Biased agonists differentially modulate the receptor conformation ensembles in Angiotensin II Type 1 Receptor

Abstract

The structural features that contribute to the efficacy of biased agonists targeting G protein-coupled receptors (GPCRs) towards G proteins or β-arrestin (β-arr) signaling pathways is nebulous, although such knowledge is critical in designing biased ligands. The dynamics of the agonist-GPCR complex is one of the critical factors in determining agonist bias. Angiotensin II type I receptor (AT1R) is an ideal model system to study the molecular basis of bias since it has multiple β-arr2 and Gq protein biased agonists as well as experimentally solved three dimensional structures. Using Molecular Dynamics (MD) simulations for the Angiotensin II type I receptor (AT1R) bound to ten different agonists, we infer that the agonist bound receptor samples conformations with different relative weights, from both the inactive and active state ensembles of the receptor. This concept is perhaps extensible to other class A GPCRs. Such a weighted mixed ensemble recapitulates the inter-residue distance distributions measured for different agonists bound AT1R using DEER experiments. The ratio of the calculated relative strength of the allosteric communication to β-arr2 vs Gq coupling sites scale similarly to the experimentally measured bias factors. Analysis of the inter-residue distance distributions of the activation microswitches involved in class A GPCR activation suggests that β-arr2 biased agonists turn on different combination of microswitches with different relative strengths of activation. We put forth a model that activation microswitches behave like rheostats that tune the relative efficacy of the biased agonists toward the two signaling pathways. Finally, based on our data we propose that the agonist specific residue contacts in the binding site elicit a combinatorial response in the microswitches that in turn differentially modulate the receptor conformation ensembles resulting in differences in coupling to Gq and β-arrestin.

Introduction

Angiotensin II (AngII) type 1 receptor (AT1R) belongs to class A G protein-coupled receptor (GPCR) superfamily of membrane proteins that is activated by the octapeptide, Angiotensin II (AngII) hormone. AngII is part of the renin-angiotensin-aldosterone system responsible for controlling blood pressure and water retention via smooth muscle contraction and ion transport, and therefore forms an important drug target (1). Binding of AngII to AT1R leads to activation of heterotrimeric Gαq (Gq) family of trimeric G proteins, with subsequent engagement of β-arrestins (β-arr 1 and 2). The β-arr mediated signaling pathway leads to pharmacologically desirable cardio protection, while over-activating the Gq coupled pathways leads to undesired cardiovascular effects (2). Therefore, agonists targeting AT1R that preferentially activate the β-arr signaling pathway over the Gq pathway commonly known as “biased agonists” are sought after for better therapeutic outcomes (3–9). There are multiple cell and tissue specific factors, (7– 9) and structural factors (6–10) that affect the agonist bias in AT1R towards either Gq protein or β-arr coupled signaling pathways. In this study, we focus on the contributions from structural factors to the ligand bias. Although β-arr2 biased agonists have perceived therapeutic advantage, designing these biased agonists remains a huge challenge. This is due to paucity in our understanding of the structural and dynamical features of AT1R conformation ensemble that contribute to agonist efficacy towards differential coupling strengths to Gq or β-arr. However, great strides have been made by recent studies on the structure and dynamics of AT1R with different peptide agonists (10–13).

Crystal structures of nanobody bound active state of AT1R with different biased agonists bound provide a critical basis for delineating the differences in the agonist binding site for βarr2 biased vs Gq biased AT1R agonists (12). A recent study (10) of AT1R using Double Electron-Electron Resonance (DEER) showed that β-arr2 biased agonists stabilize distinct receptor conformation ensembles compared to Gq protein biased agonists. This study exemplified the critical role of the conformation ensemble generated by the agonist-AT1R complex dynamics in determining the basis for differential agonist efficacy towards Gq versus β-arr2 signaling pathways. Single molecule force spectroscopy studies have shown differences in the off rates during the unbinding process of various biased agonists as well as in the number of conformational substates of AT1R when bound to Gq biased agonists compared to β-arr2 biased agonists (11). Recent all-atom Molecular Dynamics (MD) simulation studies on AT1R bound to different biased agonists proposed that AT1R adopts two distinct conformations, one that is speculated to favor β-arr2 coupling and other Gq coupling (13). MD simulation data shows that the ligand binding to the extracellular site impacts the conformational changes in the intracellular part of the receptor. However, the differences in the ligand-receptor contacts for multiple agonists that allosterically lead to intracellular conformational changes are not enumerated. Additionally, the effect of biased agonists on the class A GPCR activation microswitches compared to balanced agonists in AT1R has not been studied. Here, we used MD simulations for ten biased and unbiased peptide agonists for AT1R, in explicit lipid bilayer, to provide molecular signatures that distinguish β-arr2 biased agonists from Gq biased agonists. At the beginning of our study, we only had the crystal structure of the nanobody and the peptide agonist S1I8 bound AT1R fully active state available. We performed MD simulations for all the ten agonists bound to AT1R extracted from this conformation. Our dynamics ensemble recapitulates the residue distances recently measured using the DEER experiments. The β-arr2 biased agonists bound AT1R show a slightly narrower cavity in the intracellular region akin to the alternate conformation identified by Suomivuori et al (13) for the β-arr2 biased TRV026 bound AT1R, compared to Gq biased agonists bound receptor. We have further calculated the relative strengths of the allosteric communication from the agonist binding site to the putative Gq and β-arr2 binding sites in AT1R. This computed relative strength called the “molecular bias factor” (14) correlates well with experimentally measured ligand bias factors. Finally, inter-residue distance distributions of the activation microswitches involved in class A GPCR activation for ten different agonists, suggests that each β-arr2 biased agonist activates a unique combination of activation microswitches. Based on these results, we put forth a hypothesis that the activation microswitches behave like rheostats that tune the efficacy and potency of the biased agonists toward the Gq and β-arr2 signaling pathways.

Results

Using all-atom MD simulations, we report on the dynamic characteristics of AT1R bound to β-arr2 biased agonists namely, TRV023, TRV026, TRV027, TRV034, TRV044, TRV045 and SII, Gq protein biased agonists TRV055 and TRV056, and endogenous agonist AngII. The amino acid sequences of the peptide agonists used in this study are shown in Table S1 of the Supplementary Information. MD simulations performed in lipid bilayer were initiated from the crystal structure of the nanobody and S1I8 peptide bound fully active state of AT1R (PDB ID: 6DO1) (12). This was the only crystal structure of the active state of AT1R available at the beginning of our study. The four crystal structures of the active state of AT1R published subsequently with different agonists (12) are extremely similar with very little differences. However, these structures could not explain the differences observed in the DEER experiments (10). In this study we have shown that the ensemble of conformations combining the MD simulation trajectories starting from the inactive state structure with AngII bound, and from active state structure together recapitulate the conformational heterogeneity observed in the DEER experiments as shown in sections below. Additionally, starting our MD simulations from the S1I8 bound conformation of AT1R with β-arrestin biased agonist TRV026 bound we have shown convergence with the previous simulation study by Suomivuori et al (13) as described below. The ten peptide agonists used for this study were built into the structure using S1I8 peptide conformation as the template (details in Methods section). To understand the effect of agonist binding on the inactive state of AT1R, we also performed MD simulations with agonists AngII, TRV026 (β-arr2 biased) and TRV055 (Gq protein biased) bound to the inactive state receptor conformation taken from crystal structure of the antagonist bound inactive state (PDB:4YAY) (15).

Conformational ensembles from MD simulations of AT1R with biased agonists recapitulate DEER distance distributions and reveal structural variations in the presence of endogenous and biased agonists. Using DEER spectroscopy, Wingler et al (10) recently showed that binding of various biased agonists (analogs of AngII) differentially affect the AT1R conformations in the intracellular region. The ten pairs of spin labeled residues for which they measured the inter-residue distance distributions using DEER are in the intracellular regions of transmembrane (TM) helix TM1, TM5, TM6, TM7, intracellular loop 2 (ICL2) and helix 8 (h8) (Fig. 1B). We pooled ensembles of snapshots generated from both the inactive (R) state and the active (R*) state MD simulations of AT1R bound to AngII, TRV026 and TRV055. We added spin labels to the same modified residues that were labeled in the DEER experiments for each snapshot from the MD simulations and measured the distances between the same labeled residues that were measured using DEER in the study (See Methods).

Figure 1.

(A) Comparison between experimental (grey) and computational (green) inter-residue DEER distributions. The computational inter-residue distances were calculated from MD simulation snapshots after placing the spin labels in the residues shown in Figure B. (B) Locations of the six DEER spin label probes used on AngII Type 1 Receptor (AT1R) are indicated by colored spheres at the Cα atom of the corresponding residue, and distances measured are indicated by dashed lines. (C) The distribution of Jensen-Shannon Distance (JSD) values (Fig. S1) obtained for all ten distance comparisons for each of the three ligand-AT1R complexes, which total 30 distance distribution comparisons (Fig. S1B and Table S2), represented using the box and whisker plot. The lower the JSD value (closer to 0), the greater the similarity between the DEER and MD distance distributions.

Quantitative comparison of the inter-residue distances measured by DEER to the MD simulation ensembles:

Comparison of the distance distributions from published DEER measurements to those from our MD simulations is shown in Fig. 1A and Fig. S1B. The similarity between the distance distributions have been quantitatively evaluated using the Jensen-Shannon Distance (JSD) values (see Methods, Fig. S1A) shown in Fig. 1C and Table S2. The JSD values range from 0 to 1, with 0 indicating identical distributions and 1 the most diverse distributions (Fig. S1A).

We calculated the similarities in the distance distribution of labeled residues measured by DEER experiments shown in Fig. 1A, between AngII-AT1R complex and TRV026-AT1R complex (shown in Fig. S1C). The two inter-residue distances that show the maximum change in the DEER measurements between TRV026-AT1R and AngII-AT1R are TM1-TM6 and ICL2-TM5 (shown by green bars in Fig. S1C). DEER experiments showed that the TM1-TM6 distance distribution has a single peak compared to the two peaks observed for AngII-AT1R complex. There are two peaks in Gq biased TRV055-AT1R although the distant second peak is of greater population in AngII compared to Gq biased TRV055. The distance distributions calculated from our MD simulations also show a single peak for TRV026 close to the distance distribution from DEER experiments and two peaks for TRV055 and AngII.

To establish the statistical significance of JSD values, we scrambled the agonists and calculated the JSD by comparing all the distance distributions from DEER experiments for TRV026-AT1R system with the distributions from MD simulations of AngII-AT1R system. This scrambling of agonists clearly showed higher JSD values for those distributions from DEER experiments that show distinct differences between TRV026 and AngII (grey bars in Fig. S1C). Thus, JSD is a sensitive measure to compare the distributions from DEER experiments with those from MD simulations.

To assess how much of the similarities in distributions from the DEER experiments are recapitulated by the crystal structures, we added multiple rotamers of the spin labeled residues in TRV026 bound AT1R system and calculated the distance distributions. The JSD values between these distance distributions derived from the single crystal structure and the DEER experiments is shown in Fig. S1C (yellow bars). 9 out 10 distance distributions calculated using MD simulation trajectories are similar to the DEER experiment distributions or better than the distributions obtained using a single crystal structure.

Structural Insights on the conformational changes observed in DEER using MD simulation ensembles:

Wingler et al (10) concluded that for TRV026 bound AT1R system, the conformational changes possibly occur in TM7 and h8 that lead to changes in DEER distances. Fig. S1D shows the representative structure extracted from the most populated conformation cluster of TRV026-AT1R system in comparison to the crystal structure of TRV026 bound AT1R. It is seen that TM7 and h8 move outwards from the TM core and TM6 moves inwards accompanied by an outward shift of TM5 during the MD simulations in comparison to the crystal structure of TRV026-AT1R complex. Overall, the inter-residue distance distributions are better recapitulated for TRV055 bound AT1R followed by TRV026 and then for AngII bound AT1R. As seen in Fig. 1C, the average JSD across all ten comparisons per ligand is <0.4, with only 4 distance distributions (accounting for 10% of the distributions) having JSD ≥ 0.4. These 4 distribution distances that have JSD >0.4 are ICL2-TM6 for AngII (0.50), TM1-ICL2 for AngII (0.45), ICL2-TM5 for TRV026 (0.61) and ICL2-TM7 for TRV055 (0.46) (Table S2). MD simulations do not recapitulate these distance distributions involving ICL2. The ICL2 adopts a helical conformation upon activation in many GPCRs, but this change in conformation also depends on the nature of interaction with the binding partner proteins (G proteins or β-arr) (16). While the conformation of ICL2 in the active-intermediate state of AT1R is unknown, the ICL2 stays helical during MD simulations starting from the R* state and remains in a loop conformation with few residues sporadically transitioning into helical regions in the MD simulations starting from the agonist bound inactive R state (Fig. S2) for all the three ligands. Interestingly, the ICL2 was found to adopt both a helical and a loop conformation at different points during the long-timescale MD simulations (13). Since this region is perhaps intrinsically disordered it is challenging to capture the variations in ICL2 that affects the DEER distance distributions using all-atom MD simulations. Taken together, these results show that only a conformational ensemble consisting of a weighted combination of MD snapshots from both inactive state dynamics and active state dynamics satisfy the inter-residue distance distributions measured by DEER. The percentage of the MD snapshots from the inactive and active state ensemble are weighted differently for each agonist as shown in Fig. S3. This finding suggests that the conformation ensemble of the agonist bound receptor still comprises of the inactive state conformations which is a previously unknown finding for a peptide bound class A GPCR, AT1R (13). Previous NMR study on β2-adrenergic receptor has shown that the agonist binding activates the receptor although the inactive state is also sampled in the ensemble (17).

Spatiotemporal heatmap of ligand-AT1R contacts show differences in the binding site features of β-arr2 biased agonists versus Gq biased agonists

The heat maps in Fig. 2A and in Fig. S4A show the complete list of agonist-AT1R residue contacts observed during the MD simulations along with the calculated persistence during the MD simulations. The persistence of a ligand-protein residue contact is the percentage of MD snapshots that contain the contact. The ten agonist peptides studied here are all derivatives of the octapeptide AngII and their amino acid positions in the peptides numbered 0 to 8 (Fig. 2B). There are two types of contacts: (i) the agonist-AT1R residue contacts that are exclusively observed in the dynamics of β-arr2 biased agonists or Gq biased agonists only (Fig. 2A) and (ii) the agonist-AT1R residues contacts that are observed in the dynamics of all the agonist-AT1R complexes studied here (Fig. S4A) We call such agonist-AT1R contacts as common contacts. Both common and exclusive contacts could contribute to agonist induced coupling selectivity and bias. The common agonist-receptor contacts show differences in their temporal persistence and could thereby contribute to the functional selectivity of the agonist. As shown in Fig. S4C, the AT1R residues involved in common contacts across agonists are in ECL2, TM7, N-terminus and TM6. However, nearly half of all the exclusive contacts observed in β-arr2 biased agonist-AT1R dynamics come from the N-terminus, followed by TM7 and a smaller fraction in TM6 and TM5 (Fig. S4C). The agonist-receptor contacts observed exclusively in the Gq biased agonists are predominantly located in TM6 and ECL3, followed by an equal distribution amongst TM3, TM4, ECL2, TM5 and TM7 (Fig. S4C).

Figure 2.

(A) Heatmap showing persistence frequencies of ligand-AT1R contacts for the agonists TRV026, TRV044, TRV023, TRV034, TRV045, TRV027, AngII, SII, TRV055 and TRV056. The residue contacts are indicated by the peptide ligand sequence position (as shown in (B)) with different receptor residues. Residue contacts persistent throughout the MD simulations (persistence frequency = 100%) are represented by black cells and those contacts with persistence frequency = 20% are represented by light grey cells, with intermediate values indicated by darker shades of grey. Contacts with persistence frequencies <20% are shown as white cells. (B) Sequence alignment of the ten peptide agonists shown with corresponding sequence positions. (C) AT1R-TRV026 residue contacts seen only in the AT1R-TRV026 MD simulations. These residue contacts show increased persistence with TRV026 compared to AngII in AT1R. The difference in the value of the persistence is shown in parenthesis. Residue contacts unique to TRV026 and not present in AngII are shown in brown text. Polar contacts are indicated using red dashed lines and van der Waals contacts are shown in blue shaded triangles. (D) AT1R-TRV055 residue contacts seen only in the AT1R-TRV055 MD simulations. These residue contacts show increased persistence with TRV055 compared to AngII in AT1R. Residue contacts unique to TRV055 and not present in AngII are shown in brown text. (E) and (F) Extracellular view of the most representative conformations extracted from the MD simulations of the active state of AT1R with AngII (grey), TRV026 (blue) and TRV055 (red and salmon) bound AT1R superimposed. Representative conformations were obtained from the most populated conformation clusters of the MD snapshots clustered based on Cα atom coordinates of the TM helices. Agonists are shown in transparent ball and stick representation. AngII- and TRV026-bound AT1R showed one major conformation while TRV055 bound AT1R showed two distinct conformations as illustrated in red and salmon color in figures E and F respectively.

Fig. 2C shows the AT1R-TRV026 residue contacts that are selective to TRV026 in comparison to AngII. The numbers shown in parenthesis are the difference in the persistence frequencies of these contacts between TRV026 and AngII. All the residues shown in Fig. 2C, selective to TRV026 are present in the crystal structure within the agonist binding site, except residues C18^N-ter, N25^N-ter, T88^2.64, F170^ECL2, I271^ECL3 and Y292^7.43. These newer contacts formed during the MD simulations are exclusive to TRV026-AT1R. The residues that make contacts only to Gq biased agonist TRV055 are shown in Fig. 2D. The differences in the persistence of the residue contacts selective to TRV055 compared to AngII is generally lower than the persistence difference between contacts in TRV026 and AngII. Residue contact with L195^5.38 that is selective to TRV055 is formed during MD simulations. It is difficult to predict if mutation of these residues will affect only the binding affinity of the agonist and/or its potency and efficacy. However, mutation of the residues that we predict to be exclusive to an agonist would cause a differential effect between the agonist of choice and a reference agonist. Such information is useful for enhancing bias of an agonist.

AngII bound AT1R active state crystal structure showed weak electron density for the binding site residue Y292^7.43 suggesting flexibility in the side chain of this residue in the presence of AngII. MD simulations showed three distinct conformational clusters for F8, with the top occupied cluster containing more than 80% of the population, as shown in Fig. S4D. The residue F8 of AngII peptide makes contacts with V108^3.32, S109^3.33, L112^3.36 on TM3 and with K199^5.43 on TM5. The persistence frequency of the F8 contact with Y292^7.43 is only 30% due to the flexibility of both F8 and Y292^7.43. Movement of F8 of AngII is accommodated by the concomitant movement of L112^3.36, which leads to the rotation of TM3, repositioning N111^3.35. The flexibility of F8 and role of L112^3.36 has been observed in prior MD simulation studies (13). This mechanism explains the observation of N111^3.35 moving outwards in the crystal structure of AngII bound AT1R. Y292^7.43 forms an intermittent hydrogen bond with N111^3.35. When N111^3.35 is not involved in this hydrogen bond, it is involved in hydrogen bonding with either D74^2.50 or with S107^3.31 in an outward facing conformation. We do not observe this flexibility in the MD simulations with TRV026 bound. Instead, N111^3.35 forms a stable hydrogen bond with D74^2.50 for the entire duration of the TRV026-AT1R MD simulations.

TRV056, which is a Gq biased agonist is the longest peptide studied here consisting of nine residues. Positions 1, 2 and 8 contain amino acid substitutions amongst the ten peptides and position 0 has an Asp present only in TRV056. The Asp in TRV056 contacts R272 in ECL3. None of the other agonists show any contacts with R272. Position 1 contains amino acids with smaller side chains in all agonists except AngII and TRV056. However, the residue in position 1 in all the ten agonists make high persistence contacts with the backbone of the residues Q15, D16, and D17 in the N-terminus. Therefore, these contacts are invariant to the nature of side chain in the peptide at position 1, although it should be noted that the side chains can affect the conformation of the backbone and hence the contact frequency. Position 2 in the ten peptides have Arg for the β-arr2 biased agonists and AngII, and Gly for Gq biased agonists. It is clear from the heat map in Fig. 2A, the AT1R residues Y184 in ECL2 and D263^6.58 show persistent contacts with R in the β-arr2 biased agonists and AngII, but not in TRV055 and TRV056 that are Gq biased. Position 8 has an aromatic residue Phe in AngII, TRV055 and TRV056. The heat map in Fig. 2A shows that Y113^3.37 and W253^6.48 make strong contacts with Phe in position 8 in AngII, TRV055 and TRV056, and none of the other agonists. These contacts in TM3 and TM6 are the distinguishing contacts in the presence of AngII and the Gq protein biased agonists. When Phe is substituted in the other β-arr2 bias agonists these contacts that are Gq bias selective are lost which could in turn decrease the coupling strength to Gq compared to β-arr2.

To gain insights into the similarities and differences of the AT1R conformational changes when bound to different biased agonists, we clustered the receptor conformations from the corresponding MD simulation trajectories by their root mean square deviation (RMSD) in Cartesian coordinates (see Methods). We observed only one major conformation cluster for AngII-AT1R and TRV026-AT1R complexes, but two conformation clusters for the TRV055-AT1R complex. The extracellular view of the comparison of the representative structures of AT1R from the most populated conformation cluster of the three agonists is shown in Fig. 2E. Fig. 2F shows the extracellular view comparing the representative structures of the most populated clusters for TRV026-AT1R and AngII-AT1R to the representative structure from the second cluster for TRV055-AT1R. The β-arr2 biased TRV026 bound AT1R shows inward movement of ECL2, TM5, TM6 and ECL3 compared to AngII or TRV055 bound AT1R. While TM4 and TM7 show minor variations, the conformations of TM1, TM2 and TM3 show no major differences between the three complexes. The conformations of ECL3 and TM6 in the AngII-AT1R complex resemble those in TRV055-AT1R, whereas the positioning of TM7 in AngII-AT1R is similar to TRV026-AT1R. The ECL2 in AngII-AT1R although not in the same conformation as TRV026-AT1R, shows a similar closing in towards the agonist binding pocket. The TM4 of AngII-AT1R occupies an intermediate conformation between TRV026-AT1R and TRV055-AT1R. The major difference between the two representative conformations of TRV055-AT1R complex comes from movement of TM3, TM5 and ICL2 as shown by the green arrows in Fig. S6. In summary, in the extracellular side of the receptor, the AngII bound AT1R receptor ensemble resembles that of TRV026 bound AT1R in some features and that of TRV055 bound AT1R in others. A description of the similarities in AT1R conformations among these ligands on the intracellular side is provided in a later section.

Relative allosteric communication strengths account for differences in ligand bias

In our prior work (14) we advanced the Allosteer method to compute molecular level ligand bias, termed “computational ligand bias”. Structural features of the ligand bound GPCR ensemble as well as multiple cellular factors contribute to experimentally measured ligand bias. Here we focus on the contribution from the structural ensemble of the agonist bound AT1R. This computational ligand bias is defined as the ratio of strengths of allosteric communication from the extracellular region to the Gq protein interface and the β-arr2 interface for a test ligand compared to that of a reference ligand (Fig. 3) (18–20). Our prior studies using this method for various biased and reference agonists to β₂-adrenoceptor, κ-opioid receptor and serotonin receptors demonstrated good correlation with experimentally calculated ligand bias factors (14). We applied the same method to estimate the ligand bias of several agonists to AT1R, using AngII as the reference agonist.

Figure 3.

Calculated molecular ligand bias compared to experimental ligand bias factor of Gq and β-arr2 biased agonists of AT1R. The red dots represent Gq biased agonists, and the blue dots represent β-arr2 biased agonists. Experimental bias factor values were taken from (21).

The experimental bias factors measured using AngII as the reference agonist were taken from literature (21, 22). As seen in Fig. 3, the calculated molecular ligand bias correlates well with the experimental bias factors for both Gq protein biased and β-arr2 biased agonists (R² value of 0.87). This qualitative correlation provides evidence that an atomistic property such as allosteric communication strength is one of the important contributing factors that potentiates the ligand bias observed in agonist-receptor pairings. While we recognize the importance of cellular factors in defining ligand bias, we have only examined the role of structural and dynamical factors in molecular bias here. This method can be used to predict if a given ligand would be biased or not prior to experiments. However, identifying the complex residue networks involved in the allosteric communication from the agonist binding site to the putative Gq protein and or β-arrestin binding site is beyond the scope of this study.

Gq biased agonists stabilize two distinct AT1R conformations while β-arr2 biased agonists show shrinking of the intracellular region of AT1R

We analyzed the changes in the AT1R conformations extracted from the MD simulations in the intracellular region where it couples to Gq or β-arr2 proteins. The change in the inter-residue distances between TM3 and TM6 (measured by the distance between Cα atoms of R126^3.50 and F239^6.34) and TM3 and TM7 (measured by the distance between Cα atoms of R126^3.50 and Y302^7.53) as shown in Fig. 4A, are used as measures of the extent of conformation sampling in the intracellular regions of AT1R. The snapshots from the MD simulations for the three agonists, AngII, TRV026 (β-arr2 biased) and TRV055 (Gq protein biased) bound AT1R, projected in these two distance measures are shown in Fig. 4B and for all agonists in Fig. S5. Compared to the TRV055-AT1R and AngII-AT1R complexes, β-arr2 biased TRV026-AT1R complex exhibits decreased flexibility, while Gq protein biased TRV055-AT1R exhibits a more flexible intracellular region. While the AngII-AT1R conformation distribution is mostly centered around the crystal structure of the R* state, the conformation sampling of TRV026-AT1R complex shows shrinking of the TM3-TM6 distance to values below what was observed in the crystal structure of the R* state. The representative conformation extracted from the most populated conformation cluster of TRV026-AT1R dynamics is similar to the alternative conformation (Cα RMSD = 1.1Å for the whole receptor) reported by Suomivuori et al. (see Fig. S6) as being crucial for βarr2 binding (13). The collapse in the TM3-TM6 distance in TRV026 bound AT1R dynamics is not as predominant as seen in the inactive state of AT1R (see Fig. S6). On the contrary, the TRV055-AT1R complex shows a wider conformation sampling, with the TM3-TM6 distance in few conformations reaching ~17Å compared to the ~14Å distance in the crystal structure of AngII-AT1R R* state, which is comparable to the TM3-TM6 distance observed in the G_q/11 protein bound M₁ muscarinic acetylcholine receptor structure (23). TRV055-AT1R also exhibits significant flexibility along the TM3-TM7 distance with the most populated conformation cluster showing a decrease in the TM3-TM7 distance compared to the R* state. Taken together, the β-arr2 biased agonist TRV026 presents a more homogeneous and compact conformation sampling, indicative of a less flexible intracellular region comprising TM3, TM6 and TM7, compared to the Gq protein biased TRV055 (see Fig. S5G and S5J). However, it should be noted that the other Gq biased agonist TRV056 displays a more compact conformation sampling (Fig. S5K) and does not exhibit as much receptor flexibility as TRV055. The water density map (Fig. 4C) and distributions of number of waters (Fig. S7) for the three agonists reveal that the calculated volume of the AT1R intracellular cavity is smallest in the presence of the β-arr2 biased agonist TRV026, while it is largest in the presence of the Gq protein biased TRV055, with AngII bound AT1R occupying a state in between those occupied by the two biased agonists and is marginally closer to that of TRV026-AT1R. The intracellular region of the receptor appears to be most flexible in the presence of TRV055, followed by AngII, with TRV026 bound AT1R being the least flexible.

Figure 4.

(A) Inter-residue distances between TM3 and TM6 and between TM3 and TM7 used to measure the extent of conformational sampling in the intracellular (IC) regions of AT1R bound to different agonists. The distance between the Cα atoms of residues R^3.50 and F^6.34, and between R^3.50 and Y^7.53 were calculated. (B) Projection of MD simulation snapshots of AngII, TRV026 and TRV055 bound AT1R in the active state (R*), using TM3-TM6 along x-axis and TM3-TM7 distances along y-axis. The grey dot represents the distances corresponding the inactive state (R) AT1R crystal structure (PDB: 4YAY), while the green dot represents the distances in the nanobody bound active state crystal structure (R*) (PDB: 6OD1) (C) Map of the weighted atomic density of water molecules calculated within 4Å of the intracellular region of AT1R during the MD simulations of TRV026, AngII and TRV055 bound to the active state AT1R (R*) shown in red.

Activation microswitches behave like rheostats and show differential levels of activation in the presence of different agonists

Comparative analysis of inter-residue distances between inactive and active state structures of class A GPCRs showed significant changes in certain inter-residue distances known as “activation microswitches” (24, 25). Contraction of the inter-residue distances between N/S^3.35, D^2.50 and N^7.46 collectively known as the sodium binding site, distance between P^5.50 and F^6.44 in the TM5-TM6 interface and the distance Y^5.58-Y^7.53 are three well-characterized activation microswitches in class A GPCRs (Fig. 5A). Some of the class A GPCRs show changes in activity towards G protein coupling in response to sodium ion concentration (26). Crystal structures of some class A GPCRs showed the presence of sodium ions in the inactive state structures typically nested between residues S^3.35, N^3.39, D^2.50 and N^7.46. The sodium ion binding site characterized by these residues shrinks upon receptor activation. However, Wingler et al. (27) showed that AT1R has no sensitivity in its activity to sodium ion concentration. Their crystal structure of AngII-AT1R active state complex showed outward movement of N111^3.35 compared to TRV026 or TRV023 bound AT1R active state structures. This resulted in an expansion of the putative sodium binding site in AngII-AT1R compared to antagonist bound inactive state structure of AT1R or even the TRV026-AT1R active state structure.

Figure 5.

Class A GPCR activation microswitches show distinct patterns of activation with different biased agonists: (A) Four inter-residue distances that characterize GPCR activation known as activation microswitches are shown in the central figure. The green cartoon is the nanobody bound active state structure of AngII bound AT1R, R* (PDB: 6OS0) and the grey structure is the antagonist bound inactive state of AT1R, R (PDB: 4YAY). (B) The population density distribution of the inter-residue distances in the microswitches during the MD simulations of AT1R with various biased and balanced agonists are shown. The distances measured in AT1R are P207^5.50 – F249^6.44, Y215^5.58-Y302^7.53, D74^2.50-N295^7.46 and D74^2.50-N111^3.35. The green and grey dashed lines show the corresponding inter-residue distances in the active (PDB: 6OS0) and inactive (PDB: 4YAY) state crystal structures respectively. (C) Heatmap of microswitch distances showing the level of activation for the different agonists when no nanobody or G protein is bound to the receptor, obtained by comparing to reference values obtained from the fully active state AT1R crystal structure (PDB: 6OS0).

Activation microswitches are thought of as “on or off” binary state switches, a concept emerging from the analysis of static structures. Taking the ensemble approach of GPCR conformations, we posited that the activation microswitches for a ligand-receptor pairing could exhibit a rheostat like behavior with different levels of activation in each microswitch (28). To examine the effect of biased ligands on activation microswitches we calculated the inter-residue distance distribution from the 2μs MD simulation trajectories for each of the ten agonists (Figs. 5B and S8). For each microswitch and for each agonist, we calculated the percentage of MD snapshots (frequency) that move away from distances typified by the endogenous agonist AngII bound active state crystal structure. These frequencies show the percentage of snapshots with activation microswitch distances that are neither in the fully active state or the inactive state (Fig. 5C). All the MD simulations show increased flexibility in the receptor when bound only to an agonist and in the absence of a G protein or nanobody. As a result, in the heatmap shown in Fig. 5C we observed that different agonists show a combination of different levels of occupancy of conformations that are intermediate to active and inactive states of each microswitch despite initiating the MD simulations from the same starting conformation for all agonists. Our observations predict that these activation microswitches may function as rheostats with different levels of activation dependent on the nature of the bound agonist rather than as binary on or off switches.

AngII-AT1R complex shows the most flexibility in the distance distributions of all microswitches compared to β-arr2 or Gq biased agonists as seen in Fig. 5B and Fig. S8. For example, AngII shows a small peak beyond 11.7Å (distance in the inactive state structure PDB ID: 4YAY) in the Y^5.58-Y^7.53 microswitch distance as shown in Fig. 5B. TRV055-AT1R complex also shows similar level of flexibility as AngII in the microswitches (Fig. S8). The β-arr2 biased agonists TRV026, TRV027 and TRV044 show a peak at the contracted D^2.50-N^7.46 distance around 3.3Å similar to the distance observed in the active state crystal structures of TRV026-AT1R (3.3Å) and TRV023-AT1R (3.1Å) as opposed to 4.1Å observed in the AngII-AT1R active state structure. There is little variation in the frequency of the D^2.50-N^3.35 microswitch across the β-arr2 and Gq protein biased agonists. In support of this finding, we observed that for the β-arr2 biased agonists there is little change in the D^2.50-N^3.35 distance observed in the active state crystal structure of TRV026-AT1R complex compared to the inactive state crystal structure of AT1R. In the AngII-AT1R complex there is a minor peak in the D^2.50-N^3.35 distance distribution that comes from a small conformation ensemble showing outward movement of residue N111^3.35 in our MD simulations (MD simulations were started from the active state conformation extracted from S1I8:AT1R structure which has the N111^3.35 facing inside the TM bundle). This leads to the orange cell in the heatmap for AngII (Fig. 5C). In summary, with only the agonist bound, the activation microswitches do not show a definitive pattern among the β-arr2 or Gq biased agonists (Fig. 5C). This is understandable since the efficacy and potency of these agonists towards β-arr2 or Gq coupling arises from an ensemble of conformations with different levels of activation of the individual microswitches.

Discussion

AT1R is an ideal model system to study the dynamical basis for ligand bias since it has available, Gq protein and β-arr2 biased agonists with experimentally measured bias factors and the crystal structures of AT1R in the nanobody bound active state. The crystal structures of the active state of AT1R bound to four different agonists including the biased agonist TRV026 did not show significant differences in the intracellular region of the receptor. However, the DEER studies showed that biased agonists stabilize distinct conformations of AT1R, thus exemplifying the importance of dynamics in these systems (10). Since the AT1R crystal structures do not rationalize the DEER measurements directly, we used it to perform MD simulations in the presence of ten different agonists to extract the resulting structural differences in the receptor. The conformation ensemble consisting of a weighted combination of conformations extracted from both the inactive and active state MD ensemble of AT1R, recapitulates the inter-residue distance distributions measured by DEER. This suggests that the agonist peptide bound AT1R sample an ensemble of conformations from both the inactive and active state of the receptor akin to the findings from NMR measurements for β2-adrenergic receptor (17). This hypothesis is extensible to other class A GPCRs that agonist bound receptor samples both inactive and active-intermediate states and that the G protein coupling is required to fully activate the receptor. Biased agonists with varying efficacies and potency to Gq versus β-arrestin signaling pathways show disparate dynamic behavior and the proportion of mixing of these different conformational states vary for different agonists.

The AT1R conformation ensemble stabilized by the β-arr2 biased agonist TRV026 shows narrowing of the TM3-TM6 intracellular receptor cavity compared to the Gq biased agonist TRV055 bound AT1R. This TRV026 bound AT1R is similar to the alternative conformation reported by Suomivuori et al. (13) with a 1.1Å difference in coordinates (RMSD) of all the Cα atoms in the receptor. The spatiotemporal heat map of persistence of the agonist-AT1R contacts for all the ten peptide agonists suggests the presence of AT1R residue contacts that are selectively sampled only by β-arr2 biased agonists and others by Gq biased agonists. MD simulations predict that the β-arr2 biased agonists make contacts with N-terminus predominantly more than Gq biased agonists, a concept that can be tested out by experiments.

The β-arr2 biased ligand contacts with AT1R that are predicted to confer functional selectivity are located mainly on TM7. The distance distributions of three well-characterized activation microswitches for the ten agonists studied here suggests that each agonist-AT1R pair exhibits a combinatorial behavior in the extent of activation of microswitches. This is understandable given that every agonist-receptor pair studied here has different pharmacological properties (EC50 and Emax) towards Gq and β-arrestin signaling pathways. In summary, our study predicts that each agonist-receptor pair elicits a combinatorial response starting from differences in agonist binding site to activating a combination of microswitches that differentially modulate the receptor conformations (an ensemble comprising of inactive and active state conformations) in the intracellular sites. This is an important and biologically relevant observation made from MD simulations since GPCRs exhibit a range of receptor activity towards G protein coupling as exemplified by the continuous distribution of the inter-residue distances between TM3-TM6 and TM3-TM7 observed in the three-dimensional structures of different class A GPCRs.

Materials and Methods

Receptor and agonist preparation for AT1R-agonist system setup:

All MD simulations of wild type AT1R in the active and inactive states bound to the unbiased, β-arr2 biased as well as Gq protein biased agonists (Table S1) were performed using the GROMACS package (29) with the CHARMM36 forcefield (30) for proteins, POPC lipids, ions, and using CHARMM TIP3P water as solvent. MD simulations of ten agonists bound to the fully active state of AT1R were performed for AngII, SII, TRV023, TRV026, TRV027, TRV034, TRV044, TRV044, TRV055, and TRV056. We also performed MD simulations of the inactive state of AT1R with AngII, TRV026, TRV055 and antagonist ZD7155 bound. The initial coordinates of active and inactive states of AT1R were taken from PDBs: 6DO1 (12) and 4YAY (15), respectively. Starting agonist positions for all ten AT1R-agonist complexes were obtained by mutating the corresponding residues back from the native ligand S1I8 in PDB: 6DO1. For TRV026 (Sar¹-R²-V³-Y⁴-Y⁵-H⁶-P⁷-NH₂⁸) agonist, 5^th and 8^th residues were mutated from the sequence of S1I8 using Maestro (Schrödinger) program and minimized in Macromodel using OPLS force field with distance dependent dielectric. All other receptor-agonist structures were derived using similar protocol. The amino acid sequence of all the AngII derivative peptides used in this study and their experimental bias factors are shown in Table S1. The missing side chains were added using Prime (Schrödinger) program. Hydrogen atoms were added, and protein chain termini were capped with neutral acetyl and methylamide groups. We used OPM (orientation of proteins in membranes) structure of PDB: 6D01 for alignment of the transmembrane helices of protein structure and inserted a pre-equilibrated POPC (palmitoyl-oleoyl-phosphatidylcholine) bilayer. The dimensions of the simulation box are 80×80×80 ų, that include 130 lipids, ~15,500 waters and 0.15M NaCl. After minimizing the AT1R-ligand complex, we equilibrated using the following steps: (a) 200ps of MD using NVT ensemble followed by (b) 40 ns of MD simulations using NPT ensemble. During the NPT equilibration step the position restraints were gradually reduced from 5 to 0 kcal/mol/Ų. In the final step of equilibration, we performed 10 ns of unrestrained NPT simulations. Thus, the equilibration step for each system was 50ns long. This was followed by 5 production runs with different starting velocities each 400ns long. The snapshots were stored every 20ps and the entire 400nsx5 runs amounting to 2μs of simulation time per system was used for analysis. Tests for convergence of the MD simulations (Fig. S9), methods used to cluster the conformations and other MD analyses are provided in Supplementary information (SI).

Calculation of Jensen-Shannon Distance (JSD) values:

JSD values are calculated using the formula indicated below (Eq. 1A), which includes the calculation of Kullback-Leibler Divergence (D_KL) (Eq. 1B).

$$\begin{gathered} JSD(P\parallel Q)=\sqrt{\frac{1}{2}D(P\parallel M)+\frac{1}{2}D(Q\parallel M)} \\ where,M=\frac{1}{2}(P+Q) \end{gathered}$$ (1A)

$$D_{KL}(A\parallel B)=\sum _{x\in X}A(x)\log \frac{A(x)}{B(x)}$$ (1B)

MMM software for DEER labeling of MD simulation frames:

For each agonist, we have taken ~5000 frames (200ps time interval) aggregated from the 5 MD simulation runs (each 200ns long) and calculated the inter-residue distances for the 10 pairs of labeled residues in the DEER experiments. We pooled the MD simulation trajectories starting from the inactive state structure of AT1R and the MD simulation trajectories starting from the nanobody bound active state of AT1R for different ligands. For each agonist, and for every MD simulation snapshot we used the MMM software (31) to mutate the residue of interest to cysteine, add nitroxide labels and chose the best optimized rotamer for the spin labels and calculated the inter-label distances of interest corresponding to the experimental data. We have not used the Bayesian/Maximum Entropy reweighting approach (32) to identify a subset of conformations from the ensemble that best resemble the DEER distance distributions.

Method to calculate Allosteric Communication and Ligand Bias:

The computational methods to calculate the bias for all agonists involve the following steps; (1) preparation of GPCR-agonist complex structural models for molecular dynamics, (2) performing MD simulations in POPC bilayer, (3) using the Allosteer method to calculate mutual information (MI) in torsion angle space for all residue pairs in the protein across the MD trajectories and, (4) using Allosteer to calculate allosteric communication pipelines from the extracellular region of the receptor to the Gq protein interface (GPI) or β-arr2 coupling interface (BAI). The details of the Allosteer method and the method to calculate bias are given in references (14, 18–20). In the section below we have provided a brief description of the Allosteer method as used in this work.

Using the trajectories from MD simulations for each agonist bound AT1R, we calculated the first and second order entropies for all torsion angles of all residues in the receptor (see Fig. S10). We then calculated the shortest allosteric pathway from the extracellular region of the receptor traversing through the residues in the agonist binding site to the Gq protein or β-arr2 coupling interfaces that maximizes correlated motion. This method is called Allosteer (14, 18–20). See text in Supporting Information for more details. Although there are no crystal structures of AT1R with Gq or β-arr2 bound available, we translated this information from other solved Gq protein and β-arr2 bound GPCR complex structures. The sequence homology among GPCRs coupling to Gs, Gi or Gq proteins are low (20 to 40%) unless they belong to the same subfamily (GPCRs that bind to the same endogenous agonist such as adrenergic receptors for example). However, GPCRs have high structural homology. That is, residues in the same structural positions are commonly involved in G protein coupling. For example, the three-dimensional structures of specific G protein coupled receptors show that residues in the same structural positions are involved in coupling although not all the residue positions are the same or conserved in these receptors. This assumption that the same residue positions are involved in Gq coupling to AT1R is a caveat to our allosteric communication strength calculation.

The list of AT1R residues posited to be in the β-arr2 and Gq protein coupling interface used in this study for calculating the allosteric pipelines are listed in Table S3. We used Allosteer to calculate the ligand bias as described in the Supplementary Information.

Inter-residue distance measurements for activation microswitches:

The two pairs of inter-residue distances D74^2.50-S115^3.39 and S115^3.39-N295^7.46 that denote the sodium binding site are measured as the minimum distance between the side chain heavy atoms of the participating residues, i.e., CB, CG, OD1, OD2 for aspartate, CB, OG for serine and CB, CG, OD1, ND2 for asparagine. The distance between the Y215^5.58 and Y302^7.53 is a microswitch measured between the hydroxyl oxygen atoms, i.e., OH, of the two tyrosine residues. The PF microswitch is measured as the minimum distance between any of the carbon atoms in the side chains of P207^5.50 i.e., CB, CD, CG and of F249^6.44 i.e. CB, CG, CD1, CD2, CE1, CE2.

• Agonist bound GPCRs sample a weighted mix of active and inactive conformations

• The weighted mix of active and inactive state MD ensembles recapitulates DEER

• Biased agonists show ligand specific receptor contacts in the binding site

• β-arrrestin biased agonists show narrowing of the receptor intracellular surface

Data and Software Availability

The MD simulation trajectories for AT1R bound to ten different agonists simulated in this study have been submitted to http://gpcrmd.org website. For each system, the trajectories for five runs each 400ns long (.xtc files for gromacs) and the starting structures for the production MD runs have been submitted to this site. The software used in this study are standard GROMACS based analysis software. The MMM software that was used to generate the distribution of inter-residue distances from MD simulation trajectories were obtained from Prof. Dr. Gunnar Jeschke as mentioned in the Acknowledgments.