Activation mechanism of the class D fungal GPCR dimer Ste2

The fungal class D1 G-protein-coupled receptor (GPCR) Ste2 has a different arrangement of transmembrane helices compared with mammalian GPCRs and a distinct mode of coupling to the heterotrimeric G protein Gpa1–Ste2–Ste181. In addition, Ste2 lacks conserved sequence motifs such as DRY, PIF and NPXXY, which are associated with the activation of class A GPCRs2. This suggested that the activation mechanism of Ste2 may also differ. Here we determined structures of Saccharomyces cerevisiae Ste2 in the absence of G protein in two different conformations bound to the native agonist α-factor, bound to an antagonist and without ligand. These structures revealed that Ste2 is indeed activated differently from other GPCRs. In the inactive state, the cytoplasmic end of transmembrane helix H7 is unstructured and packs between helices H1–H6, blocking the G protein coupling site. Agonist binding results in the outward movement of the extracellular ends of H6 and H7 by 6 Å. On the intracellular surface, the G protein coupling site is formed by a 20 Å outward movement of the unstructured region in H7 that unblocks the site, and a 12 Å inward movement of H6. This is a distinct mechanism in GPCRs, in which the movement of H6 and H7 upon agonist binding facilitates G protein coupling.

The active-state structure of the class D fungal GPCR Ste2 coupled to the heterotrimeric G protein Gpa1-Ste4-Ste18 differs from that of other GPCR classes 1 . Ste2 is a homodimer that can couple to two G proteins simultaneously and its dimer interface is formed by the N terminus, extracellular loop 1 (ECL1) and transmembrane helices H1 and H7. The arrangement of transmembrane helices in Ste2 differs from receptors in other classes, with the intracellular end of H4 shifted approximately 20 Å and the α5 helix of the G protein fitting into a ledge formed by the intracellular ends of H3, H4 and H5. In addition, H5 and H6 are not curved or kinked outwards from the receptor as observed in both class A and class B GPCR G-protein-coupled states. The combination of these features suggested that the activation mechanism of Ste2 may be different from other GPCRs. To study this further we determined structures by single-particle cryo-electron microscopy (cryo-EM) of S. cerevisiae Ste2 either in the ligand-free state or bound either to an antagonist (Ste2-Ant) or an agonist , and compared them with the G-protein-coupled state 1 (Ste2-Ag-G).
The inactive-state structure Ste2-Ant was determined using the antagonist 3 [desTrp 1 Ala 3 Nle 12 ]α-factor (HALQLKPGQP(Nle)Y), hereafter referred to as 'antagonist'. The active-state structures of Ste2-Ag were determined bound to the native agonist, pheromone α-factor (WHWLQLKPGQPMY). The antagonist is a mutated form of α-factor with Trp1 deleted and the W3A mutation added 4 ; conversion of Met12 to Nle does not affect signalling but makes the peptide more resistant to oxidation. Expression and purification of wild-type Ste2 in the ligand-bound states (Extended Data Fig. 1a, b) was similar to that described previously 1 (Methods). We devised a method that we call pre-stabilization of a GPCR by weak association (PSGWAY) to purify Ste2-LF (Methods, Extended Data Fig. 1c) because of its inherent instability. The structures ( Fig. 1a-d) were determined by single particle cryo-EM (Extended Data Figs. 1d-g, 2) to overall resolutions of 3.1 Å (ligand-free) and 2.7 Å (Ste2-Ant), with two separate structures of Ste2-Ag determined to resolutions of 3.5 Å (Extended Data Figs. 3a-d, Extended Data Table 1). Two-fold symmetry (C2) was imposed on the homodimer during the last stages of cryo-EM map determination, as this improved the overall map resolution (Methods, Extended Data Fig. 2). The structures all exhibited clear density for the majority of the side chains (Extended Data Fig. 4a-d, g) and, where present, the ligands ( Fig. 1e-g, Extended Data Fig. 4f). All of the structures contained densities on the periphery of the transmembrane regions of Ste2 that were attributed to putative sterols ( Fig. 1a-d, Extended Data Fig. 4g). These were either unmodelled or, where the density was sufficiently strong, assigned putatively as cholesterol hemisuccinate (CHS) as this was present in vast molar excess throughout receptor purification; we cannot exclude the possibility that the densities may represent other sterols.

Article
(superscripts denote class D1 numbering system 1 ). Mutation of F204 5x28 and N205 5x29 in this region abrogates both ligand binding and signalling 1,5,6 , highlighting the importance of the conformational change. Upon antagonist binding, the side chains of Phe204 5x28 flip inwards by 180˚ (7 Å inward shift of the Cα atom). Lys202 5x26 has to flip in the opposite direction to prevent it clashing with the ligand (Extended Data Fig. 5b), although there is only weak density for the side chain in Ste2-Ant because it extends into the detergent micelle (Extended Data Fig. 3g, h). Upon ligand binding, Phe204 5x28 makes extensive contacts with the antagonist and to α-factor in Ste2-Ag-G. Other side chains shift to prevent clashing with the ligand (for example, Tyr128 3x29 , Asn132 5x29 , Gln135 3x36 , Asp274 ECL3 and Asp275 ECL3 ) and Asn205 5x29 changes its orientation slightly to make better contacts with the antagonist (Extended Data Fig. 5b). When mutated, all of these residues have major effects on ligand binding and/or signalling 1,7-9 . The structure of Ste2-Ant also shares similarities with Ste2-Ag-G (r.m.s.d. 1.3 Å), but there are also some marked differences ( Fig. 2a-d). The antagonist adopts the same hairpin shape as α-factor in the orthosteric binding pocket (Extended Data Fig. 5a), but makes contacts with fewer residues in the receptor (20 compared with 32 in Ste2-Ag-G; Extended Data Fig. 5c). The different sequences at the N terminus of the ligands define ligand efficacy 4 and they differ in structure substantially. The structure of the antagonist N terminus bound to Ste2 is linear, with His2 forming a hydrogen bond with Asn205 5x29 on H5 and Ala3 interacting with Thr274 ECL3 in ECL3 (Fig. 2a). By contrast, the same region in α-factor is Y-shaped, which splays Trp1 and Trp3 wide apart (Fig. 2a), with Trp3 in a similar position to His2 in the antagonist and interacting with H5 (Lys202 5x26 , Asn205 5x29 and Ile209 5x33 ), whereas Trp1 interacts with H6 (Ile263 6x55 and Tyr266 6x58 ). This is a key feature of receptor activation, because a 5.7 Å shift of the extracellular end of H6 (measured at Cα Ser267 6x59 ) is required to accommodate this conformation of α-factor and is accompanied by a similar shift in the extracellular end of H7 (7.2 Å at Cα Gly-273 ECL3 ) and a rearrangement of ECL3 (Fig. 2a). A consequence of the H6 movement upon agonist binding is a rotamer change of Tyr266, which displaces a water molecule and allows it to make interactions with α-factor and Asn205 that are not possible in the antagonist-bound state (Extended Data Fig. 8a).
To interrogate the early events of peptide binding that lead to receptor activation, we performed 3D variability analysis of the cryo-EM data (Extended Data Fig. 6a-d, Supplementary Videos 1-4). The 3D variability analysis algorithm 10 provides detailed visualization of the flexibility within a 3D reconstruction by analysing the conformational landscape of a protein molecule using a linear subspace model of 3D structures. Analyses of 3D data along the vectors of motion revealed key facets of peptide binding that were not evident in the consensus cryo-EM maps. In the Ste2-Ant state (Extended Data Fig. 6b), the receptor transitions from density maps that lack signal for the N terminus of the antagonist peptide to maps where this region becomes discernible, whereas the C terminus of the peptide is clearly visible. The stable engagement of the C-terminal region with the receptor is consistent with previous kinetic data suggesting that the C-terminus of the peptide engages with the receptor first 11 . By contrast, in the Ste2 IL -Ag state (similar to the inactive Ste2-Ant state; the superscript denotes 'inactive-like' (IL)) (Extended Data Fig. 6c), densities for both the N-terminal and C terminal sections are visible in all maps along the principal components, supporting a stronger engagement of Trp1 to cause the shift of the extracellular region of H6 and the rotamer change of Tyr266 6x58 ; by contrast, density in the middle of the peptide was weaker, presumably owing to its flexibility. In the 'active-like' (AL) agonist-bound Ste2 AL -Ag state (similar to the active Ste2-Ag-G state; Extended Data Fig. 6d), data from all principal components indicated a stable engagement of the entire peptide within the orthosteric-binding pocket. This stable interaction of the peptide upon receptor activation is consistent with ligand-binding pocket (LBP) volume analysis (Extended Data Fig. 5e), which shows approximately 1,400 Å 3 reduction from Ste2-LF to Ste2 AL -Ag. This is similar to the trend observed in a class A GPCR 12 , but is the opposite of the trend observed in a peptidergic class B GPCR, where a progressive expansion of approximately 1,200 Å 3 of the LBP occurs upon forming an active state 13 . The structural differences in the extracellular region between Ste2-Ant and Ste2-Ag-G are mirrored by substantial differences in the intracellular sections of H5, H6 and H7 (Fig. 2b). The intracellular end of H6 in Ste2-Ant is kinked outward away from the receptor core by 38˚ (12 Å shift at Cα Ser243 6x35 ) compared with its position in Ste2-Ag-G, and this section is also rotated by 170° along the helix axis (Fig. 2c). In addition, the intracellular end of H5 is 6.6 Å further out from the receptor core than in the active state. H7 adopts a markedly different conformation in the two different states (Fig. 2b-d). In Ste2-Ag-G, the intracellular end of H7 is α-helical and kinks outwards from the receptor core by 20 Å (at Cα Asn301 7x61 ) and makes contacts with the other protomer in the dimer. By contrast, in Ste2-Ant, H7 is kinked in towards the receptor core at an angle that differs by 47° from the active state and the residues Ala296 7x56 -Ser303 7x63 form a random coil. No density was observed for the 128 C-terminal residues in either structure, which are predicted to be unstructured 14 (TMHMM v2.0). The position of the intracellular end of H7 in Ste2-Ant overlaps with the position of the α5 helix 1 of the α-subunit Gpa1 in Ste2-Ag-G and thus sterically blocks G protein coupling. Notably, two putative sterol molecules reside at the kink in H7 in the inactive states of Ste2 (Extended Data Fig. 8b), but not in the active states; the sterol ergosterol is known to be important for Ste2 function 15 .

Agonist-bound states
In the cryo-EM images of ligand-free Ste2, Ste2-Ant and Ste2-Ag-G, there was only a single major conformation of the dimer. However, during processing of the Ste2-Ag dataset, two different conformations of Ste2 were found, and therefore two separate structures were determined (Extended Data Figs. 2, 3). One structure was very similar to Ste2-Ant (r.m.s.d. 0.7 Å over 263 Cα atoms, protomer A aligned) and is denoted Ste2 IL -Ag. The other structure was very similar to Ste2-Ag-G (r.m.s.d. 0.45 Å over 290 Cα atoms, protomer A aligned) and is denoted Ste2 AL -Ag (the superscript denotes 'active-like' (AL)). In addition, there were a number of classes similar to Ste2 AL -Ag that were heterogeneous and could not be refined further (Extended Data Fig. 2; Methods). The active-like states (Ste2 AL -Ag and heterogeneous Ste2 AL -Ag) represented 84% of the particles analysed. Ste2-Ant and Ste2-Ag-G represent two extremes in the transition from an inactive state to an active state, and both Ste2 IL -Ag and Ste2 AL -Ag are consistent with being intermediate states (Fig 3a-c). The cryo-EM density for Ste2 IL -Ag and Ste2 AL -Ag shows lower resolution in H5-H7 than observed in Ste2-Ant and Ste2-Ag-G, resulting in only a proportion of the side chains being modelled (Extended Data Table 2), which highlights their dynamic role in the conformational change. On the extracellular face of the receptor, Ste2 IL -Ag has a conformation part way between that of Ste2-Ant and Ste2-Ag-G. This conformation exhibits the agonist-induced displacement of H6 by 5.3 Å (measured at the Cα of Ser267 6x59 ), although this is 0.4 Å short of its position in Ste2-Ag-G, and the reorientation of ECL3. However, there is no significant displacement of the extracellular end of H7 as observed in Ste2-Ag-G and neither are major changes on the intracellular face that are necessary for G protein coupling (Fig. 3a). These additional structural changes occur in the transition from Ste2 IL -Ag to Ste2 AL -Ag (Fig. 3b). The extracellular face of Ste2 AL -Ag is almost identical to that of Ste2-Ag-G, with the full shift of H6 by 5.7 Å and H7 by 7.2 Å (at Cα Gly273 ECL3 ). On the intracellular face, H7 in Ste2 AL -Ag kinks outwards by 19.4 Å (at Cα Asn301 7x61 ) and H6 moves inwards by 11.2 Å (at Cα Ser243 6x35 ), which are only 0.6 Å and 0.7 Å short of their positions in Ste2-Ag-G. By contrast, there is a lateral movement in Ste2 AL •Ag of the intracellular end of H5 (5.1 Å at Cα Arg234 5x58 ).
Although Ste2 AL -Ag has attained a conformation where the G protein coupling site is unblocked, there are additional structural changes required for full engagement of the G protein (Fig. 3c). The transition from Ste2 AL -Ag to Ste2-Ag-G involves inward movements of H5 (4.7 Å at Cα Phe235 5x59 ), H6 (1.2 Å at Cα Phe244 6x36 ) and intracellular loop 3 (ICL3) and a lateral movement of H3, H4 and ICL2 by 1-2 Å. These movements result in repositioning of key residues, such as Asn158 ICL2 (1.8 Å at Cα), the side chain of Lys225 5x49 in H5, Leu247 6x39 in H6 (1 Å at Cα), and the side chain of Leu289 7x49 in H7, to enable favourable interactions with Gpa1. Mutations in all these residues strongly affect signalling 1 .
Together, the five cryo-EM structures of Ste2 suggest the following activation mechanism. Agonist binding to the ligand-free inactive state of Ste2 causes reorientation of the extracellular end of H5 to facilitate ligand binding, an outward movement of the extracellular end of H6 and a shift of ECL3 (Ste2 IL -Ag state), resulting in a contraction of the orthosteric binding site by an average of 27% (Extended Data Fig. 5e). This state is in equilibrium with the Ste2 AL -Ag state, in which the extracellular end of H7 has moved inwards and the intracellular end of H7 has flipped outward from the G protein binding site, with the cytoplasmic end of H6 shifting inwards. These conformational changes form the α5-helix G protein binding site, further reduce the volume of the orthosteric binding site and increase the number of receptor-ligand contacts to H5 (Extended Data Fig. 5d). Engagement with a G protein stabilizes a further contraction of the α5 helix binding site through inward movements of the intracellular ends of H3, H4 and H5, resulting in a slight increase in volume of the orthosteric binding site and minor changes on the extracellular surface of the receptor.

The dimer interface
Dimerization of Ste2 is essential for its activity 16,17 . During the activation process, the dimer interface remains intact but changes with respect to the residues making contacts and its surface area ( Fig. 4a, b). The dimer interfaces are smaller in ligand-free Ste2, Ste2-Ant and Ste2 IL -Ag (2,166 Å 2 , 2,138 Å 2 and 2,145 Å 2 , respectively) compared with Ste2 AL -Ag and Ste2-Ag-G (2,727 Å 2 and 2,519 Å 2 , respectively). Twenty-one residues in Ste2 A form interfacial contacts in all the structures determined. Another seven residues form contacts only in Ste2 AL -Ag and Ste2-Ag-G (Fig. 4a), owing to the interactions of the C-terminal end The dynamic nature of the dimer interface was not expected and suggested that there may be correlated residue movements that promote allosteric communication (cross-talk) between the two protomers. We therefore calculated statistical correlation in residue movements 18-20 for each of the dimer structures from the molecular dynamics simulation trajectories (Extended Data Fig. 7a, b) to identify the residue networks involved in allosteric communication. Of note, in Ste2-Ant and Ste2 IL -Ag, the pipeline of residues showing the strongest correlations in their movement ran from the extracellular surface of one protomer, across the dimer interface, to the intracellular surface of H7 of the other protomer (Fig. 4d). By contrast, in the active states (Ste2 AL -Ag and Ste2-Ag-G), the strongest correlated motions are within each protomer. Other weaker allosteric communication pipelines calculated for the four different states are shown in Extended Data Fig. 7b. G protein coupling affected allosteric communication within the Ste2 dimer, as seen by the differences between Ste2-Ag and Ste2-Ag-G, analogous to structural changes at the orthosteric binding site and extracellular surface observed upon coupling to a G protein 12 . Mutation of the residues that contribute to the allosteric communication can have an effect either on ligand binding (for example, Tyr101 2x63 , Tyr106 D1e1x46 , Ser184 4x64 and Tyr199 ECL2 ) and/or receptor activation (for example, Ile53 1x43 , Arg58 1x48 , Met69 1x59 , Ile80 2x42 , Ser95 2x57 , Leu102 2x64 , Asn132 3x33 , Glu143 3x44 , Ser184 4x64 , Thr199 ECL2 and Arg231 5x55 ; see Supplementary Table 1 for details and references). Further work is necessary to identify the functional consequences of the allosteric communication across the dimer interface, which may also help the understanding of allosteric communication in class A and class B GPCR dimers.

Role of conserved residues
Class A GPCRs contain the highly conserved NPXXY 7. 53 motif on H7 which facilitates G protein binding through formation of a Tyr 5.58 -Tyr 7. 53 hydrogen bond in the active state; this is coupled to outward movement of H6 (superscripts denote Ballesteros-Weinstein numbering 21 ). Class D1 GPCRs contain the highly conserved LPLSSMWA motif (residues 289 7x49 -296 7x56 ) on H7 in a similar position to the NPXXY 7. 53 motif, but functions very differently (Extended Data Fig. 8d). Pro290 7x50 is conserved in 99% of Ste2 sequences and facilitates the formation of a 3 10 -helical region, which forms the kink in H7 in the inactive state.
Mutations of residues in this motif have strong effects on the efficacy of signalling as it occupies the receptor core and functions as a central hub that mediates transition towards the active state by forming interactions with conserved residues in H3, H5 and H6 (Supplementary Table 1). The 47° kink at Pro290 7x50 in the inactive state of Ste2 is reminiscent of the 60° kink in H6 of class B GPCRs at the conserved sequence PxxG 6.50b that is essential for the outward movement of its cytoplasmic end to form the G protein-binding cleft during receptor activation 22 . However, the sequence motifs around the respective Pro residues are different, as are the mechanisms of activation of class B and class D1 GPCRs.
Another highly conserved motif in class A GPCRs is the DRY motif, in which Arg 3. 50 forms an ionic lock with H6 and stabilizes the inactive state. Ste2 does not share this motif, but Gln149 3x50 in H3 (99% conserved) occupies a position analogous to Arg 3. 50 and appears to stabilize the inactive state of Ste2, but by a different mechanism (Extended Data Fig. 8e). In the inactive state, Gln149 3x50 makes a hydrogen bond with the conserved Ser292 7x52 of the LPLSSMWA motif (residues 289 7x49 -296 7x56 ) in H7 and hydrophobic interactions with Met218 5x42 , and Leu289 7x49 packs against Leu146 3x47 . Upon activation, Ser292 7x52 is replaced with Leu289 7x49 , which then packs with Ile80 2x42 via van der Waals interactions, and this enables Leu289 7x49 to interact with Ile471 H5.25 Residues making interfacial contacts in all structures R esidues not resolved in the structure Residues making interfacial contacts in only active states R esidues making interfacial contacts >60% of the time during MD simulations (the superscript refers to the residue position in the common Gα numbering scheme for G proteins 23 ) of the 'wavy hook' in the G protein α-subunit Gpa1 along with Gln149 3x50 . Mutations in Gln149 3x50 , Ser292 7x52 , Leu289 7x49 , Leu146 3x47 and Met218 5x42 cause strong constitutive activity, whereas Ile80 2x42 exhibits negative cooperativity 24-27 with Gln149 3x50 (Supplemental Table 1). Furthermore, side-chain changes in the conserved positively charged residues of H5 (K225 5x49 and R233 5x57 ) are necessary to allow G protein coupling and cause deficient signalling when mutated 28-30 (Extended Data Fig. 8h). Pro258 6x50 is conserved in 98% of Ste2 sequences and facilitates the formation of a disordered region in H6, resulting in the C-terminal section pointing outwards in the inactive state so that it does not clash with the C-terminal section of H7 (Extended Data Fig. 8f). The kink in H6 packs against Leu291 7xx51 and Ser292 7x52 , which form a highly conserved L/I/F-S/ G/T motif that is found in 95% of Ste2 sequences. The remainder of the C-terminal portion of H6 packs against H5. Pro258 6x50 is important for stabilizing the inactive conformation as suggested by the mutations P258L and P258C, which cause sevenfold and 47-fold increases in basal signalling, respectively 28,31 (Supplementary Table 1). Furthermore, the kink in H6 is coupled to a 180° flipping of the side chain of Gln253 6x45 that allows it to interact with Trp295 7x55 of the LPLSSMWA motif (Extended Data Fig. 8g). Upon activation, H6 assumes a straight α-helix that packs between H5 and H7, and Gln253 6x45 flips towards Ser288 7x48 as the conserved Trp295 7x55 in H7 moves away to enable H7 to make contacts with both H1 and H7 in the neighbouring protomer. Consistently, Gln253 6x45 and Pro258 6x50 are two of the strongest constitutively activating mutations in Ste2 and mutations in Ser288 7x48 of the LPLSSMWA motif lead to constitutive activity 8,27 .

Discussion
The activation mechanism of mammalian GPCRs (classes A, B, C and F) has been studied intensively, most recently through the structure determination of many GPCR-G protein complexes by cryo-EM 32-34 . The paradigm for GPCR activation is that inactive states that cannot couple to a G protein undergo a conformational change facilitated by agonist binding that allows coupling to occur 2 . GPCRs appear to fall into three different categories (type I-III) depending on how the G protein binding site prevents coupling in the inactive state (Fig. 5). In class A and class B receptors (Fig. 5a), the primary block to G protein coupling is the cytoplasmic end of H6, which upon receptor activation moves away from the receptor core by 10-15 Å, forming a cleft in the cytoplasmic face of the receptor that binds the α5 helix of the G protein α-subunit 32 . In class F receptors (Fig. 5a), H6 blocks the G protein coupling site but requires a smaller movement (6 Å) to allow G protein coupling, because of the different angle of engagement of the α5 helix 33 . G protein coupling to class C receptors is very different (Fig. 5b), with the α5 helix engaging the periphery of the receptor primarily through ICL2, with the site being formed by a 5 Å movement of H3 34 . Class C receptors also provide the only previous data for how a GPCR dimer is activated, but dimerization and activation are mediated by the extracellular venus flytrap domains which are absent in all other GPCR classes. In addition, class C GPCR dimers are asymmetric and couple to a single G protein 35 , in contrast to the symmetric dimer of Ste2, which can couple simultaneously to two G proteins 1 . Here we show that the class D1 receptor Ste2 has an activation mechanism distinct from other known GPCRs (Fig. 5c). Activation of Ste2 requires both the removal of a blockage from the G protein coupling site (H7) and formation of the binding site primarily through the inward movement of H6. In addition, Ste2 provides a model for how interactions at the dimer interface can change during receptor activation of transmembrane-mediated GPCR dimers, which could have implications for understanding signalling in other GPCR dimers.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04498-3.

Data reporting
Sample size was not predetermined by statistical methods, no randomization was applied to experiments and investigators were not blinded to experimental allocation or outcome assessment.

Cloning and expression of wild-type Ste2
A construct encoding wild-type S. cerevisiae Ste2 (residues 1-431) that included a tobacco etch virus (TEV) cleavage site, eGFP, and a decahistidine tag (wtSte2-TEV-eGFP-His10) was synthesized as a gBlock gene fragment (IDT) and cloned into plasmid pAcGP67-B (BD Biosciences) by in vivo assembly in Escherichia coli XL10-Gold 43 . Expression of Ste2 in Trichoplusia ni High Five cells (Thermo Fisher Scientific; we did not test for mycoplasma) using high-titre recombinant baculoviruses prepared with the FlashBAC ULTRA system (Oxford Expression Technologies) was performed as described previously 1 . The cells were then collected by centrifugation, flash-frozen in liquid nitrogen, and stored at −80 °C until further use.
Purification of Ste2 in the agonist-bound state was performed as described above except the antagonist α-factor peptide was replaced with agonist α-factor peptide with the sequence WHWLQLKPGQPMY (Genscript).

Purification of Ste2 in the ligand-free state
To purify Ste2 in the ligand-free state, we devised PSGWAY, which involves addition of purified wild-type Gpa1-Ste4-Ste18 heterotrimer to stabilize ligand-free Ste2 before solubilization from cell membranes. Since the agonist α-factor was not present, the wild-type Gpa1-Ste4-Ste18 heterotrimer transiently stabilized Ste2 during detergent solubilization, but did not form long-term stable interactions with the receptor, thereby allowing purification of ligand-free Ste2 in reasonable yields that were not achieved in the absence of wild-type G protein heterotrimer. A similar approach has recently been used to purify the class B1 CGRP receptor in the ligand-free state for cryo-EM structure determination, however in that study a specific mutant Gα 11 protein was co-expressed in insect cells to transiently stabilize the ligand-free CGRP receptor 13 . Purification of wild-type Gpa1-Ste4-Ste18 heterotrimer was performed as described previously 1 . In brief, wild-type Gpa1 was expressed and purified from E. coli. Ste4-Ste18 was co-expressed and purified from T. ni cells (Expression Systems) using the flashBAC ULTRA system (Oxford Expression Technologies). The wild-type Gpa1-Ste4-Ste18 heterotrimer complex was formed by mixing Gpa1 and Ste4-Ste18 dimer at 1:1 molar ratio and the complex was loaded and separated on Superdex 200 10/300 GL column (GE Healthcare) pre-equilibrated with 10 mM HEPES pH 7.5, 100 mM NaCl, 10% glycerol, 1 mM MgCl 2 and 0.01 mM GDP. Peak fractions were pooled and the complex was concentrated in a 10 K molecular weight cut-off Amicon ultra centrifugal concentrator (Merck). For stabilizing ligand-free Ste2, insect cell membranes with Ste2 were incubated with 5.7 mg of purified wild-type Gpa1-Ste4-Ste18 heterotrimer supplemented with 400 mU ml −1 apyrase for 2 h. Solubilization and purification of ligand-free Ste2 was then performed as described above for the antagonist-bound Ste2, except no α-factor peptide was added throughout purification. As wild-type Gpa1-Ste4-Ste18 heterotrimer only transiently interacts with Ste2 in the absence of α-factor, the Ste2 eluate from Ni-NTA column did not contain any G protein heterotrimer as it had been eliminated in the washing steps. Only limited structural information about ligand-free states of GPCRs is currently available 13 and the PSGWAY method described here could possibly be adapted to purify ligand-free states of other GPCRs.

Vitrified sample preparation and data collection
Cryo-EM grids were prepared by applying 3 µl of purified agonist-bound Ste2 (4.7 mg ml −1 ), antagonist-bound Ste2 (3.1 mg ml −1 ) and ligand-free Ste2 (3 mg ml −1 ) onto glow-discharged holey gold (Quantifoil Au 1.2/1.3 300) mesh. The grids were blotted with filter paper for 2.5 s before plunge-freezing in liquid ethane (at −181 °C) using an FEI Vitrobot Mark IV at 100% relative humidity and 4 °C. All cryo-EM datasets were collected on an FEI Titan Krios microscopes operating at 300 kV. For the antagonist-bound Ste2 dataset, images were recorded on a K3 direct electron detector in counting mode post quantum energy filter (Gatan) operated in zero-energy-loss mode with a slit width of 20 eV to obtain dose-fractionated movies of the sample with a 100 µm objective aperture. 15,751 micrographs were recorded at a magnification of 105,000× (0.86 Å per pixel; LMB Krios3) as dose-fractionated movie frames with an exposure time of 2.08 s amounting to a total exposure of 57 e -Å −2 and defocus range set between −0.7 and −2.0 µm. For the ligand-free Ste2 dataset, images were recorded on a Falcon 4 direct electron detector (Thermo Fischer Scientific) operated in Electron Event Representation (EER) mode with a 100 µm objective aperture. 9,369 micrographs were recorded at a magnification of 96,000× (0.85 Å per pixel; LMB Krios2) in super-resolution mode with an exposure time of 8.81 s amounting Article to a total exposure of 54 e -Å −2 and defocus range set between −0.8 and −2.4 µm. For the agonist-bound Ste2 dataset, images were recorded on a K2 direct electron detector in counting mode post quantum energy filter (Gatan) operated in zero-energy-loss mode with a slit width of 20 eV to obtain dose-fractionated movies of the sample with a 100 µm objective aperture. 6,944 micrographs were recorded at a magnification of 105,000× (1.1 Å/pix; LMB Krios2) as dose-fractionated movie frames with an exposure time of 12.5 s amounting to a total exposure of 50 e -Å −2 and defocus range set between −0.9 and −2.7 µm.

Cryo-EM data processing and 3D reconstruction
Image stacks (15,751 antagonist-bound, 9,369 ligand-free and 6,944 agonist-bound Ste2 movies) were subjected to beam-induced motion correction using MotionCor2 44 by dividing each frame into 5 × 5 patches. CTF parameters were estimated from non-dose-weighted micrographs in GCTF 45 with equiphase averaging for antagonist-bound and agonist-bound Ste2 datasets and CTFFIND-4.1 in RELION3.1 46 for ligand-free Ste2 dataset. Autopicking was performed using BoxNet deep convolutional neural network implemented in Warp 47 that yielded 4,982,771 particles, 2,409,127 particles and 1,854,854 particles for the antagonist-bound, ligand-free and agonist-bound Ste2 datasets, respectively. Particles were extracted in a box-size equivalent to 190 Å and down-scaled initially to 3.3 Å per pixel. An ab initio 3D model was generated using stochastic gradient descent algorithm implemented in RELION3.1. The extracted particles were subjected to two rounds of 3D classification in C1 symmetry in RELION3.1 and the particles that displayed clear transmembrane features were selected which yielded 1,285,901 particles, 463,024 particles, and 399, 652 particles for the antagonist-bound, ligand-free and agonist-bound Ste2 datasets, respectively. These particles were consistent with a homodimeric Ste2 in all three datasets and no reliable classes consistent with a Ste2 monomer were obtained despite extensive 2D and 3D classifications. These particles were re-extracted in a box-size equivalent to 210 Å and subjected to 3D reconstruction in C1 symmetry followed by iterative rounds of Bayesian polishing, beam-tilt correction and per-particle CTF refinement in RELION-3.1. Subtraction of the detergent micelle signal was then performed and the particles were subjected to 3D classification without alignment. For the antagonist-bound Ste2 dataset, 136,877 particles that displayed high-resolution features were selected and subjected to Bayesian polishing, per-particle CTF refinement using the improved model and 3D reconstruction in C2 symmetry which yielded a map that refined to a global resolution of 2.69 Å (Fourier shell correlation (FSC) = 0.143). For the ligand-free Ste2 dataset, 67,791 particles that refined to a high-resolution were selected and the detergent micelle signal was reverted. The particle stack was subjected to Bayesian polishing and per-particle CTF refinement using the improved model. The particle stack was then exported to cryoSPARC v3.1 48 where they underwent non-uniform 3D refinement 49 in C2 symmetry which yielded a map that refined to a global resolution of 3.10 Å (FSC = 0.143). For the agonist-bound Ste2 dataset, two distinct classes were obtained from 3D classification without alignment: one class with 64,091 particles that resembled the antagonist-bound and ligand-free Ste2 in the inactive state (therefore called Ste2 IL ) and another class with 65,284 particles that resembled the G-protein heterotrimer coupled Ste2 structure in the active state (therefore called Ste2 AL ). The three other classes obtained from 3D classification without alignment were heterogenous and it was not possible to separate into distinct intermediate states upon further 3D classification. However, these heterogenous classes were more similar to the active-like state since TM7 showed relatively strong density facing the dimer interface, which is observed only in the active-like and G protein heterotrimer coupled active states of Ste2. Thus, in the presence of the agonist α-factor, the majority of particles (83.9%) correspond to an active-like state or intermediate states transitioning towards the active-like state and only a minority (16.1%) of the particles correspond to a distinct inactive-like state which is consistent with α-factor binding transitioning Ste2 into an active-like state that is conducive to G protein coupling. The distinct set of 64,091 particles (Ste2 IL state) and 65,284 particles (Ste2 AL ) were refined separately and showed high-resolution features corresponding to these distinct intermediate states. The detergent micelle signal was reverted and these separate particle stacks were subjected to Bayesian polishing and per-particle CTF refinement. The particles were then exported to cryoSPARC v3.1 where they underwent non-uniform 3D refinement in C2 symmetry that yielded maps at global resolutions of 3.53 Å and 3.46 Å for the inactive-like and active-like agonist-bound Ste2, respectively (FSC = 0.143). C2 symmetry was applied to the final reconstructions, after 3D classifications, since no significant differences between the two protomers were observed in the high-resolution cryo-EM maps obtained in C1 symmetry. In addition, application of C2 symmetry improved the resolution of the maps by 0.2-0.3 Å, which is expected to occur only if the two protomers are identical. The improved map resolution increased confidence in model building in regions with relatively lower local resolution in the intermediate state Ste2 IL -Ag and Ste2 AL -Ag structures. The refined particle stacks that contributed to each of the four final maps described above were used for 3D-Variability analysis 10 as implemented in cryoSPARC v3.1. A generous mask generated with a 5 pixel map expansion and 10 pixel soft edge was used to capture any possible motions during the 3D variability analysis. For each particle stack corresponding to one of the four distinct states of Ste2 (antagonist-bound, ligand-free, agonist-bound inactive-like, agonist-bound active-like Ste2), four orthogonal principle modes (eigenvectors of the 3D covariance) were solved. The resulting volume frame data generated in CryoSPARC v3.1 were examined in UCSF Chimera 50 as volume series and captured as movies. Local resolution was determined in cryoSPARC v3.1 that uses a local windowed FSC method similar to the blocres program of Bsoft package 51 . Postprocessed maps were generated in either RELION or cryoSPARC v 3.1. Postprocessed maps shown in Extended Data Fig. 3a-d was generated using the Deep-EMhancer algorithm 52 .

Structure determination and model refinement
The model of Ste2 receptor portion of Ste2-miniGpa1-Ste4-Ste18 complex (PDB: 7AD3) was used as an initial template and was fitted into the cryo-EM density maps in UCSF ChimeraX 36 . Portions of the receptor that differed from the initial model were rebuilt manually in COOT 53 followed by iterative rounds of refinements in CCP-EM 54 and PHENIX 55 software suites and manual model building in COOT. Restraints for NAG and CHS (monomer library ID Y01) were derived in eLBOW 56 and AceDRG 57 , respectively. C2 symmetry constraints were applied during model refinements. Comprehensive model validation was performed in PHENIX and MolProbity 58 . The final models achieved good geometry (Extended Data Table 1). No density was visible for the N-terminal residues 1-4 and C-terminal residues 304-431. For the Ste2-Ant, Ste2 AL -Ag and Ste2 IL -Ag states, density modification was performed on unfiltered half-maps using phenix.resolve_cryo_em 59 for better visualization of some weaker map regions, and where present water molecules, to assist model building in COOT; however all refinements and map validations were performed against the original EM map.

Molecular dynamics simulations
All-atom molecular dynamics simulations were performed using the CHARMM36m forcefield 62 and GROMACS MD package. The simulations were started from the cryo-EM structures of the Ste2 dimer in Ste2-Ant, Ste2 IL -Ag, Ste2 AL -Ag and Ste2-Ag-G. The structures of Ste2-Ant. Ste2 IL -Ag and Ste2 AL -Ag were prepared using Maestro protein-preparation wizard (Schrödinger Release 2021-2), including adding missing heavy atoms and all hydrogens, followed by full energy minimization. Histidine residues in the ligands were in the uncharged imidazole form. The CHS molecules from the cryo-EM structures were aligned and replaced by cholesterol using Pymol. The protein-cholesterol complexes were taken as initial input for CHARMM-GUI lipid bilayer builder [63][64][65] and placed in a 9 × 9 nm 2 POPC bilayer. The resulting protein-lipid complexes were solvated with water molecules to a thickness of 1.4 nm from the receptor along the z-axis. The charges in the systems prepared were neutralized with 0.15 M concentration of NaCl. The neutralized systems were first heated from 0 K to 310 K in NVT ensemble using a Nosé-Hoover thermostat 66 in 0.2 ns. A restraint force of 5 kcal mol−1 Å −2 was applied on the heavy atoms of protein and lipid during the heating process. The restraints force was gradually reduced to 0 kcal mol −1 Å −2 in a 30 ns equilibration protocol performed with an NPT ensemble, with a 1 kcal mol −1 Å −2 per 5 ns window. The pressure was controlled using Parrinello-Rahman method 67 and the simulation systems were coupled to a one bar pressure bath. The last frames from the equilibrium protocol were taken as the initial structure for the regular molecular dynamics production run. Five random seeds were used to assign initial velocities to start five production molecular dynamics simulation runs for each Ste2 dimer system. The production run was 400 ns long for each velocity. In all simulations, the LINCS algorithm was applied on all bonds and angles of water. The integration time step was set to 2 fs. A cut-off of 1.2 nm was used for non-bond interactions and the particle mesh Ewald method (PME) 68 was applied to treat long-range Lennard-Jones interactions. The MD trajectories were stored at every 20 ps interval. The 200 to 400 ns production trajectories of each velocity (200 ns × 5 = 1,000 ns total) were combined for each conformation of Ste2 dimer for analysis.
The inter transmembrane helix residue contact analysis was carried out using the script get_contact (https://getcontacts.github.io/), that calculates different types of residue contacts including hydrogen bond, van der Waals, salt bridge and cation-pi contacts. The definition of the range of residues in each transmembrane helix was taken from previous work 1 . Each pair of the transmembrane helices was input as two selection groups to the get_contact script to calculate the contact formed during dynamics and the frequency of each of these contacts across the whole molecular dynamics trajectories. A pair of residues is considered to form a sustained contact during simulation if the contact frequency is greater than 60%.
The interaction energies between two monomers for each system were calculated as averaged over the molecular dynamics simulation trajectories with the GROMACS energy module. The interaction energy calculation was calculated between residues in TM1 and TM7 from one monomer with the residues in TM1 and TM7 of the other monomer. The interaction energy was calculated as the sum of short-range van der Waals interaction energy and Columbic interaction energy.
Allosteric communication residue network analysis used the in-house software 20,69 Allosteer to identify the network of residues involved in allosteric communication from the extracellular ligand binding region to the intracellular G protein coupling residue site in the dimer. We calculated the mutual information in torsional angle distribution for all pairs of residues in the Ste2 dimers. Then we used graphic theory to calculate the shortest pathway with highest mutual information connecting distant pairs of residues with high mutual information. We calculated the allosteric communication pathways from extracellular residues, going through the residues in the ligand binding site to the G protein coupling site residues. Previously we have shown that mutation of the residues predicted to be in the allosteric communication pipelines have shown change in receptor coupling to G proteins and/or β-arrestin in class A GPCRs 19,69, 70 . The list of residues in the Ste2 ligand binding site used for calculating the allosteric communication pathways were chosen based on their contact frequency with the ligand during dynamics. All the residues that showed greater than 60% frequency in contacts with the ligand were taken as ligand-binding site residues. Similarly, G protein binding site residues were also defined by contact analysis on a Ste2-Ag-G simulation. The pipelines between these two groups were sorted by their strength, and their rank reflects the strength of allosteric communication. The number of allosteric communication pathways passing through each residue is defined as a 'hubscore' and this is used to describe the contribution by each residue to the strength of the allosteric communication.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.