docking protocol wasused, in which, for each ligand, the final pose from theDarwinDockstep was taken and its coordinates were matched to each of the 10low-energy CCR5 model mutants, followed by a binding-site op-timization. The complexes were evaluated, and one ligand–mutantcomplex with the lowest total energy was selected. This procedureallowed the ligand to “choose” a different conformation for eachCCR5 mutation studied.

Experimental Generation of CCR5 Mutants. Experimental mutationsin the ORF of the human CCR5 gene were carried out usingthe quick change lightning site-directed mutagenesis kit (AgilentTechnologies) following the kit’s recommendations. Briefly, forwardand reverse DNA primers (10 pmol each) encompassing the aminoacid to be mutated were mixed with kit components and 40 ng ofthe mammalian expression vector pcDNA6.2 /V5/GW/D-Topo(Invitrogen) containing the human CCR5 ORF as a template.After 18 rounds of PCR amplification, the template in the re-

action was digested with the DpnI restriction enzyme for 30 min at37 °C, and then 2 μL was transformed into XL-10 gold ultra-competent bacteria (Agilent Technologies) following the kit’s in-structions. Bacteria from the transformation reaction was platedinto LB-Ampicillin agar plates at three different densities andincubated overnight at 37 °C. Next day, for each CCR5 mutationmade, five bacterial colonies were picked and grown in LB mediaovernight. From these cultures, plasmid mini preps were performedusing QIAprep spin miniprep kit (Qiagen), and the specific

Abrol et al. www.pnas.org/cgi/content/short/1413216111 1 of 7

mutations within the CCR5 gene were verified by sequencing thecomplete ORF using specific primers.Once bacterial clones containing sequence-verified CCR5

mutants were obtained, endotoxin-free plasmid maxiprep wasdone (Qiagen) to obtain specific mutant CCR5 expression vec-tors. These vectors were later used in Maraviroc ligand-bindingassays for structure–activity relationship (SAR) studies.

Ligand-Binding Experiments.Tritium-labeledMaraviroc (99 Ci/mmol)was custom-made at Amersham-GE Healthcare Life Sciences. Toprepare membranes for ligand-binding studies, CCR5 expressionconstructs were transiently transfected into HEK-293 cells (ATCCno. CRL-1573) using the Effectene Transfection Reagent (Qiagen)according to the manufacturer’s instructions. The cells were har-vested 3 d posttransfection as follows. Transfected cells grown onT-225 flasks (two flasks per expression construct), rinsed with PBScontaining 5 mM EGTA (PBS/EGTA), and incubated in PBS/EGTA for 30 min at room temperature. Detached cells werepelleted at 300 × g for 10 min and resuspended in 8 mL of ice-cold20 mM Hepes (pH 7.3) containing 1% (wt/vol) protease inhibitorsmixture (Halt; Thermo Fisher). After 10 min on ice, the cells

were homogenized by sonication (7× 20-s bursts). Homogenateswere centrifuged at 300 × g for 10 min, and supernatants werecollected and centrifuged for 2 h at 50,000 × g. Membranes wereresuspended in 1 mL of 20 mMHepes (pH 7.3) and stored at−70 °C.Protein content was determined using the BCA Protein Assay Re-agent (Thermo Fisher).For receptor-binding studies, 10 μg of membranes were in-

cubated in 96-well polypropylene plates with tritium-labeledMaraviroc for 2 h at room temperature in 300 μL of binding buffercontaining 50 mM Hepes (pH 7.3), 1 mM CaCl2, 5 mM MgCl2,0.1% DMSO, and 0.5% BSA. Nonspecific binding was measuredin the presence of 1,000-fold excess of unlabeled Maraviroc. Thesamples were filtered through UniFilter-96 GF/C plates pre-blocked with 0.3% polyethylene imine using a Unifilter-96 Har-vester (PerkinElmer). The filters were washed four times with PBS,and radioactivity was measured using a TopCount instrument(PerkinElmer) in 40 μL of MicroScint 20 liquid scintillant. Kd andKi values were obtained from saturation and competition bindingexperiments, respectively, and calculated using GraphPad Prism(GraphPad Softaware).

1. Pollastri G, McLysaght A (2005) Porter: A new, accurate server for protein secondarystructure prediction. Bioinformatics 21(8):1719–1720.

2. Raghava GPS (2002) APSSP2: A combination method for protein secondary structureprediction based on neural network and example based learning. Available at www.imtech.res.in/raghava/apssp2. Accessed January 2, 2014.

3. McGuffin LJ, Bryson K, Jones DT (2000) The PSIPRED protein structure prediction server.Bioinformatics 16(4):404–405.

4. Kam VWT, Goddard WA, III (2008) Flat-bottom strategy for improved accuracy in proteinside-chain placements. J Chem Theory Comput 4(12):2160–2169.

5. Mayo SL, Olafson BD, Goddard WA, III (1990) Dreiding: A generic force-field formolecular simulations. J Phys Chem 94(26):8897–8909.

6. Abrol R, Bray JK, Goddard WA, III (2011) Bihelix: Towards de novo structureprediction of an ensemble of G-protein coupled receptor conformations. Proteins80(2):505–518.

7. Bray JK, Abrol R, Goddard WA, III, Trzaskowski B, Scott CE (2014) SuperBiHelix methodfor predicting the pleiotropic ensemble of G-protein-coupled receptor conformations.Proc Natl Acad Sci USA 111(1):E72–E78.

Scheme S1. Structures of Maraviroc, PF-232798, Aplaviroc, and TAK-779.

Abrol et al. www.pnas.org/cgi/content/short/1413216111 2 of 7

Fig. S1. Schematic representation of the CCR5 residues in the MVC binding site for (A) WT7 (MVC binding CCR5 conformation) with MVC, (B) X-ray CCR5conformation with MVC, (C) WT7 (PF-232798 binding CCR5 conformation) with PF, and (D) WT1 (lowest energy ranked CCR5 apo WT conformation) no ligand.

Abrol et al. www.pnas.org/cgi/content/short/1413216111 3 of 7

Fig. S2. Structural basis for the effect of the W86A mutation on the wt30 conformation and for the A90H mutation on the wt1 conformation.

Fig. S3. Comparison of predicted effect of PF binding to mutants vs. experimental data. The ±1 kcal/mol lines are shown in green.

Abrol et al. www.pnas.org/cgi/content/short/1413216111 4 of 7

Fig. S4. Comparison of predicted effect of Aplaviroc binding to mutants vs. experimental data. The ±1 kcal/mol lines are shown in green.

Fig. S5. Comparison of predicted effect of Tak-779 binding to mutants vs. experimental data. The ±1 kcal/mol lines are shown in green.

Abrol et al. www.pnas.org/cgi/content/short/1413216111 5 of 7

Table

S1.

Thetop20

predictedco

nform

ationsforhuman

CCR5ch

emokinereceptor(from

GEn

SeMBLE

)

Thet

aH

1H

2H

3H

4H

5H

6H

7Ph

iH

1H

2H

3H

4H

5H

6H

7Et

aH

1H

2H

3H

4H

5H

6H

7A

vgR

ank

CC

R5

Ener

gy

Thet

a−

10−

10−

100

00

0Ph

i−

150

0−

15−

15−

15−

15Et

a0

240

00

150

012

.8w

t10

± 3

Thet

a0

−10

−10

00

00

Phi

−15

0−

150

00

0Et

a34

522

50

034

50

014

.5w

t2Th

eta

0−

10−

100

00

0Ph

i−

150

−15

−15

00

0Et

a34

522

50

034

50

021

.3w

t3Th

eta

−10

−10

−10

00

00

Phi

−15

15−

15−

150

−15

−15

Eta

345

225

015

345

00

71.3

wt4

15 ±

3Th

eta

0−

10−

10−

100

00

Phi

−15

00

150

00

Eta

345

225

015

345

00

89.8

wt5

Thet

a−

10−

10−

1010

00

0Ph

i−

15−

1515

−15

−15

00

Eta

150

015

300

098

.5w

t6Th

eta

−10

−10

00

100

0Ph

i−

15−

1515

−15

−15

−15

0Et

a15

00

105

150

010

0.3

wt7

Thet

a−

10−

10−

100

00

0Ph

i−

150

−15

−15

0−

15−

15Et

a0

225

015

345

00

108.

5w

t8Th

eta

0−

10−

10−

100

00

Phi

−15

00

15−

150

0Et

a34

522

50

1515

00

109.

8w

t918

± 4

Thet

a−

10−

10−

1010

00

0Ph

i−

15−

150

0−

1515

0Et

a15

150

1515

00

111.

0w

t10

Thet

a0

−10

−10

00

00

Phi

−15

0−

15−

150

−15

0Et

a33

022

50

1534

50

012

2.0

wt1

1Th

eta

−10

−10

−10

100

00

Phi

−15

−15

0−

150

00

Eta

150

010

534

50

012

6.3

wt1

2Th

eta

0−

10−

100

00

0Ph

i−

150

−15

−15

0−

15−

15Et

a34

522

50

034

50

014

3.5

wt1

3Th

eta

0−

10−

10−

100

00

Phi

−15

00

0−

150

0Et

a34

522

50

150

00

145.

5w

t14

20 ±

4Th

eta

00

−10

100

00

Phi

−15

−15

−15

−15

00

0Et

a0

00

120

345

090

145.

8w

t15

Thet

a−

100

−10

100

00

Phi

−15

−15

−15

−15

00

15Et

a0

150

105

345

00

146.

5w

t16

Thet

a−

10−

10−

1010

00

0Ph

i−

15−

150

−15

015

0Et

a15

150

120

750

014

7.0

wt1

7Th

eta

−10

−10

−10

−10

00

0Ph

i−

15−

150

00

150

Eta

1515

012

022

50

0w

t18

23 ±

2Th

eta

−10

−10

−10

−10

00

0Ph

i−

15−

150

15−

15−

150

Eta

150

012

015

00

wt1

9Th

eta

−10

−10

−10

−10

00

0Ph

i−

15−

150

−15

015

0Et

a15

150

120

225

00

wt2

0w

t50

27 ±

3

Fordockingto

ligan

ds,theco

nform

ationsin

theeightshad

edrowswereselected

based

ondiversity

anden

ergy.

After

wt20,

wt50co

nform

ation’sen

ergyislistedto

show

theflatteningoftheco

nform

ational

energylandscap

e.

Abrol et al. www.pnas.org/cgi/content/short/1413216111 6 of 7

Table S2. Reordering of CCR5 conformations for W86A, A90H,and T105A mutants

W86A Avg rank A90H Avg rank T105A Avg rank

wt1 8.0 wt3 3.0 wt1 4.0wt3 11.8 wt2 6.0 wt3 7.8wt2 15.0 wt4 15.3 wt2 10.0wt30 16.0 wt5 16.3 wt11 28.3wt6 20.8 wt13 18.5 wt9 28.8wt17 26.0 wt10 21.0 wt12 29.3wt7 28.3 wt9 22.5 wt18 30.5wt8 29.3 wt27 22.8 wt4 30.5wt4 30.0 wt14 23.8 wt27 30.8wt18 30.3 wt32 24.8 wt6 31.0wt11 30.3 wt64 25.0 wt7 31.3wt48 31.0 wt15 25.0 wt20 31.5wt20 31.3 wt16 25.8 wt8 31.8wt12 31.3 wt26 27.0 wt14 31.8wt19 34.8 wt28 27.5 wt13 34.0wt14 34.8 wt31 28.5 wt15 36.0wt9 35.8 wt17 29.3 wt5 36.0wt27 36.0 wt33 30.3 wt64 37.3wt34 37.0 wt11 32.0 wt32 38.0wt13 38.0 wt74 32.5 wt30 42.5

Abrol et al. www.pnas.org/cgi/content/short/1413216111 7 of 7