Unraveling allostery within the angiotensin II type 1 receptor for Gα_q and β-arrestin coupling

Abstract

G protein-coupled receptors (GPCRs) engage both G proteins and β-arrestins and their coupling can be biased by ligands and mutations. Here, to resolve structural elements and mechanisms underlying effector coupling to the angiotensin II (AngII) type 1 receptor (AT1R), we combined alanine scanning mutagenesis of the entire sequence of the receptor with pharmacological profiling of Gα_q and β-arrestin engagement to mutant receptors and molecular dynamics (MD) simulations. We showed that Gα_q coupling to AT1R involved a large number of residues spread across the receptor, whereas fewer structural regions of the receptor contributed to β-arrestin coupling regulation. Residue stretches in transmembrane domain 4 conferred β-arrestin bias and represented an important structural element in AT1R for functional selectivity. Furthermore, we identified allosteric small molecule binding sites that were enclosed by communities of residues that produced biased signaling when mutated. Finally, we showed that allosteric communication within AT1R emanating from the Gα_q coupling site spread beyond the orthosteric AngII-binding site and across different regions of the receptor, including currently unresolved structural regions. Our findings reveal structural elements and mechanisms within AT1R that bias Gα_q and β-arrestin coupling and which could be harnessed to design biased receptors for research purposes and to develop allosteric modulators.

Introduction:

G protein-coupled receptors (GPCRs) are the largest class of membrane-bound proteins, which initiate intracellular signaling cascades in response to a diverse array of extracellular signals (1). Agonist binding causes conformational changes in the intracellular regions of the receptor leading to recruitment and activation of different transducers like heterotrimeric G proteins (Gα_s, Gα_q/11, Gα_12/13 and/or Gα_i/o) and adaptor proteins such as β-arrestins (which include β-arrestin1 or β-arrestin2 and are abbreviated as β-arrestins)(2–5). The β-arrestins not only desensitize GPCRs by binding to receptors that have been phosphorylated by G protein receptor kinases (GRKs 1–7), thereby restricting G protein-dependent signaling, but also act as internalizing and signaling adaptors (4–6). The activation of G proteins and β-arrestins thus results in the regulation of many signaling pathways controlling diverse cellular responses. Ligands can also produce distinct signaling profiles through the preferential engagement of different G proteins and β-arrestins, a phenomenon known as functional selectivity or biased signaling (7). The development of biased ligands for GPCRs holds great promise for designing more effective therapeutics (8).

The angiotensin II (AngII) type 1 receptor (AT1R), a class A GPCR, is involved in cardiovascular regulation through the control of different cell functions like growth, contractility, and apoptosis (9). Although the best-characterized cell signaling events for AT1R’s effects are mediated by Gα_q/11 and β-arrestins, AT1R also couples to Gα_i/o and Gα_12/13 (10, 11). Moreover, GRK2/3 and GRK5/6 both phosphorylate AT1R and are involved in controlling β-arrestin recruitment to the receptor, leading to the regulation of AT1R signaling and trafficking (12–15). AT1R also exhibits functional selectivity toward Gα_q and β-arrestin upon binding to peptide derivatives of AngII as compared to its endogenous ligand (10, 11). The pursuit of AT1R-biased ligands is of great interest to the management of cardiovascular diseases (16, 17). Indeed, β-arrestin-biased AT1R ligands have been shown to improve cardiac functions (such as cardiomyocyte contractility and cell survival), while antagonizing undesired Gα_q-dependent responses mediated by AngII, such as blood pressure increase and cardiac hypertrophy. However, the molecular mechanisms of biased agonism that would aid ligand design remain largely unknown. Three-dimensional structures of AT1R bound to AngII, Gα_q- or β-arrestin-biased peptide ligands have revealed conformational changes within the receptor when it transitions from an inactive to an active state (3, 18–20). However, when bound to AT1R, biased ligands do not induce distinct structural differences in the receptor as compared to AngII. Receptor dynamics studies using double electron–electron resonance (DEER) experiments revealed that biased signaling emerges from stabilization of distinct ensembles of conformations manifested in the intracellular regions of AT1R (19). Molecular dynamics (MD) simulation studies on crystal structures of the active states of AT1R also showed that β-arrestin-biased AngII peptide derivatives stabilize a distinct conformational state ensemble (21). However, little is known about how information is communicated across the receptor to ensure Gα_q or β-arrestin coupling and which residues are involved in this process. Indeed, there is a paucity of information regarding structural regions over the entire AT1R structure that regulate the allosteric coupling of these effectors. Knowledge of these domains would not only help uncover the molecular underpinnings of functional selectivity, but also assist in the design of new biased allosteric modulators of AT1R, as well as biased receptor mutants useful for advancing our understanding of AT1R biased signaling roles in vivo.

A mutational scanning study on the entire β₂-adrenergic receptor has revealed differences in the sensitivity of each residue position toward Gα_s signaling (22), underscoring the utility of global identification of signaling-sensitive structural regions and residues in GPCRs. Additionally, naturally occurring mutations in AT1R alter the receptor’s signaling landscape, favoring one pathway over the other (10, 23), thereby supporting the importance of residues in different structural regions in directing receptor responses.

To provide a more complete understanding of the role of various structural regions in AT1R involved in both β-arrestin and Gα_q coupling, we combined whole receptor mutagenesis and the pharmacological profiling of these mutant receptors using bioluminescence resonance energy transfer (BRET)-based biosensors for Gα_q and β-arrestin. Analysis of this large-scale data set with atomistic MD simulations method helped uncover mechanisms of allosteric regulation of Gα_q- and β-arrestin-mediated signaling, as well as identify ensembles of residues involved in Gα_q and β-arrestin functional selectivity that go beyond the known structured regions of the AT1R. Our analysis provides a framework for dissecting functional selectivity and identifying key regions in GPCRs for biased-drug and-receptor design.

Results

Alanine-scanning mutagenesis effects on AT1R functionality

A library of AT1R mutants was generated through alanine substitution for each of the 359 residues (or glycine substitution for alanine residues) in the coding region of human wild-type (WT) AT1R (fig. S1). In combination with Gα_q- and β-arrestin-BRET sensors, this library was used to identify the residues in AT1R involved in coupling to these pathways and using in silico MD approaches, to understand mechanisms of functional selectivity (fig. S1). Mutants expressed in HEK293 cells at levels ranging from undetectable to 170% that of WT, with a median expression of 97% and interquartile range of 36% (78%–114%), indicating that most mutations minimally affected overall receptor expression (fig. S2A). However, alanine substitutions in the extracellular loops of AT1R reduced receptor expression to the greatest extent (fig. S2B). Functional responses of mutants for Gα_q coupling and β-arrestin recruitment were measured with the PKC-c1b (PKC) and with β-arr2-RlucII and rGFP-CAAX BRET-based biosensors respectively (6, 15). Responses were measured after 5 min of agonist stimulation at room temperature, when maximal and steady-state signals are reached for both PKC and β-arrestin responses, which allowed us to record early signaling events at the plasma membrane for both pathways (24, 25), thereby limiting potential spatial and temporal biases. The PKC-BRET biosensor also has a higher signal-to-noise ratio compared to the Gα_q biosensor; therefore, it was better suited for our initial screen of AT1R alanine mutants (10). This assay measures receptor coupling to the endogenous Gα_q/11 protein family and signals mirrored that of the AT1R and mutant responses when using the Gα_q BRET biosensor. We generated concentration-response curves to obtain the half-maximal concentration (EC₅₀) and maximal response (E_max) to AngII for WT and each mutant receptor for both pathways (figs. S3 and S4, table S1 and data file S1). We first determined the minimal level of receptor expression required to achieve maximal Gα_q and β-arrestin responses by titrating the amount of AT1R cDNA for transfection (fig. S5, A to D). Maximal responses were maintained for Gα_q and β-arrestin when receptors were expressed at >50% of cDNA used in the BRET screening assays but decreased exponentially below that threshold (fig. S5, C and D). Increasing AT1R expression beyond that of WT did not further improve either response. Therefore, 30 mutants expressing less than half the level of WT were excluded from further analysis (table S1).

Structurally resolved and unresolved regions of AT1R contain sensitive residues for Gα_q and β-arrestin coupling

We mapped the functional effects of alanine substitution mutations on AngII-mediated AT1R responses to Gα_q and β-arrestin (Fig. 1A and data file S2) by considering various pharmacological parameters of mutants as compared to WT. These included E_max (as % of WT), ΔlogEC_50, and the relative activity (RA), which represents the difference in the ratio of E_max over the EC₅₀ between WT and each mutant (Fig. 1A and data file S2) and allowed us to determine the relative effect of mutants on the combined efficacy and potency responses of AngII. We also calculated potential biases for Gα_q and β-arrestin from differences in RA values between responses (ΔΔlog(E_max/EC₅₀) (Fig. 1B). Residues that significantly affected the EC₅₀, E_max and RA in a negative or positive manner with respect to Gα_q-dependent signaling (Fig. 2, A to C, and data file S3) and β-arrestin coupling (Fig. 2, D to F, and data file S3) to receptors as compared to WT are shown in the AngII-bound AT1R active state crystal structure. Analysis of these pharmacological parameters showed that residues that increased or attenuated Gα_q-dependent signaling were greater in number than those having similar effects on β-arrestin coupling and were spread across the sequence of the receptor. This generalization also held true when focusing on differences in RA between Gα_q compared to β-arrestin pathways, because more mutants led to preferential β-arrestin coupling (defined as reducing Gα_q compared to β-arrestin responses) than preferential Gα_q coupling (Fig. 1B). Residues affecting EC₅₀ of AngII for β-arrestin coupling were found in the extracellular region of the receptor, whereas residues altering E_max were found both in the extracellular and intracellular domains of AT1R, including the C-tail of the receptor (Fig. 2, D and E). On the other hand, residues that altered EC₅₀ and E_max for Gα_q coupling were distributed throughout the receptor (Fig. 2, A and B). The distribution of residues that increased or decreased RAs for Gα_q and β-arrestin resembled residue patterns affecting EC₅₀ more than that for E_max (Fig. 2, C and F). It is important to note that RA values were dominated to a greater extent by changes in the EC₅₀ rather than by those in E_max, which is more constrained by the amplitude of the response as opposed to the EC₅₀, which can span over many log units. Transmembrane (TM) region 4 (TM4) contained a string of consecutive and neighboring residues that, when mutated to alanine or glycine (L143^4.40A, I150^4.47A, I151^4.48A, I152^4.49A, W153^4.50A, L154^4.51A, G157^4.54A, A159^4.56G, and L161^4.58A), increased EC₅₀, E_max and/or RA for enhancing β-arrestin recruitment (Fig. 1B and Fig. 2G, and data file S3). Similarly, β-arrestin responses were increased by S107^3.31A and S109^3.33A mutations, which are located at the N-terminal end of TM3. The N168A mutation in the extracellular loop 2 (ECL2) selectively increased the potency, efficacy, and RA for the AngII-mediated β-arrestin response, suggesting that this extracellular domain participates in the allosteric regulation of β-arrestin coupling to AT1R. In the unresolved structure of the C-tail, the S326A mutation improved the potency, efficacy, and RA of AngII for Gα_q coupling to AT1R without affecting those for β-arrestin responses, suggesting that this residue plays a negative regulatory role for Gα_q coupling, possibly through its phosphorylation status. Similarly, the S335^C-tailA and T336^C-tailA mutations increased the E_max for Gα_q coupling without significantly affecting β-arrestin responses (table S1 and data file S3). The L337^C-tailA and K58^12.49A mutations in the intracellular loop 1 (ICL1) reduced the potency, efficacy, and RA for β-arrestin recruitment to the receptor without significantly reducing those for Gα_q responses. L337^C-tail directly contacts β-arrestin (26). T332^C-tail is phosphorylated following agonist stimulation of AT1R (27), and the T332^C-tailA mutation decreased both the E_max and RA for β-arrestin recruitment without significantly affecting AT1R coupling to Gα_q. The highly conserved DRY motif in several GPCRs is important for coupling to G proteins and to form a so-called ‘water lock’, and the D125^3.49A, R126^3.50A, and Y127^3.51A mutations in the DRY motif of AT1R led to a reduction in RA for Gα_q coupling without significantly affecting β-arrestin responses (Fig. 2G and data file S3), consistent with previous reports (28–33). Similarly, we observed that mutation of residues in the well-conserved class A GPCR NPxxY motif (N298^7.49A, P299^7.50A, and Y302^7.53A) led to a reduction in EC₅₀, E_max, and RA for Gα_q coupling without significantly affecting the β-arrestin response, also consistent with previous reports (34, 35). However, the F301^7.52A mutation reduced EC₅₀, E_max, and RA for both β-arrestin and Gα_q responses. The PI(V)F (P207^5.50, V116^3.40 and F249^6.46 in AT1R) and C(S)WxP motifs (S252^6.47, W253^6.48, I254^6.49, and P255^6.50 in AT1R), which are known as “transition switch” motifs in GPCRs, link conformational changes in the orthosteric ligand-binding pocket of receptor to those in the transmembrane domains for receptor activation (29, 32). The V116^3.40A mutation resulted only in a reduction in potency, efficacy, and RA for Gα_q coupling. The F249^6.44A mutation led to a decrease in the potency, efficacy, and RA for Gα_q coupling and in the E_max for the β-arrestin response. In the C(S)WxP motif, the S252^6.47A mutation increased E_max for Gα_q coupling, whereas the W253^6.48A mutation decreased potency, efficacy, and RA for both responses. Lastly, I254^6.49A decreased E_max and RA for Gα_q coupling.

Fig. 1. Pharmacological and bias assessments of AT1R mutants for Gαq and β-arrestin responses.

(A) LogEC₅₀ and E_max values were obtained from AngII concentration response curves on either Gα_q (using the PKC BRET sensor) or β-arrestin (using the βarr2 BRET sensor) pathways and expressed as ΔLogEC₅₀ by subtracting WT logEC₅₀ and %E_max values. Mutants that expressed at less than 15% of WT are not shown (N89A, V108A, L128A, H132A, G160A, I164A, C180A, A181G, F182A, Y184A, S189A, L300A, and P322A). Mutants showing lack of binding/activity but expressed (R167A and C101A, blue square) were assigned an arbitrary value for ΔLogEC₅₀ of 4 and ΔLog(E_max/EC₅₀) of −4, and AngII E_max was derived from the response at 10 μM AngII. Mutants showing more than 10-fold reduction in EC₅₀ compared to WT are indicated as red circles. Transmembrane domains (TMs) were specified as transparent yellow boxes. (B) ΔΔLog(E_max/EC₅₀) for each mutant was calculated by subtracting β-arrestin ΔLog(E_max/EC₅₀) values from Gα_q ΔLog(E_max/EC₅₀) values. Bias factors are expressed as 10^{ΔΔLog (Emax/EC50)}. Mutants that expressed less than 50% of WT or that showed no activity in both pathways were excluded. Blue and red circles indicate mutants showing statistically significant bias to either β-arrestin (negative values) or Gα_q (positive values) pathways (P<0.05). Red line indicates the balanced WT response. Values are means ± SEM from 3 independent experiments. Raw values are provided in data file S2.

Fig. 2. 3D mutational landscape depiction of residues affecting Gα_q and β-arrestin responses.

(A to C) Residue positions that significantly affect Gα_q coupling upon mutation as assessed by EC₅₀ in green (A), E_max in red (B) and RA (C) (inset colors). (D to F) Residue positions that significantly affect β-arrestin coupling upon mutation as assessed by EC₅₀ in green (D), E_max in red (E) and RA (F) (inset colors). Residue positions are overlaid on the AT1R structure (PDB ID: 6OS0) by displaying the Cɑ atom of each residue. (G) Bar plots of the RA values for the displayed residues above in (C) and (F). Raw values are provided in data file S3.

Regions in AT1R and residues involved in Gα_q and β-arrestin biased coupling

Functional data suggest that alanine or glycine mutations in AT1R differentially affect the coupling of receptors to Gα_q and β-arrestin, generating biased signaling. Although we found that most mutants were balanced in their effects on Gα_q and β-arrestin recruitment to AT1R, others selectively engaged either pathway (Fig. 1B). Mutations that resulted in preferential engagement of Gα_q over β-arrestin were T332^C-tailA, L337^C-tailA, A225^5.68G, S331^C-tailA, S326 ^C-tailA, K58^12.49A, V246^6.41A, and L317^8.58A. Mutations that resulted in preferential engagement of β-arrestin over Gα_q were T141^4.38A, R140^34.57A, N294^7.45A, D125^3.49A, I130^3.54A, R126^3.50A, M134^34.51A, and H256^6.51A (figs. S6, A and B, and S7, A and B, and tables S2 and S3). Many other mutations generated β-arrestin-biased receptors (figs. S6B and S7B). To ensure that the RA analysis and by extension bias calculations were not disproportionately influenced by comparing the amplified Gα_q-PKC responses to that of β-arrestin recruitment, we also evaluated the RA of mutants from the more proximal Gα_q response using a specific Gα_q/Gβγ-BRET sensor (10). The PKC sensor recapitulates the responses of the Gα_q/Gβγ sensor when assessing the behavior of different AngII biased ligands (10). The RA values obtained between PKC- and Gα_q/Gβγ-BRET sensors for mutant receptors showed good correlation (fig. S8 and data file S4). Mutants also retained similar signaling and bias profiles when their coupling was assessed with either the PKC or Gα_q-BRET sensors as compared to β-arrestin recruitment (fig. S9, A to C), suggesting that the PKC sensor was indeed an appropriate proxy for Gα_q engagement by receptors and thus for determining bias responses between pathways.

We observed that mutations in TM4 that increased RA values for β-arrestin recruitment also produced bias toward the β-arrestin pathway (I151^4.48A, W153^4.50A, L154^4.51A, A159^4.56G, and L161^4.58A) (Figs. 1B and 2B, figs. S6, A and B, and S7, A and B, and table S3). Other mutations in TM4 and TM3 also elicited β-arrestin bias by reducing Gα_q coupling. Mutating A225^5.68 to glycine in TM5 generated a Gα_q-biased receptor by decreasing β-arrestin coupling and increasing Gα_q coupling. The V246^6.41A mutation in TM6 reduced both Gα_q and β-arrestin responses as compared to WT, but decreased coupling to β-arrestin to a greater extent than that to Gα_q, thus leading to a Gα_q-biased phenotype. The T141^4.38A and R140^34.57A mutations, which are located at the C-terminal end of ICL2, were among the mutants with the greatest bias toward β-arrestin. This region is highly dynamic and undergoes conformational changes in AngII-bound AT1R but not in β-arrestin-biased ligand-bound AT1R (19). The T141^4.38A mutation is also in the vicinity of several other residues that upon mutation to alanine favored β-arrestin bias because of reduced Gα_q coupling, thereby forming a so-called functionally selective structural domain (D125^3.49, R126^3.50, I130^3.54, P133^34.50, M134^34.51, R137^34.54, and R140^34.57) (Fig. 1B and 2G, figs. S6, A and B, and S7, A and B, and table S3).

Residues with functional selectivity for Gα_q and β-arrestin coupling encompass putative small molecule binding sites

Because we identified many residues that produced functional selective coupling of AT1R to Gα_q or β-arrestin when mutated, we next determined the extent to which they are in proximity of putative small molecule binding sites that could be harnessed for developing biased modulators of AngII. We used computational analysis methods to identify putative small molecule binding sites in AT1R on snapshots from MD simulations (Fig. 3, A and B). We found two smaller binding sites located close to the AngII binding site. A third larger binding site was distinct from the AngII binding site and was surrounded by many residues that yielded β-arrestin bias when mutated (Fig. 3B) and by the Gα_q biased mutant V246^6.41A at its base (Fig. 3A). This larger binding site was also surrounded by many residues on TM4 that resulted in bias toward β-arrestin when mutated. The V246^6.41A mutant showed Gα_q bias but also an altered β-arrestin response, suggesting that targeting this region with small molecules could have differential effects on Gα_q and β-arrestin coupling to AT1R. These analyses identified critical residues and regions in AT1R that could be targeted to develop small molecules modulating receptor coupling to Gα_q and β-arrestin.

Fig. 3. Identification of small molecule binding sites in functionally selective structural domains of AT1R.

(A and B) Residue positions that upon mutation show bias toward Gα_q in red (A) and β-arrestin in blue (B). Residue positions are indicated in the AT1R structure (PDB ID: 6OS0) by displaying the Cɑ atom of each residue. Small molecule binding sites that are distinct from that of AngII are shown in black and were obtained using FindBindSite on the 5 μs MD snapshots (N=5 × 1 μs). Inset in (B) shows that the largest allosteric binding site, which has been flipped horizontally, is located centrally in the receptor within a functionally selective structural domain containing many β-arrestin-biased residues.

Residue communities proximal and distal to the orthosteric site regulate AngII binding

The observation that potencies of AngII for the Gα_q response were reduced (>10-fold) for 18 mutants (Fig. 1A and table S1) suggested that they directly and/or allosterically affected AngII binding to AT1R. Therefore, we measured the effect of these 18 mutations on AngII binding using radioligand binding studies and subsequently used MD simulations to provide a molecular understanding of how these residues interact with AngII. All mutants expressed to a similar level as WT but showed loss of AngII binding (fig. S10, A and B, and table S1). Fourteen of these mutations were in the proximity of the orthosteric binding site (fig. S10, C to E). The remaining four mutations - M134^34.51A, T141^4.38A, F301^7.52A, and F309^8.50A - were further away from the ligand binding site, suggesting allosteric regulation of AngII binding to AT1R. Although the F301^7.52A and F309^8.50A mutations affected the potency, efficacy, and RA of AngII for Gα_q and β-arrestin coupling, the M134^34.51A and T141^4.38A mutations affected mostly Gα_q potency and efficacy (Fig. 1A and data file S1).

MD simulations of AngII-bound AT1R showed that W84^2.60, Y92^ECL1, R167^4.64, K199^5.42, P285^7.36, and I288^7.39, which affected ligand binding when mutated, also made sustained contacts with AngII, as indicated by the percentage of MD snapshots that show contacts (between 26% and 87% frequency) (figs. S10, C and D, and S11A). Although mutation of Y35^1.39 to alanine led to a notable change in AngII binding affinity, Y35^1.39 does not contact AngII in the active state crystal structure of AT1R (11). However, MD simulation revealed that Y35^1.39 contacted AngII, albeit with low frequency (figs. S10D and S11A). MD simulation trajectories identified an additional 23 residues in AT1R that made direct contact with AngII in the binding site with greater than 30% frequency interactions (fig. S11A). We analyzed the effect of alanine mutations on all 23 residues predicted to contact AngII. Mutation of 5 of 23 residues in AT1R that showed high contact frequencies with AngII (V108^3.32A, C180^45.50A, A181^45.51G, F182^45.52A and Y184^ECL2A) resulted in low expression (below 50%) (table S1). Mutation of six other residues (R23^N-term, L112^3.36, W253^6.48, H256^6.51, D281^7.32 and Y292^7.43) to alanine resulted in an EC₅₀ fold change between 2.5–10-fold and a concomitant lower E_max for Gα_q or β-arrestin coupling. Lastly, mutation of 12 residues to alanine had a negligible effect (0–2.5-fold EC₅₀ and ≥90% ≤105% Emax) on Gα_q and β-arrestin responses, suggesting that substitutions at those positions minimally impacted agonist binding (data file S2). These 12 residues were disregarded in our analysis, because although they were predicted to be part of the binding site, they did not have a major impact on EC₅₀ or E_max.

The residues P82^2.58 and W94^23.50 did not form direct contacts with AngII during MD simulations and yet resulted in decreased affinity for the ligand when mutated to alanine residues (fig. S10A). These two residues were in the second shell of the AngII binding site and they interacted with W84^2.60 (figs. S10, C and D, and S11B). Furthermore, P82^2.58 is part of the conserved S/TxP^2.58 motif and orients other residues in TM2 for ligand binding in other GPCRs (36, 37). C18^N-term, C101^3.25, and C274^7.25 are located outside of the AngII binding pocket and form disulfide bonds (fig. S10C), suggesting that they play a role in overall protein stability or local conformation. Mutation of G22^N-term and D278^7.29 to alanine impaired ligand affinity because these residues indirectly and directly interacted with C274^7.25 (figs. S10E and S11B). Moreover, W94^23.50 and C101^3.25 also interacted with each other. This suggests that G22^N-term, D278^7.29 and W94^23.50 contribute to an extra lock on the disulfide bridges in AT1R by preventing AngII from leaving its binding pocket (figs. 10E and S11B)(20).

Because alanine substitutions for M134^34.51, T141^4.38, F301^7.52 and F309^8.50 impaired AngII binding but because these residues were distantly located from the binding pocket (fig. S10, A and C), we performed additional MD simulations on these mutants. We found that these residue positions allosterically weakened the binding site residue contacts with AngII (fig. S12). The inter-residue distances between the N-terminus residues of AngII and residues in ECL2, as well as between the C-terminus residues of AngII and residues in TM7, increased in the mutants as compared to WT (fig. S12), leading to a weakening of AngII binding. Such allosteric regulation of agonist binding has been observed in other class A GPCRs (38, 39). We observed that the F301^7.52A and F309^8.50A mutations increased the inter-residue distances between the N-terminus of AngII and residues in ECL2 (Y184^ECL2, E185^ECL2 and S186^ECL2), and the F309^8.50A mutation increased the inter-residue distance between the C-terminus of AngII (P7 and F8) and residues Y292^7.43 in TM7. For the F309^8.50A mutant, the distance between Y4 and M284^7.35 also increased. The residue at position 34.51 in both the muscarinic acetylcholine receptor 1 (M1R) and histamine receptor 1 interacts with Gα_q/11 (PDB: 6OIJ and 7DFL) (40, 41), consistent with our findings. Residue 4.38 interacted with Gα_q in our AT1R-Gα_q model and with arrestin in three-dimensional structures of GPCR:arrestin complexes (PDB: 4ZWJ, 6TKO, 6U1N, 6UP7) (42–45). The residues involved in G protein (fig. S10C) or β-arrestin binding support their involvement in the allosteric regulation of AngII binding to AT1R.

Overall, these analyses provide an extended map of residues in the ligand binding site and elsewhere in AT1R that allosterically modulate AngII binding. Consequently, this information may be useful for future ligand design.

Identification of residues in the intracellular regions of AT1R that regulate G proteins and β-arrestin coupling

Next, we used the GPCR:Gα_q interface residues from a subset of available three-dimensional structures of Gα_q/11-coupled class A GPCRs to determine how corresponding residues located in the AT1R:Gα_q interface affected coupling to this G protein. A AT1R-Gα_q 3D structure was unavailable at the beginning of this project (46). Consequently, we selected other GPCR-bound Gα_q/11 structures, including the CHRM1 muscarinic acetylcholine receptor (PDB ID: 6OIJ) and the H₁ histamine receptor, (PDB ID: 7DFL) because these had the fewest mutations in their Gα_q/11 proteins compared to other GPCR-bound Gα_q/11 structures. The multiple sequence alignment highlights which residue positions are part of this putative Gα_q coupling interface and how the corresponding residues in other receptors overlap with those in AT1R (fig. S13A). From this analysis, 19 residues were inferred to be in the AT1R:Gα_q interface (fig. S14, A and B). Twelve residues out of 19 (63%) showed a notable change in RA responses to Gα_q when mutated in AT1R (Fig. 4A, Fig. 2G and data file S3; comparable results were also obtained when changes in EC₅₀ and E_max were used instead), suggesting that Gα_q binds to a similar interface in the AT1R as in these other Gα_q-coupled class A GPCRs. Mutations in other corresponding residue positions in GPCR:Gα_q complexes proposed to affect Gα_q coupling to AT1R did not change RA responses. However, because these other residues were in the vicinity of other ones that modulated RA responses when mutated, we generated a homology-based structural model of the AT1R:Gα_q complex and performed MD simulations (Fig. 4B and fig. S14D). Analysis of the MD simulation trajectories showed 17 residue positions making persistent contact with Gα_q (fig. S14D). Mutations in 12 out of 17 residue positions located in the AT1R:Gα_q interface resulted in changes in RA (comparable results were also observed when considering changes in EC₅₀) and 13 out of 17 residue positions affected E_max (data file S2). These findings show that the dynamics of the AT1R:Gα_q model robustly recapitulated the effect of interface residues on Gα_q coupling.

Fig. 4. Putative, modeled, and proposed residues affecting Gα_q and β-arrestin coupling interfaces.

(A) Putative Gα_q coupling interface with red surfaces showing the positions that affect Gα_q coupling. (B) Modeled Gα_q coupling interface with red surfaces showing the positions that affect Gα_q coupling. (C) Proposed β-arrestin coupling interface shown as blue surfaces using the experimentally measured E_max and RA as the surrogate parameters. White surfaces represent the putative or predicted residue positions that contact with Gα_q or β-arrestin (full list can be found in fig. S14, B to C). Residues are shown in the AngII-bound active state AT1R structure (PDB ID: 6OS0). Modeled Gα_q coupling interface was obtained using the 5 μs (N=5 × 1 μs) MD simulation trajectories of the AT1R-Gα_q model.

Because G proteins are promiscuous in their coupling to many GPCRs, including for AT1R (10, 11, 47, 48), we measured the extent to which key mutations in AT1R that negatively impacted its coupling to Gα_q also affected Gα_i activation. We selected 12 residues from our modeled Gα_q coupling interface that reduced Gα_q coupling when mutated (Fig. 4B) and tested their coupling to Gα_i using the Gα_i3(α/βγ)-BRET sensor (10) (fig. S15A and table S4). Mutation of seven out of twelve residues (R126^3.50A, I130^3.45A, R137^34.54A, R140^34.57A, Y215^5.58A, Y302^7.53A, and G306^8.47A) significantly reduced both Gα_q and Gα_i coupling as compared to WT (fig. S15B). The Y215^5.58A mutation reduced both Gα_q and Gα_i coupling to AT1R (49). Polar interactions between the Gα_i family protein Gα_o and R121^3.50 and N444^8.47 also occur in the muscarinic acetylcholine M₂ receptor-G protein structure (50). The remaining 5 mutants (P133^34.50A, M134^34.51A, L138^34.55A, T141^4.38A and I238^6.33A) affected only Gα_q coupling (fig. S15C). These findings show that although Gα_q and Gα_i share common contacts with AT1R, other residues within the receptor are selective for Gα_q interaction, consistent with G proteins selectivity and promiscuity observed in many GPCRs (47, 51–53).

We applied a similar approach to understand the AT1R:β-arrestin interface, which was modeled from structures of the β₁-adrenergic receptor (PDB ID: 6TKO), the muscarinic acetylcholine receptor M2 (PDB ID: 6U1N), the neurotensin receptor 1 (PDB ID: 6UPT) and the rhodopsin receptor (PDB ID: 4ZWJ) that bound arrestin (fig. S13B). Here, the picture differed. Analysis of these structures suggested 33 residues to be in the putative AT1R:β-arrestin coupling site (fig. S14, A and C). However, only one of these putative residues showed a significant change in EC₅₀ and 9 showed changes in RA and E_max when mutated, suggesting that β-arrestin couples differently to AT1R as compared to these other class A GPCRs (Fig. 4C and data file S3). Together, these results imply that AT1R couples to Gα_q in a manner analogous to other Gα_q-coupled class A GPCRs, although some intrinsic specificity interactions may exist in each GPCR. In contrast, β-arrestin is a far more promiscuous GPCR-interacting adaptor because it binds to different receptors distinctly.

Allosteric communication from the Gα_q coupling interface spread beyond the AngII binding site

We next took advantage of our AT1R:Gα_q model to map the network of residues in the receptor outside the AngII binding site or in the Gα_q coupling interface that affected Gα_q activation upon mutation. We posited that some of these residues could be involved in the allosteric communication from the Gα_q coupling interface to various structural domains of AT1R. We used the Allosteer computational method (22) on the MD simulation trajectories to identify AT1R residues involved in such allosteric communication. Forty-six residues which are spread across the receptor that are predicted to be involved in such allosteric communication emanating from the Gα_q coupling site in AT1R, also affected RAs for Gα_q coupling (Fig. 5). As predicted by Allosteer, the number of residues involved in the allosteric communication was quantified with the area under the receiver-operating characteristic curve (AUC) of 0.66, which is significant compared to randomly chosen residues. The AUC value quantifies the rate of recovery of the true positives over false positives (figs. S16 and S17). With the alanine scanning mutagenesis approach and MD simulations, our findings reveal that allosteric communication in AT1R extends to structural regions beyond the orthosteric AngII binding site. Such a finding is also consistent with the fact that many class A GPCRs have been crystallized with ligands bound in different structural regions and not just in the orthosteric extracellular regions (23), thus suggesting the possibility of allosteric binding sites in AT1R.

Fig. 5. Allosteric communication in AT1R.

The residues (Cɑ atoms in red) are predicted to be involved in allosteric communication starting from the Gα_q coupling interface residues and going to various structural regions of AT1R. The lines in red represent the allosteric communication pipelines. The Gα_q structure shown in red colored surface representation was transferred from the H₁ histamine receptor-Gα_q structure (PDB ID: 7DFL) to visualize the coupling site of the G protein with AT1R. Residues are shown in the AngII bound active state AT1R structure (PDB ID: 6OS0). Analysis was performed on the 5 μs (N=5 × 1 μs) MD simulation trajectories.

Mechanisms by which mutations lead to Gα_q and β-arrestin bias in AT1R.

We next explored some of the mechanisms by which receptor bias could be achieved. The Gα_q-or β-arrestin-biased mutants were grouped as: (i) coupling to Gα_q or β-arrestin2 was unaltered while the other pathway was decreased, (ii) coupling to both Gα_q and β-arrestin2 was decreased but one pathway was decreased more than the other, and (iii) coupling to either Gα_q or β-arrestin2 was increased and the other pathway was decreased or unaltered (fig. S7, A and B). However, bias for mutants in each category may not be achieved by the same mechanism.

For Gα_q-biased receptor mutants, five mutations were identified in the C-tail, which may have affected their phosphorylation states by GRKs and β-arrestin binding at the expense of Gα_q coupling or as in the case of S326 ^C-tailA, which alleviated the inhibitory restraint of phosphorylation on Gα_q coupling, thereby eliciting bias toward this pathway. K58^12.49A, V246^6.41A and A225^5.68G are in ICL1, TM6 and TM5 respectively. K58^12.49A was a Gα_q biased mutant with decreased β-arrestin coupling that retained its coupling to Gα_q. K58^12.49 is expected to have a direct interaction with β-arrestin (Fig. 4C and fig. S14C) because this residue interacts with β-arrestin in three dimensional structures of class A GPCRs complexed, thereby explaining its Gα_q bias when mutated. This is the only residue that overlaps with the putative residue positions affecting β-arrestin coupling. The V246^6.41A mutant was Gα_q biased such that both Gα_q and β-arrestin coupling were weakened but β-arrestin coupling was reduced to a greater extent than that of Gα_q. V/M/L^6.41 in Class A GPCRs like β₂-adrenergic, muscarinic M₂, adenosine A2A, μ-opioid, and rhodopsin receptors undergo contact rearrangements with residues in TM5 when going from inactive to active states for G protein coupling (27). This suggests that V246^6.41 is involved in AT1R structural rearrangements that lead to bias. Hubscore is the number of allosteric communication pathways going through a particular residue, and analysis of MD simulations of V246^6.41A with AngII bound showed increased allosteric communication (hubscore 9) to the Gα_q coupling interface compared to WT (hubscore 2) (figs. S18, A and B, and S19, A and B). Similarly, allosteric communication in A255^5.68G increased compared to WT, potentially explaining its bias toward Gα_q (fig. S19, C and D).

For β-arrestin biased mutants, we identified 48 mutations that were outside the resolved regions of the three-dimensional structure of AT1R (Fig. 1B and fig. S6A). The rest of the biased mutations were scattered throughout the receptor except for those in ICL1, ECL1, ICL3, and ECL3. T141^4.38A and M134^34.51A were β-arrestin biased mutants for which β-arrestin coupling was maintained as in WT with a decrease in Gα_q coupling (fig. S7B). These residues directly interact with Gα_q in our model (fig. S14D), which explains their impaired Gα_q coupling when mutated. In contrast to AT1R, the corresponding residue T148^4.38 in the β₁-adrenergic receptor contacts β-arrestin and not Gα_s (43), suggesting that engagement of Gα_q compared to that of Gα_s proteins differs from that of β-arrestin in these receptors. The N294^7.45A mutant demonstrated impaired Gα_q coupling and a minor change in β-arrestin coupling (data file S2). N294^7.45 is in the core of the TM domain distant from the Gα_q coupling interface. Analysis of the allosteric communication using MD simulations showed that the involvement of the residue in communicating with the Gα_q coupling interface was nullified by alanine substitution (hubscore 0) as compared to WT (hubscore 61) (fig. S19, E and F).

For the F301^7.52A mutant, in which Gα_q and β-arrestin coupling were decreased but that of Gα_q was more heavily impaired, this mutation, although distant to the AngII binding site, decreased the EC₅₀, E_max and the ligand binding affinity. MD simulations showed that the F301^7.52A mutation allosterically weakened the binding site residue contacts with AngII by altering ligand-receptor contacts (fig. S10, A and C, and fig. S12), suggesting that this mutation, although it impaired both Gα_q and β-arrestin coupling by affecting the AngII binding affinity, had a greater effect on Gα_q. The L154^4.51A and L161^4.58A mutations in TM4 resulted in little to no attenuation in Gα_q coupling with a small increase in β-arrestin coupling. DEER spectroscopy combined with MD studies (47) has assigned flexibility to TM4 in this region which could lead to an ensemble ICL2 conformations favoring β-arrestin2 coupling. Our MD simulations of the AT1R bound to AngII also showed differential flexibility in this region of TM4 that leads to an ensemble of ICL2 conformations (fig. S20, A and B). We therefore speculate that this flexibility led to preferred β-arrestin coupling. Overall, our computational analysis recapitulated mechanisms by which residue positions can create bias toward Gα_q or β-arrestin when mutated to alanine (or glycine).

Discussion

AT1R is an important drug target and a prototypical receptor for studying functional selectivity because of the availability of Gα_q- and β-arrestin-biased ligands and of three-dimensional structures of the active state of AT1R with AngII and β-arrestin-biased ligands (19, 20, 54, 55). Although the active state crystal structures do not show distinct ligand-specific conformational changes, DEER experiments with specifically labeled residue pairs, have revealed differences in receptor conformations in the intracellular regions when bound to biased ligands (19). However, how different structural domains across the entire AT1R, including structurally unresolved regions of the receptor, regulate Gα_q and β-arrestin coupling is unknown. Such knowledge is not only essential for understanding the mechanisms of functional selectivity at GPCRs, but also for identifying druggable structural regions that can be used for designing orthosteric and/or allosteric modulators that bias responses for therapeutic use.

We used a combination of global alanine scan of AT1R with functional assays and MD simulations (including the Allosteer method) to characterize the roles of different structural elements of the receptor regulating Gα_q and β-arrestin coupling (fig. S1). Our findings reveal that multiple residues across AT1R regulated Gα_q coupling whereas residues that affected β-arrestin coupling were found in fewer structural elements of the receptor. We also identified residues at the receptor:G protein interface that were commonly involved in binding of Gα_q and Gα_i to AT1R and those that were selective for Gα_q coupling, opening the possibility for generating Gα_q- or Gα_i-biased AT1R mutants. A key finding from our analyses is that multiple structural regions in AT1R, beyond that of the AngII binding site, regulate ligand-specific coupling to Gα_q or β-arrestin. We showed that allosteric communication in AT1R was regulated by an ensemble of allosteric communication pipelines that began at the Gα_q coupling site and spread beyond the orthosteric ligand binding site, showing that multiple residues had a diffuse effect on receptor activity. This finding also suggests the possibility that AT1R has other putative ligand binding sites and opens opportunities for identifying new AT1R allosteric modulators that confer functional selectivity towards Gα_q or β-arrestin (26). Indeed, we found structural domains in AT1R containing cryptic small molecule binding sites that could be used for designing allosteric and biased modulators of the endogenous AngII ligand. This is also supported by our findings that many residues in proximity to these cryptic binding sites also elicited biased responses in AT1R upon their mutation.

The Gα_q and β-arrestin biased responses of AT1R mutants resulted from differential effects and mechanisms imparted by the mutations on these pathways. Some residues that elicited bias toward β-arrestin when mutated did so by attenuating the engagement of AT1R with Gα_q more so than that with β-arrestin, hence underscoring the greater structural and allosteric dependencies specific to AT1R:Gα_q coupling than for AT1R:β-arrestin coupling. These differential dependencies also emerged in our analysis of AT1R:Gα_q and AT1R:β-arrestin interfaces, which showed that the structural arrangement of receptor:effector complexes better agreed with that of other class A GPCR:Gα_q than with that of GPCR:β-arrestin complexes. This analysis suggests that β-arrestin recognizes a more divergent ensemble of conformations among GPCRs and even within AT1R itself and provides a rationale as to why it may be more challenging to develop Gα_q-biased ligands for AT1R (10, 56) because altered conformational changes in the receptor are more permissive for β-arrestin than Gα_q recognition. Notwithstanding these observations, mutation of other residues in AT1R to alanine resulted in increased relative activity and bias towards β-arrestin without substantially affecting receptor coupling to Gα_q. This effect was observed with mutation of several residues in TM4, supporting the importance of this structural element in AT1R for β-arrestin binding. This finding is also consistent with the structure of the rhodopsin-arrestin complex, which shows important conformational differences in TM4 as compared to the Gα C-terminus-bound rhodopsin structure (42, 57).

Our findings raise questions about the role of the G protein in regulating β-arrestin biases because some GRKs that phosphorylate GPCRs to promote β-arrestin engagement require G protein activity and because β-arrestin desensitize G protein responses (4, 6). Many β-arrestin-biased mutant receptors reduced Gα_q (and Gα_i) or had no effect on G protein activity but still favored efficient β-arrestin recruitment. Moreover, for some Gα_q-biased mutants, although we observed a decrease in β-arrestin recruitment, Gα_q activity was unaffected. These observations suggest that the functional selectivity of specific receptor mutants results from intrinsic properties and mechanisms mediated by the substituted residues. For β-arrestin-biased mutants with substantially reduced Gα_q coupling, the presumed reduction in GRK2/3-mediated phosphorylation that requires G protein activation for translocation to the plasma membrane would have limited rather than favored β-arrestin recruitment to these mutant receptors. However, GRK5/6, which are associated with the plasma membrane adjacent to activated receptors, may have enabled the maintenance of effective receptor phosphorylation and β-arrestin recruitment to these mutants, as well as bias (13, 14). Consistent with this possibility, GRK5/6 regulates the activity of AT1R-biased ligands and impeding Gα_q activity promotes a switch in dependency from GRK2/3 to GRK5/6 in the regulation of β-arrestin recruitment to AT1R (13, 14, 58). However, we cannot exclude that those mutant receptors that showed reduced G protein activity may have sampled distinct conformations that enabled phosphorylation by GRK2/3, or that the reduced G protein activity and GRK-mediated phosphorylation may have partially contributed to limit the observed β-arrestin biases of some of these mutants. The reduced β-arrestin recruitment at some Gα_q-biased mutants may also have contributed to such functional selectivity.

AT1R alanine-scanning mutagenesis enabled the identification of residues in both the structurally resolved and unresolved regions of AT1R that allosterically regulated Gα_q and β-arrestin coupling and conferred bias to either Gα_q or β-arrestin when mutated. Many of these residues could not have been inferred from static structures and might have been overlooked in MD simulations alone. Indeed, MD simulations on AT1R have been used to identify isolated residues involved in G protein signaling and in β-arrestin coupling, effects that were confirmed through targeted mutagenesis (49). Although our results are consistent with these findings, we found that these residues did not confer the strongest bias when mutated to alanine. We identified multiple residues that elicited stronger bias towards Gα_q or β-arrestin coupling, underscoring the need to use multidisciplinary approaches as undertaken here to probe elements important for functional selectivity. The combination of techniques used here could be extended to other GPCRs.

Some other considerations are also worth highlighting. Substitutions with alanine at sites of previously identified naturally occurring mutations A163^4.60T, T282^7.33M, C289^7.40W (10, 23), the constitutively active mutation N111^3.35G (59), and the G protein compromised mutation D74^2.50N (60) yielded different signaling profiles, implying that the chemical properties of the substituted amino acids at these functionally relevant positions may play similarly important roles in determining the receptor’s activity and bias. Therefore, biased residue positions identified here may have differential effects on AT1R functional selectivity regulation to Gα_q or β-arrestin when mutated to other amino acids. Similarly, we may have overlooked residues involved in AngII, Gα_q, and β-arrestin binding to AT1R and allosteric regulation because of the lack of early functional effects of the alanine mutants. When assessing β-arrestin responses, we used a plasma membrane, proximity translocation assay that did not discriminate between different known structural arrangements of β-arrestin (tail and core arrangements) when binding to AT1R (61). Indeed, β-arrestins couples to the phosphorylated C-terminal tail of GPCR forming a tail arrangement, whereas the core arrangement involves β-arrestin’s engagement with both the C-terminal tail and TMs of the receptor. Because different β-arrestin conformations can carry out distinct functions (internalization and β-arrestin signaling compared to desensitization of G protein signaling), it is also possible that the mutations studied here differentially affected the susceptibility of the receptor to GRK-dependent phosphorylation (GRK2/3 compared to GRK5/6) and β-arrestin functions, thereby producing distinct biases at the level of these β-arrestin-mediated responses. Biased mutants might also show altered phenotypes when assessed in cells owing to cell type-specific differences in effector/adaptor/kinase complements (G proteins, GRKs, and/or β-arrestins). Lastly, we cannot exclude for some AT1R mutants that other cell signaling events may have contributed to regulating their bias phenotypes. During the review of our study, a 3D structure of AT1R-Gα_q was published (46), which aligned well with our AT1R-Gα_q homology model (Fig. 4B). However, our combined approaches identified additional important functional residues in AT1R involved in Gα_q coupling regulation.

In summary, the combination of systematic AT1R mutagenesis and MD simulations as well as the use of MD simulations with the Allosteer method helped us to identify important structural domains and specific residues in AT1R that bias coupling to either Gα_q or β-arrestin when mutated. We identified allosteric sites and specific residues that could be used to probe for allosteric modulators and to design AT1R-biased receptors. Further research may focus on these receptor domains and residues which could be targeted for deep mutational scanning using residues other than alanine to assess their chemical tolerance in generating biased signaling, as well as identifying other domains and residues in AT1R involved in functional selectivity among G proteins.

Materials and Methods

Reagents:

Dulbecco’s Modified Eagle Media (DMEM), fetal bovine serum (FBS), and other cell culture additives were purchased from Gibco, Life Technologies. Linear polyethyleneimine MW 25000 (PEI) was purchased from Polysciences. Human angiotensin II [Asp¹-Arg²-Val³-Tyr⁴-Ile⁵-His⁶-Pro⁷-Phe⁸], poly-L-ornithine, poly-L-lysine, ANTI-FLAG^® M2-Peroxidase (HRP) antibody, and SIGMAFAST OPD were purchased from Sigma-Aldrich. 16% paraformaldehyde (PFA) was purchased from ThermoFisher Scientific. BSA (Bovine Serum Albumin) was purchased from Fisher Bioreagents. Coelenterazine 400a was purchased from Nanolight Technology. Q5 high-fidelity DNA polymerase, Gibson Assembly Master mix, DpnI, and other PCR reagents were purchased from New England BioLabs. 96-well plate DNA clean-up and 96-well plasmid DNA miniprep kits were purchased from Bio Basic and Qiagen, respectively.

DNA constructs:

The signal peptide-FLAG-tagged human AT1R construct used to generate the alanine substitution AT1R library was described previously (62). PKC-c1b, Gα_q polycistronic (GFP10-Gγ₁, Gα_q(118)-RlucII, and Gβ₁ on a single transcript), Gα_i3-RlucII, GFP10-Gγ₂, Gβ₁, β-arr2-RlucII, and rGFP-CAAX were described previously (10, 25, 63).

Generation of the alanine substitution AT1R library:

A two-fragment PCR approach was used for the mutagenesis, as described previously (64). Briefly, site-directed mutagenesis primers (mutant-forward and reverse (mutant-F and mutant-R) were generated with 18 bp of Gibson homology for Gibson assembly recombination to introduce alanine mutations in the signal peptide-FLAG-tagged-AT1R construct. Mutations were made through stepdown PCR, in which two separate PCRs (primer pairs of a mutant-F primer with a forward primer from the ColE1 origin of replication sequences of pcDNA vector (Ori-F), and a mutant-R primer with the reverse Ori primers (Ori-R)) were run to split the vector in half. Each half of PCR samples were then purified, and Gibson assembled. Re-annealed vectors were transformed into E. coli and colonies were picked and amplified. Plasmid DNA was purified using QIA prep 96 plus miniprep kits (Qiagen) and sent for sequencing to verify successful mutagenesis (Genome Quebec CES). Primers are listed in data file S5.

Cell culture and transfections:

HEK293SL cells were cultured in DMEM supplemented with 10% FBS and 20 μg/ml gentamicin. Cells were authenticated (GeneCopoeia^™) and periodically tested for mycoplasma contaminations (ABM^™, Mycoplasma PCR Detection kit). Cells were grown in 5% CO₂ and 90% humidity. Cells were seeded at a density of 2.0×10⁴ cells per well in poly-ornithine-coated white (for BRET) or clear (for ELISA) 96-well plates and simultaneously transfected with receptor and BRET sensor constructs using PEI transfection reagent. Briefly, 1 μg of total DNA in 100 μl of PBS was mixed with 100 μl of PBS containing 3 μl of 1 mg/ml PEI. For Gα_q signaling, 60 ng of AT1R DNA with 60 ng of PKC-c1b sensor DNA or 150 ng of AT1R DNA and 250 ng of Gα_q polycistronic DNA were used. For the β-arrestin sensor, 150 ng of AT1R, 8 ng of βarr2-RlucII, and 40 ng of rGFP-CAAX DNA were used. For ELISA, 250 ng of AT1R DNA was used. For the Gα_i3 assay, we used 150 ng of AT1R DNA with 24 ng of Gα_i3-RlucII, 60 ng of GFP10Gγ₂, and 60 ng of Gβ₁. Empty vector DNA was used to make up 1 μg of the total DNA amount. After 20 min incubation, the DNA/PEI complexes were dispensed into cells in 96 well plates (15 μl/well). Assays were performed 48 h post-transfection.

Whole-cell ELISA:

To measure cell surface expression of receptors, HEK293SL cells were transfected with WT or mutant receptors and seeded on poly-L-lysine-coated, clear 96-well plates (Fisher Scientific). 48 hours later, cells were washed with PBS and fixed with 4% PFA. Cells were blocked with 2% BSA in PBS and incubated with anti-FLAG M2-HRP antibody (1:5000). The cells were washed four times with PBS and 100 μl of SIGMAFAST OPD solution was added into each well. After 10 minutes, 25 μl of 3 M HCl was added to stop the reaction. The plate was read at an absorbance of 492 nm using a Synergy 2 microplate reader (Bio-Tek). To obtain a specific signal, non-specific signal from empty vector-transfected cells was subtracted.

BRET measurements:

To measure Gα_q, Gα_i3 or β-arrestin activation, HEK293SL cells were co-transfected with WT or mutant receptors and with either PKC-c1b, Gα_q polycistronic, Gα_i3RlucII/GFP10Gγ₂/Gβ₁, or βarr2-RlucII and the plasma-membraned-anchored rGFP-CAAX BRET sensors. Cells were seeded and transfected in poly-ornithine-coated 96-well white plates. 48 hours later, cells were incubated with Tyrode’s buffer (140 mM NaCl, 2.7 mM KCl, 1 mM CaCl₂, 12 mM NaHCO₃, 5.6 mM D-glucose, 0.5 mM MgCl₂, 0.37 mM NaH₂PO₄, 25 mM HEPES, pH 7.4) at room temperature for 1 h. Cells were subsequently stimulated at room temperature with various concentrations of AngII for 5 min to induce maximal receptor activation. The Rluc substrate coelenterazine 400a (2.5 μM; NanoLight Technology) was added 6 min before BRET measurement. BRET measurements were performed using the Synergy 2 microplate reader with donor filter 410/80 nm and acceptor filter 530/30 nm. BRET ratios were calculated by dividing the intensity of the signal emitted by the acceptor over the signal emitted by the donor and plotted as 12-point concentration-response curves.

Radioligand binding:

HEK293SL cells were maintained in DMEM supplemented with 10% FBS, 100 IU/ml penicillin, and 100 μg/ml streptomycin at 37°C in a humidified 5% CO₂ atmosphere. Cells were transiently transfected with recombinant receptors using PEI at a 3:1 ratio (w/w) with plasmid DNA. HEK293SL cells in 100 mm dishes transiently expressing each receptor (4 μg) were washed once with PBS, subjected to one freeze-thaw cycle, and gently scraped into washing buffer (25 mM Tris-HCl, pH 7.4, 100 mM NaCl, 5 mM MgCl₂). Cells were centrifuged for 15 min 2500 x g at 4°C and resuspended in binding buffer (25 mM Tris-HCl, pH 7.4, 100 mM NaCl, 5 mM MgCl₂, 0.1% bovine serum albumin, 0.01% bacitracin). Dose displacement experiments were done by incubating broken cells (20–40 μg of protein) for 1 h at room temperature with 0.8 nM [¹²⁵I]-AngII as tracer and increasing concentrations of AngII. After a 1 h incubation period, the bound radioactivity was separated from free ligand by filtration with GF/C filters presoaked for at least 1 hour in binding buffer at 4°C. Receptor-bound radioactivity was evaluated by γ counting.

AT1R-agonist MD simulations:

MD simulations were performed using the GROMACS package (version 2016/2019) with the CHARMM36 forcefield for proteins, POPC lipids, ions, and using CHARMM TIP3P water as solvent (65, 66). We performed MD simulations starting from the active state crystal structure of the AngII-bound AT1R (PDB id: 6OS0). The cytochrome b562 BRIL fusion protein was removed and the intracellular loop 3 (10 residues) was generated using Prime (Schrödinger). Missing sidechains and hydrogen atoms were added, protein chain termini were capped with neutral acetyl and methyl amide groups, and histidine-protonated states were assigned using Maestro (Schrödinger). The simulation box was created using CHARMM-GUI. We used the OPM (orientation of proteins in membranes) structure of PDB: 6OS0 for alignment of the transmembrane helices of protein structure and inserted a pre-equilibrated POPC (palmitoyl-oleoyl-phosphatidylcholine) bilayer. Final system dimensions were 75 × 75 × 118 Å, including 131 lipids, 13037 waters and 150 mM NaCl. After minimizing the AT1R-ligand complex, we equilibrated it using an NVT ensemble (1 ns long) and consequently with an NPT ensemble where we gradually reduced the position restraints from 5 to 0 kcal/mol/Ų (each step was 5 ns long). In the last step of equilibration, we performed 5 ns of unrestrained NPT simulations before running a total of 5 production MD simulations, each 1000 ns long. The snapshots were stored every 20 ps and the entire 1000 ns*5 runs amounting to 5 μs of simulation time was used for analysis. Mutant receptors (M134A, T141A, I151A, L154A, A159G, L161A, A225G, V246A, N294A, F301A and F309A) were generated based on the WT AT1R MD simulation. For the AT1R bound Gα_q model, a superimposition of H1-Gα_q (PDB ID: 7DFL) with AT1R-AngII (PDB ID: 6OS0) was performed in PyMOL by aligning the H1R and AT1R, after which the nanobody and H1R were removed. Mutations in the G protein were reverted to human WT Gα_q Uniprot ID: P50148), after which the complex was minimized using Maestro (Schrödinger). MD conditions for the mutants were similar to the WT setup except for the AT1R bound Gα_q model, in which the size of the box was 120 × 120 × 165 Å, 388 lipids and 52155 waters were used, and the unrestrained NPT was 50 ns. For the mutants, we performed 3 production MD simulations, each 400 ns long (total simulation time used for analysis was 1.2 μs), whereas for the AT1R-Gα_q, this was 5 runs each 1000 ns. Trajectories were visualized with VMD (67) and PyMOL (Molecular Graphics System, Version 2.0 Schrödinger) and analyzed using the GROMACS package. The convergence of the calculations was confirmed by calculating the change in root mean square deviation (RMSD) of the coordinates of the backbone atoms of the residues in the transmembrane region (residues 25–57 (TM1), 62–90 (TM2), 98–131 (TM3), 142–166 (TM4), 193–228 (TM5), 238–267 (TM6), 274–305 (TM7)) and ligand (fig. S19).

Calculating allosteric communication pipelines:

The Allosteer method uses the MD trajectories to calculate the allosteric communication pathways and to identify the network of residues involved in allosteric communication (67–70). Briefly, Allosteer involves calculating the correlation (mutual information; MI) in torsion angle distribution for every pair of residues in a protein. For each pair of distant residues (> 10 Å) with high mutual information (in the top 15% of MI values), we used the shortest-path algorithm by Dijkstra (71) as implemented Bioinformatics ToolBox in MATLAB (The MathWorks, Natick, MA) to calculate the shortest pathway with maximum mutual information. The nodes in the pathways are residues that have high MI but need not have any type of direct interaction between them because they are not necessarily neighbors. The pathways were clustered into pipelines based on their mutual proximity in the receptor structure. Each residue in the protein had several allosteric pathways passing through the residue which is the hubscore for that residue. The residues that mode up the allosteric communication pipelines modulated the strength of coupling of the G protein by the GPCR. We calculated the allosteric hub score for every residue in AT1R for allosteric communication pathways starting from the G-protein coupling site to various structural regions within the receptor (67, 70). The receiver operating characteristic (ROC) curve (to predict the robustness of our allosteric communication analysis) was calculated using the true positives (experimentally affected and computationally predicted), false positives (experimentally not affected and computationally predicted), true negatives (experimentally not affected and computationally not predicted) and false negatives (experimentally affected and computationally not predicted) (fig. S20). Among the false positives, we eliminated the residues that showed high Shannon entropy or the ones that belonged to the top quartile of Shannon entropy (<75%) (fig. S20). The rationale for this was that residue positions that show high entropy are more tolerant to mutations and may show changes in Gα_q or β-arrestin coupling upon mutations to residues other than alanine. The Shannon entropy was calculated using (72) the full sequence alignment of the Angiotensin receptor family. This included a total of 18 sequences (AT1R and AT2R as well as the distinct species). The alignment was performed using the full sequence alignment function of GPCRdb (73).

Residue group definition and calculation of contact heatmap:

We obtained the lists of β-arrestin and Gα_q protein interface residues from the 3D structures for rhodopsin-β-arrestin1 (PDB ID: 4ZWJ), β₁-adrenergic receptor-β-arrestin1 (PDB ID: 6TKO), M₂-β-arrestin1 (PDB ID: 6U1N) and NTS1-β-arrestin1 (PDB ID: 6UPT) for the β-arrestin interface residues, and M₁ muscarinic receptor-Gα₁₁ (PDB ID: 6OIJ) and H₁ histamine receptor-Gα_q (PDB ID: 7DFL) for the Gα_q protein interface residues. We identified the residues in the receptor that are within 7Å of the Gα_q protein or β-arrestin2 using GPCRdb and Get_Contacts (https://getcontacts.github.io/), respectively (fig. S14, A to C). Residues belonging to the extracellular loops and the top turn of the transmembrane helix in the extracellular region were considered in the calculation of allosteric communication pipelines. Also, the residues within 7Å of the agonist binding site and contacting the agonists in more than 40% of the snapshots from MD simulations were chosen as the ligand binding site residues (using Get_Contacts). Contact frequencies between the biased residues and their interactions were set at more than 20%. The contact frequency cut-off between the receptor and Gα_q protein was set to 34%. Frequencies of ligand-receptor and G protein-receptor interactions were displayed as a heatmap using GraphPad Prism 9.

Method to identify druggable binding sites using the code FindBindSite:

FindBindSite (FBS) (74) was used to identify putative druggable small molecule binding pockets where potential allosteric modulators may bind. FBS is a workflow that involves docking a small (60,000) and diverse library of small molecules to the entire protein structure. We used Glide to dock the 60,000 molecules library to AT1R. After docking this diverse library of 60,000 molecules, we clustered the regions with the highest docked ligand density and favorable dock score and rank the sites. FBS then outputs the rank of each site as well as the centers of the clusters that show ligand occupancy above the cut-off ligand density. FBS showed a hit rate of 71% with high confidence and 93% with lower confidence for the 41 proteins used for predicting druggable binding sites on the protein–protein interface.

Relative activity (RA) and bias calculation:

To quantify the activity of mutant AT1R receptors relative to the WT receptor, we used a “relative activity (RA)” scale (75). This scale uses a ratio of the maximal response to the EC₅₀ value for an agonist. Data were normalized to the E_max of AngII for WT receptor in a specific signaling pathway, and EC₅₀ and E_max were estimated from the nonlinear regression curve fitting equations in GraphPad Prism. RA of the mutant receptors compared to the reference receptor (WT) for a specific pathway was determined from the difference between their log(E_max/EC₅₀) values using Eqs. 1 and 2:

$$\mathit{\text{Relative}}\mathit{\text{Activity}}\left(RA\right)=\frac{E\mathit{\text{max}}{\mathit{\text{EC}}}_{50}\mathrm{\text{'}}}{{EC}_{50}Ema{x}^{\mathrm{\text{'}}}}=\left(\frac{\mathit{\text{Emax}}}{{\mathit{\text{EC}}}_{50}}\right)\left(\frac{Ema{x}^{\mathrm{\text{'}}}}{{\mathit{\text{EC}}}_{50}\mathrm{\text{'}}}\right)$$ (1)

$$\log \left(\mathit{\text{RA}}\right)=\Delta \log \left(\frac{\mathit{\text{Emax}}}{{\mathit{\text{EC}}}_{50}}\right)=\mathit{\text{log}}{\left(\frac{\mathit{\text{Emax}}}{{\mathit{\text{EC}}}_{50}}\right)}_{\mathit{\text{mutant}}}-log{\left(\frac{E\mathit{\text{max}}}{{\mathit{\text{EC}}}_{50}}\right)}_{\mathit{\text{WT}}}$$ (2)

Herein we defined “Relative activity (RA)” as Δlog(E_max/EC₅₀). To quantify bias between Gα_q and β-arrestin pathways, the difference in RA for the Gα_q and β-arrestin pathways (bias, ΔΔLog(E_max/EC₅₀)), was calculated by subtracting RA values of β-arrestin pathway from that of the Gα_q pathway of same mutants (eq 3).

$$\mathit{\text{bias}}=\Delta \Delta \mathit{\text{log}}\left(\frac{\mathit{\text{Emax}}}{{\mathit{\text{EC}}}_{50}}\right)=\Delta \mathit{\text{log}}{\left(\frac{\mathit{\text{Emax}}}{{\mathit{\text{EC}}}_{50}}\right)}_{\mathit{\text{Gq}}\mathit{\text{pathway}}}-\Delta log{\left(\frac{\mathit{\text{Emax}}}{{\mathit{\text{EC}}}_{50}}\right)}_{-\mathit{\text{arr}}\mathit{\text{pathway}}}$$ (3)

Data analysis and statistics:

Log of EC₅₀ and E_max where derived from BRET concentration-response curves using GraphPad Prism 9 software (GraphPad Software Inc.). Curves represent the best fits. The data from a typical BRET experiment followed a normal probability distribution and 3 experiments were considered sufficient to accurately estimate the standard deviation. Student’s t-tests were therefore used to compare differences in the LogEC₅₀, E_max and Log(E_max/EC₅₀) of each mutant to WT in the same pathway, and relative activities (RA) of each mutant between pathways (Fig. 1B, Fig. 2, figs. S6, S7, S15B and C, tables S2, S3 and S4, and data files S1, S2 and S3). P values < 0.05 were considered significant.

Data and Materials Availability:

MD data have been deposited in GPCRmd (https://submission.gpcrmd.org/home/; Submission ID 1479, Dynamics ID 1284). The Allosteer codes is available upon request here: https://www.cityofhope.org/research/beckman-research-institute/computational-and-quantitative-medicine/vaidehi-lab/vaidehi-lab-software. The FBS code can be obtained by contacting the corresponding authors. All other data needed to evaluate the conclusions are in the manuscript or the Supplementary Materials. BRET biosensors for academic research are available from S.A.L. under a material transfer agreement (MTA) with McGill University and Université de Montréal, Canada.