A structural chemogenomics analysis of aminergic GPCRs: lessons for histamine receptor ligand design

Abstract

Background and Purpose

Chemogenomics focuses on the discovery of new connections between chemical and biological space leading to the discovery of new protein targets and biologically active molecules. G-protein coupled receptors (GPCRs) are a particularly interesting protein family for chemogenomics studies because there is an overwhelming amount of ligand binding affinity data available. The increasing number of aminergic GPCR crystal structures now for the first time allows the integration of chemogenomics studies with high-resolution structural analyses of GPCR-ligand complexes.

Experimental Approach

In this study, we have combined ligand affinity data, receptor mutagenesis studies, and amino acid sequence analyses to high-resolution structural analyses of (hist)aminergic GPCR-ligand interactions. This integrated structural chemogenomics analysis is used to more accurately describe the molecular and structural determinants of ligand affinity and selectivity in different key binding regions of the crystallized aminergic GPCRs, and histamine receptors in particular.

Key Results

Our investigations highlight interesting correlations and differences between ligand similarity and ligand binding site similarity of different aminergic receptors. Apparent discrepancies can be explained by combining detailed analysis of crystallized or predicted protein-ligand binding modes, receptor mutation studies, and ligand structure-selectivity relationships that identify local differences in essential pharmacophore features in the ligand binding sites of different receptors.

Conclusions and Implications

We have performed structural chemogenomics studies that identify links between (hist)aminergic receptor ligands and their binding sites and binding modes. This knowledge can be used to identify structure-selectivity relationships that increase our understanding of ligand binding to (hist)aminergic receptors and hence can be used in future GPCR ligand discovery and design.

Linked Articles

This article is part of a themed issue on Histamine Pharmacology Update. To view the other articles in this issue visit http://dx.doi.org/10.1111/bph.2013.170.issue-1

Introduction

The aim of chemogenomics is to derive predictive links between the chemical structures of biologically active molecules and the protein targets with which these molecules interact (Klabunde, 2007; Rognan, 2007). Based on the assumption that similar proteins bind similar ligands (and vice versa) knowledge on the molecular determinants of protein-ligand interactions can be used to identify novel ligands for a given target or a novel target for a given ligand. Generally, chemogenomic analyses are based on the comparison of the molecular and structural properties of ligands, protein targets or ligand-protein complexes (Keiser et al., 2007; Klabunde, 2007; Rognan, 2007; Jacoby, 2009; Garland and Gloriam, 2011b). The family of GPCRs is a particularly interesting system for chemogenomic analyses for several reasons: (i) GPCRs are targeted by ∼30% of the currently marketed drugs (Overington et al., 2006); (ii) large experimental GPCR ligand binding data sets are available (Knox et al., 2011; Gaulton et al., 2012); (iii) GPCRs share ligands within and between receptor subfamilies [Jacoby et al., 2006; Brianso et al., 2011; Besnard et al., 2012; Sanders et al., 2012) (and with other protein families (Morphy and Rankovic, 2005; Keiser et al., 2009; de Graaf et al., 2013)], (iv) GPCRs consist of seven transmembrane helices (7TM) that share a similar fold (Katritch et al., 2012) (Figure 1A), which allows the definition of a generic GPCR ligand binding site consisting of a small set of residues at conserved locations in the TM helices (Figure 1B,C) (Surgand et al., 2006; Gloriam et al., 2009). Until the past few years, chemogenomic analyses of GPCR-ligand interactions have been limited to protein information derived from sequence alignments (Attwood and Findlay, 1994; Kolakowski, 1994) or GPCR homology models based on the bovine rhodopsin crystal structure, the first (and for long time only) solved GPCR crystal structure (Palczewski et al., 2000). These studies have been used to identify privileged GPCR ligand scaffolds (Jacoby et al., 2006; Johansson et al., 2013) and complementary structural or sequence motifs in GPCR binding sites (Bondensgaard et al., 2004; Garland and Gloriam, 2011a; Surgand et al., 2006) that offer insight in GPCR ligand selectivity profiles (Besnard et al., 2012; de Graaf et al., 2013), the construction of ligand- and protein-based virtual screening models for GPCR ligands (Klabunde et al., 2009; Weill and Rognan, 2009) and receptor deorphanization (Gloriam et al., 2011; Weill, 2011). The increased number of GPCR crystal structures in the past 5 years (Jacobson and Costanzi, 2012), in particular of the aminergic GPCR subfamily (27 structures for 6 of the 42 aminergic receptors) (Surgand et al., 2006; Katritch et al., 2012), now for the first time allow to combine chemogenomic studies with high-resolution structural analyses of GPCR-ligand complexes. In addition to the emerging information on GPCR structures, also more complete GPCR ligand data sets are becoming more and more accessible to further push the limits of (structural) chemogenomic investigation of GPCR-ligand interaction space. The binding affinities of large numbers of small molecule ligands against many individual GPCRs have been determined over the past decades, but only in the past few years these data can be systematically analysed in publically accessible libraries of protein-target annotated ligands [e.g. ChEMBL (Gaulton et al., 2012), DrugBank (Knox et al., 2011), BindingDB (Chen et al., 2001)]. Secondly, the first consistent and complete experimental screening data of (small fragment-like) ligand libraries against multiple GPCR targets are being reported (Besnard et al., 2012; de Graaf et al., 2013). Our structural chemogenomics study combines these emerging data and insights on the aminergic GPCR family, a receptor family that has been extensively investigated by ligand structure-activity/affinity relationships (SAR) and site-directed protein mutagenesis studies (Table 1, Supporting Information Table S1) (Paolini et al., 2006; Shi and Javitch, 2002; Surgand et al., 2006). We will show how this analysis can be used to elucidate the molecular determinants of ligand binding to a particular subfamily of aminergic GPCRs, namely the histamine receptor family, important players in allergy, acid secretion, inflammation and CNS disorders (Engelhardt et al., 2009; Kiss and Keseru, 2012; Kuhne et al., 2011; Parsons and Ganellin, 2006; Simons and Simons, 2011).

Figure 1.

(A) Conserved TM–fold of crystallized aminergic GPCRs (one crystal structure per receptor). The co-crystallized ligand doxepin (1) in H₁R is depicted using black carbon atoms. (B) Top view of the H₁R with doxepin (1, black carbon atoms). Magenta spheres depict C-alpha atoms from the binding pocket residues. The side chain of the key ionic anchor D^3.32 is displayed in black. Both the major and minor binding pocket are highlighted. (C) Sequence alignment of putative binding site residues of the human H₁R, human H₂R, human H₃R, human H₄R, human M₂R, rat and human M₃R (the pocket residues for M₃R rat and human are identical); human α₁AR, turkey α₁AR, human α₂AR and human D₃R. The lower case character preceding the receptor abbreviation indicates the species, h (human), r (rat) and m (turkey). Binding site residues are assigned based on the basis of 30 residues proposed by (Surgand et al., 2006) plus an additional 24 residues based on the six aminergic crystal structures and SDM studies (Table 1, Supporting Information Table S1) (Vroling et al., 2011). Residues in contact with the ligand in the crystal structure are coloured cyan. Magenta highlights residue W^7.40, which is an aminergic family specific conserved residue. The conserved residue D^3.32 is coloured red. Capital letters at the bottom indicate a partially conserved residue (>75%) and * indicates a fully conserved residue. All cysteines that form a disulphide bridge are highlighted in yellow.

Table 1.

Binding affinities from site directed mutagenesis studies on histamine receptors and crystallized aminergic GPCRs

Ligand	Reference	WT	2.57	2.61	2.64	3.24	3.28	3.32	3.33	3.36	3.37	3.40	4.57	45.49	45.50	45.51	5.39	5.42	5.43	5.46	6.48	6.51	6.52	6.55	7.31	7.39	7.42
	Histamine H1 receptor		V	N	Y	L	W	D	Y	S	T	I	V	K	C	E	K	T	A	N	W	Y	F	F	N	I	G
8	Moguilevsky et al. (1995)	5.9																5.2 (A)		3.6 (A)
19	Moguilevsky et al. (1995)	7.0																8.0 (A)
8	Gillard et al. (2002)	5.9															4.7 (A)	5.2 (A)
18	Gillard et al. (2002)	8.5															7.9 (A)	8.7 (A)
19	Gillard et al. (2002)	7.1															6.3 (A)	8.2 (A)
16	Nonaka et al. (1998)	8.6						7.3 (A)										8.4 (A)		8.4 (A)
1	Nonaka et al. (1998)	9.2						6.7 (A)
8	Ohta et al. (1994)	4.8						N.D. (A)										4.2 (A)		<2.0 (A)
8	Bruysters et al. (2004)	4.3						N.D. (A)									3.9 (A)			2.6 (A)			N.D. (A)	2.7 (A)
8	Wieland et al. (1999) a	4.1															3.5 (A)						3.7 (M)	4.5 (M)
17	Wieland et al. (1999) a	8.0															6.3 (A)
19	Wieland et al. (1999) a	7.2															6.3 (A)
8	Leurs et al. (1995) a	4.5															3.8 (A)
9	Leurs et al. (1994) a	9.1																9.3 (A)		8.3 (A)
12	Leurs et al. (1994) a	7.2																7.4 (A)		6.5 (A)
8	Bakker et al. (2004)	4.2					<2.2 (A)											3.7 (A)		3.0 (A)
	Histamine H2 receptor		V	S	Y	F	Y	D	V	C	T	I	S	K	C	K	G	D	G	T	W	Y	F	F	N	L	G
8	Gantz et al. (1992) b	-						↓ (N)										↓ (A)		↓ (A)
	Histamine H3 receptor		C	Y	Y	L	W	D	Y	C	T	A	Y	H	C	Y	L	A	S	E	W	Y	T	M	P	F	L
8	Uveges et al. (2002)	7.8															7.7 (A)	7.2 (Q)		4.5 (A)
29	Yao et al. (2003)	9.4									9.1 (A)	10.5 (V)
21	Yao et al. (2003)	7.2									8.2 (A)	8.4 (V)
22	Yao et al. (2003)	6.1									7.6 (A)	7.4 (V)
	Histamine H4 receptor		S	Y	H	I	W	D	Y	C	T	V	N	E	C	E	L	T	S	E	W	Y	S	T	K	F	Q
31	Schultes et al. (2012)	7.1															6.5 (V)			6.7 (Q)
32	Schultes et al. (2012)	7.7															6.4 (V)			7.5 (Q)
33	Schultes et al. (2012)	8.3															7.7 (V)			7.2 (Q)
34	Schultes et al. (2012)	8.7															7.2 (V)			6.7 (Q)
29	Istyastono et al. (2011b)	8.1										8.0 (A)	7.8 (Y)					8.6 (A)		7.3 (Q)				8.8 (M)			9.0 (L)
30	Istyastono et al. (2011b)	7.5										7.2 (A)	6.5 (Y)					7.9 (A)		6.8 (Q)				7.9 (M)			7.9 (L)
8	Jongejan et al. (2008)	7.7											N.D. (A)							5.0 (Q)
48	Jongejan et al. (2008)	6.3											N.D. (A)							6,2 (Q)
48	Lim et al. (2010)	6.4											4.7 (H)				7.0 (V)		6.6 (L)
32	Lim et al. (2010)	8.3											7.7 (H)				7.1 (V)		7.8 (L)
8	Shin et al. (2002)	7.8						N.D. (A)					7.5 (T)					7.3 (A)	7.5 (A)	N.D. (A)			7.1 (F)
8	Shin et al. (2002)	7.8						N.D. (E)					7.2 (A)							6.8 (D)			7.5 (A)
8	Shin et al. (2002)	7.8						N.D. (N)												N.D. (N)
	Dopamine D3 receptor		V	V	L	C	F	D	V	C	T	I	S	V	C	S	V	S	S	S	W	F	F	H	S	T	G
35	Dorfler et al. (2008)	7.1						6.0 (E)														<4.1 (W)
37	Dorfler et al. (2008)	8.4						5.9 (E)														6.6 (W)
38	Dorfler et al. (2008)	8.6						5.8 (E)														7.5 (W)
39	Ehrlich et al. (2009)	6.1		5.6 (F)	6.1 (F)			6.2 (E)																7.2 (A)
40	Ehrlich et al. (2009)	7.6		6.1 (F)	6.4 (F)			6.2 (E)																7.9 (A)
35	Alberts et al. (1998)	8.4								7.0 (S)
36	Lundstrom et al. (1998)	8.6				8.8 (H)																		7.6 (L)	8.7 (A)	9.1 (V)
37	Lundstrom et al. (1998)	8.3				8.9 (H)																		8.4 (L)	8.6 (A)	9.3 (V)
35	Lundstrom et al. (1998)	7.2				7.1 (H)																		6.0 (L)	7.1 (A)	7.9 (V)
41	Lundstrom et al. (1998)	8.6				9.1 (H)																		7.9 (L)	9.1 (A)	9.4 (V)
35	Sartania and Strange (1999)	7.4																5.7 (A)	7.1 (A)	7.5 (A)
	Muscarinic M2 receptor		S	Y	Y	V	W	D	Y	S	N	V	W	E	C	Y	T	T	A	A	W	Y	N	V	P	Y	C
44	Gregory et al. (2010)	10.0		9.6 (A)			9.6 (A)		8.4 (A)	9.7 (A)						10.0 (A)						9.49 (A)
42	Gregory et al. (2010)	5.1		4.9 (A)			4.5 (A)		2.5 (A)	3.8 (A)						5.0 (A)						3.06 (A)
49	Gregory et al. (2010)	5.7		7.1 (A)			8.8 (A)		5.2 (A)	5.7 (A)						6.2 (A)						6.27 (A)
47	Gregory et al. (2010)	6.2		5.9 (A)			6.3 (A)		5.8 (A)	5.7 (A)						6.2 (A)						5.85 (A)
45	Heitz et al. (1999)	9.6					9.6 (A)	N.D. (A)					N.D. (A)				9.1 (A)	9.3 (A)			N.D. (A)	N.D. (A)	N.D. (A)
44	Heitz et al. (1999)	9.7					9.4 (A)	N.D. (A)					8.5 (A)				9.0 (A)	9.8 (A)			9.1 (A)	N.D. (A)	N.D. (A)
42	Hulme et al. (1995)	5.6						3.6 (E)
	Muscarinic M3 receptor		S	F	Y	A	W	D	Y	S	N	V	W	E	C	F	T	T	A	A	W	Y	N	V	P	Y	C
45	Schmidt et al. (2003) c	9.3															10.0 (A)			8.5 (T)				9.6 (I)
42	Wess et al. (1993) c	6.1																			4.8 (F)
45	Wess et al. (1991) c	10.3	9.3 (A)						9.9 (F)								10.4 (A)	10.3 (A)				10.1 (F)				9.7 (F)
42	Wess et al. (1991) c	5.8	5.4 (A)						4.7 (F)								4.7 (A)	4.2 (A)				4.1 (F)				4.4 (F)
42	Bluml et al. (1994) c	5.7																					5.0 (A)
43	Bluml et al. (1994) c	10.5																					9.5 (D)
46	Bluml et al. (1994) c	11.0																					6.5 (D)
	β2 adrenoceptor		V	G	H	W	W	D	V	V	T	I	S	C	C	D	A	S	S	S	W	F	F	N	R	N	G
7	Dohlman et al. (1990)	9.9												8.5 (V)	8.5 (V)
50	Dohlman et al. (1990)	8.4												6.8 (V)	6.2 (V)
52	Dohlman et al. (1990)	8.2												6.9 (V)	6.9 (V)
53	Dohlman et al. (1990)	9.4												7.3 (V)	7.2 (V)
7	Fraser (1989)	8.0												7.2 (S)	5.2 (S)
50	Fraser (1989)	6.7												4.7 (S)	4.3 (S)
56	Fraser (1989)	9.7												7.7 (S)	6.2 (S)
55	Fraser (1989)	7.7												5.5 (S)	5.0 (S)
53	Liapakis et al. (2000)	6.9																6.5 (T)	5.4 (A)
7	Suryanarayana and Kobilka (1993)	6.3																								4.2 (T)
54	Suryanarayana and Kobilka (1993)	5.7																								4.0 (T)
51	Ambrosio et al. (2000)	6.7																	5.3 (A)	5.5 (A)
7	Wieland et al. (1996)	6.6											6.7 (A)											5.6 (L)
50	Wieland et al. (1996)	5.0											5.1 (A)											4.8 (L)
54	Wieland et al. (1996)	6.8																						5.2 (L)

The annotated binding affinities (pIC₅₀/pK_i) for the mutants are followed by their respective mutation in brackets. Human (unless otherwise specified) SDM binding affinity data was collected from GPCRdb (Vroling et al., 2011) and an exhaustive literature search. The complete overview is available in the supporting information (Table S1) from which a subselection of single-point mutations within the predefined binding pocket is presented here.

^aSDM study on guinea pig H₁R.

^bOnly functional SDM data available for histamine (8) in H₂R (Gantz et al., 1992).

^cSDM study on rat M₃R.

Conserved bioaminergic GPCR ligand binding site

While the crystal structures of aminergic, adenosine, chemokine, lipid, opioid, opsin and peptide GPCR subfamilies show differences in helical bends [e.g. TM2 in CXCR4 (Wu et al., 2010)] and relative orientations of helices [e.g. TM5 in β₂AR (Katritch and Abagyan, 2011)], the overall protein fold around the TM binding pocket is well conserved, in particular when comparing GPCRs of the same family (Figure 1A) (Katritch et al., 2012). Generally, two subpockets in which ligands can bind are defined within the GPCR TM bundle, i.e. a minor pocket consisting of TMs 1, 2, 3, and 7 and a major pocket consisting of TMs 3, 4, 5, and 6 (Figure 1B) (Surgand et al., 2006).

Throughout this manuscript, we use the Ballesteros-Weinstein residue numbering scheme (Ballesteros and Weinstein, 1995) that is based on the presence of several highly conserved residues among class A GPCRs: N^1.50 in TM 1, D^2.50 in TM2, R^3.50 in TM3, W^4.50 in TM4, P^5.50 in TM5, P^6.50 in TM6 and P^7.50 in TM7. D^3.32, for example, is part of TM3 and is located 18 residues before the highly conserved R^3.50 (Supporting Information Figures S1 and S3). We have defined the ligand binding pocket in the TM domain of aminergic GPCRs by considering 54 positions: 30 residues defined by Surgand et al. (2006) based on the bovine rhodopsin crystal structure and 24 additional residues that are accessible from the ligand binding pocket in six aminergic GPCR crystal structures (Figure 1C) and have been investigated in site-directed mutagenesis studies (Table 1, Supporting Information Table S1).

GPCRs have been divided in seven classes (Kolakowski, 1994) (class A-F and O), five families (Fredriksson et al., 2003) (rhodopsin, glutamate, secretin, adhesion and frizzled/taste), and many subfamilies (Surgand et al., 2006). Up to now crystal structures have been published for 15 different class A GPCRs, including six GPCRs targeted by biogenic amines (i.e. aminergic GPCRs): human histamine H₁ receptor, human dopamine D₃ receptor, human muscarinic M₂ receptor, rat muscarinic M₃ receptor, turkey β₁ adrenoceptor and human β₂ adrenoceptor (Jacobson and Costanzi, 2012; Katritch et al., 2012). These GPCRs are abbreviated to H₁R, D₃R, M₂R, M₃R, β₁AR and β₂AR, respectively and will be used throughout this paper (Alexander et al., 2011). For β₁AR and β₂AR multiple crystal structures with different ligand types are available [i.e. antagonists, inverse agonist and (partial/full/biased) agonist bound], whereas for the other aminergic GPCRs only one crystal structure bound to an inverse agonist or antagonist has been solved (Jacobson and Costanzi, 2012; Katritch et al., 2012). An overview of all aminergic GPCR crystal structures can be found in Supporting Information Table S2.

The conserved fold of GPCRs has enabled the construction of protein homology models including histamine receptor models (Wieland et al., 1999; Kelley et al., 2001; Jongejan et al., 2005; 2008; Schlegel et al., 2007; Jojart et al., 2008; Kiss et al., 2008b; Igel et al., 2009; Strasser et al., 2009; Lim et al., 2010; Werner et al., 2010; Istyastono et al., 2011b; Schultes et al., 2012; Sirci et al., 2012; Seifert et al., 2013) to predict GPCR-ligand interactions (de Graaf and Rognan, 2009; Kooistra et al., 2013). GPCR homology models have furthermore been successfully used to discover new ligands by structure-based virtual screening (de Graaf and Rognan, 2009; Kooistra et al., 2013), as demonstrated for H₃R (Schlegel et al., 2007; Sirci et al., 2012) and H₄R (Kiss et al., 2008a; Istyastono, 2012). Most aminergic GPCR models so far have been constructed based on the bovine rhodopsin (Palczewski et al., 2000) or β₂AR crystal structure (Figure 2E) (Cherezov et al., 2007), but the recently solved GPCR crystal structures now in principle offer more (closely related) templates to construct higher resolution homology models (Istyastono et al., 2011b; Schultes et al., 2012). For example, H₄R models based on the H₁R crystal structure could better explain the fact that 2-aminopyrimidines (33 and 34) can accommodate larger substituents than indolecarboxamides (32 and 31) (Figure 4), than H₄R homology models based on the β₂AR crystal structure template (Schultes et al., 2012). However, β₂AR- and H₁R-based H₄R homology models performed comparable in retrospective virtual screening studies (Istyastono, 2012), and in recent prospective virtual screening runs against the D₃R crystal structure and a D₃R homology model similarly high numbers of novel ligands were discovered (Carlsson et al., 2011). The recent GPCR crystal structures have nevertheless opened up new opportunities for structure-based virtual screenings studies to identify (more fragment-like) ligands with higher hit rates, as illustrated for the aminergic H₁R (de Graaf et al., 2011), β₂AR (Kolb et al., 2009), and D₃R (Carlsson et al., 2011) receptors. The H₁R crystal structure was for example used in combination with a customized virtual screening approach for the identification of novel fragment-like compounds (including compound 13, see Figure 4B) (de Graaf et al., 2011).

Figure 2.

Binding mode of (A) doxepin (1, magenta carbon atoms) in human H₁R (PDB code 3RZE (Shimamura et al., 2011)), (B) (R)-3-quinuclidinylbenzilate (2, green carbon atoms) in human M₂R (PDB code 3UON (Haga et al., 2012)), (C) tiotropium (3, orange carbon atoms) in rat M₃R (PDB code 4DAJ (Kruse et al., 2012)), (D) (S)-carvedilol (4, blue carbon atoms) in turkey β₁AR (PDB code 4AMJ (Warne et al., 2012)), (E) (S)-carazolol (5, red carbon atoms) in human β₂AR (PDB code 2RH1 (Cherezov et al., 2007)) and (F) (S)-eticlopride [(6, brown carbon atoms in D₃R (PDB code 3PBL (Chien et al., 2010)]. The yellow ribbons represent parts of the backbone of transmembrane (TM) helices 2, 3, 5, 6 and 7. Selected binding site residues are depicted as ball-and-sticks with light grey carbon atoms. Oxygen, nitrogen, sulphur, hydrogen and chlorine atoms are coloured red, blue, yellow, cyan and green, respectively. Hydrogen bonds are depicted by black dashes. Polar hydrogen atoms of the ligand are shown, but are omitted for the pocket residues. The labels for W^6.48 are omitted for all structures as well as F^6.52 for H₁R, β₁AR, β₂AR and D₃R for clarity purposes. (G) Molecular interaction fingerprint (IFP) (Marcou and Rognan, 2007) bitstrings describing the binding poses of 1–6 (A–F), encoding different interaction types (negatively charged, positively charged, H-bond acceptor, H-bond donor, aromatic face-to-edge, aromatic-face-to-face and hydrophobic) for each residue in the binding site. For reasons of clarity, only the bit strings of residues D^3.32, 5.42, 5.46, 6.52, 6.55, and 7.39 are shown. All binding modes are presented in a similar fashion throughout the manuscript. 2D structures of the molecules are presented in Figure 3A.

Figure 4.

Proposed binding modes, based on SAR and SDM studies, of (A) clozapine (48) (magenta carbon atoms) in human H₁R, (B) SBVS hit VUF13816 (13) (de Graaf et al., 2011) (green carbon atoms) in human H₁R, (C) R-cetirizine (18) (green carbon atoms) and S-cetirizine (19) (orange carbon atoms) in human H₁R, (D) VUF5228 (30) (blue carbon atoms) in a human H₄R homology model (Istyastono et al., 2011b), (E) JNJ7777120 (32) (red carbon atoms) and (F) aminopyrimidine 33 (brown carbon atoms) (Schultes et al., 2012). Rendering and colour-coding are the same as in Figure 2. H₄R homology models were build using the H₁R (E, F) and β₂AR (D) crystal structures as templates. (G) Molecular interaction fingerprint (IFP) bit strings describing the binding poses of 48, 13, 18, 19, 30, 32, 33 (A–F), encoding different interaction types with D^3.32, 5.42, 5.46, 6.52, 6.55 and 7.39 (colour-coding as described in Figure 2). 2D structures of the molecules are presented in Figure 3B.

Structural chemogenomics analyses of (hist)aminergic ligand binding sites

Over several decades mutagenesis studies have been extensively performed on GPCR targets in order to identify residues that are important for ligand binding (Shi and Javitch, 2002; Surgand et al., 2006; Vroling et al., 2011). Arguably, mutagenesis studies were needed to compensate for the lack of useful GPCR crystal structures. With the emerging GPCR crystal structures, mutation data should be revisited and evaluated in the new context of structural biology. Binding affinities from mutation studies of the TM binding pocket (Figure 1C) of histamine receptors (i.e. H₁R, H₂R, H₃R and H₄R) and crystallized aminergic GPCRs (i.e. D₃R, M₂R, M₃R, β₁AR and β₂AR) are reported in Table 1 and Supporting Information Table S1. It must be noted that mutations can not only have an effect on ligand binding, but can also have an effect on the structure of the binding site/protein. Supporting Information Table S1 contains 1420 reported single point mutations for 128 individual amino acid positions in the biogenic amine GPCRs (Supporting Information Figure S3). Most of the data (47%) correspond to residues located in the major pocket (between TM3, TM4, TM5 and TM6), while only 23 and 7% of the data are associated with residues located in the minor pocket (between TM1, TM2, TM3 and TM7) and in the second extracellular loop (EL2) respectively. The conserved residue D^3.32, the main interaction anchor of aminergic GPCRs, has been studied in 104 mutation studies. TM5 and TM6 have also been frequently studied (306 and 325 data points, respectively), in particular positions 5.39 (53), 5.42 (92), 5.46 (62), 6.34 (115), 6.52 (51) and 6.55 (41). Based on ligand binding pockets in the aminergic GPCR crystal structures and the analysis of the available aminergic receptor mutation data (Table 1, Supporting Information Table S1) and SAR, four important ligand interaction hot spots in aminergic GPCRs will be discussed that systematically cover the different regions of the GPCR ligand binding site: (i) the conserved ionic interaction anchor D^3.32; (ii) the aromatic cluster in TM6; (iii) functional selectivity via TM5; (iv) allosteric contacts with the minor pocket and extracellular loops. We illustrate how systematic mining of aminergic GPCR-ligand interaction space can give insights into how conserved and selective aminergic GPCR interaction hot spots in different regions of the receptor binding site can accommodate different chemical scaffolds observed in aminergic GPCR ligands and histamine receptor ligands in particular.

The conserved ionic interaction anchor D^3.32

The negatively charged and conserved aspartate residue in TM3 (D^3.32) of aminergic GPCRs is generally proposed as a key anchor for the basic moieties of aminergic ligands (Shi and Javitch, 2002; Surgand et al., 2006). The aminergic GPCR crystal structures (Figure 2) show subtle differences in the binding sites and ligand binding modes around this conserved D^3.32 residue. Structural chemogenomics analyses can help to explain the ligand and receptor dependent effects of mutation of D^3.32 and rationalize structure-affinity relationships of basic amine groups in aminergic GPCR ligands.

Mutation of D^3.32 in aminergic receptors, including histamine receptors (Gantz et al., 1992; Ohta et al., 1994; Nonaka et al., 1998; Shin et al., 2002; Bruysters et al., 2004; Jongejan et al., 2008), often leads to a significant decrease in ligand binding affinity (Table 1, Supporting Information Table S1). This is in line with the currently available crystal structures of aminergic GPCRs (H₁R, M₂R, β₁AR, β₂AR, D₃R) in which co-crystallized ligands (Figure 2) make an ionic interaction as well as a hydrogen bond with this residue. The quaternary amine moiety of tiotropium (3) in M₃R cannot form a hydrogen bond (Figure 2C), but is nevertheless within ionic interaction distance (4.3 Å) from the carboxylate group of D148^3.32. The experimentally supported ionic/hydrogen bond interaction with D^3.32 has been used to (i) guide the in silico prediction of ligand binding modes in (hist)aminergic receptor crystal structures and homology models [e.g. Figures 4 and 5 (Istyastono et al., 2011b; Schultes et al., 2012)]; and (ii) to select docking poses in structure-based virtual screening studies to identify new (hist)aminergic receptor ligands (e.g. H₁R ligand VUF13816, 13, Figure 4B) (de Graaf et al., 2011). Although the orientation of D^3.32 changes only slightly between crystal structures, it should be noted that the residues around the D^3.32 vary between aminergic GPCRs, creating distinct microenvironments in this area of the binding site. Such microenvironment can still be a subtle but important determinant of ligand selectivity. For example, the large aromatic Y^7.39 residue in M₂R and M₃R covers a deep hydrophobic binding pocket that fits the quinuclidine and N-methylscopine amine moieties in 2 (Figure 2B) and 3 (Figure 2C) respectively. This ligand binding mode is further accommodated by an alternative rotamer conformation of W^6.48 (that is not observed in the other aminergic GPCR crystal structures) that creates more space for these relatively bulky and rigid ring systems (Figure 2). Quinuclidine is indeed identified as a common substructure in muscarinic receptor specific ligands (van der Horst et al., 2009). The ethanolamine group in (S)-carvedilol (4, Figure 2D) and (S)-carazolol (5, Figure 2E) is a common substructure in (selective) beta-adrenergic receptor ligands (van der Horst et al., 2009) that can form a tight hydrogen bond network with D^3.32 and the medium sized polar N^7.39 residue in β₁AR and β₂AR. N^7.39 is a beta-adrenergic receptor-specific and stereoselective (Rosenbaum et al., 2007; Katritch et al., 2009; Seifert and Dove, 2009) recognition site for ethanolamines (e.g. the affinity of R-isoproterenol, 7, is 40 fold higher than S-isoproterenol, 50, see Table 1) and therefore an attractive feature to include in protein-based pharmacophore screening studies (Sanders et al., 2011; 2012). Interestingly, while the D₃R mimicking N322^7.39T mutation in β₂AR diminishes ligand binding (Suryanarayana and Kobilka, 1993), the beta-adrenergic receptor mimicking F369^7.39N mutation in the alpha-2-adrenergic receptor promotes stronger binding for aryloxyalkylamine ligands such as S-alprenolol (58), pindolol (57) and R-propranolol (55) (Suryanarayana et al., 1991). All these data indicate the important role of this residue in aminergic subtype selectivity (Ericksen et al., 2012). Most H₁R ligands contain linear alkylamines [like doxepin (1), Figure 2A] or cyclic amines such as piperazines and piperidines (Simons and Simons, 2011). Most high affinity H₃R ligands (Celanire et al., 2005), as well as compounds that have so far been tested in clinical trials contain cyclic amines including pyrrolidines and (homo)piperidines (Kuhne et al., 2011). Many H₄R receptor ligands contain a N-methylpiperazine moiety, but also other basic cyclic amines are an alternative to N-methylpiperazines including azetidines, aminopyrrolidines and piperazines (Engelhardt et al., 2009; Smits et al., 2009; Istyastono et al., 2011a). While H₁R SAR studies show a preference for tertiary amines over secondary and primary amines (Shah et al., 2009), N-methylation can be used as a subtle chemical switch to modulate H₃R affinity (Smits et al., 2012) and H₃R/H₄R selectivity (Lim et al., 2005; Govoni et al., 2006). Changing piperazine to a N-methylpiperazine (Smits et al., 2012) increases H₃R affinity while maintaining similar affinity for H₄R, but changing immepip (23) into methimepip (24) (Lim et al., 2005), or changing imbutamine (25) to N, N-dimethylimbutamine (26) (Govoni et al., 2006) increases H₃R selectivity over H₄R.

Figure 5.

Binding mode of (A) isoproterenol (7, orange carbon atoms) in the turkey β₁AR (PDB code 2Y03 (Warne et al., 2012)). Proposed binding modes of endogenous agonists, based on SDM studies, of (B) dopamine (35, blue carbon atoms) in the human D₃R and the proposed binding modes of histamine (8, magenta carbon atoms) in the different human histamine receptors; (C) H₁R, (D) H₂R, (E) H₃R and (F) H₄R. Minor changes in the H₁R crystal structure (i.e. rotation of K191^5.39 and N198^5.46) allow for accommodation of histamine. The H₂R, H₃R and H₄R models are based on the H₁R crystal structure. Rendering and colour-coding are the same as in Figure 2. (G) Molecular interaction fingerprint (IFP) bit strings describing the binding poses of 7–8 and 35 (A–F), encoding different interaction types with D^3.32, 5.42, 5.46, 6.52, 6.55 and 7.39 (colour-coding as described in Figure 2). Two-dimensional structures of the molecules are presented in Figure 3A and B.

This preference for specific amine moieties might be explained by differences between the histamine receptor binding sites in amino acid composition in the region close to D^3.32, including positions 7.39 (I in H₁R, F in H₃R and H₄R) and 7.42 (G in H₁R, L in H₃R, and Q in H₄R). In silico guided mutagenesis studies have indeed identified Q/L^7.42 as one of the molecular determinants of H₃R over H₄R selectivity (Istyastono et al., 2011b). Residues at position 7.39 however, have so far not yet been subjected to mutagenesis studies in histamine receptors (Table 1 and Supporting Information Table S1).

It should finally be noted that negatively charged residues like D^3.32 also play an important role in binding the basic amine moieties of small molecule ligands in other GPCR subfamilies (Surgand et al., 2006), as demonstrated by the crystal structures of delta (Granier et al., 2012), kappa (Wu et al., 2012), and mu (Manglik et al., 2012)-opioid receptors (i.e. D^3.32) and the chemokine receptor CXCR4 (Wu et al., 2010) (i.e. D97^2.63 and E288^7.39). As a result, several positively ionized privileged structures, like orthoalkoxy-N-phenylpiperazines (Rognan, 2007), are known to bind both aminergic and opioid receptors, most probably via ionic interactions with the shared D^3.32 residue. This is emphasized once more by ligand overlap between aminergic and opioid receptors, e.g. H₁R and the μ/κ/δ-opioid receptors share multiple ligands (32%, 29% and 25% of 57, 42, 32 tested ligands respectively). Fluphenazine (62) and Chlorpromazine (63) for example have affinity for these three opioid receptors and they were also found to have affinity for 22 out of 26 and 19 out of 24 tested aminergic receptors respectively. A similar pattern was obtained by (Poulain et al., 2001) who screened a panel of 47 compounds against multiple aminergic as well as opioid receptors and showed that several ligands, including ligand 64, bind to the NOP receptor (nociceptin), μ/κ/δ-opioid receptors, H₂R, β₁AR, M₂R, D₂R and D₃R amongst others.

From all these combined data and chemogenomic analyses, it can be concluded that D^3.32 is a key interaction partner in (hist)aminergic receptors, but that subtle differences in the binding site around this residue can in principle be used to design ligands with specific selectivity profiles, as demonstrated by position F/N^7.39 in alpha and beta-adrenergic receptors (vide supra).

The aromatic cluster in TM6

Aminergic GPCRs contain a conserved cluster of aromatic residues in TM6: W^6.48, F/Y^6.51 and F^6.52 (Figures 1C and 2) that play an important role in the binding of agonists and antagonists to aminergic receptors (Table 1, Table Supporting Information and Figure S5), including histamine receptors (Wieland et al., 1999; Shin et al., 2002; Bruysters et al., 2004). The new GPCR crystal structures, however, show that the binding modes of aromatic ring systems of aminergic GPCR ligands in this conserved aromatic region is ligand dependent and chemogenomics analyses suggest that small differences in this region can determine receptor selectivity.

Mutation studies of W^6.48, F/Y^6.51 and F^6.52 have been used to guide the construction of three-dimensional aminergic receptor-ligand models, including histamine receptor models (Wieland et al., 1999; Jongejan et al., 2005; 2008; Schlegel et al., 2007; Kiss et al., 2008b; Strasser et al., 2009; Lim et al., 2010; Istyastono et al., 2011b; Schultes et al., 2012; Sirci et al., 2012), that propose essential apolar interactions between this aromatic cluster and the different aromatic ring systems in the ligands (e.g. Figures 5). These binding mode hypotheses are confirmed by all currently available crystal structures of aminergic GPCRs in which ligands indeed make aromatic stacking and/or hydrophobic interactions with this aromatic cluster (Figure 2). In addition to the conserved D^3.32 ionic link, this conserved aromatic and hydrophobic pocket contributes to the low selectivity of many aminergic ligands (Figures 6C and 7), including the atypical antipsychotic drug clozapine (48, Figure 4A) which has considerable affinity for many aminergic receptors (Selent et al., 2008), including H₁R, H₄R, D₃R, M₂R, M₃R and to a less extent for β₂AR (Bolden et al., 1992; Schotte et al., 1996; Lim et al., 2005). Nevertheless, several subtle and local differences in the pharmacophoric properties and shape of this hydrophobic pocket in aminergic receptors can be used to explain and obtain receptor selectivity for specific (hetero)aromatic ligand scaffolds. For example, most high affinity H₁R ligands contain (tricyclic) hydrophobic aromatic systems that are proposed to form multiple complementary stacking interactions with W428^6.48, F432^6.52 and F435^6.55, as demonstrated for doxepin (1) in the H₁R crystal structure (Shimamura et al., 2011) and predicted for levocetirizine (18) (Figure 4C). Mutation of these aromatic residues in TM6 of H₁R diminish histamine (8) and [³H]mepyramine (14) binding (Table 1, Supporting Information Table S1) (Wieland et al., 1999; Bruysters et al., 2004). Several high affinity H₄R ligands on the other hand contain heteroaromatic systems including indoles (e.g. JNJ7777120 (32), Figure 4E, Table 1) and aminopyrimidines (e.g. compound 33, Figure 4F, Table 1) that are complementary to the larger and more polar binding pocket between S^6.52 and T^6.55 and form hydrogen bonds and/or ionic interactions with the H₃R/H₄R specific E^5.46 residue, as demonstrated by in silico guided mutagenesis studies (Jongejan et al., 2008; Schultes et al., 2012). Doxepin (1) adopts a butterfly-shape in the H₁R, and interacts extensively with W428^6.48 in the lower binding pocket (Figure 2A) (Shimamura et al., 2011). Similarly, clozapine (48), the SBVS hit VUF13816 (13) (de Graaf et al., 2011) and R- (18) and S-cetirizine (19) are proposed to all adopt this butterfly conformation in the H₁R binding site (Figure 4A–C). While most aminergic receptors contain a phenylalanine at position 6.52 (74%), M₂R and M₃R do not contain a phenylalanine at position 6.52 as the other receptors, but a polar (and less bulky) asparagine residue. This difference in local amino acid composition results in a clear change in ligand binding mode: (R)-3-quinuclidindyl-benzilate (2) and tiotropium (3) adopt an inverse butterfly-shape in M₂R and M₃R crystal structures respectively (Figure 2B,C) (Haga et al., 2012; Kruse et al., 2012), compared to doxepin (1) in the H₁R co-crystal structure (Figure 2A) (Shimamura et al., 2011), and probably also other tricyclic aromatic ligands in aminergic receptors that contain F^6.52 (Supporting Information Figure S2). Mutation studies demonstrate the essential role of this asparagine residue in binding (-)scopolamine (46), the N404^6.52D mutant has a 28 000-fold lower affinity than wild-type M₃R) (Bluml et al., 1994). Mutation studies suggest that the aromatic rings of the endogenous agonists histamine (8) and dopamine (35) interact with F435^6.55 (H₁R, Figure 5C) (Wieland et al., 1999; Bruysters et al., 2004) and H^6.55 (D₃R, Figure 5A) (Lundstrom et al., 1998) respectively.

Figure 6.

(A) The percentage (%) of the pairwise sequence identity between the ligand binding pockets (i.e. the selected 55 residues) of the histamine receptors as well as the aminergic receptors with a crystal structure available. The percentage (%) of the pairwise sequence identity for the TM helices is described in Supporting Information Table S3. (B) The average ligand similarity as calculated by EDprints (Kooistra et al., 2010) by comparing ligands from the ChEMBL database for each of the histamine receptors as well as the aminergic receptors with a crystal structure available (the scores are based on the average of the highest similarity scores). (C) The ligand overlap by comparing 9903 ligands from the ChEMBL database with annotated affinity for one or more of the discussed aminergic receptors (expressed as a percentage of the total number of ligands with experimentally determined binding affinity, i.e. if the experimentally determined radioligand displacement K_i or IC₅₀ value is 10 μM or lower, for one or both receptors). (D) The structural distance between the ligand binding pockets as calculated by SiteAlign (Schalon et al., 2008) (i.e. the selected 54 residues using distance-3) for the crystallized aminergic GPCRs. The gradient from blue to white to red indicates a low to high similarity of sequences (A), similarity of ligands (B), ligand overlap (C) and similarity of structures (D) respectively. One has to be warned for the values with a grey colour (B, C) as this indicates a low number of ligands available for this analysis (n < 45) therefore they might be misleading.

Figure 7.

Examples of ligand overlap within histamine receptor subtypes (H₁R/H₃R/H₄R) as observed in the ChEMBL database (Gaulton et al., 2012) and the VU-MedChem fragment library (de Graaf et al., 2013). Compounds with annotated affinities (K_i/IC₅₀) and a confidence factor of 8 or higher for H₁R, H₃R or H₄R were retrieved from the ChEMBL database and considered as ligands for a receptor if the affinity for that receptor was ≤10 μM. The median affinity was used when multiple values were reported. Two-dimensional structures of some subtype selective as well as subtype unselective histamine ligands are presented.

The rotameric state of W^6.48, a conserved residue in aminergic receptors and most other GPCRs [e.g. A_2AR (Jaakola et al., 2008) and CB₁R (Singh et al., 2002)], has been postulated to be associated with the activation state of GPCRs via a ‘rotamer toggle switch’ (Schwartz et al., 2006; Holst et al., 2010), like proposed for H₁R (Jongejan et al., 2005; Sansuk et al., 2011), H₄R (Jongejan et al., 2008) and β₂AR (Shi et al., 2002). The W^6.48 side chain shows relatively small conformational changes; however, when comparing the currently available agonist and antagonist bound GPCR crystal structures. Interestingly, W^6.48 adopts a different conformation (oriented more perpendicular to TM6) in the inactive antagonist bound M₂R and M₃R crystal structures compared to all other crystal structures (Figure 2B,C) (Katritch et al., 2012). The χ2 angle of W^6.48 rotates 60 degrees (compared to W^6.48 in H₁R) while the χ1 remains similar, which is compatible with the bulky quinuclidine and N-methylscopine amine groups and the ‘inverted’ butterfly shape of the benzene/thiophene rings of the M₂R/M₃R ligands 2 and 3 as discussed above (Figure 2B,C). Interestingly, the chemically related clozapine (48) acts as an inverse agonist on H₁R, but is a full agonist of H₄R (Lim et al., 2005; Bakker et al., 2007), suggesting slightly different binding modes and receptor activation mechanisms for these histamine receptors. Taken together, this data clearly shows the importance of the aromatic cluster in (hist)aminergic receptors.

Selective binding to TM5

The sequence diversity of the binding site residues in TM5 is relatively high among aminergic receptors (Supporting Information Figure S1). Structural and chemogenomics analyses of aminergic GPCRs indicate that interactions with these residues are important determinants of ligand function and receptor selectivity. In particular, positions 5.39 (A/G/K/L/V), 5.42/5.43 (A/D/G/S/T) and 5.46 (A/E/N/S/T) have been associated with functional and receptor selectivity in aminergic receptors (Figures 1C and 2) (Shi and Javitch, 2002), as demonstrated for histamine receptors (Gantz et al., 1992; Leurs et al., 1994; Ohta et al., 1994; Moguilevsky et al., 1995; Shin et al., 2002; Uveges et al., 2002; Bruysters et al., 2004; Jongejan et al., 2008). Natural aminergic receptor agonists (e.g. histamine 8 (Figure 5C–F), dopamine 35 (Figure 5B), adrenaline 54) are proposed to interact with polar residues at positions 5.42 and/or 5.46, as shown in the β₁AR (biased) agonist bound crystal structures (Figures 2D and 5A) (Warne et al., 2011; 2012) as well as the agonist (BI-167107, 60) bound β₂AR crystal structure (Rasmussen et al., 2011a; 2011b), and supported by mutagenesis studies (e.g. compound 8, 35, 42 and 53 in Table 1 and Supporting Information Table S1) (Gantz et al., 1992; Wess et al., 1993; Ohta et al., 1994; Moguilevsky et al., 1995; Sartania and Strange, 1999; Liapakis et al., 2000; Gillard et al., 2002; Shin et al., 2002; Uveges et al., 2002; Bakker et al., 2004; Bruysters et al., 2004; Jongejan et al., 2008). Interestingly, mutagenesis studies suggest that similar agonists adopt aminergic receptor specific binding modes. While the catechol hydroxyl groups of beta-adrenergic receptor agonist isoproterenol (7) form hydrogen bonds with both S228^5.42 and S232^5.46 in the β₁AR crystal structure (Figure 5A) (Warne et al., 2012), mutagenesis studies indicate that the catechol moiety of dopamine (35) interacts with S192^5.42 but not with S196^5.46 in D₃R (Table 1) (Sartania and Strange, 1999). Mutation studies also suggest that histamine (8) has similar binding modes in H₁R, H₃R, and H₄R by donating a hydrogen bond to N^5.46 (H₁R) (Leurs et al., 1994; Ohta et al., 1994; Moguilevsky et al., 1995; Bruysters et al., 2004) or E^5.46 (H₃R, H₄R) (Shin et al., 2002; Uveges et al., 2002; Jongejan et al., 2008) with its N^τ imidazole nitrogen atom (Figures 3 and 5), but adopts a different binding orientation in H₂R in which N^τ and N^π form hydrogen bonds with D186^5.42 and T190^5.46 respectively (Figures 3 and 5) (Gantz et al., 1992). In H₃R and H₄R, the negatively charged E^5.46 can furthermore form a stronger ionic/hydrogen bond with N^τ than N198^5.46 in H₁R, explaining the higher affinity of histamine for H₃R and H₄R compared to H₁R. Although there is a negatively charged aspartate at position 5.42 in the H₂R, it has been hypothesized that an increase in distance to the N^τ nitrogen atom might lead to the reduced affinity for this receptor compared to the H₃R and H₄R (Shin et al., 2002). The symmetric distributions of complementary pharmacophore features in H₂R, H₃R and H₄R binding sites (i.e. D^3.32 in TM3 and D98^5.42/E206^5.46/E182^5.46 in TM5) and histamine receptor ligands that contain two basic groups, makes binding mode prediction challenging (Lorenzi et al., 2005; Schlegel et al., 2007; Jongejan et al., 2008; Kiss et al., 2008b; Ishikawa et al., 2010; Istyastono et al., 2011b; Schultes et al., 2012). The binding modes of several H₄R ligands, including isothioureas (e.g. 29, 30), indolecarboxamides (e.g. 31, 32) and aminopyrimidines (e.g. 33, 34), in H₄R have been investigated by combining complementary in silico and in vitro approaches (Istyastono et al., 2011b; Schultes et al., 2012). Extensive SAR, mutagenesis, docking and MD simulation studies indicated that indolecarboxamides and aminopyrimidines form an H-bond with D94^3.32 via their piperazine amine moiety, while making H-bond interactions with E182^5.46 with their indole and aminopyrimidine groups respectively (Figure 4D–F) (Jongejan et al., 2008; Schultes et al., 2012). A comparable ligand-steered, experimentally supported protein-modelling approach combining 3D-quantitative structure-activity relationship (QSAR), MD simulations, SAR and mutagenesis studies indicated that clobenpropit (29) can bind H₄R in two distinct binding modes (forming H-bonds with D94^3.32 and E182^5.46 with their imidazole and isothiourea groups). The addition of a cyclohexyl group to the clobenpropit isothiourea moiety, however, allows VUF5228 (30) to adopt only one specific binding mode in the H₄R binding pocket, in which its imidazole interacts with D94^3.32 and its isothiourea group interacts with E182^5.46 (Figure 4D) (Istyastono et al., 2011b). Indolecarboxamides (31 and 32) and aminopyrimidines (33 and 34) show predominantly one binding mode in which the methylpiperazine moiety forms a hydrogen bond with D94^3.32 (Figure 4E,F) (Schultes et al., 2012). Mutation studies based on a guinea pig H₁R homology model identified K200^5.39 as an important interaction site for the zwitterionic ligands acrivastine (17) and (to lesser extent) S-cetirizine (19) (Wieland et al., 1999). This H₁R-specific anion-binding subpocket, consisting of the residues K179^45.49, K191^5.39 and H450^7.35, was indeed confirmed by the H₁R crystal structure in which this pocket is occupied by a phosphate ion (Shimamura et al., 2011). L175^5.39 is an important interaction point for the chlorine substituents of indolecarboxamide 32 and aminopyrimidines 34 in H₄R that forms a subpocket between the extracellular region of TM5 and EL2 which determines subtle differences in SAR between these two ligand classes (Schultes et al., 2012). The same L175^5.39 residue furthermore is an important molecular determinant of H₄R species selectivity that explains differences in binding affinities of JNJ7777120 (32) and clozapine (48) for human (L175^5.39) and monkey (V175^5.39) H₄R orthologs (Lim et al., 2010).

Figure 3.

2D-structures of (A) co-crystallized ligands in the crystal structures (Figure 2) and (B) ligands shown in Table 1 or mentioned in the text. Most of the ligands in panel B are ordered according to their primary aminergic GPCR target: histamine receptors H₁R (8, 11–20), H₂R (8), H₃R/H₄R (8–10, 20–34), dopamine receptor D₃R (35–41), muscarinic receptors M₂R/M₃R (42–49), beta-adrenergic receptors β₁AR/β₂AR (50–60), but it should be noted that many of the ligands bind to multiple aminergic GPCR subfamilies (e.g. 48, 61, 62, 63).

Stereoselective binding is observed for R- (18) and S-cetirizine (19) and R- (11) and S-chlorpheniramine (12), where both the R enantiomers have a higher affinity for H₁R. Interestingly, the T194^5.42A mutant increases binding of S-cetirizine (19), but not of R-cetirizine (18) (Table 1) (Leurs et al., 1994; Gillard et al., 2002). This stereoisomer specific mutational effect can be explained by docking studies in the H₁R crystal structure which indicate that only S-cetirizine (19) is sterically hindered by T194^5.42 (Figure 4C).

Although polar residues in TM5 of aminergic receptor are proposed to specifically interact with agonists, mutation studies and crystal structures indicate that also some antagonists can form hydrogen bonds interactions with some of these residues. For example, co-crystallized antagonists (S)-carvedilol (4) and S-carazolol (5) make polar interactions with TM5 (S^5.42) in β₁AR and β₂AR respectively, but not with S^5.46 (Cherezov et al., 2007; Warne et al., 2012). It is therefore the combination of interactions that make small molecules behave as either (inverse) agonists or antagonists (de Graaf and Rognan, 2008).

Allosteric interactions with the minor pocket and extracellular loops

Many GPCRs have allosteric sites and ligands binding there can alter orthosteric ligand affinity and/or efficacy, thereby offering additional opportunities for specific modulation of GPCR signalling (May et al., 2007). Currently all ligands in aminergic GPCR co-crystal structures bind primarily in the major pocket between TM3-6, like doxepin (1) in H₁R (Figure 2A). Some ligands also extend towards the minor pocket between TM1-3 and TM7 (Figure 1B,C), including (S)-carvedilol (4) (Figure 2D), (S)-bucindolol (59), and BI-167 107 (60) in beta-adrenergic receptors (Rasmussen et al., 2011a,b; Warne et al., 2012). The methoxyphenoxy moiety of (S)-carvedilol (4) for example is stacked between the aromatic rings W134^3.28 and W364^7.40 and makes hydrophobic contacts with G115^2.61, L118^2.64 and V119^2.65 in the minor pocket of β₁AR (Figure 2D). In silico modelling studies suggest that also other large aminergic receptor ligands can extend towards the minor pocket, as for example proposed for (R)- (18) and S-cetirizine (19) (Figure 4C). Mutation studies indicate that residues in this region can determine ligand affinity and selectivity for aminergic GPCRs (Table 1), and can for example explain species selectivity of HP-HA (15) and VUF4669 (20) for guinea pig (S84^2.61) versus human (N84^2.61) H₁R (Bruysters et al., 2005). Mutation studies in the dopamine D₂ and D₄ receptors showed that D₂/D₄ receptor selectivity is not determined by specific single amino acids but rather by a cluster of divergent aromatic residues in TM2, TM3 and TM7 (Simpson et al., 1999). The ELs of GPCRs covering the TM binding site (in particular EL2) show a high diversity in residue composition as well as length (Supporting Information Figure S1) (Peeters et al., 2012). The currently available GPCR crystal structures indeed confirm the high structural diversity in EL2 regions (Peeters et al., 2012; Wheatley et al., 2012). Mutagenesis studies have first of all shown that the conserved disulphide bridge between C^3.25 in TM3 and C^45.50 in EL2 is essential to maintain a proper fold of this loop over the GPCR ligand binding site (Shi and Javitch, 2002; de Graaf et al., 2008; Peeters et al., 2011). The crystal structure of β₂AR for example explains that only mutation of C191^45.50 but also mutation of C190^45.49 causes a significant decrease in ligand binding affinity (Fraser, 1989; Dohlman et al., 1990) as the latter residue forms another disulphide bridge with C184^45.43 that stabilizes a short helix in EL2 (Cherezov et al., 2007). Mutation studies have furthermore identified several residues (especially in the region around C^45.50) that are of great importance for ligand binding to aminergic GPCRs (including M₁R, M₂R, M₃R, α_1AAR, β₂AR, D₂R, 5-HT_1D and H₄R, Table 1 and Supporting Information Table S1) ((Lim et al., 2008; de Graaf et al., 2008) and references therein). For example, systematic mutation of 10 residues of the EL2 of D₂R indicated several residues that were involved in ligand binding (Shi and Javitch, 2004), and this was later confirmed by the crystal structure of the highly homologous D₃R (Chien et al., 2010). Residues of EL2 in M₂R (Huang et al., 2005) and M₃R (Krejci and Tucek, 2001) are also involved in ligand binding. Furthermore, based on the recently solved H₁R crystal structure it can be hypothesized that the positively charged and H₁R-specific lysine residue at position 45.49 might be involved in binding of second-generation antihistamines such as R- (18) and S-cetirizine (19) (vide supra) (Figure 4C).

Despite the apparent role of the minor pocket and extracellular loops in aminergic receptor ligand recognition (de Graaf et al., 2008; Shi and Javitch, 2002), it should be noted that the IT1t bound CXCR4 crystal structure (Wu et al., 2010) is currently the only GPCR crystal structure that shows a ligand bound solely in the minor pocket. Furthermore, there are several co-crystal structures solved of (the N-terminal) extracellular domains of (especially class B and class C) GPCRs bound to polypeptide ligands, but so far no GPCR crystal structure has been solved in which a small molecule is bound to only the extracellular loop region. Mutation studies indicate that the minor pockets of certain GPCR subfamilies [including chemokine receptors (Scholten et al., 2012) and prostanoid receptors (Kedzie et al., 1998)] can indeed be specifically targeted by small molecules that do not bind the major pocket. The minor binding site and/or extracellular loop regions is therefore an interesting site to allosterically modulate aminergic receptors. Only few small molecule allosteric modulators of aminergic receptors have been reported (e.g. for muscarinic receptors (Gregory et al., 2010; Mohr et al., 2010), dopamine receptors (Hoare et al., 2000), alpha-adrenergic receptors (Leppik et al., 1998), and beta-adrenergic receptors (Steinfeld et al., 2011), and serotonin receptors (Im et al., 2003; Mohr et al., 2010). M₂R is so far the only aminergic GPCR for which the extracellular allosteric binding site has been systematically mapped (located between TM2 (Y80^2.61), EL2 (E172^45.46, D173^45.47, E175^45.49, Y177^45.51) and TM7/EL3 (N419^7.32, T423^7.36) (Gregory et al., 2010). This has facilitated the design of dualsteric ligands (Mohr et al., 2010); compounds that bind simultaneously to both the orthosteric and allosteric binding pockets. Dualsteric ligands for, for example, H₁R (Bruysters et al., 2005), H₂R (Birnkammer et al., 2012) and H₁R/H₃R (Procopiou et al., 2011) have been described, but to the best of our knowledge true allosteric modulators of histamine receptors (that do not bind the orthosteric site) have so far not been identified. The emerging GPCR crystal structures now offer new opportunities to further explore allosteric binding sites in aminergic receptors, and to try to target these pockets with structure-based ligand discovery and design approaches. In particular fragment-based ligand design approaches, including growing, merging and linking of fragments can be an attractive strategy to target different binding sites (Smits et al., 2008; de Kloe et al., 2009). It should be noted however that (orthosteric and allosteric) ligand binding sites in aminergic GPCRs can be rather flexible, as suggested by long-term (Dror et al., 2011; Kruse et al., 2012) and random acceleration molecular dynamic simulation (Wang and Duan, 2009) studies of β₂AR, M₂R and M₃R crystal structures that identified several (transient) ligand access and exit channels. Such computational simulations can be used to map the binding pathway in order to steer experimental studies (e.g. mutagenesis studies, biophysical measurements) to ultimately derive structure kinetic relationships (Miller et al., 2012) and rationally achieve receptor subtype selectivity, as has for example been suggested for the adrenergic and muscarinic receptors (Dror et al., 2011; Kruse et al., 2012). Interestingly, the H₁R antagonist levocetirizine (18) (Figure 4C) shows a decrease in binding affinity and a reduced dissociation half-life at the H₁R K191^5.39A mutant (Gillard et al., 2002), while recently an analogue of VUF14480 (Supporting Information Figure S7) was designed that covalently binds to C^3.36 (but not S^3.36) in H₄R (Nijmeijer et al., 2013). These H₁R and H₄R ligands and mutants are valuable tools to study structure kinetic relationships in histamine receptors.

Links between aminergic GPCR binding site, ligand similarity and receptor-ligand selectivity

The previous sections have provided an extensive analysis of receptor-ligand interaction hotspots and binding modes in aminergic GPCR binding sites within the context of receptor mutagenesis studies and ligand structure-affinity and selectivity relationships. This final section brings together systematic bioinformatics and chemoinformatics analyses of aminergic receptor binding site similarity ( Figure 6A,D), ligand similarity (Figure 6B) and experimental receptor-ligand selectivity data (Figures 6C and 7). This integrative analysis gives complementary insights into the possibilities and challenges in the discovery and design of ligands with specific selectivity profiles within and between aminergic GPCR subfamilies. Our analysis highlights the current gaps in experimentally determined GPCR-ligand selectivity profiles and illustrates how complete all-against-all ligand-protein affinity matrices can give new insights into the atomic details of (selective) protein-ligand interactions. The affinity profiles across aminergic GPCRs (Figures 3B and 7) and ligand similarity analyses (Figure 6B) allow the identification of affinity cliffs (two chemically similar fragments of which one shows affinity and the other shows no (or much lower) affinity for a specific protein target) and selectivity cliffs (two chemically similar fragments of which one has affinity for a set of protein targets and the other shows no affinity for at least one of these targets) that can be used for (in silico guided) hit optimization. Furthermore, our ligand similarity analysis reflects to what extent the binding pockets of different aminergic GPCRs can accommodate/have been probed with different ligand scaffolds, and offer opportunities and challenges to identify new ligand chemotypes for specific receptors. This combined chemogenomics analysis provides important information for the development of histamine receptor ligands, and can offer new opportunities to develop, for example, dual H₁R-H₃R or H₁R-H₄R ligands with synergistic anti-inflammatory properties (Thurmond et al., 2008; Kuhne et al., 2011; Leurs et al., 2011).

Ligand binding site similarity

Within the sequence identity matrix of crystallized aminergic GPCRs (H₁R, β₁AR, β₂AR, M₂R, M₃R, D₃R) and the other three (H₂R, H₃R, H₄R) histamine receptors (Figure 6A) three clusters can be identified: a histamine receptor cluster, a muscarinic receptor cluster and a cluster including β₁AR, β₂AR and D₃R.

It should be noted that the histamine receptors are not as closely related (especially H₂R and H₄R) as the muscarinic or the beta-adrenergic receptors. The binding sites of H₃R and H₄R are relatively similar and explain the difficulty to develop H₃R/H₄R selective ligands (Celanire et al., 2005; Istyastono et al., 2011a). Figure 6A shows however that the overall sequence similarities between H₁R and M₂R/M₃R and between H₂R and β₁AR/β₂AR are significantly higher than the binding site similarities between H₁R and H₃R/H₄R and between H₂R and H₃R/H₄R. These data are in line with previous phylogenetic analyses of human GPCRs (Vassilatis et al., 2003; Surgand et al., 2006; Gloriam et al., 2009). The H₁R crystal structure furthermore has a relatively high structural similarity (Schalon et al., 2008) to the crystal structures of other aminergic receptors (Figure 6D). Although the overall sequence similarity between D₃R and β₁AR and β₂AR is relatively high, the subtle differences of the pharmacophoric properties of T369^7.39 (in D₃R) versus N^7.39 (in β₁AR and ADBR2) is apparently crucial for dopamine/beta-adrenergic receptor binding selectivity (vide supra), as reflected by the slightly lower structural similarity (Figure 6D) and the low ligand overlap (Figures 6C and 7, vide infra). The high but expected sequence similarity in sequence and structure for both M₂R-M₃R and β₁AR-β₂AR strongly highlight their close kinship and once more underlines the difficulties in developing receptor subtype selective ligands (Baker, 2010; Mohr et al., 2012).

Ligand similarity and ligand selectivity profiles

Similarity analyses of the sequences and structures of the receptors (Figure 6A,D) often correlate with chemical similarity and receptor selectivity of ligand sets (Figures 6B,C and 7). H₁R for example shares the highest sequence identity with M₂R and H₂R (41%), which is correlated with a high ligand similarity and ligand overlap with both targets (Supporting Information Figure S4). Multiple ligands, in particular tricyclic amines (i.e. doxepin, 1, clozapine, 48 and maprotiline, 65) but also other ligands like benztropine (66) and orphenadrine (67), have been found to bind the H₁R, H₂R as well as both muscarinic receptors (Supporting Information Figure S6). Within the histamine subfamily, the highest sequence identity, ligand similarity and ligand overlap (Figure 6A–C) are however observed between H₃R and H₄R. Imidazole containing compounds are considered to be non-selective H₃R and H₄R binders as illustrated by the ligand structures in Figure 3B, and the ligand overlap analyses presented in Figure 7. For example, 90 and 80% of the imidazole containing compounds binding at least H₃R or H₄R in the ChEMBL database and VU-MedChem fragment sets, respectively, bind both receptors. It should be noted however, that the relative affinities of specific imidazole containing ligands for H₃R and H₄R can be very different. While histamine has equal affinity for H₃R and H₄R, methylsubstitution of the amine group (N_α-methylhistamine, 9) gives a 100-fold selectivity for H₃R, while methyl substitution of the four position of histamine (4-methyl-histamine, 10) gives more than 100-fold selectivity for H₄R (Istyastono et al., 2011a). Other interesting examples of strong changes in H₃R/H₄R selectivity by subtle changes in chemical structure are immepip (23) (35-fold selectivity for H₃R) versus the methyl-substituted analogue methimepip (24) (3500-fold selectivity for H₃R) (Istyastono et al., 2011a) and ligands 27 (10-fold selectivity for H₃R) versus 28 (15-fold selectivity for H₄R) that differ by only one carbon atom in the linker between their imidazole and triazole moieties (Wijtmans et al., 2011). Another example of a steep H₃R/H₄R selectivity cliff is the addition of a cyclohexyl group to clobenpropit (29), that is VUF5228 (30), which results in more than 60-fold selectivity for H₃R compared to clobenpropit (29) (Istyastono et al., 2011b). Interestingly, the isoquinoline fragment 69 only binds H₁R, while several chemically related quinazoline and aminopyrimidine-containing fragment-like compounds have been reported to show submicromolar affinities for H₃R and H₄R (Smits et al., 2008; 2012; de Graaf et al., 2013; Schultes et al., 2012), illustrating how small changes in heteroaromatic ring systems can affect binding of specific histamine receptors. Stereoisomers 71 and 72 represent interesting examples of a stereoisomer specific histamine receptor selectivity switch (Figure 7) (Ishikawa et al., 2010). While 71 has a 4-fold selectivity for H₁R over H₃R (and 6-fold selectivity over H₄R), 72 has a 20-fold selectivity for H₃R over H₁R (and 93-fold selectivity over H₄R).

Several H₄R/H₃R selectivity hotspots have been identified by combining (3D) QSAR, protein homology modelling, molecular dynamics simulations and site-directed mutagenesis studies (i.e. N/Y^4.57, T/M^6.55, Q/L^7.42) that give insights into the molecular determinants of H₄R/H₃R ligand selectivity (Istyastono et al., 2011b). Another interesting example is the histamine receptor selectivity cliff between the pyrrolidine fragment 68, which binds all three histamine receptors, and the related piperidine fragment 23, which only has affinity for H₃R and H₄R (Figure 7). Finally, it should be noted that there are also examples where chemical similarity and selectivity analyses of ligand sets show more pronounced differences between receptors than suggested by the receptor sequence and structure similarity analyses (Figure 6A,D). This is exemplified for the beta-adrenergic ligand similarity and overlap that share many ligands that contain the beta-adrenergic privileged scaffold ethanolamine (432 out of 532) that in 14 out of the 15 cases (with the exception of carvedilol) do not bind other aminergic GPCRs that share sequence similarity, like D₃R (Figure 6B,D). The discovery of high affinity ligands for adrenergic receptors that do not contain this scaffold is therefore challenging, but might be facilitated by experimental screening of diverse sets of fragment-like libraries (de Graaf et al., 2013) or structure-based virtual screening studies (Kolb et al., 2009).

Data (in)completeness in chemogenomic analyses

The ligand overlap analyses presented in Figures 6 and 7 provide insight in the molecular determinants of receptors binding and selectivity, but also illustrate that data incompleteness is the Achilles' heel of chemogenomic analyses and predictions (Mestres et al., 2008). While there is a wealth of ligand affinity data available for individual receptors and receptor selectivity data for specific receptor pairs are available (e.g. H₃R and H₄R), there are many (large) gaps in the all-against-all aminergic GPCR ligand affinity matrix that could give new insights into the molecular determinants of ligand selectivity. For example, Figure 7 shows the ligand overlap of 280 ligands that were tested on H₃R and H₄R, while for only 56 of these 280 ligands also the H₁R affinity is known (Figure 7D). This is even more clearly illustrated by Supporting Information Table S3 describing the number of compounds tested on different receptor combinations. Supporting Information Table S3 shows that there are almost no H₃R and H₄R ligands for which binding data for M₂R, M₃R, β₁AR, β₂AR or D₃R is available (coloured grey in Figure 6). Complete all receptor against all ligand selectivity profiles are valuable data sets to guide chemogenomic analyses and enable the identification of protein-ligand interactions (Besnard et al., 2012; de Graaf et al., 2013; Stumpfe and Bajorath, 2012). However, up to now, most chemogenomic studies have focused on the analysis of large, lead-like molecules from high-throughput screening data or (often incomplete) published ligand affinity data (Lipinski et al., 2001; Macarron, 2006). Fragment-like molecules can probe their respective chemical space more efficiently (Murray et al., 2012), which allows for the construction of more accurate maps of protein-ligand binding sites as well as the construction of complete selectivity profiles (de Graaf et al., 2013). Using our in-house VU-MedChem library complete screenings against multiple proteins (e.g. GPCRs and an ion channel) have been performed giving unique insights in ligand overlap but also selectivity, which is already observed on a fragment level (de Graaf et al., 2013). This complete fragment library, containing 1010 compounds, has been screened against H₁R, H₃R and H₄R (Figure 7) and include 120 fragment hits that bind individual or a combination of histamine receptors (vs. 56 ligands present in ChEMBL database for which H₁R, H₃R, and H₄R is reported). These fragment screening data provide additional surprising examples of subtle selectivity switches, as illustrated by the H₃R selective fragment 70 (Figure 7), which is a substructure of JNJ777120 (32) which has more than 300-fold selectivity for H₄R over H₃R (Lim et al., 2005). Systematic determination of ligand affinities against multiple protein targets (Besnard et al., 2012; de Graaf et al., 2013) furthermore can yield more surprising cases of ligand overlap between aminergic GPCRs and other GPCR subfamilies or even other protein families, like ligand-gated ion channels, kinases or phosphodiesterases (Keiser et al., 2009; Lin et al., 2012; Sanders et al., 2012; de Graaf et al., 2013). These cases offer useful information for the consideration of potential side effects associated with anti-targets (Vaz and Klabunde 2008) and/or offer new opportunities for designing ligands with polypharmacological effects (Besnard et al., 2012; Morphy and Rankovic, 2005).

Conclusion

In summary, we have performed chemogenomic studies to identify links between (hist)aminergic receptor ligands and their binding sites and binding modes. This GPCR receptor subfamily shares a conserved structural fold around the transmembrane binding pocket and a large amount of experimental ligand affinity data is available to derive protein and ligand SAR. The increasing number of aminergic GPCR crystal structures now for the first time allowed us to combine ligand affinity data, receptor mutagenesis studies, and amino acid sequence analyses to high-resolution structural analyses of GPCR-ligand interactions. This combined approach enabled us to identify the molecular and structural determinants of ligand affinity and selectivity in several regions of the crystallized aminergic GPCRs as well as all histamine receptors. Further investigations highlighted interesting correlations and differences between ligand similarity/overlap between different receptors and ligand binding site sequence/structure similarity. Apparent discrepancies could be explained by detailed analysis of crystallized or predicted protein-ligand binding modes and local differences in essential pharmacophore features in the ligand binding sites of different receptors. Currently, chemogenomic analyses are however limited by the incompleteness of experimental (all ligand against all protein) affinity data sets. Fragment-based approaches, however, can be used to overcome this problem as they probe their respective chemical space more efficiently, which allows systematic experimental screening of the same small chemical library against multiple protein targets. Chemogenomic analysis of these bioaffinity profiles can be used to identify structure-selectivity relationships that improve our understanding of ligand binding to (hist)aminergic receptors and hence can be used in future GPCR ligand design.

Conflicts of interest

The authors state no conflict of interest.

Supporting information

Additional Supporting Information may be found in the online version of this article at the publisher's web-site:

Table S1 Binding affinities from site directed mutagenesis studies on histamine receptors and crystallized aminergic GPCRs (see separate excel file). The two-dimensional structures of all the compounds are depicted in Figures 3 and Supporting Information Figure S7. The single-letter amino acid codes indicated in the grey rows are the human wild-type residues. ^aSDM study on guinea pig H₁R. ^bOnly functional SDM data available for histamine (8) in H₂R. ^cSDM study on rat M₃R.

Table S2 Aminergic GPCR crystal structures available in the PDB.

Table S3 The number of ligands for which the activity for both receptors is known (as annotated in the ChEMBL).

Table S4 The percentage (%) of the pairwise sequence identity between the TM domains of the histamine receptors as well as the aminergic receptors with a crystal structure available. The gradient from blue to white to red indicates a low to high sequence similarity.

Figure S1 Full sequence alignment of the histamine and all crystallized aminergic GPCR receptors.

Figure S2 Sequence alignment of the selected pocket residues for all aminergic GPCR receptors.

Figure S3 Snakeplot (http://www.ssfa-7tmr.de/ssfe/snakes/snakes_designer.php) showing the helices of H₁R. All B&W numbers are given within the circles as well as the amino acids in H₁R, all B&W positions within the TM helices that were mutated during single-point SDM studies for the aminergic GPCRs from Supporting Information Table S1 are shown in orange. The most conserved residues (i.e. the numbers as indicated by TM.50 in the Ballesteros-Weinstein numbering scheme) are circled in green.

Figure S4 Venn diagram visualizing the overlap between the ligands of H₁R and all crystallized GPCR receptors according to ligand binding data obtained from the ChEMBL.

Figure S5 (A) front and (B) bottom view displaying the changes in the orientation of the TM helices of β₂AR upon activation. The structure of the inactive (PDB-code 2RH1) and active and (PDB-code 3P0G) β₂AR structure are shown in red and light blue respectively.

Figure S6 Venn diagram visualizing the overlap between the ligands of H₁R, H₂R, M₂R and M₃R.

Figure S7 2D structures for most of the ligands for which SDM data is annotated in Table S1. The structures omitted here are depicted in Figure 3.