Structure and dynamics of the M3 muscarinic acetylcholine receptor.

Acetylcholine, the first neurotransmitter to be identified, exerts many of its physiological actions via activation of a family of G-protein-coupled receptors (GPCRs) known as muscarinic acetylcholine receptors (mAChRs). Although the five mAChR subtypes (M1-M5) share a high degree of sequence homology, they show pronounced differences in G-protein coupling preference and the physiological responses they mediate. Unfortunately, despite decades of effort, no therapeutic agents endowed with clear mAChR subtype selectivity have been developed to exploit these differences. We describe here the structure of the G(q/11)-coupled M3 mAChR ('M3 receptor', from rat) bound to the bronchodilator drug tiotropium and identify the binding mode for this clinically important drug. This structure, together with that of the G(i/o)-coupled M2 receptor, offers possibilities for the design of mAChR subtype-selective ligands. Importantly, the M3 receptor structure allows a structural comparison between two members of a mammalian GPCR subfamily displaying different G-protein coupling selectivities. Furthermore, molecular dynamics simulations suggest that tiotropium binds transiently to an allosteric site en route to the binding pocket of both receptors. These simulations offer a structural view of an allosteric binding mode for an orthosteric GPCR ligand and provide additional opportunities for the design of ligands with different affinities or binding kinetics for different mAChR subtypes. Our findings not only offer insights into the structure and function of one of the most important GPCR families, but may also facilitate the design of improved therapeutics targeting these critical receptors.

The mAChR family consists of five subtypes, M1-M5, which can be subdivided into two major classes (Fig 1a). The M1, M3, and M5 receptors show selectivity for G proteins of the Gq/11 family, whereas the M2 and M4 receptors preferentially couple to Gi/o-type G proteins[2-4]. The development of small molecule ligands that can selectively act on specific mAChR subtypes has proven extremely challenging, primarily due to the high degree of sequence similarity in the transmembrane (TM) core of these receptors[2-4]. More recently, considerable progress has been made in targeting drugs to non-classical (allosteric) binding sites of certain mAChR subtypes[5].

Figure 1

Major structural features of the M3 receptor

a, Analysis of muscarinic receptor sequences divides them into two classes. b, The overall structure of the M3 receptor (green) is similar to that of the M2 receptor (orange). The M3-bound ligand, tiotropium, is shown in spheres c, Comparison of the intracellular surfaces shows divergence in the cytoplasmic end of transmembrane helix 5. d, Comparison of the extracellular surfaces shows less deviation, with near perfect conservation of backbone fold of extracellular loops e, A solvent accessible surface for the M3 receptor bound to tiotropium (spheres) shows the receptor covering the ligand with a tyrosine lid (outlined in red). f, M3 receptor structure colored by sequence conservation among the five mAChR subtypes. Poorly conserved regions are shown with larger backbone diameter. The orthosteric and allosteric sites are indicated in blue and red, respectively, and the ligand tiotropium is shown in spheres.

Within the mAChR family, the M3 subtype mediates many important physiological functions, including smooth muscle contraction and glandular secretion[3,4,6-9]. Central M3 receptors have also been implicated in the regulation of food intake[7], learning and memory[8], and the proper development of the anterior pituitary gland[9]. Selective drugs targeted at this receptor subtype may prove clinically useful[4,6-9], and non-selective muscarinic ligands are already widely used in current practice.

Due to the profound physiological importance of the M3 receptor and its longstanding role as a model system for understanding GPCR function[3,10], we used the T4 lysozyme (T4L) fusion protein strategy[11] to obtain crystals of Rattus norvegicus M3 receptor-T4L fusion protein (Supplementary Fig. 1) by lipidic cubic phase crystallization. Diffraction data from more than 70 crystals were merged to create a data set to 3.4 A resolution and to solve the structure by molecular replacement. The M3 receptor structure, together with that of the M2 receptor[12], affords for the first time an opportunity to compare two closely related mammalian receptors with divergent G protein coupling selectivities.

The overall structure of the M3 receptor is similar to that of M2 (Fig. 1b-d). Surprisingly, structural conservation includes intracellular loops (ICLs) 1 and 2, and extracellular loops (ECLs) 1–3, which share highly similar overall folds despite low sequence conservation (Fig. 1f). Like the M2 receptor, the M3 receptor exhibits unique mAChR features, including a large extracellular vestibule as part of an extended hydrophilic channel containing the orthosteric binding site (Fig. 1e). Also like M2, the M3 receptor features a pronounced outward bend at the extracellular end of TM4 (Fig. 1d; Supplementary Fig. 2b). This bend, not seen in any other GPCR family crystallized to date, is stabilized by a hydrogen bond from the Q207 (Q163 in M2) side chain to the L204 backbone peptide carbonyl (Supplementary Fig. 2b). This bond is part of a polar interaction network involving four residues absolutely conserved within the mAChR family, suggesting that this unusual feature is important to mAChRs in general. Indeed, mutagenesis of Q207 in M3 impaired both ligand binding and receptor activation[13].

The M3 receptor was crystallized in complex with tiotropium (Spiriva), a potent muscarinic inverse agonist[14,15] used clinically for the treatment of chronic obstructive pulmonary disease. The M2 receptor was crystallized in complex with R-(–)-3-quinuclidinyl benzilate (QNB) which, like tiotropium, is a non-subtype-selective mAChR blocker[14,16]. The two ligands bind in remarkably similar poses (Fig. 2b), and it is likely that this pose represents a conserved binding mode for structurally similar anticholinergics. In the M3 receptor, as in M2, the ligand is deeply buried within the TM receptor core (Fig. 2a, d) and is covered by a lid comprised of three conserved tyrosines, Y1483.33, Y5066.51, and Y5297.39 (Fig. 2a; superscripts indicate Ballesteros-Weinstein numbers[17]). The ligand is almost completely occluded from solvent and engages in extensive hydrophobic contacts with the receptor. A pair of hydrogen bonds are formed from N5076.52 to the ligand carbonyl and hydroxyl, while D1473.32 interacts with the ligand amine.

Figure 2

Orthosteric binding sites of the M2 and M3 receptors

In all panels, the M3 receptor is green with its ligand tiotropium in yellow, while the M2 receptor and its ligand QNB are shown in orange and cyan, respectively. a, Tiotropium binding site in the M3 receptor. A 2Fo-Fc map contoured at 2Fo-Fc is shown in wire. b, Chemical structures of ligands. A red arrow indicates the tropane C3 atom used as a tracking landmark in Fig. 3. Superimposing the receptor structures reveals that the two ligands adopt highly similar poses (bottom). c, There is a Phe (M2)/Leu (M3) sequence difference between the M2 and M3 receptors near the binding site. d, This produces an enlarged binding pocket in the M3 receptor. e, A displacement of M3 Y5297.39 is seen. f, This may arise from a sequence difference at position 2.61 (Tyr80 in M2 and Phe124 in M3).

Reflecting the difficulty in developing subtype-selective orthosteric ligands, the residues forming the orthosteric binding pocket are absolutely conserved among the five mAChR subtypes (Fig. 1f). However, this conservation at the amino acid level does not preclude the existence of differences in the three-dimensional architecture of the orthosteric site between the different mAChR subtypes. In fact, comparison of the structures of the M3 and M2 receptor ligand binding sites reveals for the first time structural divergences that might be exploited in the development of subtype-selective ligands.

One such difference derives from the replacement of Phe181 in ECL2 of M2 with Leu225 in M3 (this residue is leucine in all mAChRs except M2). This creates a pocket in M3 not found in M2 (Fig. 2c,d). A second difference is a 2.8 A shift of Tyr5297.39 relative to the position of the corresponding M2 residue (Tyr426; Fig. 2e). This feature may derive from a difference in the identity of the residue in position 2.61 (Phe124 in M3 and Tyr80 in M2; Fig. 2f). This residue interacts directly with TM7, influencing the position of this helix and the residues within it, including Tyr5297.39. Notably, the residue at position 2.61 is not a part of the orthosteric binding pocket, but is positioned near a probable allosteric binding site[12]. Since tiotropium and QNB are structurally similar but not identical, the observed binding site differences must be interpreted with some degree of caution. However, site-directed mutagenesis studies with M1 and M3 receptors support the concept that the residue at position 2.61 plays a role in receptor activation[18,19], and ligand binding selectivity[20]. This site does not appear to play a role in determining antagonist dissociation rates, since mutation of M3 F2.61 to tyrosine or of M2 Y2.61 to phenylalanine had no effect on dissociation rates for [3H]NMS or [3H]QNB.

We used molecular dynamics simulations to characterize the pathway by which tiotropium binds to and dissociates from the M2 and M3 receptors. Similar techniques have previously been shown to correctly predict crystallographic ligand binding poses and kinetics in studies of β-adrenergic receptors[21]. In both the M2 and M3 receptors, our simulations indicate that as tiotropium binds to or dissociates from the receptor, it pauses at an alternative binding site in the extracellular vestibule (Fig. 3, Supplementary Fig. 3). Intriguingly, this site corresponds to an allosteric site that has been previously identified by mutagenesis[12], a finding consistent with pharmacological studies showing that orthosteric ligands can act as allosteric modulators at the M2 receptor[22].

Figure 3

Molecular dynamics of ligand binding

Simulations suggest that the tiotropium binding/dissociation pathway for both receptors involves a metastable state in the extracellular vestibule. a, When tiotropium is pushed out of the binding pocket of M3, it pauses in the extracellular vestibule. Spheres represent positions of the ligand’s C3 tropane atom at successive points in time. b, When tiotropium is placed in solvent, it binds to the same site in the extracellular vestibule. Our simulations are insufficiently long for it to proceed into the orthosteric binding pocket; the agonist ACh, a much smaller molecule, bound spontaneously to the orthosteric site in similar simulations (see Supplementary Methods) c, Schematic free energy landscape for binding/dissociation. d, Common binding poses for tiotropium in the extracellular vestibule of M2 (orange) and of M3 (green). Non-conserved residues that contact the ligand are shown in thin sticks. The location of the orthosteric site is indicated by tiotropium in spheres.

Tiotropium adopts different preferred allosteric binding poses in M2 and M3 (Fig. 3d, Supplementary Fig. 4). These metastable binding poses, which appear independently in both binding and dissociation simulations, may represent the first structural view of a clinically used “orthosteric” GPCR ligand binding to an experimentally validated allosteric site. Conceivably, therapeutic molecules could be rationally engineered to act independently as both allosteric and orthosteric ligands (in contrast to previously described bitopic ligands that bind at both sites simultaneously[23]). Tiotropium dissociates from M3 receptors more slowly than from M2 receptors, a phenomenon thought to provide clinically important “kinetic selectivity” of this drug for M3 receptors despite similar equilibrium binding affinities for both subtypes[14]. In simulations with tiotropium bound, the portion of ECL2 nearest the binding pocket proved more mobile in M2 than in M3 (Supplementary Fig. 5), likely due to multiple sequence differences between the two receptor subtypes. This increased mobility disrupts a hydrophobic cluster involving a thiophene ring of tiotropium, the ECL2 residue Phe181(M2)/Leu225(M3), and Tyr3.33, facilitating movement of Phe181/Leu225 away from the orthosteric site and rotation of Tyr3.33 toward TM4. In simulations of ligand dissociation, such motions clear a path for tiotropium’s egress from the orthosteric site to the extracellular vestibule. The increased mobility of ECL2 in M2 thus appears to facilitate tiotropium’s traversal of the largest energetic barrier on the binding/dissociation pathway (Fig. 3c). Experimental measurements with wild-type and mutant receptors (M3 L225F and M2 F181L) suggest that the Leu225/Phe181 sequence difference alone is insufficient to explain the difference in off-rates, (for practical reasons these measurements were performed with QNB rather than tiotropium; see Online Methods).

One of the most interesting features of the M2 and M3 receptors is the fact that the two highly similar receptors display pronounced differences in G-protein coupling specificity. For this reason, the M2/M3 receptor pair has long served as an excellent model system to identify features contributing to GPCR-G protein coupling selectivity[3]. Since no simple sequence elements have been identified as general determinants of coupling specificity across GPCR families[24], it is likely that recognition depends on features such as overall conformation in addition to specific inter-residue contacts.

The M2 and M3 receptor structures show a significant difference in the position of the cytoplasmic end of TM5 and of ICL2 (Fig. 4a, b). The highly conserved tyrosine residue at position 5.58 (M3 Tyr2505.58, M2 Tyr2065.58) shows a clear deviation between the two receptors, pointing toward the core of the protein in M2, and away from the receptor toward the surrounding lipid bilayer in M3. Interestingly, mutagenesis studies have identified a tetrad of residues (‘AALS’ in M3, ‘VTIL’ in M2) located on the cytoplasmic end of TM6 that are critical in determining G protein coupling selectivity[25,26]. In both structures, these residues interact directly with TM5 (Fig. 4a), and in the β2 adrenergic receptor-Gs complex[27] two of the four corresponding residues make contact with the carboxy terminal helix of Gαs. M3 Tyr2545.62 at the bottom of TM5 also plays a role in activation of Gq/11[28]. In the M2 receptor structure, the corresponding residue (Ser2105.62) is displaced by approximately 4 A relative to Tyr2545.62 in M3 (Fig. 4a).

Figure 4

G protein coupling specificity determinants

a, The M3 receptor shows displacement of TM5 relative to its position in M2, and a conserved tyrosine (M3 Tyr2505.58) adopts different positions in the two receptors. Four TM6 residues near TM5 (AALS in M3, VTIL in M2) have been shown to be important coupling specificity determinants. b, ICL2 is also divergent between the two structures. Four residues previously implicated as specificity determinants[25] are shown, with residue numbers for M2 followed by M3 c, Plot of interhelical distances for crystallographically unique inactive GPCR structures published to date. Distances were measured between Cα atoms of TM5 residue 5.62 and TM3 residue 3.54 (x-axis), and TM5 residue 5.62 and TM6 residue 6.37 (y-axis). GPCRs cluster by coupling specificity, although squid rhodopsin is an exception. GPCRs coupling preferentially to Gi/o and those coupling to the homologous G protein Gt cluster together. d, Structural alignment of mammalian Gi/o-coupled and Gq/11-coupled receptor structures.

When we compared the position of TM5 in the M2 and M3 receptors to that in other GPCR structures, we found that it is M2-like in all Gi/o-coupled receptors, while the two mammalian Gq/11-coupled receptors solved to date exhibit another conformation (Fig. 4c, d). An important caveat here is that these structures have been solved using the T4L fusion strategy, and we cannot completely exclude the possibility that this approach perturbs the conformation of TMs 5 and 6. However, in molecular dynamics simulations of M2 and M3 receptors without T4L, each of the receptors adopts a set of conformations that includes its own crystallographically observed conformation (Supplementary Fig. 6, 7). These simulations suggest that the observed conformations are unlikely to be artifacts of the crystallization methodology, though the crystal structures likely represent only one conformation among many adopted by the receptors in a biological context.

The structure of the M3 receptor, together with that of the M2 receptor[12], offers a unique opportunity to directly compare the structural properties of two members of a mammalian GPCR subfamily endowed with different G-protein coupling selectivities. Examination of the M3 structure has provided the first structural evidence of differences between ligand binding sites of mAChR subtypes that could be exploited for the design of more selective therapeutics. Moreover, computational studies have identified a pathway by which the COPD drug tiotropium may bind to and dissociate from the M3 receptor, offering the first structural view of an orthosteric GPCR ligand binding to an experimentally validated allosteric site. This information should facilitate the rational design of new muscarinic drugs exhibiting increased receptor subtype selectivity, potentially improving treatment for a wide variety of important clinical disorders.

Methods summary

The M3 muscarinic receptor-T4 lysozyme fusion protein was expressed in Sf9 insect cells and purified by nickel affinity chromatography followed by FLAG antibody affinity chromatography and then size exclusion chromatography. It was crystallized using the lipidic cubic phase technique, and diffraction data were collected at the GM/CA-CAT beamline at the Advanced Photon Source at Argonne National Lab. The structure was solved by molecular replacement using merged data from 76 crystals. All-atom classical molecular dynamics simulations with explicitly represented lipids and water were performed using the CHARMM force field[29] on Anton[30]. Ligand-binding simulations included no artificial forces. Dissociation studies included a time-varying biasing term that gradually forces the ligand away from its crystallographic position, but not along any prespecified pathway or direction. Full details are provided in the online methods.

Online Methods

Expression and purification of M3 muscarinic receptor

The wild-type M3 mAChR contains several long, likely poorly ordered regions, including the extracellular amino terminal domain and the third intracellular loop, making it a challenging candidate for crystallographic studies. To alleviate this problem, the M3 receptor from Rattus Norvegicus was modified to include a TEV protease recognition site in the amino terminus and a hexahistidine tag at the carboxy terminus. Moreover, the third intracellular loop (residues 260–481) was replaced with T4 lysozyme residues 1-161 in a manner described previously[11], with two different fusions tested. These modifications are diagrammed in figure S1, which also shows the final crystallization construct.

The pharmacological properties of the construct were tested and compared to those of the wild-type receptor (Fig. S8, supplementary table 1; see below for methods details). Both constructs showed almost identical affinity for antagonists, while the crystallization construct (M3-crys) showed somewhat higher affinity for the agonist acetylcholine than the wild-type construct. A similar observation has been noted previously in the β2 adrenergic receptor[11]. Studies with membranes prepared from transfected COS-7 cells showed that TEV cleavage of M3-crys (to remove most of the N-terminal tail) had no significant effect on ligand binding affinities (Fig. S9). Moreover, the wild-type receptor and M3-crys, either cleaved with TEV or left uncleaved, showed very similar [3H]-QNB dissociation rate kinetics (Fig. S10). As expected, the crystallization construct failed to stimulate agonist-dependent phosphoinositide hydrolysis in transfected COS cells (data not shown), likely because essential G protein interacting regions in intracellular loop 3 were omitted from the construct and also because the T4 lysozyme fusion protein sterically blocks G protein association.

The crystallization construct was expressed in Sf9 cells using the baculovirus system in the presence of 1 μM atropine. M3 receptors expressed in Sf9 cells are known to exhibit functional and pharmacological properties similar to M3 receptors expressed in mammalian cells[31]. Infection was performed at 4 × 106 cells per mL and flasks were shaken at 27 °C for 60 hr following infection.

Cells were harvested by centrifugation, then lysed by osmotic shock in the presence of 1 μM tiotropium bromide (obtained from W & J PharmaChem, Inc., Silver Spring, MD, USA), which was present in all subsequent buffers. Receptor was extracted from cells using a dounce homogenizer with a buffer of 0.75 M NaCl, 1% dodecyl maltoside (DDM), 0.03% cholesterol hemisuccinate (CHS), 30 mM HEPES pH 7.5, and 30% glycerol. Iodoacetamide (2 mg/mL) was added to block reactive cysteines at this stage. Nickel-NTA agarose was added to the solubilized receptor without prior centrifugation, stirred for two hr, and then washed in batch with 100 × g spins for 5 min each. Washed resin was poured into a glass column, and receptor was eluted in 0.1% DDM, 0.03% CHS, 20 mM HEPES pH 7.5, 0.75 M NaCl, and 250 mM imidazole.

Nickel resin-purified receptor was then loaded by gravity flow over anti-FLAG M1 affinity resin. Following extensive washing, detergent was gradually exchanged over 1.5 hr into a buffer in which DDM was replaced with 0.01% lauryl maltose neopentyl glycol (MNG), and the NaCl concentration was lowered to 100 mM. MNG has been shown to be more effective at stabilizing muscarinic receptors than DDM[32]. Receptor was eluted with 0.2 mg/mL FLAG peptide and 5 mM EDTA. TEV protease (1:10 w/w) was added and incubated with receptor for 1.5 hr at room temperature to remove the flexible amino terminal tail. Receptor was then separated from TEV by size exclusion chromatography (SEC) on a Sephadex S200 column (GE Healthcare) in a buffer of 0.01% MNG, 0.001% CHS, 100 mM NaCl, and 20 mM HEPES pH 7.5. Tiotropium was added to a final concentration of 10 μM following SEC. The resulting receptor preparation was pure and monomeric (Fig. S11). Purification of unliganded M3 receptor was also possible by this procedure, but the resulting preparation was polydisperse and unsuitable for crystallographic study.

Crystallization and data collection

Purified M3 receptor was concentrated to 60 mg/mL, then mixed with 1.5 parts by weight of a 10:1 mix of monoolein with cholesterol (Sigma) using the two syringe reconstitution method[33]. The resulting lipidic cubic phase mix was dispensed in 15 nL drops onto glass plates and overlaid with 600 nL precipitant solution using a Gryphon LCP robot (Art Robbins Instruments). Crystals grew after 2 - 3 days in precipitant solution consisting of 27 – 38 % PEG 300, 100 mM HEPES pH 7.5, 1 % (w/w) 1,2,3-heptanetriol, and 100 mM ammonium phosphate. Typical crystals are shown in figure S12.

Data collection was performed at Advanced Photon Source GM/CA-CAT beamlines 23ID-B and 23ID-D using a beam size of 5 or 10 microns for most crystals. Diffraction quality rapidly decayed following exposure, and wedges of typically 5 degrees were collected and merged from 76 crystals using HKL2000[34]. Diffraction quality ranged from 3–4 A in most cases, with strong anisotropy evident in many frames. Most crystals tested showed evidence of epitaxial twinning, though in most cases one of the two twins dominated the observed diffraction pattern, allowing processing as a single crystal. A more extensive discussion of the twinning is given below. Some contamination of diffraction measurements due to the twin-related reflections was unavoidable, leading to slightly poorer merging statistics than is typical for datasets collected from many small crystals (Supplementary table 2). Despite this, maps were generally of high quality and electron density was easily interpretable (Figs. S13, S14), in part due to the availability of non-crystallographic symmetry.

Analysis of /<σF> along each of the three reciprocal space axes indicated that the diffraction was strong in two directions, and weak in the third direction, along the reciprocal space axis c* (Fig. S15). Using /<σF> greater than 3 as a guideline suggested a resolution cutoff of better than 3.2 A along a* and b*, and of 4.0 A along c*. We therefore applied an ellipsoidal truncation along these limits, and then applied an overall spherical truncation at 3.4 A due to low completeness in higher resolution shells. Fortunately, 4-fold non-crystallographic symmetry (NCS) allowed for improved map quality with map sharpening followed by NCS averaging, largely alleviating the effects of anisotropic diffraction and epitaxial twinning to give highly interpretable maps (Fig. S13, S14) and allowing details of ligand recognition to be clearly resolved (Supplementary table 3).

The structure of the M3 receptor was solved using the structure of the M2 muscarinic receptor (companion manuscript) as the search model in Phaser[35]. The model was improved through iterative refinement in Phenix[36] and manual rebuilding in Coot guided by both NCS averaged and unaveraged maps. NCS restraints were applied in initial refinement stages, and omitted in final refinement cycles to account for differences between NCS-related copies. The quality of the resulting structure was assessed using MolProbity[37], and figures were prepared using PyMOL[38].

Epitaxial twinning

Crystals of the M3 receptor showed hallmarks of epitaxial twinning, such as mixed sharp and split spots, poor indexing, and many unpredicted reflections in some frames. In some cases diffraction from two distinct lattices was clearly visible, with a small fraction of reflections exactly superimposed from both lattices (Fig. S16). In most cases one lattice dominated the diffraction pattern to such an extent that it could be easily processed as a single crystal. Intriguingly, the two indexing solutions were not equivalent cells but rather were two enantiomorphic P1 cells (supplementary table 2).

As one of these two cells gave significantly better diffraction data than the other, data processing and refinement were only pursued in this case. Within the asymmetric unit, two layers of receptors and two layers of T4 lysozyme are present, but each of these four layers exhibits a different lattice packing (Fig. S17, S18). The order in which these layers are stacked in the crystal defines a unique direction along c, the axis normal to the membrane plane. Since P1 is a polar space group, the positive direction along c is uniquely defined, and the two possible orientations of the stacked layers of membrane relative to the positive direction along c distinguish the two twin crystal forms.

Transient expression of wild-type and modified M3 receptors in COS-7 cells, membrane preparation, and TEV treatment

COS-7 cells were cultured as described previously[39]. About 24 h prior to transfections, ~1 × 106 cells were seeded into 100 mm dishes. Cells were transfected with 4 μg/dish of receptor plasmid DNA using the Lipofectamine Plus kit (Invitrogen, Carlsbad, CA), according to the manufacturer's instructions. The mammalian expression plasmid coding for the wild type rat M3 receptor has been described previously[40]. The coding sequence of the modified M3 receptor construct used for crystallization studies (M3-crys; see Fig. S8, supplementary table 1) was inserted into the pcDNA3.1(-) vector. Transfected cells were incubated with 1 μM atropine for the last 24 h of culture to increase receptor expression levels[39]. COS-7 cells were harvested ~48 h after transfections, and membranes were prepared as described in detail by Ward et al.[39]

Membranes prepared from COS-7 cells transiently expressing M3-crys were resuspended in TEV protease digestion buffer (50 mM NaCl, 10 mM HEPES pH 7.5, and 1 mM EDTA) and incubated overnight with TEV protease (homemade, final concentration: 1 μM) at 4 ΰC with rotation. Efficient removal of the N-terminal tail of Mΰ-crys by TEV was confirmed by SDS-PAGE and immunoblotting using a monoclonal anti-FLAG antibody directed against the N-terminus of M3-crys. TEV-treated membranes were resuspended in either buffer A (25 mM sodium phosphate and 5 mM MgCl2, pH 7.4) for radioligand binding studies or in sodium- potassium-phosphate buffer (4 mM Na2HPO4, 1 mM KH2PO4, pH 7.4) for [3H]QNB dissociation assays (see below).

Radioligand binding studies

[3H]N-methylscopolamine ([3H]NMS) saturation and competition binding studies were carried out essentially as described previously[41]. In brief, membrane homogenates prepared from transfected COS-7 cells (~10 20 μg of membrane protein per tube) were incubated with the muscarinic antagonist/inverse agonist, [3H]NMS, for 3 h at 22 °C in 0.5 ml of binding buffer containing 25 mM sodium phosphate and 5 mM MgCl2 (pH 7.4). In saturation binding assays, we employed six different [3H]NMS concentrations ranging from 0.1 to 6 nM. In competition binding assays, we studied the ability of tiotropium, atropine, or acetylcholine to interfere with [3H]NMS (0.5 nM) binding. Incubations were carried out for 20 hr in the case of tiotropium in order to achieve equilibrium binding[42] (3 hr for all other ligands). Nonspecific binding was assessed as binding remaining in the presence of 1 μM atropine. Binding reactions were terminated by rapid filtration over GF/C Brandel filters, followed by three washes (~4 ml per wash) with ice-cold distilled water. The amount of radioactivity that remained bound to the filters was determined by liquid scintillation spectrometry. Ligand binding data were analyzed using the nonlinear curve-fitting program Prism 4.0 (GraphPad Software Inc., San Diego, CA).

Atropine sulfate and acetylcholine chloride were from Sigma-Aldrich (St. Louis, MO). Tiotropium bromide was purchased from W&J PharmaChem, Inc. (Silver Spring, MD). [3H]NMS (specific activity: 85.0 Ci/mmol) was obtained from PerkinElmer Life Sciences (Waltham, MA).

[3H]-QNB dissociation rate assays

[3H]-QNB (PerkinElmer; specific activity: 50.5 Ci/mmol) dissociation rate assays were carried out as described previously[43]. Measurements were carried out at 37 oC in a total volume of 620 μl using a buffer consisting of 4 mM Na2HPO4 and 1 mM KH2PO4 (pH 7.4). Membranes prepared from transfected COS-7 cells (final protein concentration: 10 μg protein/ml) were prelabeled with 1 nM [3H]-QNB for 30 min. Dissociation of the labeled ligand was initiated by the addition of atropine (final concentration: 3 μM). Incubations were terminated by filtration through GF/C Brandel fiber filters that had been pretreated with 0.1% polyethyleneimine, followed by two rinses with ice-cold distilled water. The amount of radioactivity that remained bound to the filters was determined by liquid scintillation spectrometry.

Molecular dynamics

In all simulations, the receptor was embedded in a hydrated lipid bilayer with all atoms, including those in the lipids and water, represented explicitly. Simulations were performed on Anton[30], a special-purpose computer designed to accelerate standard molecular dynamics simulations by orders of magnitude.

System setup and simulation protocol

Simulations of the M2 receptor were based on the crystal structure of the QNB–M2 complex, and simulations of M3 were based on the structure of the tiotropium–M3 complex (chain A). These crystal structures were determined using a T4 lysozyme (T4L) fusion strategy, in which intracellular loop 3 (ICL3) of each receptor was replaced by T4L; the T4L sequence was omitted in our simulations. All chain termini were capped with neutral groups (acetyl and methylamide). Residues 6.31–6.33 near the intracellular end of TM6 were unresolved in the M3 crystal structure, and residues 6.27–6.30 were resolved in an unstructured conformation packed against T4L. Residues 6.27–6.36 were modeled manually as a helical extension of TM6, with side chains then placed using Prime. Hydrogens were added to the crystal structures using Maestro (Schrodinger LLC, New York, NY), as described in previous work[44]. All titratable residues were left in the dominant protonation state at pH 7.0, except for Asp692.50 in M2 and Asp1142.50 in M3, which were protonated. Asp692.50 and Asp1142.50 correspond to rhodopsin Asp832.50, which is protonated during the entire photocycle[45].

Prepared protein structures were inserted into an equilibrated POPC bilayer as described in previous work[46]. Sodium and chloride ions were added to neutralize the net charge of the system and to create a 150 mM solution.

Simulations of the M3 receptor initially measured 80 × 80 × 87 A3 and contained 163 lipid molecules, 26 sodium ions, 41 chloride ions, and approximately 9,897 water molecules, for a total of ~56,000 atoms. Simulations of the M2 receptor initially measured 79 × 79 × 85 A3 and contained 156 lipid molecules, 24 sodium ions, 35 chloride ions, and approximately 9,165 water molecules, for a total of ~53,000 atoms. To simulate M2 with tiotropium bound, we removed the co- crystallized ligand, QNB, and docked in tiotropium using Glide (Schrodinger LLC, New York, NY).

All simulations were equilibrated using Anton in the NPT ensemble at 310 K (37 °C) and 1 bar with 5 kcal mol 1A 2 harmonic position restraints applied to all non-hydrogen atoms of the protein and the ligand (except for the tiotropium–M2 complex, where the ligand was unrestrained); these restraints were tapered off linearly over 50 ns. All bond lengths to hydrogen atoms were constrained using M-SHAKE[47]. A RESPA integrator[48] was used with a time step of 2 fs, and long-range electrostatics were computed every 6 fs. Production simulations were initiated from the final snapshot of the corresponding equilibration runs, with velocities sampled from the Boltzmann distribution at 310 K, using the same integration scheme, long-range electrostatics method, temperature, and pressure. Van der Waals and short-range electrostatic interactions were cut off at 13.5 A and long-range electrostatic interactions were computed using the k-space Gaussian Split Ewald method[49] with a 32 × 32 × 32 grid, σ = 3.33 A, and σs = 2.33 A.

Spontaneous binding of tiotropium and acetylcholine

We performed simulations where tiotropium was placed arbitrarily in the bulk solvent (at least 40 A from the entrance to the extracellular vestibule) and allowed to diffuse freely until it associated spontaneously with the M2 or M3 receptor, following methodology described in previous work[21]. In these simulations (Supplementary table 4, conditions D and E), the co-crystallized ligand was removed and four tiotropium molecules were placed in the bulk solvent. A tiotropium molecule bound to the extracellular vestibule at least once in each simulation. In the longer simulations, tiotropium bound to and dissociated from the extracellular vestibule multiple times. Tiotropium assumed several different poses when bound to the extracellular vestibule of either M2 or M3 (Fig. S4). Tiotropium never entered the orthosteric binding pocket, presumably because the simulations were not of sufficient length.

The fact that tiotropium associated with and dissociated from the vestibule multiple times, but did not enter the binding pocket, suggests that tiotropium must traverse a larger energetic barrier to enter the binding pocket of M2 or M3 from the extracellular vestibule than to enter the vestibule from bulk solvent. This contrasts with earlier simulations on alprenolol binding to the β2-adrenergic receptor, in which the largest energetic barrier (by a small margin) was between the bulk solvent and the extracellular vestibule[21]. This difference likely reflects the fact that ligands must pass through a much tighter passageway to enter the binding pocket of the M2 and M3 receptors from the vestibule than is the case for the β2-adrenergic receptor. Tiotropium lost the majority of its hydration shell as it entered the vestibule (Fig. S19), as observed previously for ligands binding to β-adrenergic receptors[21].

We followed a similar protocol in a simulation of the M3 receptor in the presence of the agonist acetylcholine, a smaller molecule which might be expected to bind faster (Supplementary table 4, condition F). Indeed, an acetylcholine molecule bound in the orthosteric binding pocket after 9.5 μs and remained there for the remainder of the 25-μs simulation. Although acetylcholine quickly passed through the extracellular vestibule en route to the binding pocket, it did not exhibit metastable binding in the vestibule. Acetylcholine exhibited significant mobility in the binding pocket, likely reflecting the low affinity of the crystallized inactive state for agonists.

Forced dissociation of tiotropium

To identify the entire binding/dissociation pathway, we “pushed” tiotropium out of the binding pocket of both the M2 and M3 receptors[50,51]. Production simulations were initiated from configurations of the corresponding unbiased trajectory. These simulations employed a time-dependent harmonic biasing potential, U(t): where t is time, k is a force constant in units of kcal/mol/A2, d is the distance between the center-of-mass of the heavy atoms of tiotropium and the center-of- mass of the protein Cα atoms, and d0(t) varied linearly over 1.0 μs from 9.6 A to 33 A for M2 and from 8.6 A to 32 A for M3. This biasing term does not impose any preferred direction of ligand exit. We performed seven such simulations for each of M2 and M3,with k = 5, starting from configurations extracted from the tiotropium bound simulations of M2 and M3. Each initial configuration was separated in time by 36 ns. Results were similar across all simulations.

Force field parameters

The CHARMM27 parameter set[29] with CMAP terms[52] and a recently introduced correction to charged side-chain electrostatics[53] was used for all protein molecules and salt ions in conjunction with the CHARMM TIP3P[54] water model and a modified CHARMM lipid force field[55]. Force field parameters for tiotropium and QNB were obtained from the CHARMM ParamChem web server[56], version 0.9.1 beta. QNB was simulated in its protonated (ammonium) state. To evaluate the assigned partial charges assigned by ParamChem, we performed a quantum mechanical computation of the electrostatic potential at a collection of points surrounding each ligand (in vacuo at the HF/6-31G* level of theory using MOLPRO[57]), and compared it to the potential generated by the assigned charges.

Analysis protocols

Trajectory snapshots, each containing a record of all atom positions at a particular instant in time, were saved every 180 ps during production simulations. Distance measurements were computed using the HiMach parallel analysis framework[58]. VMD[59] was used to visualize trajectories and to produce Fig. 3a and Fig. 3b.

To determine the most common vestibule-bound poses of tiotropium shown in Fig. S4, we performed a clustering analysis on the 14.2-μs spontaneous binding simulation of M2 (Table S4, condition D) and the 16.0-μs spontaneous binding simulation of M3 (Table S4, condition E). We performed k-means clustering on the set of trajectory snapshots in which a tiotropium molecule was in the extracellular vestibule, using the positions of atoms indicated in Fig. S4c. Clusters representing highly similar poses were merged.