Pharmacological hallmarks of allostery at the M4 muscarinic receptor elucidated through structure and dynamics

Allosteric modulation of G protein-coupled receptors (GPCRs) is a major paradigm in drug discovery. Despite decades of research, a molecular level understanding of the general principals that govern the myriad pharmacological effects exerted by GPCR allosteric modulators remains limited. The M4 muscarinic acetylcholine receptor (M4 mAChR) is a well-validated and clinically relevant allosteric drug target for several major psychiatric and cognitive disorders. Here, we present high-resolution cryo-electron microscopy structures of the M4 mAChR bound to a cognate Gi1 protein and the high affinity agonist, iperoxo, in the absence and presence of two different positive allosteric modulators, LY2033298 or VU0467154. We have also determined the structure of the M4 mAChR-Gi1 complex bound to its endogenous agonist, acetylcholine (ACh). Structural comparisons, together with molecular dynamics, mutagenesis, and pharmacological validations, have provided in-depth insights into the role of structure and dynamics in orthosteric and allosteric ligand binding, global mechanisms of receptor activation, cooperativity, probe-dependence, and species variability; all key hallmarks underpinning contemporary GPCR drug discovery.


Introduction 47
Over the past 40 years, there have been major advances to the analytical methods that allow 48 for the quantitative determination of the pharmacological parameters that characterise G 49 protein-coupled receptor (GPCR) signaling and allosteric modulation (Figure 1A,B). These 50 analytical methods are based on the operational model of agonism (Black and Leff, 1983) and 51 have been extended or modified to account for allosteric modulation (Leach et al., 2007), 52 biased agonism (Kenakin, 2012), and even biased allosteric modulation (Slosky et al., 2021). 53 Collectively, these models and subsequent key parameters (Figure 1B) are used to guide 54 allosteric drug screening, selectivity, efficacy and ultimately, clinical utility, and provide the 55 foundation for modern GPCR drug discovery (Wootten et al., 2013). Yet, a systematic 56 understanding of how these pharmacological parameters relate to the molecular structure 57 and dynamics of GPCRs remains elusive. 58 2008). Despite LY298 being one of the best characterized M4 mAChR PAMs, its therapeutic 94 potential has been limited by numerous factors including its chemical scaffold, which has 95 been difficult to optimize with respect to its molecular allosteric parameters (Figure 1) and 96 variability of response between species (Suratman et al., 2011;Wood et al., 2017a). In the 97 search for better chemical scaffolds, the PAM, VU0467154 (VU154), was subsequently 98 discovered. VU154 showed robust efficacy in preclinical rodent models, however, it also 99 exhibited species selectivity that prevented its clinical translation (Bubser et al., 2014). 100 Collectively, LY298 and VU154 are exemplar tool molecules that highlight the promises and 101 The signaling assays and use of an operational model of allosterism also allowed for the 183 determination of the functional cooperativity () exerted by the PAMs (Figure 2I; S1D), 184 which is a composite parameter accounting for both binding () and efficacy () modulation. 185 Notably, VU154 displayed lower positive functional cooperativity with ACh than LY298. 186 Strikingly, VU154 had negligible functional modulation with Ipx in contrast to the 187 cooperativity observed with ACh in the TruPath assay. The 10-fold difference in  values for 188 VU154 between ACh and Ipx highlights the dependence of the orthosteric probe used in the 189 assay (i.e. probe dependence); on this basis, VU154 would be classified as a NAL (not a PAM) 190 with Ipx in the TruPath assay (Table S1). and VU154 appear to have a slight negative to neutral effect on agonist efficacy in the Gi1 199 Trupath and pERK1/2 assays (Table S1), suggesting that the predominant allosteric effect 200 exerted by these PAMs is mediated through binding modulation. 201 202 Collectively, our extensive analysis on the pharmacology of LY298 and VU154 with ACh and 203 Ipx offers detailed insight into the key differences between these ligands across a range of 204 pharmacological properties: ligand binding, probe dependence, efficacy, agonist-receptor-205 transducer interactions, and allosteric modulation (Figure 1, Table S1). We hypothesised that 206 structures of the human M4 mAChR in complex with different agonists and PAMs combined 207 with molecular dynamic simulations could provide high resolution molecular insights into the 208 different pharmacological profiles of these ligands.  242 to 387 of the third intracellular loop to improve receptor expression and purification, and  214   made complexes of the receptor with Gi1 protein and either the endogenous agonist, ACh, or  215 Ipx. Due to the higher affinity of Ipx compared to ACh (Schrage et al., 2013), we utilised Ipx to 216 form additional M4R-Gi1 complexes with or without the co-addition of either LY298 or VU154. 217 In all instances, complex formation was initiated by combining purified M4 mAChR 218 immobilized on anti-FLAG resin with detergent solubilized Gi1 membranes, a single-chain 219 variable fragment (scFv16) that binds Gi and G, and the addition of apyrase to remove 220 Recently, cryo-EM structures of M4R-Gi1 complexes bound to ipx, ipx and the PAM LY2119620, 256 and an allosteric agonist c110, were determined . Surprisingly, comparison 257 of the M4R-Gi1 complex structures reveal larger differences in the position of key orthosteric 258 and allosteric site residues than the M 1 R-G 11 and M 2 R-G oA complex structures (Figure S4-5). 259 Unfortunately, the quality of density in the EM maps around the orthosteric and allosteric 260 sites of these M4R-Gi1 structures  was poor resulting in several key residues 261 being mismodelled in each site ( Figure S5). Therefore, differences between the M4R-Gi1 262 structures are highly likely to not be due to genuine differences, and as such we compared to 263 the M1R-G11 and M2R-GoA complex structures in this study (Maeda et al., 2019). in common with the other mAChR subtypes, is buried within the TM bundle in an aromatic 275 cage that is composed of four tyrosine residues, two tryptophan residues, one phenylalanine 276 residue, and seven other polar and nonpolar residues ( Figure 4C). Notably, all 14 of these 277 residues are absolutely conserved across all five mAChR subtypes, underscoring the difficulty 278 in developing highly subtype-selective orthosteric agonists (Burger et al., 2018). Both ACh and 279 Ipx have a positively charged trimethyl ammonium ion that makes cation- interactions with 280 Y113 3.33 , Y416 6.51 , Y439 7.39 , and Y443 7.43 ( Figure 4C) (superscript refers to the Ballesteros and 281 Weinstein scheme for conserved class A GPCR residues) (Ballesteros and Weinstein, 1995). isoazoline group of Ipx that makes a - interaction with the conserved residue W413 6.48 288 ( Figure 4D). The residue W413 6.48 is part of the CWxP motif, also known as the rotamer toggle 289 switch, a residue that typically undergoes a change in rotamer between the inactive and 290 active states of class A GPCRs (Shi et al., 2002). 291

292
To investigate the structural dynamics of the M4 mAChR, we performed three independent 293 500 ns GaMD simulations on the ACh-and Ipx-bound M4R-Gi1 cryo-EM structures ( Table S3). 294 GaMD simulations revealed that ACh undergoes higher fluctuations in the orthosteric site 295 than Ipx (Figure 4E,F). Similarly, the interactions of N117 3.37 , W164 4.57 , and W413 6.48 with Ipx 296 were more stable than those with ACh ( Figure 4I-P). In the ACh-bound structure, W413 6.48 297 was in a conformation that more closely resembled the inactive-state tiotropium-bound 298 structure (Figure 4C,D). GaMD simulations also showed that W413 6.48 sampled a larger 299 conformational space in the ACh-bound structure than in the Ipx-bound structure ( Figure  300 4L,P). The predominate χ2 angle of W413 6.48 was approximately 60 • and 105 • in the ACh-bound 301 and Ipx-bound simulations, respectively, corresponding to the cryo-EM conformations. 302 303 Located above ACh and Ipx is a tyrosine lid formed by three residues (Y113 3.33 , Y416 6.51 , and 304 Y439 7.39 ) that separates the orthosteric binding-site from an extracellular vestibule (ECV) at 305 the top of the receptor and the bulk solvent ( Figure 4C). In the inactive conformation, the 306 tyrosine lid is partially open due to Y416 6.51 rotating away from the binding pocket to 307 accommodate the binding of bulkier inverse agonists such as tiotropium. In contrast, mAChR 308 agonists are typically smaller in size than antagonists and inverse agonists, and this is reflected 309 in a contraction of the size of the orthosteric binding pocket from 115 Å 3 when bound to 310 tiotropium to 77 and 63 Å 3 when bound to ACh and Ipx, respectively ( Figure 4G,H) (Tian et 311 al., 2018). Together, the smaller binding pocket of Ipx and more stable binding interactions 312 with nearby residues that include W413 6.48 likely explain why Ipx has greater than 1,000-fold 313 higher binding affinity than ACh. 314  Table S3. 328

Structure and dynamics of PAM binding and allosteric modulation of agonist affinity 329
The M4R-Gi1 structures of LY298 and VU154 co-bound with Ipx are very similar to the Ipx-and 330 ACh-bound structures, as well as to prior structures of the M2 mAChR bound to Ipx and the 331 PAM, LY2119620 ( Figure   there being differences between the collected data sets that led to the structures. To support 366 these findings, we compared the GaMD simulations of all four cryo-EM structures ( Table S3). 367 Notably, VU154 underwent considerably higher fluctuations than LY298 with RMSDs ranging 368 from 1.5-15 Å for VU154 and 0.8-2.1 Å for LY298 relative to the cryo-EM structures ( Figure  369 5D,E). Therefore, the GaMD simulations corroborate our 3DVA results and suggests that 370 complexes bound to agonists with high affinity or co-bound with agonists and PAMs with high 371 positive cooperativity will exhibit lower dynamic fluctuations. 372

373
To investigate why the binding of LY298 was more stable than VU154, we examined the ligand 374 interactions with the receptor. There are three key binding interactions that are shared 375 between both PAMs and the M4 mAChR: 1) a three-way -stacking interaction between 376 F186 45.51 (ECL2 residues have been numbered 45.X denoting their position between TM4 and 377 TM5 with X.50 being a conserved cysteine residue), the aromatic core of the PAMs, and 378  Table S3. 400 401 A potential fourth interaction was observed with residue Q184 45.49 and the amide nitrogen of 402 the PAMs; however, the GaMD simulations suggest that this interaction is relatively weak 403 ( Figure 5K,O), consistent with the fact that mutation of Q184 45.49 to alanine had no effect on 404 the binding modulation of LY298 or VU154 ( Figure S6; Table S4). In addition, each PAM has 405 at least one unique binding interaction with the receptor (Figure 5F,G). For LY298, this is an 406 interaction between the fluorine atom and N423 6.58 that appeared to be stable during 407 simulation and, when mutated to alanine reduced the binding modulation of LY298 (Figure 408  Figure 5R). These results indicate that the binding of LY298 is more stable 418 than VU154 due to LY298 being able to form stable binding interactions with key residues in 419 the ECV. This provides a likely explanation for why LY298 was able to exert greater positive 420 binding cooperativity on orthosteric agonists than VU154. 421 422 A molecular mechanism of probe dependence 423 As highlighted above, PAMs, LY298 and VU154, displayed stronger allosteric binding 424 modulation with ACh than Ipx, an example of probe dependence (Figure 2, S1D) with ACh in the corresponding cryo-EM structures (Table S3 and Figure S7). In the absence of 430 PAM, ACh was more dynamic than Ipx with root-mean-square fluctuations (RMSF) of 2.13 Å 431 versus 0.88 Å, reflective of the fact Ipx binds with higher affinity than ACh ( Figure S7L). In the 432 presence of LY298 or VU154 the dynamics of ACh binding was decreased, with RMSFs reduced 433 to 1.23 Å and 1.82 Å, respectively, and with LY298 having the greatest effect ( Figure S7L). This 434 is in line with LY298 having more cooperativity with ACh than VU154 ( Figure 2C). In 435 comparison to ACh, there was a modest increase in the dynamics of Ipx with the addition of 436 LY298 or VU154, likely reflecting the fact Ipx binding to the receptor was already stable (Figure 437 S7D-F). These results provide a plausible mechanism for probe dependence, at least with 438 regards to differences in the magnitude of the allosteric effect depending on the ligand 439 bound. Namely, PAMs manifest higher cooperativity when interacting with agonists, such as 440 ACh, that are inherently less stable on their own when bound to the receptor, in contrast to 441 more stable ligands such as Ipx. 442 443

Structural and dynamic insights into orthosteric and allosteric agonism 444
In addition to the ability to allosterically modulate the function of orthosteric ligands, it has 445 become increasingly appreciated that allosteric ligands may display variable degrees of direct 446 agonism in their own right, over and above any allosteric modulatory effects ( As previously discussed, agonist binding decreases the size of the orthosteric binding site 457 ( Figure 4G,H). The primary driver of this decrease was the tyrosine lid residue Y416 6.51 , which 458 underwent a large rotation towards Y113 3.33 creating a hydrogen bond that seals off the 459 tyrosine lid (Figure 4C). The closure of the tyrosine lid was further reinforced by a change in 460 the rotamer of W435 7.35 to a planar position that sits parallel to the tyrosine lid allowing for a 461 - interaction with Y416 6.51 and a positioning of the indole nitrogen of W435 7.35 to 462 potentially form a hydrogen bond with the hydroxyl of Y89 2.61 (Figure 6B). The contraction of 463 the orthosteric pocket by the inward movement of Y416 6.51 also led to a contraction of the 464 ECV with a 5 Å inward movement of the top of TM6 and ECL3. As a consequence, the top of 465 TM5 was displaced outward by 4 Å forming a new interface between TM5 and TM6 that was 466 stabilized by a hydrogen bond between T424 6.59 and the backbone nitrogen of P193 5.36 along 467 with aromatic interactions between F197 5.40 and F425 6.60 ( Figure 6B). These interactions were 468 specific to the active state structures and appear to be conserved as they were also present 469 in the M1 and M2 mAChR active state structures (Maeda et al., 2019). In addition to the 470 movements of TM5 and TM6, there was a smaller 1 Å inward movement of ECL2 (Figure 6B). 471 The binding of LY298 and VU154 had minimal impact on the conformation of most ECL 472 residues, implying that the reorganisation of residues in the ECV by orthosteric agonists 473 contributes to the increased affinity of the PAMs (Figure 2E outward movement of TM6 that typifies GPCR activation and creation of the G protein binding 487 site. In comparison to the ECV residues (Figure 6B), beyond the rotamer toggle switch residue 488 W413 6.48 , there are no discernible differences between the agonist and PAM-agonist-bound 489 structures, suggesting a shared activation mechanism for residues below W413 6.48 (Figure 6C). Gi1-LY298, M4R-VU154, and M4R-LY298 systems, respectively. See Table S3. 499 500 As indicated above, LY298 also displays robust allosteric agonism in comparison to VU154 501 ( Figure 2G,H). To probe whether the allosteric agonism of LY298 could be related to its ability 502 to better stabilize the M4 mAChR in an active conformation in comparison to VU154, we 503 performed additional GaMD simulations on the LY298-Ipx-and VU154-Ipx-bound M4R-Gi1 504 structures with the agonist Ipx removed (3x 500 ns) and with both Ipx and the G protein 505 removed (3x 1,000 ns) (Figure 6, Table S3). In GaMD simulations, LY298 underwent lower 506 RMSD fluctuations than VU154 before dissociating from the receptor (Figure 6D-G). Similarly, 507 the conformations of W435 7.35 and W413 6.48 were better stabilized in the LY298-Ipx-bound 508 systems, indicating that LY298 more strongly promotes an active receptor conformation 509 Similarly, LY298 has activity at the M2 mAChR. However, the allosteric properties of VU154 526 are differentially affected by the species of the receptor (Wood et al., 2017a, 2017b). At the 527 human M4 mAChR, LY298 displays robust binding modulation, functional modulation, and 528 allosteric agonism, while VU154 has comparatively weaker allosteric properties (Figure 1, 529 Table S1). Conversely, at the mouse M4 mAChR, VU154 has a high degree of positive binding 530 modulation, functional modulation, and allosteric agonism that is comparable to LY298 at the 531 human M4 mAChR ( Figure S7, Table S1). Therefore, we aimed to determine if our prior 532 findings could be used to explain the selectivity of VU154 between the human and mouse 533 receptors. 534

535
The amino acid sequences of the human and mouse M4 mAChRs are highly conserved, with 536 most of the differences occurring between the long third intracellular loop and the N-and C-537 termini. As shown in Figure 7A, only three residues differ between the human and mouse M4 538  Table S1). For LY298, there were no statistically significant differences in 547 binding or function between species and across the mutants that were more than 3-fold in 548 effect. In contrast, VU154 had a 10-fold higher binding affinity for the Ipx-bound mouse M4 549 mAChR (compare Figure 2EG with Figure 7B). The affinity of VU154 increased by 2.5-fold at 550 the D432E and T433R mutants and the triple mutant matched the affinity of the mouse 551 receptor ( Figure 7B). In functional assays, similar results were observed for VU154 with Ipx at 552 the mouse M4 mAChR, with significant increases in the efficacy (B -corrected for receptor 553 expression), transduction coefficients (B/KB), and the functional modulation () (Figure 7B, 554 S8, Table S1). Relative to the WT M4 mAChR, the efficacy, transduction coefficients, and 555 functional modulation of VU154 increased for all of the mutants (Figure S8), however, none 556 of the values fully matched the mouse receptor. Nevertheless, these results indicate that 557 V91L, D432E, and T433R play a key role in mediating the species selectivity of VU154. 558   Our prior findings suggest the robust allosteric activity of LY298 at the human M4 mAChR was 576 due to stable interactions with the receptor (Figure 5). As a proof-of-principle, we questioned 577 if GaMD simulations would produce a stable binding mode for VU154 with D432E and T433R 578 mutations to the VU154-Ipx-bound M4R-Gi1 cryo-EM structure that was similar to our 579 previously observed stable binding pose of LY298 (Figure 5, Table S3) (Figure 7D- (Figure 4G,H), and GaMD simulations showed that Ipx formed more 603 stable interactions with the receptor (Figure 4I-P). These observations likely explained why 604 Ipx exhibited greater than 1,000-fold higher binding affinity than ACh (Figure 2B) (Figure 6). These findings were not contradictory to our above 665 findings that ACh was more efficacious than Ipx despite having weaker interactions with the 666 receptor, because when the ground state affinity of the ligands was accounted for in the 667 transduction coupling coefficients the rank order of agonism was Ipx >> ACh ~ LY298 > VU154 668 ( Figure 2B). Furthermore, these results were in accordance with the observations of Kenakin 669 and Onaran that ligands with the same binding affinity can also have differing efficacies (and 670 vice-versa). In addition, the mechanism of agonism for allosteric ligands that bind to the ECV 671    represent the mean ± SEM of 3 or more independent experiments performed in duplicate. 1620 Similar data were observed for competition binding with Ipx instead of ACh. See Table S4.   Table S3. 1661 1662