Residues remote from the binding pocket control the antagonist selectivity towards the corticotropin-releasing factor receptor-1

Abstract

The corticotropin releasing factors receptor-1 and receptor-2 (CRF₁R and CRF₂R) are therapeutic targets for treating neurological diseases. Antagonists targeting CRF₁R have been developed for the potential treatment of anxiety disorders and alcohol addiction. It has been found that antagonists targeting CRF₁R always show high selectivity, although CRF₁R and CRF₂R share a very high rate of sequence identity. This has inspired us to study the origin of the selectivity of the antagonists. We have therefore built a homology model for CRF₂R and carried out unbiased molecular dynamics and well-tempered metadynamics simulations for systems with the antagonist CP-376395 in CRF₁R or CRF₂R to address this issue. We found that the side chain of Tyr^6.63 forms a hydrogen bond with the residue remote from the binding pocket, which allows Tyr^6.63 to adopt different conformations in the two receptors and results in the presence or absence of a bottleneck controlling the antagonist binding to or dissociation from the receptors. The rotameric switch of the side chain of Tyr356^6.63 allows the breaking down of the bottleneck and is a perquisite for the dissociation of CP-376395 from CRF₁R.

Introduction

The corticotropin-releasing factor (CRF) receptor-1 (CRF₁R) and CRF receptor-2 (CRF₂R) are family B G-protein-coupled receptors (GPCRs) composed of seven transmembrane helices (TM1 –TM7) linked by three intracellular loops (ICL1 –ICL3) and three extracellular loops (ECL1 –ECL3)¹. CRF₁R and CRF₂R belong to the subfamily of CRF receptors and have been identified to be widely distributed throughout the central nervous system and periphery nervous system and act as key regulators of the hypothalamus-pituitary-adrenal axis^2,3,4. It is believed that a well-balanced opposing action between CRF₁R and CRF₂R is responsible for the initiation of and the recovery from an elicited stress response and that a failed adaptation of the two receptors could lead to neuropathology, including anxiety and depression. Recent studies have revealed that CRF₁R and CRF₂R are involved in stress-associated anxiety and depression-like behavior in a more complicated way^4,5,6. Selectively blocking of CRF₁R or CRF₂R with an antagonist is an effective way to treat the neuropathology. Efforts have been made to develop antagonists with high selectivity towards CRF₁R or CRF₂R. Antagonists targeting CRF₁R were among the first allosteric GPCR ligands to be evaluated clinically for treating depression and anxiety related disorders⁷.

In a GPCR subfamily, residues in the ligand binding pocket of the GPCRs are highly conserved, which can lead to the side effects posed by off-target effects⁸. It is interesting to note that sequence conservation in the subfamily of CRF receptors is even higher than in most of the other GPCR subfamilies. CRF₁R and CRF₂R show very high sequence conservation on the helices TM5 and TM6 and the residues that directly interact with the antagonists are identical. However, the antagonist CP-376395 in the crystal structure of CRF₁R shows a 1000 fold lower affinity towards CRF₂R than towards CRF₁R⁹. It has been shown that residues along the ligand binding/dissociation pathway of a target can affect the efficacy of a drug through influencing the binding kinetics of the drug towards its target¹⁰. Therefore, we assume that residues remote from the binding pocket play a role for the selectivity of the antagonist CP-376395.

To study the selectivity of the antagonist CP-376395 towards the receptors CRF₁R and the role of the remote residues in the selectivity, we built a homology model of CRF₂R with CRF₁R as the template and carried out unbiased molecular dynamics simulations and well-tempered metadynamics simulations for both CRF₁R and CRF₂R with CP-376395 binding to them. The dissociation of CP-376395 from CRF₁R or CRF₂R was observed in the well-temped metadynamics simulations. We found that the hydrogen bond between His228^3.40 and Tyr356^6.63 in CRF₁R, which is absent in CRF₂R, plays a pivotal role in controlling the difference of the binding of CP-376395 towards CRF₁R and CRF₂R (Throughout this paper, the superscript on a residue represents the Wootten generic residue numbering¹¹).

Results and Discussion

Homology modeling of CRF₂R

CRF₁R and CRF₂R belong to the same family and share a high sequence identity. The identity rate is 73% if only the transmembrane parts of the receptors are considered. The sequence alignment of CRF₂R to CRF₁R is shown in Figure S1. We can see that the most conserved residues match each other (Table S1). A Richardson plot of the modeled CRF₂R structure indicates that 98% of the residues are located in the allowed regions, reflecting that the structure is geometrically reasonable (Figure S2)¹². The root mean square deviation (RMSD) between the crystal structure of CRF₁R and the modeled structure of CRF₂R is 0.01 Å (Figure 1).

Figure 1

The crystal structure of CRF₁R and the modeled structure of CRF₂R.

The protein structures are shown in the cartoon mode and the antagonist CP-376395 is shown in the stick mode. The structures of CRF₁R and CRF₂R are colored in green and cyan, respectively. The antagonist CP-376395 is colored in yellow. (a) Alignment of the crystal structure of CRF₁R and the modeled structure of CRF₂R; (b) Key residues in the antagonist binding pocket of CRF₁R; (c) Key residues in the antagonist binding pocket of CRF₂R.

Comparison of the structures of CRF₁R and CRF₂R

The residues in the antagonist binding pocket of CRF₁R are very similar to those in the corresponding pocket of CRF₂R, and, in particular, the residues in the first shell around CP-376395 are conserved (Figure 1). The trimethyl-mesitylene motif on CP-376395 forms a T-shaped π-π stacking interaction with the residue Phe313^5.51 in CRF₁R or with the residue Phe280^5.51 in CRF₂R. The nitrogen atom on the dimethylpyridine group of the antagonist forms a hydrogen bond with the side chain of the residue Asn312^5.50 in CRF₁R (Figure 1b) or with the residue Asn279^5.50 in CRF₂R (Figure 1c)¹. It has been suggested that residues Phe232^3.44 and Tyr356^6.63 in CRF₁R work as a bottleneck for the binding of CP-376395 to the binding pocket (Figure 1b). Residues Phe199^3.44 and Tyr323^6.63 in CRF₂R are located in the corresponding region of the bottleneck residues in CRF₁R to restrict the binding of CP-376395 to CRF₂R (Figure 1c). Although the residues directly interacting with the antagonist CP-376395 as described above are conserved in these two receptors, CP-376395 is highly selective towards CRF₁R, with the K_i values 12 nM towards CRF₁R and >10000 nM towards CRF₂R⁹. Some residues along the suggested antagonist binding pathways are different (Figure S3). Such differences lead us to assume that the residues remote from the binding pocket likely affect the binding kinetics to control the selectivity of CP-376395 towards CRF₁R. Therefore, we performed unbiased molecular dynamics simulations and well-tempered metadynamics simulations to explain the differences.

MD simulations

As we can see from the RMSD curves in Figure 2a, CRF₁R is rather stable with the RMSD value less than 2.0 Å during the whole simulation. The RMSD value of CRF₂R is always larger than that of CRF₁R but is still less than 2.5 Å during the whole simulation. We attribute the slightly large RMSD value of CRF₂R to the relaxation of the modeled structure. To evaluate the flexibility of the residues in CRF₁R and CRF₂R, the root mean square fluctuation (RMSF) values of the two receptors are calculated. The RMSF values indicate that in CRF₁R and CRF₂R, TM1-TM7 are much more stable than the loops connecting these helices. The RMSF values of the two receptors calculated from the MD simulation trajectories were found to follow the same trend of those calculated from the B-factors of the x-ray crystallography structure of CRF₁R (Figure S4).

Figure 2

(a) RMSD values of the backbone atoms with respect to the first snapshots in the simulations; (b) Evolution of the tilt angle of TM7 with the pivot point at Ser^7.47 in the simulations. (c) Structural representation of the key residues Ser^1.50, Ser^7.47 and Phe^7.51 which control the kink of TM7 on CRF₁R and CRF₂R. The structures of CRF₁R and CRF₂R are colored in green and cyan, respectively. The key residues are shown as sticks.

The kink of TM7

Helices TM1-TM7 form a bundle for the binding of the antagonist and for the signal transmission in the GPCRs. A sharp kink of TM7 has been observed in the crystal structure of CRF₁R¹. We found that this kink is preserved during the MD simulation and that the kink angle is changing gradually from 135° to 130° with the pivot point at Ser382^7.47 (Figure 2b and 2c). An analysis of the structure of CRF₁R revealed that the kink is stabilized by the conserved residue Ser130^1.50 which forms hydrogen bonds with the backbones of Ser382^7.47 and Phe386^7.51. The two hydrogen bonds were preserved during the MD simulation (Figure S5).

The structure of CRF₂R was built based on the crystal structure of CRF₁R. Therefore, the sharp kink on TM7 and the two hydrogen bonds between the corresponding residues are kept in the modeled CRF₂R structure (Figure 2b and 2c). The hydrogen bond between Ser127^1.50 and Phe353^7.51 is also preserved during the whole MD simulation, but the hydrogen bond between Ser127^1.50 and Ser349^7.47 broke in the beginning of the MD simulation (Figure S5). As a result, a smaller kink angle of TM7 was observed in CRF₂R (Figure 2b and 2c). In family A GPCRs, such a kink has also been observed on TM7 with the residue Pro^7.50 as the pivot point^13,14.

The hydrogen bond between His^2.50 and Glu^3.50

Family B GPCRs lack the sequence motifs to form the conserved ionic lock connecting TM3 and TM6 as found in the inactive conformation of family A GPCRs¹. In the family B GPCRs, a hydrogen bond between the residues His^2.50 and Glu^3.50 is believed to be involved in the receptor activation (Figure 3a)¹⁵. In CRF₁R, the distance between N^δ of His184^2.50 and C^δ of Glu238^3.50 is about 4 Å after 8ns of the simulation and a hydrogen bond is formed between the side chains of His184^2.50 and Glu238^3.50 at the beginning of the simulation and is preserved in the simulation (Figure 3b). This is in line with the observation obtained by Bai. et al¹⁶. It can also be seen from Figure 3b that the distances between N^δ of His184^2.50 and C^δ of Glu238^3.50 are predominantly around 4 Å. In contrast, the hydrogen bond between His^2.50 and Glu^3.50 is not stable in the simulation of CRF₂R. The N^δ of His152^2.50 and C^δ of Glu205^3.50 became closer in about 5 ns to 7 ns but was separated thereafter with most of the distances distributed around 6 Å as indicated in Figure 3b. The ionic lock between TM3 and TM6 in family A receptors is supposed to interconvert between the two states corresponding to the formation and breaking of the ionic lock^17,18,19,20. Our simulation results also revealed that the potentially important hydrogen bond between His^2.50 and Glu^3.50 can interconvert between the formation and breaking of the hydrogen bond in family B GPCRs. Additionally, it is interesting for us to observe that in CRF₂R, Glu205^3.50 formed an ionic lock with Arg148^2.46 after 10 ns of the simulation and this ionic lock is preserved thereafter. The corresponding ionic lock in CRF₁R was formed between Glu238^3.50 and Arg180^2.46 after 28 ns of the simulation and is preserved thereafter (Figure 3c). The free energy landscapes obtained from the metadynamics simulations clearly indicate that the locked state has a free energy of 5 kcal/mol lower than the unlocked state (Figure S6). This reflects that the system prefers to stay in the locked state. Thus, our results are in agreement with the observations from the long MD simulations^18,21.

Figure 3

(a) The final structures of the two receptors from the unbiased MD simulations viewed from the intracellular side. Residues His^2.50, Arg^2.46 and Glu^3.50 are shown in the stick mode; (b) Evolution of the distances between N^δ of His^2.50 and C^δ of Glu^3.50 in the two receptors with the distributions shown in the middle panel; (c) Evolution of the distances between C^ζ of Arg^2.46 and C^δ of Glu^3.50 in the two receptors with the distributions shown in the middle panel.

Ligand fluctuations in the unbiased MD simulations

CP-376395 binds to an unexpected site located in the cytoplasmic half of the receptor CRF₁R, which is about 18 Å away from the putative agonist binding site of the receptor and is about 13–23 Å away from the corresponding small ligand binding site of family A GPCRs¹⁶. Consistent with the observation of Bai et al., small RMSD values (~0.5 Å) of the ligand were obtained from our simulation of CRF₁R. In contrast, CP-376395 shows RMSD values in the binding pocket of CRF₂R larger than in that of CRF₁R as indicated in Figure 4a. The conserved residue Asn^5.50 forms an essential hydrogen bond with the nitrogen on the pyridine ring of CP-376395 and mutation of this residue to Ala results in a complete loss of ligand binding¹. This key hydrogen bond is preserved during the simulations of both CRF₁R and CRF₂R and stabilizes the aryloxy moiety of CP-376395 (Figure 4b). Thus, the larger RMSD value of CP-376395 in CRF₂R is mainly contributed by the fluctuation of the exocyclic alkylamino group.

Figure 4

(a) The RMSD values of the ligand with respect to the first snapshots of the simulations; (b) The distance between the side chain of Asn^5.50 and the nitrogen on the pyridine ring of CP-376395; (c) The distance between the backbone carbon atoms on Tyr^6.63 and His228^3.40 (CRF₁R) or between the backbone carbon atoms on Tyr^6.63 and Val228^3.40 (CRF₂R); (d) The distance between the side chain oxygen atoms on Tyr^6.63 and Gln^5.50; (e) Cross-section view of CRF₁R with the bottleneck formed; (f) Cross-section view of CRF₂R without the bottleneck. In (e) and (f), the antagonist binding pockets are colored in magenta.

The exocyclic alkylamino group is located adjacent to the residues Phe^3.44 and Tyr^6.63 of CRF₁R (Figure 1b) and CRF₂R (Figure 1c), which have been suggested to be the bottleneck for the binding of the antagonist to CRF₁R¹. The residue Tyr356^6.63 is located within the hydrogen bond distance of His228^3.40 in CRF₁R. However, the residue Val195^3.40 of CRF₂R is corresponding to His228^3.40 of CRF₁R and lacks the hydrogen bond donor or acceptor atoms on its side chain to form a hydrogen bond with Tyr323^6.63. In our simulations, the distance between the side chain oxygen atom of Tyr356^6.63 and C^α of His228^3.40 of CRF₁R was preserved during the MD simulation, reflecting that the hydrogen bond formed between the side chains of Tyr356^6.63 and His228^3.40 was stable (Figure 4d). In contrast, the distance between the side chain of Tyr323^6.63 and C^α of Val195^3.40 in CRF₂R became much larger in the first couple of nanoseconds and remained unchanged in the following simulation with a restoration of the initial value between 20 ns and 25 ns. The increase of the distance alters the conformation of the side chain of Tyr323^6.63 in CRF₂R (Figure 4d and Figure S7) and this conformational change is not taking place randomly (Figure S8). Interestingly, the rotameric change of this residue generates a hydrogen bond between the side chains of Tyr323^6.63 and Gln269^6.44. This hydrogen bond was preserved until the end of the 50 ns MD simulation to stabilize the rotameric change of Tyr323^6.63 (Figure 4c and Figure S7). Such rotameric switch results in the breaking down of the bottleneck (Figure 4f), while such a bottleneck is preserved during the simulation of CRF₁R (Figure 4e). We thus assume that such a difference controls the CP-376395 to dissociate from the antagonist binding pocket in CRF₁R and CRF₂R.

Metadynamics simulations

In our well-tempered metadynamics simulations, CP-376395 left the antagonist binding pocket in CRF₁R and CRF₂R and explored the binding pathways to exit the receptors through the helices bundle (Supporting videos). The free energy surfaces (FESs) for CP-376395 leaving the antagonist binding pockets of CRF₁R and CRF₂R are displayed in Figure 5 and Figure 6, respectively.

Figure 5

Metastable states in the dissociation of CP-376395 from CRF₁R.

(a) The binding free energy surface for the dissociation of CP-376395 from CRF₁R as a function of the Z-component of the vector connecting the nitrogen on the dimethylpyridine group of CP-376395 and C^Y on the Asn312^5.50 and χ₁ torsional angle of Tyr356^6.63.The three main energy basins B0-B2 found in the metadynamics simulation are highlighted in b–d, respectively; (b)–(d) Structural characterization of the metastable states B0–B2.

Figure 6

Metastable states in the dissociation of CP-376395 from CRF₂R.

(a) The binding free energy surface for the dissociation of CP-376395 from CRF₂R as a function of the Z-component of the vector connecting the nitrogen on the dimethylpyridine group of CP-376395 and C^γ on the Asn279^5.50 and χ₁ torsional angle of Tyr323^6.63. The three main energy basins B0–B2 found in the metadynamics simulation are highlighted in b–d, respectively; (b)–(d), Structural characterization of the metastable states B0–B2.

Metastable poses of CP-376395 in CRF₁R

For CP-376395 in CRF₁R, we observed three energy minima (basin B0, basin B1 and basin B2) as displayed in the FES (Figure 5a). The deepest free energy minimum in the FES is depicted as basin B0 in Figure 5b and corresponds to the conformation that CP-376395 adopts in the x-ray crystallography structure. This conformation corresponds to the energetically most favorable pose for the binding of CP-376395 in the antagonist binding pocket of CRF₁R. The aryloxy moiety of CP-376395 is holding tightly in the binding pocket by strongly hydrogen-bonding to Asn312^5.50 and by hydrophobic interactions with Leu316^5.54, Ile319^5.57, Thr345^6.42, Leu348^6.45 and Leu349. The exocyclic alkylamino group keeps interacting with Gly353^6.50, Phe232^3.44, Leu309^5.47 and Tyr356^6.63. The mesityl group on CP-376395 forms a T-shaped π-π stacking interaction with Phe313^5.51.

In the metadynamics simulation, CP-376395 moved along the helix bundle to the extracellular side half of the receptor CRF₁R and reached a position sufficiently close for interacting directly with the bottleneck residues Tyr356^6.63 and Phe232^3.44 (Figure 5c). The mesityl group of CP-376395 is placed in a hydrophobic cavity surrounded by Leu309^5.47, Leu348^6.55, Leu349^6.56, Leu352^6.59 and Gly353^6.60. The dimethylpyridine group of the antagonist is surrounded by the residues Ile306^5.44, Gln302^5.40 and Tyr356^6.63. Met305^5.43 also interacts with the dimethylpyridine group. The pyridine ring of CP-376395 forms a face to face π-π stacking interaction with the benzene ring of Phe232^3.44. This π-π stacking interaction can stabilize the ligand in the binding site corresponding to the basin B1. In this basin, the torsional angle χ₁ of the residue Tyr356^6.63 is around π (or - π), while χ₁ of Tyr356^6.63 is about -1 in basin B0. Such a difference indicates that a rotameric change of Tyr356^6.63 is required for CP-376395 moving from the site corresponding to the basin B0 to that corresponding to the basin B1.

Once all the metastable states in the helix bundle were filled, the antagonist CP-376395 crossed the helix bundle to reach the extracellular vestibule of CRF₁R to form a metastable vestibule-bound state, where a local minimum (basin B2) was found as displayed in Figure 5d. The vestibule-bound state is the first step for CP-376395 to enter the antagonist binding site. In this local minimum, the antagonist CP-376395 is stabilized by the hydrophobic interactions with the residues Ile306^5.44, Phe360^6.67 and Tyr356^6.63 and an additional amount of interaction energy can be gained from the relative closeness of the residues Asp298^3.36 and Lys291.

Metastable poses of CP-376395 in CRF₂R

We observe three energy minima (basin B0, basin B1 and basin B2) during the dissociation of the antagonist CP-376395 from CRF₂R as displayed in Figure 6a. Basin B0 represents the energetically most stable site. This basin approximately corresponds to the crystallized conformation of CP-376395 in CRF₁R (Figure 6b). At the basin, the hydrophobic interactions between CRF₂R and CP-376395, which have been observed in the unbiased MD simulation, are kept. The dimethylpyridine group adopts a conformation with the nitrogen atom on the ring hydrogen-bonded to Asn279^5.50, which allows CP-376395 to hold tightly with the antagonist binding pocket. The hydrophobic interactions originating from Leu283^5.54, Ile286^5.57, Leu315^5.55 and Leu316^5.56 further stabilize CP-376395 in this region. The exocyclic alkylamino group keeps interacting with Gly313^6.50, Phe199^3.44, Leu276^5.47 and Tyr323^6.63.

With the action of metadynamics, the antagonist CP-376395 moved along the helix bundle and reached another minimum displayed as basin B1 (Figure 6c). The traverse from basin B0 to basin B1 results in a breaking of the hydrogen bond between the nitrogen on the dimethylpyridine group of CP-376395 and the side chain of Asn279^5.50. This hydrogen bond was found to be preserved during our 50 ns of the unbiased MD simulation, reflecting the inability of the unbiased MD in traversing energy barriers to sample the conformation states at different energy minima. Here, the dimethylpyridine group of CP-376395 resides in the pocket surrounded by Phe191^3.36, Val195^3.40, Ile321^6.61 and Met324^6.64. The mesityl moiety on CP-376395 is in a highly hydrophobic cage defined by Leu316^5.56, Leu319^6.59, Phe199^3.44 and Leu276^5.47. It is interesting to observe the π-π interaction between Tyr323^6.63 and the pyridine ring on CP-376395 to stabilize the pose of CP-376395 in CRF₂R.

We observed another metastable state (basin B2) residing at about 10 Å away from the residue Asn279^5.50 (Figure 6d). As shown in Figure 6a, a large region surrounding basins B1 and B2 is energetically favorable, with the torsional angle ranging from −1 to −π and the distance kept at about 10 Å. This means that the torsional angle can change between 1 and −π with a very low energy barrier. Interestingly, the π-π interactions between Tyr323^6.63 and the pyridine ring on CP-376395 are kept in this energetically favorable region. On the other hand, the hydrogen bond between Tyr323^6.63 and Asn269^5.40 is broken in basins B1 and B2. We thus suggest that the π-π interactions between Tyr323^6.63 and the pyridine ring on CP-376395 compensate for the energy required for the breaking of the hydrogen bond between Tyr323^6.63 and Asn269^5.40.

Comparison of the ligand dissociation pathways and energy profiles

Three energy minima were identified for the dissociation of CP-376395 from both CRF₁R and CRF₂R as displayed in Figure 5 and Figure 6, respectively. As depicted in the figures, the basin B0 corresponds to the region located about 1–2 Å away from the key residue Asn^5.50 with the torsional angle χ₁ of the residue Tyr^6.63 stabilized at −1 in CRF₁R or −π (or π) in CRF₂R. With CP-376395 moving from the basin B0 to the basin B1, the torsional angle χ₁ of the residue Tyr356^6.63 in CRF₁R switches from −1 to −π (or π) while the corresponding angle in CRF₂R stays at −1 (Figure S). Additionally, Phe^3.44 displays no significant conformational change in neither the unbiased MD simulations nor in the metadynamics MD simulations of CRF₁R or CRF₂R. These observations strongly support our suggestion that the conformational change of Tyr356^6.63 plays a pivotal role for CP-376395 binding to or dissociation from the antagonist binding site of CRF₁R. With the formation of the hydrogen bond between the side chains of the residues Tyr323^6.63 and Gln269^6.44 in CRF₂R, the crucial role of Tyr^6.63 which controls the antagonist binding to or to dissociation from the binding site in CRF₁R does not occur in CRF₂R.

Bai et al. explored the dissociation pathway of CP-376395 from CRF₁R by using random acceleration molecular dynamics simulations¹⁶. They found that breaking of the hydrogen bond between CP-376395 and Asn312^5.50 results in the first energy barrier for the dissociation. This observation was confirmed from our study. Besides, we found that breaking of the hydrogen bond formed between Tyr356^6.63 and His228^3.40 also contributes to this barrier. The barrier is about 5 kcal/mol from our simulation, while it is 9.9 kcal/mol from Bai's work. Bai and coworkers also found the second energy barrier for the dissociation of CP-376395 along the pathway they detected because there exist two hydrogen bonds formed by CP-376395 with His228^3.40 and Gln302^5.40 in the pathway. However, these two hydrogen bonds were not observed along the CP-376395 dissociation pathway in our study. We found that CP-376395 forms π- π stacking interactions with Tyr356^6.63, which was not described by Bai. et al. The π-π stacking interactions, together with the interactions between CP-376395 and the remaining residues, form the second energy barrier of about 4 kcal/mol. This barrier is much lower than the one obtained by Bai et al., which is 11.4 kcal/mol.

Implications for the drug design

The binding free energy for CP-376395 towards CRF₁R, averaged over those from the first and repeated runs (−11.42 and −11.27 kcal/mol, respectively), is −11.35 kcal/mol, while that for CP-376395 towards CRF₂R, averaged over those from the two simulations (−8.18 kcal/mol and −8.03 kcal/mol, respectively) is −8.10 kcal/mol (Figures S9 and S10). The binding free energy for the CP-376395 towards CRF₁R is in agreement with the experimental value (−10.87 kcal/mol) while that for the CP-376395 towards CRF₂R is a little larger than the experimental result (weaker than −6.86 kcal/mol)⁹. The decrease of the antagonist binding affinity from CRF₁R to CRF₂R is in agreement with the experimental results.

Mutation of His228^3.40 has also been performed to evaluate the effect of His228^3.40 on the binding of the antagonist NBI27914, which shares a similar scaffold to CP-376395, towards CRF₁R²². Compared to the 1000-fold higher binding affinity of CP-376395 towards CRF₁R than towards CRF₂R, the His288Val mutation only leads to a 40-fold lower binding affinity of NBI27914 towards CRF₁R. Such a difference likely comes from two aspects. One aspect is that the exocyclic alkylamino group in NBI27914 is bigger than that in CP-376395, which results in the hydrogen bond between His228^3.40 and Tyr356^6.63 being less stable in the binding of NBI27914 to CRF₁R. Another aspect is that the decrease of the binding affinity of CP-376395 to CRF₂R is likely contributed partially by the difference in the rotameric properties of Tyr^6.63 in CRF₁R and CRF₂R.

Recent studies have indicated that the binding kinetics of a ligand towards its target could be one of the most crucial factors for sustainable drug efficacy and in some cases, even more important than the binding affinity in determining the drug efficacy^23,24. The recently published crystal structure of the smoothened receptor by Stevens's group revealed that there are multiple distinct binding sites for the ligand in the helix bundle of the receptor^25,26. Both smoothened receptor and CRFR family receptors possess deep ligand binding cavities. This allows us to suggest that the sites along the pathway of an antagonist binding to the CRFR family of receptors, especially those corresponding the basins and saddle points we discovered in our metadynamics simulations, are important in the design of new drug candidates with attenuated side effects and chemoresistance.

Conclusion

In this work, we have carried out homology modeling to build the structure of CRF₂R with the crystal structure of CRF₁R as the template. Based on the crystal structure of CRF₁R and the homology model of CRF₂R, we performed unbiased MD simulations as well as well-tempered metadynamics simulations to investigate the origin of the selectivity of the antagonist CP-376395 towards the two receptors. From the unbiased MD simulations, we found that in CRF₁R the oxygen atom on Tyr356^6.63 forms a hydrogen bond with the side chain of His228^3.40 which allows the formation of a bottleneck consisting of the residues Phe232^3.44 and Tyr356^6.63, while in CRF₂R, the side chain oxygen on Tyr323^6.63 is hydrogen bonded to the side chain of Gln269^6.44, leading to a lack of such a bottleneck. The existence of the bottleneck in CRF₁R and its absence in CRF₂R provide an explanation for the origin of the high selectivity of the antagonist CP-376395 towards CRF₁R. The metadynamics simulations provided an even stronger support for that explanation. The rotameric switch of the side chain of Tyr356^6.63 results in the breaking down of the bottleneck in CRF₁R and is a prerequisite for the dissociation of CP-376395 from CRF₁R, but it is not required for the dissociation of CP-376395 from CRF₂R as indicated by the FES. Thus, our studies provide important structural information in explaining the origin of the high selectivity of CP-376395 towards CRF₁R.

Methods

Homology modeling of CRF₂R

The crystal structure of CRF₁R (PDB entry ID: 4K5Y)¹ was used as the template for the homology modeling of the structure of CRF₂R. ClustalW2 was used for the sequence alignments and manual adjustments were carried out to guarantee no gaps in the secondary structures²⁷. Prime 3.5 was used to build the model of CRF₂R^28,29.

Unbiased molecular dynamics simulations

System preparation. Two systems, one containing the crystal structure of CRF₁R and the other the modeled structure of CRF₂R, were built for the simulations. A POPC (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine) bilayer with the surface area of 75 Å × 75 Å on the X-Y plane was constructed using VMD¹³. For each system, the receptor was first embedded into the POPC bilayer using our in-house program according to the orientations provided by the OPM database³⁰. The antagonist CP-376395 was placed in the antagonist binding pocket. A box of 75 × 75 × 100 ų with water molecules was then used to solvate the protein. Lipid molecules within 0.85 Å of the heavy atoms on the protein structure and water molecules in the bilayer were removed. Thereafter, sodium and chloride ions were added to produce the neutral system of 0.15 M NaCl. The resulting systems are summarized in Table 1.

Table 1 Systems prepared for the MD simulations

Simulation details. MD simulations were performed using Gromacs 4.6.5^31,32 with the CHARMM36 parameters for the proteins, lipids and ions and the TIP3P model for water. Force field parameters for the ligand molecule were generated with the CHARMM General Force Field (version 2b8) interface (version 0.9.7.1 beta)³³ and were listed in Appendix 1. Three steps were used to equilibrate each system. In the first step, the system was subject to a 50000-step energy minimization with 1000.0 kJ/mol/nm as the force threshold. Then, the system was relaxed by an MD simulation of 100 ps with 1 fs as the time step using the NVT ensemble. In the last step, the system underwent an NPT MD simulation for 1 ns with the time step of 2 fs for equilibration.

After the equilibration run, each system was simulated for 50 ns using the NPT ensemble with the temperature and pressure set to 310 K and 1 bar, respectively. The Nose-Hoover thermostat and the Parrinello-Rahman pressure coupling were applied during the simulation. The bonds containing hydrogen atoms were constrained with the LINCS algorithm and a time step of 2 fs was used. The cubic periodic boundary conditions were applied. The cut-offs for the electrostatic and van der Waals interactions were set to 12 Å, with the long range electrostatic interactions recovered by the Particle Mesh Ewald summation.

Metadynamics simulations

Theory. Metadynamics^34,35,36 has been successfully applied in describing the selectivity of a ligand towards different targets³⁷. We performed metadynamics simulations to detect the residues that are relevant to the selectivity of CP-376395 towards CRF₁R and CRF₂R. In a metadynamics simulation, an additional history-dependent biased potential V_G(S, t) was introduced into the system,

where t represents time, S represents the collective variables, ω is the energy rate and σ_i controls the width of the Gaussian for the ith collective variable. With the evolution of the system, the wells in the FES of the collective variables are filled up with the biased potential V_G. The underlying free energy −F(S) is assumed to be estimated from the biased potential once all the wells have been filled after a sufficiently long time,

The correctness of the relationship as shown in equation 2 has proven to be empirical by extensive tests under the assumption that the stochastic dynamics in the collective variable space is memoryless in the absence of the bias. Under the assumption, the error in FES construction has proven to be:

Where D is the intrinsic system diffusion coefficient in the collective variable space, k_B is the Boltzmann constant and T is the temperature of the system³⁸.

In fact, if one is interested in reconstructing the free energy surface from a metadynamics simulation, the simulation should be stopped once the biased potential fulfills the underlying FES in the region of interest. If the simulation does not stop as soon as the system exits from the minima, the biased potential would overfill the minima and push the system to high energy regions with respect to the collective variable space. To solve the problem, a “well-tempered” and “smoothly converging” algorism is introduced by Barducci et al³⁹. In the well-tempered metadynamics, the deposition rate for the biased potential decreases by rescaling the Gaussian height (W) over the simulation time

where V_G(S, t) is the biased potential at the current position and current time, τ_G is the deposition stride and ΔT is a temperature-like parameter. The underlying free energy is a scaled approximation to the V_G(S, t), with

With respect to the standard metadynamics, the biased potential decreases as 1/t when the simulation proceeds, which allows to smoothly converge to an approximation of F(S).

Simulation details. We have carried out 50 ns well-tempered metadynamics simulations for the systems of CRF₁R and CRF₂R with the antagonist CP-376395 in their antagonist binding pockets. For each system, the metadynamics simulations were carried out two times, with the last snapshot from the unbiased MD simulation used as the initial structure for the simulations. The metadynamics simulations were carried out using plumed 2.02 implemented in Gromacs 4.6.5. The collective variables were selected based on the unbiased MD simulations. The residues Phe^3.44 and Tyr^6.63 have been suggested to be working as the bottleneck for the binding of CP-376395 to its binding pocket¹. We have observed a rotational switch of Tyr323^6.63 in CRF₂R in the unbiased MD simulation. This rotational switch opens the bottleneck that controls the binding of CP-376395 to the antagonist binding pocket. Thus, the χ₁ torsional angle of Tyr^6.63 was selected to control the opening and closing of the bottleneck and used as the first collective variable. The nitrogen on the dimethylpyridine group of the antagonist forms a hydrogen bond with the side chain of Asn^5.50 in CRF₁R and CRF₂R. This hydrogen bond plays a pivotal role in the binding of CP-376395 to the antagonist binding pockets. In addition, a dissociation pathway along the Z-axis from the antagonist binding pocket to the extracellular side of the receptor CRF₁R was proposed base on the random acceleration MD simulations¹⁶. We thus select the Z-component of the vector connecting the nitrogen on the dimethylpyridine group of CP-376395 and C^Y on the Asn^5.50 as the second collective variable. For the metadynamics simulations, the biasing potential was added every 250 steps, with the width and height of the Gaussian hills set to 0.05 and 0.3 kJ/mol, respectively and ΔT = 2700.