Internal water molecules are essential for the function of many membrane proteins such as channel and transport proteins1. For G-protein-coupled receptors (GPCRs), which form the largest class of membrane proteins transmitting extracellular signals across the plasma membrane, internal water molecules have been proposed to play an important role for receptor activation but mechanistic models remain speculative2. Previous studies revealed that water molecules from the cytoplasmic region of rhodopsin (Rho) are involved in the hydrolytic release of retinal upon Rho activation by light3,4. Crystal structures of Rho and the beta-2 adrenergic receptor (β2AR) revealed further details. Distinct conformational changes occur during activation of these receptors, including reorganizations of inner protein hydrogen bonding networks, which could be in contact with internal waters5,6. An important step forward came from the recent high-resolution crystal structures of the A2A adenosine receptor (A2AR) and the delta opioid receptor (δOR), which revealed internal ordered water molecules that could be crucial for GPCR activation7,8. On the basis of their crystal structures, Liu et al.7 proposed that the A2AR in the non-activated state contains a nearly continuous internal water channel, which is disrupted in the activated receptor. However, this detailed suggestion on the extension of the internal water channel is not conclusive for several reasons: (1) As is shown in this article, the A2A receptor’s internal water molecules are quite mobile, which exclude determination of their precise spatial localization within the protein’s structural framework. (2) To calculate the receptor’s intrinsic water pathway, Liu et al.7 increased the radius of water molecules and connected all water molecules with each other because it is not possible to distinguish between a continuous and a discontinuous water channel from crystal structures.

However, it is important to know the exact spatiotemporal distribution of water inside of GPCRs and their contact sites with the receptor’s polypeptide backbone and side chains to finally understand the relationship of an internal water channel to the receptor’s function. Internal waters can strongly interfere with internal hydrogen bonding networks of a protein’s backbone and side chains, thereby strongly influencing its conformational and structural changes. Notably, the details of how the intrinsic water molecules couple to the protein structure and thereby affect GPCR activation at the atomic level cannot be solved by static X-ray structures. Alternatively, molecular dynamics (MD) simulations can resolve atomic details of structural transitions of receptors and their internal waters, which are experimentally not accessible8,9,10,11. Here, by using a total of 32-μs all-atom MD simulations, we studied with high spatial and temporal resolution important steps during the activation of three different receptors from class A GPCRs to elucidate how water molecules enter the inner space of a receptor, couple to and influence this protein’s structural transitions. Contrary to Liu et al.7, our simulations revealed that in the inactivated state of the A2AR the internal water pathway is interrupted by a gate of hydrophobic amino acids, which upon receptor activation evoke conformational changes opening a continuous internal water channel. By analyzing all presently available GPCR crystal structures we were able to identify common functional roles of internal waters within the members of class A GPCRs.


Hydrophobic layers of amino acid residues in the A2AR

We started our MD investigations from the 1.8-Å resolution A2AR (PDB 4EIY) crystal structure7, which shows a comprehensive network of 57 solvent molecules inside the receptor together with a sodium ion at the allosteric site. According to the crystallographic b-factors (Fig. 1a) many of the receptor’s intrinsic water molecules are mobile, a property further manifested by the poor electron density map in several regions of the receptor, especially those close to Y2887.53 in the NPxxY motif and D522.50 at the allosteric site. Meanwhile, another 1.8-Å resolution GPCR crystal structure, that of the DOR (PDB 4N6H), was released12. This also contains internal waters just as the A2AR does (Fig. 1b).

Figure 1: Water molecules and hydrophobic layers of amino acid residues in the crystal structures of adrenergic and opioid receptors.
figure 1

(a) A2AR (PDB 4EIY) and (b) δOR (PDB 4N6H). The precision of defining the position of individual water molecules (depicted as coloured spheres) is given by the crystallographic b-factors (see colour code). A crystallographic electron density map of water molecules at 1.5σ level is pictured with COOT software56. 2FO-FC map is in blue grid and FO-FC map is in red grid.

To reveal how water molecules located inside the receptor are distributed spatially and temporally and how they couple to the protein’s structural fluctuations during GPCR activation, we performed all-atom long-timescale MD simulations starting from particular crystal structures of the A2AR: (a) ZMA antagonist bound to A2AR (ZMA/A2AR) representing an inactivated GPCR, (b) agonist NECA docked to inactive A2ARI (NECA/A2ARI) as well as activated A2ARII crystal structures (NECA/A2ARII) representing a GPCR in the meta state between a inactivated and activated receptor, and (c) NECA bound to A2AR in presence of a C-terminal helix fragment of the Gs protein α-subunit (Gα) at the cytoplasmic side (NECA/A2AR/Gα) representing an activated GPCR/G-protein complex7. Two parallel MD simulations of 1.6 μs duration were performed for each of the above-mentioned cases (Supplementary Table 1).

Although many water molecules were found inside the receptor during our MD simulations in the case of ZMA/A2AR, two distinct water-free layers of hydrophobic amino acid side chains were resolved based on water density maps (Fig. 2a). These were also revealed from the water molecule distribution in the crystal structure (Fig. 1a). The first hydrophobic layer (HL1) was located between the orthosteric and the allosteric sites, and had a thickness of ~7 Å (Fig. 2a). Interestingly, water molecules from the bulk next to the bound agonist NECA only rarely diffused into the deep pocket of the receptor. The second hydrophobic layer (HL2) of ~12 Å thickness was located close to the highly conserved NPxxY motif in TM7. In this hydrophobic zone, three interconnected water molecules (Fig. 1a) bound to residues between Y2887.53 and Y1975.58 in the crystal structure, were broken in the MD simulations and only one water molecule was left connected to Y1975.58. This agrees well with the waters’ high mobility indicated by their high b-factors and poor localization in the density maps of the crystal structures.

Figure 2: Location of the hydrophobic residue layers based on maps of water density inside the A2AR.
figure 2

Final MD structural details of the hydrophobic layer and average water density are shown for (a) ZMA/A2AR, (b) NECA/A2ARI, (c) NECA/A2ARII and (d) NECA/A2AR/Gα.

For the activated GPCR in the NECA/A2ARI complex, many more water molecules were located inside the receptor (Fig. 2b) than for the inactive GPCR in the ZMA/A2AR complex. In ZMA/A2AR, water molecules from the bulk directly accessed HL1, and in NECA/A2ARI, the NECA agonist was connected to the NPxxY motif through a network of continuous water molecules involving a sodium ion inside the receptor. In addition, the conformation of Y2887.53 underwent a single flip-transition during MD simulations in the time range of 0.3–0.6 μs (Fig. 3c,d). Superposing the final structure of NECA/A2ARI from MD simulations (after a residue flip) with the NECA/A2AR crystal structure12, we found that the residue Y2887.53 adopted identical conformations (Supplementary Fig. 1a). Moreover, the HL2 observed in the ZMA/A2AR complex was retained in the NECA/A2ARI complex, but the thickness was reduced to ~10 Å (Fig. 2b). The HL2 was also observed within two independent additional 1.6 μs MD simulations, starting from the NECA/A2ARII structure13 (Fig. 2c).

Figure 3: Three distinct rotamer states of Y2887.53 at the NPxxY motif.
figure 3

(a) Free-energy surface of the agonist NECA bound to A2AR in absence of Gα. (b) Free-energy surface of agonist NECA bound to A2AR in presence of Gα at the cytoplasmic site. (c) Time dependence of angle χ1 of Y2887.53 during MD simulations of ZMA/A2AR, NECA/A2AR and NECA/A2AR/Gα. (d) Time dependence of angle χ2 of Y2887.53 for the same complexes.

Notably, during a simulation of the NECA/A2AR/Gα complex with Gα present at the cytoplasmic G-protein-binding site, both HLs were no longer detected in the receptor. Instead, a continuous water channel formed spanning from the ligand orthosteric binding site towards the G-protein interaction zone (Fig. 2d). Y2887.53 also performed a flip-rotation transition during the course of MD simulations of NECA/A2AR/Gα (Figs 2d and 3c,d), but adopted a conformation distinct from those in crystal structures of ZMA/A2AR and NECA/A2AR. Superpositioning the final NECA/A2AR/Gα structure with that of fully activated β2AR in complex with G protein (PDB 3SN6)5 revealed that Y2887.53 shared the same conformation (Supplementary Fig. 1b).

HLs in Rho and the β2AR

To further test our results on HLs in A2AR, the MD simulations were also performed for two additional receptors, Rho and the β2AR. As before, 2 × 1.6-μs all-atom MD simulations were performed for three different forms of each receptor. For Rho, we simulated the inactive receptor structure14 comprising the covalently bound 11-cis-retinal (PDB 1U19; cRET/Rho) as well as the activated receptor structure with bound all-trans-retinal15 (PDB 2X72) in the presence (tRET/Rho/Gα) and absence of Gα (tRET/Rho). Similarly, in the β2AR, we investigated the structure of the complex (PDB 2RH1) between the partial inverse agonist carazolol and β2AR16 (CAR/β2AR) and between the full agonist BI-167107 and β2AR (PDB 3SN6)17 with Gα (BIA/β2AR/Gα) and without Gα (BIA/β2AR) (Supplementary Table 1).

Since the HLs changed during MD simulations, we took snapshots from the final 0.3 μs simulations for statisitical analysis. In inactive Rho (Supplementary Fig. 2a), we also identified two HLs near the orthosteric site and the NPxxY motif showing a thickness of ~5 Å and ~10 Å, respectively. However, in activated Rho without Gα at the cytoplasmic site (Supplementary Fig. 2b), HL1 increased to ~7 Å while HL2 disappeared and was replaced by water molecules from the intracellular region. Interestingly, a continuous water channel was found in the structure of the tRET/Rho/Gα complex (Supplementary Fig. 2c), where both, HL1 and HL2 were accessed by water during MD simulations.

Similarly, two layers of hydrophobic amino acid residues were observed in the complex with antagonist CAR/β2AR (Supplementary Fig. 2d), having a thickness of ~5 Å and ~12 Å, respectively. Different from both A2AR and Rho, lack of the Gα-protein in the activated β2AR led to an influx of water molecules from the cytoplasmic side of the receptor, finally forming a continuous water pathway. Furthermore, reorganization of the water molecules also induced Y3267.53 at the β2AR NPxxY motif to flip back to the inactive conformation (Supplementary Figs 2e and 3). These observations agree with the previous findings of Dror et al.17 that resetting the fully activated β2AR into the inactivated state requires tenths of a microsecond.

The HL as a common feature of GPCRs

To further investigate whether the existence of the second layer of hydrophobic residues at the highly conserved NPxxY motif is a common feature of family A GPCRs, we employed a three-dimensional sequence alignment based on all 20 released structures of different members of family A GPCRs. As depicted in Fig. 4, residues in this second layer were mainly hydrophobic (Val, Leu, Ile, Phe and Pro). Small polar residues (Thr, Ser, Tyr and Met) were a few times also found in this region bound to other hydrophobic residues by their non-polar moieties. There was only one exception, with an Arg in chemokine receptor type 4 (CXCR4) at position 2.43 in transmembrane helix 2 (TM2)17. But even in this particular case, the hydrophobic part in the side chain of Arg in the crystal structure was still in a conformation pointing towards the hydrophobic zone, leaving the positively charged moiety directed towards the cytoplasmic region. The recently released 1.8-Å high-resolution DOR crystal structure (Fig. 1b) also fits into the above findings12.

Figure 4: The residues comprising the conserved hydrophobic layer located near the NPxxY motif in class A GPCRs.
figure 4

Left: alignment of amino acid residues (depicted in colour code and size according to frequency) in the hydrophobic layer HL2 near the NPxxY motif of 20 class A GPCRs. Right, top: location of HL2 inside a non-activated class A GPCR. Residues are coloured according to their type. Right, bottom: The most conserved residues, P7.50 and Y7.53 are coloured orange in the image of the GPCR structure.

Three different conformations of Y7.53

From long time scale MD simulations on A2AR, Rho and β2AR, we observed that Y7.53, a highly conserved residue in class A GPCRs (Fig. 4), underwent various defined conformational transitions (Fig. 3). In the crystal structures of the analyzed receptors we observed three unique rotamer configurations of Y7.53, namely YI, YII and YIII (Fig. 3), which correlate with the functional state of GPCRs (Table 1). The YI conformation is found in many antagonist-bound GPCR structures such as ZMA/A2AR (ref. 7), or in inverted agonist-bound receptors such as CAR/β2AR (ref. 16). The YII rotamer is present in agonist-bound receptors, such as BIA/β2AR/Gα, with Gα bound at the cytoplasmic site. The YIII conformation appears in agonist-bound receptors without stabilization by Gα such as NECA/A2AR, or arrestin-biased ligand-bound receptors as in the crystal structure of 5-HT2BR (ref. 18). All three conformations have been distinguished in our simulations, including both classical unbiased long timescale MD simulations as well as in metadynamics simulations (biased). In the classical unbiased MD simulations (Fig. 3c,d), starting from the antagonist-bound crystal structure7, the residue Y2887.53 was stabilized at the YI conformation during the whole 1.6-μs MD simulation of ZMA/A2AR. However, in the simulations of NECA/A2AR, a flip-rotation transition to the YIII conformation took place in the 0.3–0.6 μs time range and this rotamer state remained until the end of the simulations. The YIII conformation found in the MD simulations is identical to that in the previously solved crystal structure of A2AR (PDB 2YDV)13. Interestingly, when both the agonist NECA and the Gα protein were present, a transition of YI to YII state took place in the 0.3–0.5 μs time range and this rotamer state remained stable until the end of these simulations.

Table 1 The Y7.53 rotamer and the internal water channel structures in relation to the activity states of different ligand/GPCR complexes*.

To investigate the free-energy differences between the different conformations of Y7.53 we performed a well-tempered metadynamics simulation to estimate free-energy surface (FES). It was done for the agonist NECA-bound A2AR receptor, in the absence (Fig. 3a) and presence (Fig. 3b) of Gα at its binding site on the receptor. In the metadynanics simulations, two distinct YIII conformations were sampled, YIIIa and YIIIb, which differed in the χ1 angle. In the absence of Gα, YI, YII and YIIIa showed similar FES values of −33, −38 and −37 kJ mol−1, respectively, implying that under these conditions various states can co-exist in the receptor. It is important to note that an alternative YIII conformation YIIIb, with a FES value of −22 kJ mol−1, was also sampled. YIIIb showed a different χ1 angle with the aromatic ring flipped towards TM1 and TM2. Interestingly, in the presence of Gα, the FES for all states, except YIIIa, decreased by −15 kJ mol−1, and finally stabilized at values of −34, −47 and −28 kJ mol−1 for YI, YII and YIIIb, respectively. Compared with structures without Gα, the FES values indicate that the YII state is substantially more stable than any other rotamer configuration. Our findings are also supported by the observations of Nygaard et al.19 that G proteins can stabilize the fully active state of GPCRs, preventing the receptor from switching to different states. These observations tempted us to investigate how the Y7.53 conformations correlate with a continuous water pathway and the activation of GPCRs. Internal, structured water molecules play a critical role in receptor activation; some of these waters connect two TM helices and some form a H-bond network between receptor and ligand20. Closer examination revealed that these water molecules are mobile.

The activation of GPCRs is featured by a movement of TM helixes, which opens a large space at the cytoplasmic region for binding a G protein21,22. Therefore, the activated state of a GPCR can be characterized by the volume change at the G-protein-binding site. According to our simulations, the antagonist-bound complex ZMA/A2AR (Fig. 5), representing an inactive receptor, shows a relatively small average volume of the G-protein-binding region of VG ~700 Å3. In NECA/A2ARI and NECA/A2AR,II the A2AR remains in the meta state18, which is between an inactive and active state, showing VG values of ~1,100 Å3. It is interesting that in the meta state GPCR crystal structures of NECA/A2AR and arrestin/5-HT2BR, the Y7.53 adopt the YIII conformation20,22. Moreover, it is known that upon NECA binding the adenosine receptor also signals via arrestin23. These findings imply that arrestin-biased activated GPCRs are featured by YIII rotamer. We suggest that our final state of 1.6 μs MD simulation of NECA/A2AR/Gα represents the initial activation stage of A2AR with a partially activated receptor. For NECA/A2AR/Gα, the receptor’s volume of the binding region for the G protein is VG~1600 Å3, which is considerably larger than for ZMA/A2AR and NECA/A2AR.

Figure 5: Average volume (VG) of the G-protein-binding site in A2AR at different functional states.
figure 5

Non-activated, meta state and activated A2AR are characterized by different volumes. The inactive states show the lowest VG values, the activated receptors have the largest VG values, and VG values of the meta states are between those of inactive and activated GPCR states.

In Rho, the VG of the inactive state is also small (Supplementary Fig. 4), adopting a value of ~800 Å3. Interestingly, the volumes VG of active tRET/Rho and tRET/Rho/Gα are both in the range of ~1,300 Å3. The comparison of the VG value in the complex CAR/β2AR with the VG values in A2AR and Rho, show that this volume in a receptor’s inactive state is much smaller than in the active one. Noticeably, the VG of BIA/β2AR is comparable with VG of BIA/β2AR/Gα, although no Gα is present at cytoplasmic region in BIA/β2AR. This is in agreement with the previous finding24 that β2AR requires tens of microseconds to reset back to inactive states from a fully activated structure. However, we still observed that VG of BIA/β2AR already shrunk to ~700 Å3 compared with fully activated BIA/β2AR/Gα (Supplementary Fig. 4).

Gα stabilizes the YII state of Y7.53 in various GPCRs

To investigate how G protein stabilizes the YII conformation in the fully activated receptor state, we simulated 2 × 1.6 μs for Rho, β2AR and A2AR, with and without Gα (Fig. 6). In the Gα-bound receptor state, Y7.53 was tightly stabilized by residues of the G protein. Specifically, the Y3067.53 YII rotamer in Rho (Fig. 6a) was bound to R1353.50 and L349 from the Gα protein, Y3267.53 in YII state in the β2AR (Fig. 6b) was stabilized by R1313.50 and Y391 from Gα, and the residue Y2887.53 (YII) in the A2AR (Fig. 6c) was stabilized by I2927.57 and Y391 from Gα. Analogously, the YII state was also seen in two recently published crystal structures of GPCRs with bound agonists, the 5-HT1BR (ref. 25) and the β2AR (ref. 16) (Supplementary Fig. 5): the YII conformation in these structures was stabilized by L1048 from the fusion protein and by R1473.50, which is present in the G-protein-binding site of 5-HT1BR. In the β2AR, the residue I104 from the nanobody Nb80 and residue R1313.50 stabilized the active YII state of the receptor. When Gα was absent in agonist-bound Rho (Fig. 6d), Y3067.53 was still retained by the hydrophobic moieties of three methionine residues including M2576.40, M2536.36 and M3097.56, which have been shown by mutagenesis experiments26,27 to play a role in the activation of Rho. Our MD results are also in agreement with a recently released crystal structure of octylglucoside solubilized Rho without G protein28, revealing that the active state of Rho does not need to be stabilized by a G protein. In A2AR and β2AR, which are partially activated even in the basal state, the YII state was not maintained and finally stabilized in YI (Fig. 6e) and YIII (Fig. 6f) conformations in complexes with bound agonists and in absence of Gα.

Figure 6: Stabilization of of Y7.53 in the NPxxY motif.
figure 6

(a) In Rho, Y3067.53 is stabilized in the YII active state by R1353.50 of TM3 and L349 of Gα. (b) In the β2AR, Y3267.53 is stabilized in the YII active state by R1313.50 of TM3 and Y391 of Gα. (c) In the A2AR, Y2887.53 is stabilized in the YII active state by I2927.57 of TM7 and Y391 of Gα. (d) In absence of Gα, Y3067.53 is stabilized in the YII active configuration by the three Met residues: M2576.40, M2536.36 and M3097.56. (e) In absence of Gα, Y3267.53 switched back to the inactive YI configuration. (f) In absence of Gα, Y2887.53 switched to the alternative YIII configuration.

The mechanism of water penetration into GPCRs

In our present work we found that GPCRs in their activated state comprise a continuous internal TM water channel. In our previous projects we investigated how water molecules enter the interior of the receptor8,29. For formyl peptide receptor 1 (FPR1) and opioid receptors (ORs), using microsecond MD simulations, we showed that the influx of water molecules occurs towards the allosteric site, which is different from that in the human sphingosine-1-phosphate receptor 1 (S1PR1)30. Jastrzebska et al.3 reported that water molecules can reach the opsin binding region from the intracellular side during the retinal isomerization in Rho, which is also seen in our MD simulations of Rho. These observations imply that there are two possible water penetration routes. In some receptors, such as Rho and S1PR1, the intracellular water penetration occurs when the receptors are being activated, whereas for other receptors, such as ORs and A2AR, the extracellular water penetration occurs during the activation process.

Our analysis of 20 crystal structures of different GPCRs showed that the allosteric ligand binding regions of Rho and S1PR1 are entirely covered by large N-termini and ECL2 loops (NEL; Fig. 7a, left blue sphere), preventing bulk water molecules from moving towards the central inner space of the receptor. However, for all other receptors, the NELs are much smaller, leaving the orthosteric site widely open (Fig. 7a, right blue sphere). Closer inspection of the interaction surface area (ISA) between NEL and the remaining moiety of GPCRs (Fig. 7a) further confirmed these differences: Rho and S1PR1 show ISA values of 3,500–4,300 Å2, which are much larger than the ISA values of all other receptors ranging between 1,200 and 2,400 Å2. Furthermore, the ligand binding layers in Rho and S1PR1 are more hydrophobic than those in the other GPCRs (Fig. 7b). This strong hydrophobic property of Rho and S1PR1 could further block the water influx from the extracellular bulk environment. Notably, many GPCRs used in X-ray crystallography were strongly modified to allow crystallization, for example, sometimes the flexible N-termini were cut or additional protein sequences were inserted into certain receptor loops. Nevertheless, it is still acceptable to compare the ISA contact areas in the crystal structures, since the flexible modified regions may not contribute much to the contact area as they might be disordered under physiological conditions.

Figure 7: Contact areas between EL2 and the N-terminus in GPCRs.
figure 7

(a) Inspection of crystal structures of 20 different GPCRs show two receptor clusters: Receptors similar to Rho and S1PR1 have a high contact area of 3,500–4,300 Å2. The remaining receptors have a low contact area of 1,200–2,400 Å2. (b) Hydrophobicity profile of the ligand binding site in rhodopsin (top) and in the μ-OR (bottom); colours: orange, hydrophobic surface; blue, polar surface.


More than 800 human GPCRs allow cells to recognize diverse extracellular stimuli and transduce signals across the plasma membrane to regulate central physiological processes. Currently, >20 GPCR crystal structures have been solved, but details of their activation are still not fully understood. On the basis of a total of 32-μs all-atom MD simulations, we here analyzed the activation of three different receptors of class A GPCRs. We discovered a specific common feature of the receptors—an intrinsic water pathway, which in the receptors’ resting state is interrupted by a HL of amino acid residues and upon agonist binding, opened to a continuous intrinsic water channel (Fig. 8). Water from the bulk can enter into the receptor during activation from two different directions: from the extracellular (Fig. 8a) and from the intracellular side (Fig. 8b) depending on the receptor type. Three distinct rotamer conformations, namely YI, YII and YIII, of a highly conserved Tyr (Y7.53) residue in the NPxxY motif correlate with the formation of the internal water pathway and the functional state of GPCRs. The YI state represents the inactive state of receptor with the first layer of hydrophobic amino acid residues located close to the NPxxY motif, the second layer of hydrophobic amino acids next to Y7.53 and a water pathway is interrupted (Fig. 8c). The YII state indicates fully activated GPCR state with a G protein bound and a continuous water pathway connecting Y7.53 to extracellular or intracellular side (Fig. 8a,b). The YIII state corresponds to the meta state of the receptor comprising a HL next to Y7.53 (Fig. 8d). These findings contribute to better understanding of a role of internal waters in the activation process of GPCRs and provide important new directions for MD-based design of compounds targeting GPCRs.

Figure 8: A schematic view of three different rotamer conformations of the highly conserved Y7.53 residue.
figure 8

(a) The YII conformation is preferred in activated GPCRs; during formation it is accessed by extracellular water molecules forming a continuous internal water network. (b) For receptors similar to Rho and S1PR1 the YII state is associated with the cytoplasmic influx of water. GPCRs are shown in grey and the G-protein subunits are displayed in different colours. (c) The YI conformation is preferred in inactive GPCRs with two hydrophobic layers blocking a continuous water pathway. (d) The YIII conformation is preferred in GPCR meta states with one hydrophobic layer located next to the NPxxY motif and blocking formation of a continuous water pathway.


Loop filling and refinements

Since the intracellular loop ICL2 for each receptor was missing because of the insertion of a fusion protein in this region, the loop refinement protocol in Modeller V9.10 was used to fulfil and refine this area31. A total of 20,000 loops were generated for each receptor and a conformation with the lowest Discrete Optimized Protein Energy (DOPE) score was chosen for receptor construction. Repaired models were submitted to Rosetta V3.4 for loop refinement with kinematic loop modelling methods32. Kinematic closure (KIC) is an analytic calculation inspired by robotics techniques for rapidly determining the possible conformations of linked objects subject to constraints. In the Rosetta KIC implementation, 2N-6 backbone torsions of an N-residue peptide segment (called non-pivot torsions) are set to values drawn randomly from the Ramachandran space of each residue type, and the remaining six φ/ψ torsions (called pivot torsions) are solved analytically by KIC.

Protein structure preparations

All protein models were prepared in Schrodinger suite software under OPLS_2005 force field. Hydrogen atoms were added to repaired crystal structures at physiological pH (7.0) with the PROPKA (ref. 33) tool in Protein Preparation tool in Maestro33,34,35 to optimize the hydrogen bond network. Constrained energy minimizations were conducted on the full-atomic models, with heavy atom coverage to 0.4 Å. Models of the receptor/Gα complexes were prepared as previously reported21.

Ligand structure preparations

All ligand structures were obtained from the PubChem36 online database. The LigPrep module in Schrodinger 2012 suite software was introduced for geometric optimization by using OPLS_2005 force field. Ionization states of ligands were calculated with the Epik tool37 employing Hammett and Taft methods in conjunction with ionization and tautomerization tools.

Protein-ligand docking

The docking procedure was performed by Glide38,39. Ligand molecules were initially placed in the binding pocket in a pose similar to that noted in the ligand-bound crystal structure. Cubic boxes centered on the ligand mass centre with a radius 8 Å for all ligands defined the docking binding regions. Flexible ligand docking was executed for all structures. Twenty poses per ligand out of 20,000 were included in the post-docking energy minimization. The best scored pose for each ligand was chosen as the initial structure for MD simulations.

MD simulations

The membrane system was built by the g_membed40 tool in Gromacs41 V4.6.3 with the receptor crystal structure pre-aligned in the Orientations of Proteins in Membranes (OPM) database42,43. Pre-equilibrated 132 POPC lipids coupled with 9,200 TIP3P water molecules in a box ~70 Å × 70 Å × 98 Å were used for building the protein/membrane system. We modelled the protein using Amber99SB-ILD* force field44 with improved side-chain torsion potentials of Amber99SB force field. For the POPC lipids, we used the Amber Slipids (Stockholm lipids) force field parameter set, which in recent MD simulations of lipid bilayers yielded excellent agreement with experimental data (X-ray and neutron diffraction; NMR order parameters)45,46. For modelling the ligand we used the Amber GAFF small molecule force field47. The parameters of covalently bound 11-cis-retinal were obtained from a previous report 48. The ligand geometry was submitted to the GAUSSIAN 09 programme49 for optimization at Hartree-Fock 6-31G* level, when generating force filed parameters. The system was gradually heated from 0 to 310 K followed by 1 ns initial equilibration at constant volume and temperature set at 310 K. Next, an additional 40 ns constrained equilibration was performed at constant pressure and temperature (NPT ensemble; 310 K, 1 bar), and the force constant was trapped off gradually from 10 to 0 kcal mol−1. All bond lengths to hydrogen atoms were constrained with M-SHAKE. Van der Waals and short-range electrostatic interactions were cut off at 10 Å. Long-range electrostatic interactions were computed by the Particle Mesh Ewald summation scheme. MD simulations results were analyzed in Gromacs41 and VMD50.

Average water density calculation

Water density was calculated in Volmap plugin in VMD50. Volmap creates a map of the weighted atomic density at each gridpoint. This is done by replacing each atom in the selection with a normalized Gaussian distribution of width (standard deviation) equal to its atomic radius. The Gaussian distribution for each atom is then weighted using an optional weight read from one of the atoms’ numerical properties, and defaults to a weight of one. The various Gaussians are then additively distributed on a grid. The meaning of final map will depend of the weight of mass. The average water density was calculated based on the final 0.3-μs frame of each long time scale MD simulation. Final output results were visualized in VMD.

Contact map calculation

The contact map for residues at G-protein-binding site was done by g_mdmat in Gromacs that makes distance matrices consisting of the smallest distances between residue pairs. Frames (0.3 μs) of each long time scale MD simulation was used for contact map calculations.

Metadynamics simulations

Free-energy profiles of the systems were calculated by using well-tempered metadynamics in Gromacs V4.6.3 (ref. 41) with Plumed V1.3 patches (ref. 51). Metadynamics adds a history-dependent potential V(s, t) to accelerate sampling of the specific collective variables (CVs) s(s1, s2, …, sm)52. V(s, t) is usually constructed as the sum of multiple Gaussians centred along the trajectory of the CVs (equation 1).

Periodically during the simulation, another Gaussian potential, whose location is dictated by the current values of the CVs, is added to V(s, t) (ref. 21). In our simulations, the dihedral angles of Y2887.53, χ1 and χ2, were assigned as the CVs s1 and s2, while the width of Gaussians, σ, was set as 5 degrees. The time interval τ was 0.09 ps. Well-tempered metadynamics involves adjusting the height, wj, in a manner that depended on V(s, t), where the initial height of Gaussians w was 0.03 kcal mol−1, the simulation temperature was 310 K and the sampling temperature ΔT was 298 K. The convergence of our simulations was judged by using the free-energy difference between states X and A at 10-ns intervals. Once the results stopped changing over time, the simulation was considered as converged53. Each metadynamics simulation lasted for 140 ns, and the results were analyzed upon convergence.

Reweighting algorithm

From a converged metadynamics run we can calculate directly the canonical probability distribution of the CVs at a given temperature. On the contrary, the statistics of other degrees of freedom is somehow distorted by the application, during the simulation, of a time-dependent external potential on the CVs. In well-tempered metadynamics, the reconstruction of the distribution of variables different from the CVs is particularly simple since for long time periods the amount of bias added decreases to zero and the system approaches equilibrium. The algorithm described by Bonomi et al.54 consists of three different steps: (1) Accumulate the histogram of the CVs plus the other variables of interest between two updates of the bias potential. (2) When a new Gaussian is added, evolve the histogram as equation (2):

where P(R,t) is the biased probability distribution, V(S(R),t) the time derivative of the bias potential and the average in the exponent is calculated in biased ensemble. (3) At the end of the simulation, the unbiased distribution PB(R) can be recovered from the histogram collected by using a standard umbrella sampling reweighting as equation (3):

Contact area and binding volume calculations

Contact areas between the ECL2/N-terminal and the left space of receptors were calculated by the contact_surface module in Pymol. It calculates individual or global contact areas between a receptor molecule and a (multimodel) bundle of ligand partner structures. The exact contact surface area values are in Å2. If only a single global contact surface is calculated, a selection named ‘contact’ is created that includes all receptor atoms within 3.9 Å of any ligand atom to illustrate the approximate contact surface. The average G-protein-binding volumes at the cytoplasmic region were carried out in POVME (ref. 55). After filling the defined box with grid points at a 2-Å resolution, the grid points that overlapped with any atom based on van der Waals clashes were removed by ContiguousSeedSphere and ContiguousSeedBox in POVME. A spherical region positioned at the box centre with a 5-Å radius was defined as the core region. Only points that were contiguous with points in the core region were considered. Then the volume was calculated as the summation of the volumes occupied by each remaining point.

Additional information

How to cite this article: Yuan, S. et al. Activation of G-protein-coupled receptors correlates with the formation of a continuous internal water pathway. Nat. Commun. 5:4733 doi: 10.1038/5733 (2014).