Structural basis for recognition of antihistamine drug by human histamine receptor

The histamine receptors belong to the G protein-coupled receptor (GPCR) superfamily, and play important roles in the regulation of histamine and other neurotransmitters in the central

nervous system, as potential targets for the treatment of neurologic and psychiatric disorders. Here we report the crystal structure of human histamine receptor H3R bound to an antagonist

PF-03654746 at 2.6 Å resolution. Combined with the computational and functional assays, our structure reveals binding modes of the antagonist and allosteric cholesterol. Molecular dynamic

simulations and molecular docking of different antihistamines further elucidate the conserved ligand-binding modes. These findings are therefore expected to facilitate the structure-based

design of novel antihistamines.

The biogenic amine histamine plays important pathophysiological roles in both the central nervous system (CNS) and periphery tissues, such as allergy, gastric acid secretion,

neurotransmission, and immune response1. The action of histamine is mediated through four subtypes of G protein-coupled receptors (GPCRs), H1R, H2R, H3R, and H4R2. Antagonists of H1R and H2R

have been clinically used for the treatment of allergies and gastric acid-related diseases, and the H3R inverse agonist Pitolisant (Wakix®) was approved for the treatment of narcolepsy3.

While H4R antagonists are still in the clinical trials for their potential therapeutics in immune-related diseases4. Structures of H1R in complex with the agonist and antagonist have been

determined5,6, providing the molecular mechanisms for ligand recognition and facilitating the structure-based design of novel drugs targeting H1R. However, the molecular mechanisms for

ligand recognition with other histamine receptors were still elusive, due to the lacking of the H2R, H3R, and H4R structures.

H3R is expressed mainly in the brain and acts as an auto- or hetero-receptor in the histaminergic neurons7. As an auto-receptor, H3R modulates the histamine release by the negative

feedback8. While, as a hetero-receptor, H3R regulates the release of various neurotransmitters such as dopamine, γ-aminobutyric acid (GABA), and acetylcholine9. It was suggested that H3R was

associated with several physiological progresses such as sleeping and wakefulness, learning and memory, feeding, and cerebral ischemia10,11,12. Therefore, H3R is a potential target for the

treatment of neurologic and psychiatric disorders, such as sleep disorders, Parkinson’s disease, schizophrenia, Alzheimer’s disease, and cerebral ischemia13,14. The imidazole antagonist of

H3R showed poor penetration through the blood–brain barrier and unwanted interactions with hepatic cytochrome P45015. Thus, great efforts have been devoted to the development of

non-imidazole H3R antagonists15. Here we determine the crystal structure of human H3R bound to a non-imidazole antagonist PF-03654746 at 2.6 Å resolution. The structure, together with the

computational and functional assays, reveals the critical interactions for the ligand binding, as well as the unexpected cholesterol binding at the allosteric site, which could accelerate

the structure-based design of novel antihistamines.

To obtain the stable human H3R proteins for structure determination, the flexible regions of the N-terminal residues 1–26, intracellular loop 3 (ICL3) residues 242–346, and C-terminal

residues 433–445 were truncated, and a thermostabilized apocytochrome b562RIL (BRIL) was inserted at the N-terminus. Additionally, a mutation of S1213.39K (superscript indicates residues

numbers according to the Ballesteros–Weinstein scheme16) at the putative allosteric Na+ binding site was introduced to improve the homogeneity and thermostability of H3R as described in

several GPCR structures determination17,18,19,20,21,22 (Supplementary Fig. 1b, c). In our calcium mobilization assays, the crystallized construct of H3R with S1213.39K mutation could be

activated by histamine with ~3-fold lower efficacy but inhibited by PF-03654746 with ~18-fold higher efficacy (Supplementary Fig. 2, Supplementary Table 1), which was in consistent with our

results that the crystallized H3R-PF-03654746 proteins showed significantly improved homogeneity and thermostability (Supplementary Fig. 1). The crystal structure of H3R in complex with the

antagonist PF-03654746 was determined at 2.6 Å resolution (Fig. 1, Supplementary Fig. 1, Supplementary Table 3).

a Membrane view of H3R–PF-03654746 structure. H3R was shown in forest green ribbons. PF-03654746 was shown in a magenta sphere. The disulfide bond was shown as orange sticks. b, c Structural

comparison of H3R (forest green) with inactive H1R (gray, PDB ID: 3RZE) and active H1R (pink, PDB ID: 7DFL) from extracellular view (b) and intracellular view (c). d Intracellular view

showing a salt-bridge interaction (yellow dashed line) between D1313.49 and R1323.50. The red arrows indicated movements of TMs5/6 and ECL2 in the H3R structure compared to the H1R inactive

structure.

The H3R structure consisted of the canonical seven transmembrane helical bundles (TMs1–7) connected by three extracellular loops (ECLs1–3) and three intracellular loops (ICLs1–3) with an

amphipathic helix 8 (Fig. 1a). The ECL2 of H3R was stabilized by the conserved disulfide bridge between C1073.25 and C188ECL2, and the second disulfide bridge was found between C384ECL3 and

C388ECL3 (Fig. 1a, b). Compared with the inactive H1R structure5, the extracellular tips of TM6 and TM7 in H3R moved inwards by 2.3 and 3.5 Å, respectively (Fig. 1b). Additionally, the first

section of ECL2 shifted towards TM3 by 11 Å and extended from the receptor core, otherwise the antagonist PF-03654746 would clash with ECL2 if it adopted a similar conformation to that in

H1R (Fig. 1b). At the intracellular side, the TM6 of H3R showed an outward movement of 2.8 Å compared to the inactive H1R, whereas the active H1R showed the TM6 outward movement of 12 Å

(Fig. 1c). Moreover, the ICL2 of H3R was found to form an additional helix (Fig. 1c, d). Notably, the Y3.51 of D3.49–R3.50–Y3.51 motif in H3R was substituted by F1333.51, with the salt

bridge formed between D1313.49 and R1323.50, which was a key feature of the inactive state of GPCRs23 (Fig. 1d).

In our H3R structure, PF-03654746 occupies a shallow pocket at the extracellular side, with clear densities for both the receptor and ligand (Fig. 2a). Although the orthosteric binding

pocket of H3R is relatively shallow, an extended binding pocket (EBP) was found around TMs2/7 and ECL2 in H3R, compared to other aminergic receptors24,25 (Fig. 2a). The ligand-binding pocket

of H3R is constituted by the residues mainly from TMs2/3/6/7 and ECL2 (Fig. 2b). At the extracellular side, the carbonyl and N-ethyl-carboxamide moieties of PF-03654746 extends into the EBP

by forming hydrophobic and hydrogen interactions with E3957.36 and Y912.61, respectively (Fig. 2b). In our calcium mobilization assays, the E3957.36A mutant could fully abolish the

PF-03654746 inhibition, while the Y912.61A mutant could significantly decrease the PF-03654746 inhibition by ~46-fold (Supplementary Fig. 3a, Supplementary Table 1). Both Y2.61 and E7.36 are

located in the minor pocket of aminergic GPCRs, which were shown to determine the ligand affinity and selectivity26. Additionally, the 3-fluoro-phenyl moiety of PF-03654746 formed

hydrophobic interaction with F193ECL2 (Fig. 2b). Mutating F193ECL2 to alanine could completely abolish the PF-03654746 inhibition (Supplementary Fig. 3a, Supplementary Table 1). This

phenylalanine on ECL2 was suggested to determine the ligand specificity among the aminergic receptors27,28. Moreover, the hydrophobic interaction with PF-03654746 is seen with Y3746.51 (Fig.

 2b). Mutagenesis of Y3746.51A could fully abolish the PF-03654746 inhibition (Supplementary Fig. 3a, Supplementary Table 1). Notably, the fluorine atom of 3-fluoro-cyclobutane of

PF-03654746 engages a hydrogen bond with C18845.50, and the amine moiety of pyrrolidine of PF-03654746 forms a salt bridge with D1143.32 at the bottom of the pocket (Fig. 2b), which is

highly conserved in the aminergic receptors28. Surprisingly, both D1143.32A and C18845.50A mutations displayed similar PF-03654746 inhibition on the histamine-induced calcium mobilization

compared to the wild-type (Supplementary Fig. 3a, Supplementary Table 1). However, the D1143.32A and C18845.50A mutants showed ~6-fold and ~4-fold reduction of histamine activation,

indicating these two residues might be involved in the binding of both histamine and PF-03654746 (Supplementary Fig. 3a, Supplementary Table 1). Indeed, D3.32 forms hydrogen bonds with

histamine in H1R6.

a Vertical cross section showing a shallow binding pocket in H3R. The extending binding pocket (EBP) of H3R-PF03654746 is shown in a red ellipse. |2Fo|−|Fc| electron density map for the

PF-03654746 contoured at 1.0σ. b Detailed interactions of PF-03654746 in the H3R ligand-binding pocket. H3R was shown in gray ribbons, with critical residues for ligand-binding as cyan

sticks and PF-03654746 as magenta sticks. Hydrogen bonds were shown as yellow dashed lines. c Surface representation of cholesterol-binding site with cholesterol shown in yellow spheres. d

Detailed interactions of cholesterol with H3R. Residues critical for cholesterol binding were shown as orange sticks and cholesterol was shown as yellow sticks. The hydrogen bond was shown

as yellow dashed lines.

Cholesterol has been observed in many GPCR structures for its regulatory roles29,30,31,32,33, at the classical cholesterol consensus motif (CCM)34, as well as diverse binding

sites35,36,37,38,39,40. In the adrenergic receptor β2AR, two cholesterols bound at the CCM stabilizing the receptor conformation34, while two other cholesterols were observed around helix 8

and TM1, modulating the β2AR dimerization39. In the histamine receptors, the cholesterol-binding site was not identified previously. In our structure, the electron density of a cholesterol

molecule is observed around TM1 and TM7 of H3R (Fig. 2c). Cholesterol forms extensive hydrophobic interactions in the extrahelical pocket consisting of F29N-term, L371.35, M411.39, L401.38,

L441.42, T3967.37, Y3937.33, and W3997.40. Especially, the β3-hydroxy head group of cholesterol interacts with E3957.36 through hydrogen bonding (Fig. 2d). Notably, E3957.36 also

participates in the polar interactions with PF-03654736 (Fig. 2b). Our functional assays showed that mutating the negatively charged E3957.36 to uncharged alanine or positively charged

arginine had little effects on the histamine activation, while completely abolishing the PF-03654746 inhibition, indicating that cholesterol binding to E3957.36 might not be critical for

agonist binding and H3R activation, but might potentially to affect antagonist binding and H3R inhibition through an allosteric mode (Supplementary Fig. 3a, Supplementary Table 1).

To investigate the effects of cholesterol binding on H3R, molecular dynamics (MD) simulations were performed on H3R/PF-03654746 complex in the presence and absence of the crystal cholesterol

molecule. Two systems, H3R/PF-03654746/cholesterol (hereafter referred to as CHL) and H3R/PF-03654746 (hereafter referred to as PF), were embedded in the palmitoyl oleoyl

phosphatidylcholine (POPC) bilayer with a duration of 2000 ns, respectively, and each system was replicated to perform three independent simulations. A free-energy landscape was built to

analyze the conformational changes in six 2-μs MD trajectories. RMSDresidues and RMSDPF, representing the root mean square deviations (RMSD) of orthosteric site residues and that of

PF-03654746, respectively, were used as two collective variables of the landscape (Supplementary Fig. 4b). The small value of these parameters means the more approaching to the starting

crystal conformation, while the larger value indicates obvious movements for both protein and PF-03654746.

The free-energy landscape showed three main minima corresponding to three states of the complexes: crystal-like state, state 2, and state 3 (Supplementary Fig. 4a). The crystal-like state

contained snapshots from simulations CHL1, CHL2, and PF3 and displayed the smallest RMSDPF and RMSDresidues, representing the closest conformation to crystal structure. It is associated with

the lowest free energy and is therefore the most stable. With larger RMSDPF and RMSDresidues, snapshots in simulation CHL3 formed state 2, and complexes from PF1 and PF2 fell into state 3.

Both states were different from the crystal conformation and are characterized by higher free-energy values. In the crystal-like state, the PF-03654746-binding geometry was similar to that

in the crystal structure, especially in the middle and bottom of the binding pocket (Supplementary Fig. 4a), where salt bridges with D1143.32 and hydrophobic interactions existed in every

system. In the EBP, PF-03654746 was not that stable and adopted slightly different conformations, forming hydrogen bonds with Y912.61 in CHL1 and CHL2 systems or with Y942.64 in the PF3

system. Though PF-03654746 maintained the stable salt bridge with D1143.32 in state 2, its conformation changed in the middle and external parts of the pocket and only occasionally

interacted with A190ECL2. For state 3, PF-03654746 totally lost its binding pose and rarely interacted with D1143.32, resulting in a random orientation in each MD trajectory. It’s noteworthy

that the cholesterol molecule in CHL3 was not so stable as in CHL1 and CHL2 and eventually dissociated from its binding site at the TM1–TM7 interface (Supplementary Fig. 5a, c), so

cholesterol-bound complexes only existed in simulations CHL1 and CHL2, and both of them were stabilized into the crystal-like conformations. Considering that one out of four

cholesterol-unbound simulations also reproduced the crystal binding mode of PF-03654746, we came to the conclusion that cholesterol at the TM1–TM7 groove was not very stable and not the

determining factor for complex stability, but bound cholesterol facilitated PF-03654746 present in the crystal pose at a higher frequency.

A significant phenomenon observed is that the conserved W3997.40 played an essential role in stabilizing the cholesterol binding and ligand–H3R interactions. W3997.40 predominantly

maintained the original rotameric state (RI-I, χ1 ≈ −80° and χ2 ≈ 100°) in CHL1 and CHL2 (Supplementary Fig. 5b), and cholesterol resided stably in its site, forming a parallel π–π stacking

with W3997.40 (Supplementary Fig. 5a, c). But in the CHL3 simulation, the side chain of W3997.40 flipped out of the TM1–TM7 cleft and pointed outward to the lipids at about 400 ns, resulting

in a new rotamer conformation (RT-II, χ1 ≈ 175° and χ2 ≈ 100°) (Supplementary Fig. 5b, d). The side chain flipping reduced π–π stacking and caused a big steric hindrance for the bound

cholesterol. As a result, cholesterol gradually dissociated from the cleft (Supplementary Fig. 5a, c). Lacking the stabilization of cholesterol, the side chain of W3997.40 turned to another

conformation (RT-III) at about 1200 ns, and RMSDPF and RMSDresidues in CHL3 greatly increased at the same time (Supplementary Fig. 4b). The observations above predicted that cholesterol

regulated the complex dynamics by stabilizing W3997.40 in RI-I state. To verify the role of W3997.40 in ligand binding, we further analyzed the rotameric states of W3997.40 in

non-cholesterol system. As expected, W3997.40 in PF1 and PF2 underwent a certain conformational change, while W3997.40 of PF3 predominantly displayed RI-I state throughout the simulation,

which should contribute to the stable conformation of H3R/PF-03654746 complex obtained in this trajectory (Supplementary Fig. 5f).

To explore how W3997.40 influenced the ligand binding, we examined its interactions with surrounding residues in the crystal structure. W3997.40 formed T-shape π–π stackings with Y912.61,

which was important for the PF-03654746 binding (Supplementary Fig. 5e). Indeed, mutation of W3997.40A could completely abolish the PF-03654746 inhibition, while had little effects on the

histamine activation (Supplementary Fig. 3a, Supplementary Table 1), indicating that cholesterol might affect the PF-03654746 binding mediated by the cholesterol–W3997.40–Y912.61–PF-03654746

interactions. W3997.40 and D1143.32 are completely conserved, and W4027.43 is highly conserved among monoamine receptors. Experiments have independently indicated the importance of W7.40

for the ligand binding in several GPCRs41,42. Therefore, our study provided additional support for this idea and suggested a relevance between cholesterol and the W3997.40–W4027.43–Y912.61

motif.

More importantly, cholesterol facilitated rearrangements of the TM1–TM7 interface and stabilized a polar network of cholesterol–E3957.36–R27N-term. By making extensive hydrophobic contacts

with the extrahelical part of TM1 and TM7, cholesterol joined TM1 and TM7 tightly like a ‘glue’ and promoted the formation of E3957.36–R27N-term salt bridge (Supplementary Fig. 6a–c).

Meanwhile, the hydroxy of cholesterol established a stable hydrogen bond with the carboxyl group of E3957.36 in our simulation, as indicated by the time dependences of their distance

(Supplementary Fig. 5a). Hence, cholesterol–E3957.36–R27N-term polar network remained in CHL1 and CHL2, like in the crystal structure (Supplementary Fig. 6a). In the cholesterol-unbound

simulations, only PF3 possessed the stable E3957.36–R27N-term salt bridge and similar compact conformation in TM1–TM7 interface. As for CHL3, PF1, and PF2, they showed declining stability of

TM1 and TM7, as well as the E3957.36–R27N-term interaction (Supplementary Fig. 6b–e), consistent with their unstable complex states. Accordingly, the tight TM1–TM7–N-term contacts seemed to

be favorable for ligand binding and cholesterol stabilized this receptor conformation through both hydrophobic and electrostatic interactions.

To explore the binding modes of different H3R antagonists, molecular docking studies were used to predict the binding conformations of other 9 H3R antagonists (Fig. 3). PF-03654746 was first

re-docked into the protein to verify the reliability of the docking simulation, which showed the RMSD