“Selective” serotonin 5-HT_2A receptor antagonists

Abstract

Blockade of the serotonin 5-HT_2A G protein-coupled receptor (5-HT_2AR) is a fundamental pharmacological characteristic of numerous antipsychotic medications, which are FDA-approved to treat schizophrenia, bipolar disorder, and as adjunctive therapies in major depressive disorder. Meanwhile, activation of the 5-HT_2AR by serotonergic psychedelics may be useful in treating neuropsychiatric indications, including major depressive and substance use disorders. Serotonergic psychedelics and other 5-HT_2AR agonists, however, often bind other receptors, and standard 5-HT_2AR antagonists lack sufficient selectivity to make well-founded mechanistic conclusions about the 5-HT_2AR-dependent effects of these compounds and the general neurobiological function of 5-HT_2ARs. This review discusses the limitations and strengths of currently available “selective” 5-HT_2AR antagonists, the molecular determinants of antagonist selectivity at 5-HT_2ARs, and the utility of molecular pharmacological and computational methods in guiding the discovery of novel unambiguously selective 5-HT_2AR antagonists.

Graphical Abstract

1. Introduction

Serotonin (5-hydroxytryptamine, 5-HT) is an endogenous signaling molecule that regulates nearly all neurocognitive functions by activating at least 14 genetically-encoded 5-HT receptors, along with dozens of variants generated by post-transcriptional editing or alternative splicing (1, 2). Thirteen genetically encoded 5-HT receptors are G protein-coupled receptors (GPCRs) that, owing to their plasma membrane localization and extensive control over intracellular signaling, are highly accessible and versatile drug targets. Approximately one-third of drugs approved by the United States Food and Drug Administration target GPCRs and ~4% bind to 5-HT_2A receptors—arguably the most studied 5-HT receptor (3, 4).

Early antipsychotic medications, known to block dopamine D₂-like GPCRs in the striatum, were coincidentally instrumental in the discovery of 5-HT_2ARs in the cortex. For example, radioligand binding studies with rat brains showed that both spiperone (Figure 1) and haloperidol bound striatal dopamine receptors, whereas only spiperone had high affinity for a population of cortical receptors labeled by the promiscuous (i.e., nonselective) serotonergic compound lysergic acid diethylamide (LSD), suggesting that spiperone-bound 5-HT receptors (5). Electrophysiology experiments then showed that at least two 5-HT receptors in the brain modulate neuronal firing (6). LSD was subsequently reported to bind both receptors in the rat frontal cortex: those demonstrating higher affinity for [³H]5-HT or [³H]spiperone, termed 5-HT₁R and 5-HT₂R, respectively (7). The discovery of distinct 5-HT receptors naturally led to investigations into their behavioral and physiological functions, creating demand for selective antagonists to block 5-HT₂R-mediated effects (8).

Figure 1:

Structures of antagonists, with varying degrees of selectivity, commonly employed to probe the 5-HT_2AR.

Among the first behavioral activities to be associated with the 5-HT₂R was the ability of laboratory animals to discriminate the subjective effects elicited by serotonergic psychedelics, including LSD, mescaline, and 2,5-dimethoxy-4-methyl-amphetamine, which could be blocked by the 5-HT₂R antagonists pirenperone and ketanserin (described below) (9–12). Soon after this, 5-HT was found to effectuate phosphoinositide turnover in the rat cerebral cortex, and the potencies of 5-HT₂R antagonists to block this effect were correlated with their affinity for these sites (13). The 5-HT₂R was later designated as the 5-HT_2AR after the discovery of 5-HT_2B and 5-HT_2CRs (14–20), classified together based on extensive similarities in their ligand binding affinities, signal transduction mechanisms, and structural homology (60–70% amino acid identity within structurally conserved regions).

Decades have passed since the seminal discoveries described above, and antagonism of 5-HT_2ARs has become a mainstay of antipsychotic drug action. In fact, so called selective 5-HT_2AR antagonists can effectively treat dimensions of psychosis (21–25), without untoward side effects associated with the off-target binding of more promiscuous antipsychotic agents. Additionally, there is new widespread interest in identifying mechanisms of rapid, persistent neuronal and behavioral plasticity elicited by psychedelics that share activation of 5-HT_2ARs, but diverge in their activities at other targets (26–30). The delineation of such mechanisms inherently relies on the use of unambiguously selective 5-HT_2AR antagonists to block these effects, yet the utility of currently available antagonists is limited by their selectivity, pharmacological characterization, and structural diversity.

Central to this review, defining the fundamentally vague pharmacological term “selective” remains a formidable challenge. For example, a ligand may bind target A with 10-fold greater potency than targets B and C and be defined as “selective” for target A, albeit, the ligand affinities at targets D and E may not have been considered. In other words, a ligand may be referred to in the literature as “selective” irrespective of the number of off-target proteins against which it has been compared. Thus, it is critical to define under what context a ligand may be “selective”. In this review, we discuss the discovery and relative selectivity of putatively selective 5-HT_2AR antagonists. We define here ligands that block 5-HT-elicited Gα_q/11-signaling via 5-HT_2ARs as “antagonists” owing to a paucity of data at alternative G protein or β-arrestin1/2 signaling pathways. Likewise, there is limited evidence of physiological, therapeutic, or structural distinctions between neutral antagonists and inverse agonists, leading us to refer to neutral antagonists and inverse agonists alike as “antagonists”, unless otherwise specified. We also discuss structural mechanisms underlying selective antagonist binding to 5-HT_2ARs, and the potential of molecular modeling and molecular dynamics to advance the design of unambiguously selective 5-HT_2AR antagonists.

2. Putatively selective 5-HT_2AR antagonists

2.1. Spiperone (spiroperidol)

Spiperone, a butyrophenone discovered by Janssen Pharmaceuticals, was developed as an antipsychotic and binds central dopamine (31) and 5-HT (32) receptors with high affinity. After its affinity for 5-HT receptors was reported, several other antipsychotics were also found to bind 5-HT receptors (5, 7). Spiperone was later found to have >500-fold higher affinity at human 5-HT_2ARs than the highly homologous 5-HT_2B and 5-HT_2CRs (33, 34), making it the first commercially-available compound for clearly labeling and distinguishing 5-HT_2ARs from 5-HT_2B and 5-HT_2CRs. However, its high affinity at dopamine D₂-like receptors, along with its moderate (K_i = 30–300 nM) to high (K_i ≤ 30 nM) affinity at α_1A-, α_1B-, α_1D-adrenergic, 5-HT_1A, and 5-HT₇Rs (Table 1) limits its use for interrogating the physiological and behavioral role of 5-HT_2ARs (9).

Table 1:

Binding affinities of various 5-HT_2AR antagonists at relevant off-target GPCRs.

	Spiperone	Ketanserin	M100,907	Pimavanserin Altanserin		Ritanserin	Pirenperone	Risperidone
GPCR	Affinity (Ki, nM)
5-HT 2A	1.2 a,e,g	1.13 a,e,g	0.47 a,e,g	0.24 a,e,g,(80)	0.3 b,e,g,(170)	0.22 a,e,g	1.1 a,e,h,(171)	0.3 a,e,g,(80)
5-HT 2B	1,114a,e,g	234a,e,g,(172)	261a,e,h	437a,e,g,(80)	-	0.14a,e,h	61a,e,h,(171)	15.5a,e,g,(80)
5-HT 2C	923a,e,g	88a,e,g	100a,e,g	2.8a,e,g,(80)	6b,e,g,(170)	0.24a,e,g	77a,e,h,(171)	19a,e,g,(80)
5-HT 1A	17a,e,g	>1,000a,e,h	>1,000a,e,h	>1,000a,e,g,(78)	>1,000b,f,h,(173)	309a,e,g	485b,f,g	296a,e,g
5-HT 1B	>1,000b,f,g	>1,000a,e,g	>1,000a,e,g	>1,000a,e,g,(78)	>1,000b,f,h,(173)	194a,e,h	>1,000b,f,g	53.6a,e,g
5-HT 1D	>1,000a,e,h	111a,e,g,(174)	>850d,f,h,(58)	>1,000a,e,g,(78)	-	36.8a,e,h	-	29a,e,g
5-HT 1E	>1,000a,e,h,(175)	>1,000a,e,h,(175)	>850h,j,(58)	>1,000a,e,g,(78)	-	>1,000a,e,h	-	>1,000a,e,h
5-HT 1F	>1,000a,e,h,(176)	>1,000a,e,h,(176)	>850h,j,(58)	>1,000a,e,g,(78)	-	-	-	>1,000a,e,h,(177)
5-HT 3	>1,000c,f,g	>1,000c,f,g	>1,000a,e,g	>1,000a,e,g,(78)	-	>1,000b,f,g,(178)	-	>1,000b,e,g,(177)
5-HT 4	-	>1,000e,(179)	>1,000b,f,g,(58)	>1,000a,e,g,(78)	-	>1,000a,e,g	-	-
5-HT 5A	>1,000a,e,h	>1,000b,e,h	>1,000a,e,h	-	-	77a,e,h	-	206a,e,h
5-HT 6	>1,000b,f,g	>1,000a,e,h	>1,000a,e,h	>1,000a,e,g,(78)	>1,000a,e,h,(170)	67a,e,h	-	>1,000a,e,h
5-HT 7	110a,e,h	794a,e,h,(180)	226b,f,g,(58)	>1,000a,e,g,(78)	15a,e,g,(170)	29.6a,e,g	6.5a,e,h,(181)	4.8a,e,g
H 1	272c,f,g,(43)	1.8c,f,g,(43)	>1,000a,b,e,g,(57, 68)	>1,000a,e,g,(80)	20b,f,g,(173)	<40b,f,g,(95)	-	18.2a,e,g,(80)
H 2	-	-	-	>1,000i,j,(78)	-	4.5a,e,g	-	120a,e,g
α 1A	20a,e,g,(182)	28b,f,g,k,(35)	128a,e,g,(67)	>1,000i,j,(78)	4.55b,f,g,k,(173)	80.7a,e,g	20b,f,g,k,(183)	5a,e,g
α 1B	3a,e,g,(182)		425a,e,g,(67)	>1,000i,j,(78)		223a,e,g		9a,e,g
α 1D	8a,e,g,(182)		-	>1,000i,j,(78)		86.5a,e,g		-
α 2A	135b,f,g,k,(182)	-	>1,000a,e,g,(67)	>1,000i,j,(78)	99b,f,g,k,(96)	95.8a,e,g	20b,f,g,k,(183)	16.5a,e,g
α 2B		199a,e,g,(184)	271a,e,h	>1,000i,j,(78)		>1,000a,e,g		8.5a,e,g,(177)
α 2C		-	>1,000a,e,g,(67)	-		269a,e,g		1.3a,e,h
D 1	220a,e,g,(185)	190a,e,g,(185)	>1,000a,e,g	>1,000i,j,(78)	-	344a,e,g	-	249a,e,g
D 2	0.28a,e,g,(80)	>1,000b,f,g	>1,000a,e,g	>1,000a,e,g,(78)	62b,f,g,(173)	>1,000a,e,g	-	4.35a,e,g
D 3	0.14a,e,g,(80)	-	>1,000b,e,g,(58)	>1,000i,j,(78)	-	57.7a,e,g	-	12.2a,e,g
D 4	1.39a,e,g	-	540a,e,g,(58)	>1,000i,j,(78)	-	51.5a,e,g	-	4.7a,e,g
D 5	>1,000a,e,g,(185)	>1,000a,e,g,(185)	>1,000a,e,g	-	-	163a,e,g	-	290a,e,g
sigma	632b,f,g,k,(186)	-	87b,f,g,k,(58)	>1,000i,j,(78)	-	-	-	-

Note: Unless otherwise noted by reference (in parenthesis), K_i values represent PDSP certified values or an average obtained from the NIMH Psychoactive Drug Screening Program database, https://pdspdb.unc.edu/pdspWeb/; search conducted 01/18/2022. K_i values were derived from ^ahuman, (when available), ^brat/mouse, ^cguinea pig, or ^dcow receptors expressed in ^eheterologous cell systems or ^fnative tissue, using ^gantagonist or ^hagonist radioligands. Some values have undefined ⁱradioligands ^jspecies or ^kreceptor subtypes. ‘-’ denotes ligand affinities that are, to the best knowledge of the authors, not reported in the literature.

2.2. Ketanserin (R 48,648)

Since its discovery by Janssen Pharmaceuticals in 1981 (35), the quinazoline derivative ketanserin is among the most widely used tools for probing 5-HT_2AR function in preclinical research (26–28, 36), and the sole antagonist used to delineate the 5-HT_2AR-dependent effects of serotonergic psychedelics in humans (37–41). Although ketanserin was the first 5-HT_2AR antagonist discovered that lacks high affinity for other serotonin and dopamine receptors, it is less appreciated that it has high affinity at several aminergic receptors, including α_1A-, α_1B-, α_1D-adrenergic, and histamine H₁ receptors (35, 42–44), as well as, moderate affinity at α_2B-adrenergic and 5-HT_2C receptors (Table 1). These off-target activities limit the utility of ketanserin as a specific tool for assessing 5-HT_2AR activity. The off-target activity of ketanserin at adrenergic and histaminic receptors is particularly confounding because α₁-adrenergic receptors colocalize with 5-HT_2AR transcripts, and α₁-, α₂-adrenergic and histamine H₁ receptors share signal transduction mechanisms with the 5-HT_2AR and modulate the excitability of serotonergic neurons in the dorsal raphe nucleus (45–49).

To demonstrate that ketanserin interacts with high affinity at sites independent of the 5-HT_2AR, we present an autoradiography experiment wherein mouse brain slices were incubated with 1 nM [³H]ketanserin in the presence or absence of 1 μM M100,907 (Figure 2), another putatively selective 5-HT_2AR antagonist (discussed below). Our results demonstrate that [³H]ketanserin prominently labels sites in the striatum that are not specific to 5-HT₂-type receptors, consistent with earlier work in rat brain using other 5-HT₂-type receptor antagonists (50, 51). Dopaminergic, adrenergic, or histaminergic antagonists were also ineffective at blocking striatal [³H]ketanserin binding (50, 51). Further experiments using rat brains (52), bovine chromaffin granule membranes (53), or heterologous cell systems (54) demonstrate that [³H]ketanserin binds the vesicular monoamine transporter 2 with moderate affinity (K_i, K_d = 22–540 nM). Therefore, studies investigating the effects of 5-HT_2AR agonists should use caution when interpreting results obtained with ketanserin as a pharmacological tool since its pharmacodynamic effects are incompletely understood.

Figure 2:

[³H]Ketanserin binds with high affinity (1 nM) a large population of targets in the striatum that are not 5-HT₂-type receptors. Note that co-incubation with the putatively selective 5-HT_2AR inverse agonist, M100,907 (1 μM, a concentration chosen to occupy and block the vast majority of 5-HT_2ARs, in order to reveal off-target binding of [³H]ketanserin), blocked binding of [³H]ketanserin in the cortex, but not the striatum. [³H]ketanserin (42.5 Ci/mmol) autoradiography (10-week film exposure) of coronal brain sections from FVB.129P2-Pde6b⁺ Tyr^c-ch/AntJ (sighted FVB) mice.

2.3. M100,907 (MDL100,907, Volinanserin)

Driven by work suggesting that 5-HT_2AR antagonists elicit antipsychotic-like behavioral effects with a low risk of extrapyramidal symptoms compared to antipsychotics that primarily block D₂Rs, researchers at Hoechst Marion Roussel in the early 1990s developed the piperidine derivative M100,907 (Figure 1) for the treatment of schizophrenia (55, 56). Until this time, 5-HT_2AR antagonists demonstrated only modest selectivity (< 30-fold) over 5-HT_2C, α₁-adrenergic, H₁, and dopamine D₂ receptors. Thus, Hoechst Marion Roussel broke ground by reproducibly showing that M100,907 had sub-nanomolar affinity at 5-HT_2ARs, with >100-fold selectivity over these off-target GPCRs, along with numerous other receptors and ion channels (57, 58).

An important caveat when considering the selectivity of M100,907 is that much of the work characterizing its binding profile is derived using a variety of species, tissue sources, and radioligands (57, 58). Thus, the apparent ligand affinity may be impacted by differences in the ligand binding pocket between species (59–61), plasma membrane composition (62), and/or radiolabeling discrepancies in native tissue (e.g., [³H]spiperone binds non-selectively to several dopamine and serotonin receptors, Table 1). Moreover, GPCRs exist in a dynamic equilibrium of conformational states often determined by the prevalence of guanine nucleotides and G proteins (63, 64). Agonists tend to recognize these conformational states with variable affinity, making radioligand-based affinity measurements sensitive to differences in the conformation stabilized by the radioligand and competitor ligand. For example, 5-HT has much higher affinity for [³H]5-HT-labeled than [³H]ketanserin-labeled 5-HT_2ARs (K_i = 3.8 and 144 nM, respectively) (65). Accordingly, the off-target binding of M100,907 is less certain than one might assume by glancing at available data in Table 1, and since competitive radioligand displacement data provides little information on ligand functional activity, some skepticism is warranted regarding the selectivity of M100,907 at 5-HT_2ARs (see ref. (66) for review on defining ligand activity).

Among the currently known off-targets of M100,907 are sigma, 5-HT_2C, α_1A-, and α_2B-adrenergic receptors, with lower affinity at the latter receptors than is observed for ketanserin (Table 1). While selectivity over H₁Rs has been challenged (67), most work supports that M100,907 is selective at binding 5-HT_2A over H₁Rs. (57, 58, 68). Autoradiography studies further support 5-HT_2AR engagement by [³H]M100,907 at 0.4 nM in rat brain, exhibiting dense labeling in the frontal cortex, motor cortex, and claustrum (51). Rodent studies indicate that M100,907 does not antagonize apomorphine induced-climbing or phenylephrine lethality, indices of in vivo D₂ and α₁-adrenergic receptor engagement, respectively, and may only engage 5-HT_2CRs at elevated doses (58). Notably, some studies report M100,907 is not a potent modulator of mouse locomotor activity (55, 58, 69), while others report robust locomotor suppressing effects (70, 71), suggesting that different animal strains and/or test conditions are involved in the confounding results reported for M100,907 behavioral pharmacology.

In preclinical studies, M100,907 exhibited promising antipsychotic-like behavioral effects in models of psychomotor stimulation and sensorimotor gating deficits, with a favorable side-effect profile (55, 58, 72, 73). Positron emission tomography (PET) studies in humans confirmed that M100,907 exhibited high levels of apparent 5-HT_2AR occupancy in the cortex (74) for over 24 hours, far outlasting its presence in plasma (75). However, clinical trials for schizophrenia were discontinued due to a failure to meet endpoints (76). In summary, M100,907 appears to offer advantages over ketanserin for evaluating 5-HT_2AR function in vivo. However, the lack of broad receptor panel screening data with a consistent tissue-source suggests exercising caution when interpreting data reported for M100,907.

2.4. Pimavanserin (ACP-103, Nuplazid®)

Since their development, so-called “atypical” antipsychotic medications (e.g., clozapine) have been noted to demonstrate distinct therapeutic and side-effect profiles compared to their “typical” counterparts (e.g., haloperidol), which were defined primarily by potent dopamine D₂R antagonist activity. While differences in D₂R occupancy at clinically effective doses or polypharmacology (e.g., partial agonism of 5-HT_1ARs) likely contribute to these differences, scientists at Acadia Pharmaceuticals identified functionally distinct effects at the 5-HT_2AR as being, at least in part, responsible for atypical antipsychotic therapeutic outcomes (68). Key to this was the development of a sensitive, high-throughput chemical genomics platform (Receptor Selection and Amplification Technology^™; R-SAT^™) enabling reliable detection of 5-HT_2AR constitutive activity (i.e., ligand-independent G protein activation), historically difficult to detect through 5-HT_2ARs. By screening numerous antipsychotics via the R-SAT^™ platform, it was found that nearly all atypical antipsychotics, as well as spiperone, ketanserin, and M100,907, demonstrated inverse agonist activity (i.e., a concentration-dependent reduction in constitutive activity) at 5-HT_2ARs (68). This finding invigorated a high-throughput screening (HTS) campaign that identified several hits for potent 5-HT_2AR inverse agonism (77). After thorough selectivity screening and iterative rounds of lead optimization, pimavanserin (ACP-103, Figure 1) was identified as a novel putatively selective 5-HT_2AR inverse agonist with preclinical efficacy in multiple models of psychosis, and was devoid of untoward motor side effects (78). Notably, pimavanserin exhibits Gα_q-mediated inverse agonist activity in clonal cells expressing wild type (WT) 5-HT_2ARs or constitutively activated 5-HT_2AR variants (79, 80). A recent study using human and mouse brain tissue, however, reported that pimavanserin may act as a neutral antagonist and inverse agonist of Gα_q/11- and Gα_i1-mediated signal transduction, respectively, in vivo, apparently via 5-HT_2ARs (81). Nevertheless, although the therapeutic relevance of inverse agonism at 5-HT_2ARs is uncertain, pimavanserin is approved to treat hallucinations and delusions in Parkinson’s disease patients (22, 82). Pimavanserin also has shown efficacy in clinical trials to treat dementia-related psychosis (23, 24), negative symptoms of schizophrenia (25), and major depression (83).

A broad affinity profiling screen of over 75 GPCRs, ion channels, transporter proteins, and enzymes, indicated that pimavanserin had >1,000-fold selectivity for 5-HT_2ARs over all other targets except for 5-HT_2CRs (~12-fold), where it also acts as an inverse agonist (78, 80). Pimavanserin also was found to bind L-type calcium channels with moderate to low affinity (K_i = 310 nM) (78). The off-target activity at 5-HT_2CRs is especially noteworthy for researchers investigating mechanisms sensitive to midbrain dopamine release, which might be oppositely influenced by 5-HT_2A and 5-HT_2CRs (67, 84–86); see refs. (87) and (88) for investigations into the effects of 5-HT_2CRs on midbrain dopamine release.

Several studies report that pimavanserin may not impact motor function in humans (22, 83, 89, 90), likely due to its nil affinity at D₂ and H₁Rs (77, 91). In rodents, however, pimavanserin causes robust locomotor suppression (78, 80). This activity appears to be at odds with the role of 5-HT₂-type receptors in modulating rodent locomotor activity. For example, 5-HT_2AR activation enhances locomotor activity whereas 5-HT_2CR activation suppresses locomotor activity in mice (84). Conversely, antagonism of 5-HT_2ARs normalizes the locomotor enhancing effects of 5-HT_2CR antagonists (85). Therefore, a potent and selective 5-HT_2A/5-HT_2CR inverse agonist such as pimavanserin might be expected to have neutral effects on locomotor activity in mice. A PET study in pigs also questions the selectivity of pimavanserin, which was not competitively displaced from the thalamus by unlabeled pimavanserin or ketanserin (92), suggesting the parent compound or a metabolite exhibits off-target binding, high non-specific binding, or a slow dissociation rate. Nevertheless, of the ligands discussed here, pimavanserin is the most thoroughly characterized putatively selective 5-HT_2A-preferring 5-HT_2A/5-HT_2CR antagonist/inverse agonist available.

2.5. Other 5-HT_2AR antagonists

In addition to spiperone and ketanserin (35, 93), scientists at Janssen Pharmaceuticals also discovered several other highly potent 5-HT_2AR antagonists with some selectivity for 5-HT_2ARs, including altanserin, pirenperone, risperidone, and ritanserin (9, 94–97) (Figure 1). However, while these ligands demonstrate high affinity at, and some selectivity for, 5-HT_2ARs, neither is likely to engage 5-HT_2ARs selectively at behaviorally active doses due to their high affinity at off-target receptors (Table 1). Altanserin, a PET ligand for labeling 5-HT_2ARs in vivo (98), is a possible exception to this shortcoming, although its in vitro characterization is sparse. Nevertheless, studies with altanserin (99, 100) and pirenperone (9–11, 101) have informed our understanding of 5-HT_2AR pharmacology, while structural biology studies with risperidone (79) and ritanserin (102) have begun clearing a path to the rational design of novel and selective 5-HT_2AR antagonists (described below).

3. Molecular determinants of polypharmacology and subtype selectivity at 5-HT_2A GPCRs

3.1. Polypharmacology of aminergic ligands targeting 5-HT_2ARs

The 5-HT_2AR orthosteric binding pocket—where the endogenous ligand 5-HT binds—shares significant homology with numerous aminergic GPCRs (receptors for neurotransmitters with basic amine moieties, e.g., acetylcholine, dopamine, histamine, norepinephrine, serotonin), and this overlap is critical to understanding ligand polypharmacology. Within transmembrane domain 3 (TM3) sits the amino acid residue D^3.32 (Ballesteros-Weinstein numbering system (103)), which is conserved in the binding pocket of all aminergic receptors. This conserved residue is negatively charged and serves to anchor the positively charged amine moiety present in most aminergic receptor ligands within the binding pocket via an ionic bond (Figure 3). Accordingly, point mutation of D^3.32 to uncharged residues drastically diminishes the binding affinity of most agonists and antagonists at aminergic GPCRs, including 5-HT_2ARs (104–107).

Figure 3: Molecular determinants of nonselective binding at aminergic receptors.

Structures of a risperidone-bound 5-HT_2AR (light green, PDB: 6A93) and ritanserin-bound 5-HT_2CR (dark green, PDB: 6BQH) show aromatic and hydrophobic interactions with conserved residues forming a hydrophobic cleft deep in the binding pocket, and an ionic bond between the ligand’s protonated amine and the negatively charged side chain of D^3.32. The whole 5-HT_2AR is shown (left) for perspective, with each transmembrane domain numbered.

The side chains of I^3.40, F^5.47, F^6.44, W^6.48, F^6.51, and F^6.52 form a highly conserved hydrophobic cleft at the base of the orthosteric binding pocket (Figure 3) and contribute to promiscuous antagonist binding at 5-HT_1B, 5-HT_2A, 5-HT_2C, D₂, and H₁Rs (79, 108). Interestingly, flexible-structured antagonists are more sensitive to manipulations in this region of the binding pocket, likely because they extend deeper than more rigid-structured antagonists such as those of the ergoline class (109, 110). For example, the affinity of flexible ligands such as risperidone and pimavanserin is attenuated by >900-fold at point mutated W336L^6.48 (tryptophan to leucine) and F339A^6.51 (phenylalanine to alanine) 5-HT_2ARs (79). Similarly, the affinity of ritanserin at highly homologous 5-HT_2CRs with W324L^6.48 or F327L^6.51 point mutations is attenuated by >10,000- and ~100-fold, respectively (102). In contrast, the affinity of the relatively structurally-rigid ergoline antagonist mesulergine (Figure 4) is attenuated by only 7-fold at W324L^6.48 5-HT_2CRs (102). Recently, the side chains of F243^5.47 and F340^6.52 within the 5-HT_2AR were found capable of expanding the binding pocket to accommodate various antagonist structural moieties (111). As such, mutation of either position to alanine or leucine attenuates the affinity (~5–46-fold) of many antagonists, including: spiperone, ketanserin, pimavanserin, risperidone, aripiprazole, and cariprazine (79, 104, 111). Although the affinity of ketanserin at F340A^6.52 5-HT_2ARs was not different from WT 5-HT_2ARs (111).

Figure 4:

Structures of ergoline-based 5-HT_2AR antagonists variably substituted at the N(1) position (denoted by *).

A subset of residues in the hydrophobic cleft forms the P^5.50–I^3.40–F^6.44 motif, a molecular ensemble wherein side-chain rearrangements are understood to mediate activation of class A GPCRs (112), along with other key, conserved motifs, including CWxP, E/DRY, and nPxxY (113). Accordingly, I^3.40 and F^6.44 are critical for agonist function and antagonist binding at 5-HT_2A and 5-HT_2CRs (79, 102, 104, 114). Indeed, the agonist activity of 5-HT is abolished at I163A^3.40 5-HT_2ARs, as is the affinity of mesulergine (79). Moreover, the affinity of pimavanserin and risperidone is attenuated at F332L^6.44 5-HT_2ARs (10- and 17-fold, respectively) (79), and of ritanserin at I142F^3.40 and F320L^6.44 5-HT_2CRs (20- and 9-fold, respectively) (102).

3.2. Molecular determinants of subtype-selectivity at 5-HT_2ARs

Perhaps the most intuitive mechanism of achieving selectivity is the formation of ligand-receptor contacts with amino acid residues that differ across target and off-target receptors. Identifying such residues allows medicinal chemists to rationally design ligands that interact with these residues to realize a degree of selectivity. By combining site-directed mutagenesis, molecular modeling, and structural biology techniques, several non-conserved residues in the 5-HT_2AR have been identified and offer significant insight into the design of subtype-selective ligands.

3.2.1. S242^5.46

The most thoroughly characterized 5-HT_2AR amino acid residue, S242^5.46 (A225^5.46 and A222^5.46 in 5-HT_2B and 5-HT_2CRs, respectively), has been implicated in species differences in ligand affinity (60, 61, 115), subtype-selective binding (60, 61, 115), binding kinetics (116, 117), and biased signaling (106). Importantly, S^5.46 diverges between humans (serine) and rodents (alanine) and is sufficient to explain species differences in the affinity of ergoline-based antagonists, tryptamine-based agonists (60, 115, 116, 118), as well as a 2-aminotetralin-based agonist (61). Such interspecies variations could be due to limited evolutionary pressure to conserve sequence identity for a non-deleterious mutation (119). Although, drug development programs designing ligands that engage interspecies variant residues (intentionally or not) would necessitate the development of humanized animal models.

The species-dependent affinity of 5-HT_2AR ligands was first shown by Kao et al. (60) using the N(1)-methyl ergoline antagonist mesulergine, which has a lower affinity at cloned human 5-HT_2ARs (S242^5.46, K_d = 174 nM) compared to 5-HT_2ARs in the rat cortex (A242^5.46, K_d = 8.52). Notably, the high affinity of mesulergine at human S242A^5.46 point mutated 5-HT_2ARs (K_d = 3 nM) recapitulated its affinity at WT rat 5-HT_2ARs (60). Johnson et al. (115) extended these findings to various N(1)-substituted ergolines. For example, LY86057, LY108742, and LY53857 differ only in their substitution of the N(1)-position (i.e., −H, −CH₃, and −CH[CH₃]₂, respectively; Figure 4), yet LY86057 (−H) had a higher affinity at human 5-HT_2ARs than LY108742 (−CH₃) or LY53857 (−CH[CH₃]₂). In contrast, LY108742 (−CH₃) and LY53857 (−CH[CH₃]₂) had higher affinity at rat 5-HT_2ARs than LY86057 (−H). However, these trends were reversed when rat 5-HT_2ARs were point mutated to recapitulate human 5-HT_2ARs (A242S^5.46); e.g., the affinity of LY86057 (−H) better resembled its affinity at human 5-HT_2ARs, whereas the affinity of LY108742 (−CH₃) and LY53857 (−CH[CH₃]₂) depreciated by ~4-fold. Such relationships likely stem from a combination of factors, including the availability of favorable electrostatic interactions between the indole N(1)-H and the side chain oxygen of S^5.46, as well as steric tolerance and opportunities for hydrophobic interactions between the indole N(1)-alkyl moieties and the side chain of A^5.46.

The residue at position X^5.46 also diverges between 5-HT_2A (serine) and the highly homologous 5-HT_2B and 5-HT_2CRs (alanine). In an elegant study, Almaula et al. (116) tested the hypothesis that S242^5.46 mediates subtype-selective binding to human 5-HT_2A over human 5-HT_2CRs using exchange mutations to generate S242A^5.46 5-HT_2ARs and A222S^5.46 5-HT_2CRs. Ergolines with an N(1)-H moiety (i.e., ergonovine, LSD, and lisuride) that preferentially bound WT 5-HT_2ARs over WT 5-HT_2CRs, demonstrated lower affinity at S242A^5.46 5-HT_2ARs, and higher affinity at A222S^5.46 5-HT_2CRs. In contrast, ergolines with an N(1)-CH₃ moiety (i.e., mesulergine) that bound preferentially to 5-HT_2CRs over 5-HT_2ARs, demonstrated higher affinity at S242A^5.46 5-HT_2ARs, and lower affinity at A222S^5.46 5-HT_2CRs. These findings suggest that the affinity of ergolines at 5-HT_2A and 5-HT_2CRs depends on hydrogen bonding as well as steric and hydrophobic interactions between the N(1)-substituent and the side chains of S/A^5.46.

While the side chain of S/A^5.46 explains differences in the structure-activity relationships (SAR) of ergolines between human and mouse or rat 5-HT_2ARs, and between human 5-HT_2A and 5-HT_2CRs, the effect of S^5.46 vs. A^5.46 does not reliably extend to other chemotypes. For example, tryptamine derivatives are not reliably impacted by the S/A^5.46 side chain (115, 120). Similarly, we and others have shown that the affinity of various 5-HT_2AR antagonists is not different at S242A^5.46 5-HT_2ARs, including, ketanserin, pimavanserin, risperidone, and certain 2-aminotetralins (79, 80, 115, 116, 121). Taken together, S242^5.46 is not a general determinant of subtype-selective antagonist binding at 5-HT_2ARs over 5-HT_2B and 5-HT_2CRs.

3.2.2. G238^5.42

The residue G^5.42 is unique to 5-HT₂-type receptors and is occupied by larger residues including alanine, cysteine, serine, or threonine in other aminergic GPCRs (4). The importance of this to drug design was highlighted in the crystal structure of a ritanserin-bound 5-HT_2CR, which indicates that one of the 4-fluorophenyl groups on ritanserin contacts the backbone of TM5 at G218^5.42 (102). Given that ritanserin demonstrates some selectivity to bind 5-HT₂-type receptors over other aminergic receptors, Peng and coworkers tested the hypothesis that G218^5.42 was in part responsible for the preferential binding of ritanserin to 5-HT₂-type receptors. They generated a G218S^5.42 5-HT_2CR to mimic the corresponding S199^5.42 residue in 5-HT_1ARs—for which ritanserin has low affinity (K_i = 309 nM, Table 1). Ritanserin showed 60-fold lesser affinity at G218S^5.42 5-HT_2CRs, supporting the hypothesis that G218^5.42 contributes to its preferential binding at 5-HT₂-type receptors. This hypothesis was further supported by a complementary experiment with clozapine, which exhibits appreciable affinity (K_i < 100 nM) for at least 26 receptors including the 5-HT_1AR and has an embedded 4-benzylidenepiperazine core like ritanserin. The affinity of clozapine, however, was attenuated by a mere 2–3-fold at G218S^5.42 5-HT_2CRs (102, 122, 123).

The role of G^5.42 in mediating selective binding at 5-HT₂-type receptors has since been extended to the 5-HT_2AR by Kimura et al. (79) and was recently replicated and expanded upon by our lab (80). Using a combination of molecular modeling and site-directed mutagenesis, this body of work suggests that the selectivity of pimavanserin to bind 5-HT_2A/5-HT_2CRs stems from the isobutoxybenzyl moiety occupying a cavity between TM4 and TM5 afforded by the small side chain of G^5.42 (79, 80). Generation of a G238S^5.42 5-HT_2AR confirmed this binding pose, as pimavanserin showed a near-complete loss in affinity and antagonist activity at the receptor variant (79, 80). Moreover, we used novel 2-aminotetralin derivatives to demonstrate that this effect was driven by the size of the ligand motif extending into the cavity between TM4 and TM5, and this was predictive of ligand affinity over several other aminergic GPCRs (80). In contrast, risperidone, which has high affinity at multiple aminergic receptors and binds to 5-HT_2ARs in an extended conformation spatially distinct from the side cavity, was only modestly affected by the G238S^5.42 mutation (79, 80).

Certain lipids are known to potentiate 5-HT-mediated G protein activation through the 5-HT_2AR (124, 125). Very recently, the structural basis of this effect was delineated using crystal structures of 5-HT_2ARs bound to the agonists 5-HT, psilocin, lisuride, or LSD (126). In each structure, the lipid monoolein snaked into the cavity adjacent to G238^5.42, between TM4 and TM5. The lipids monoolein, oleamide, oleoylethanolamide, and 2-oleoyl glycerol were all shown to modestly activate WT, but not G238S^5.42 5-HT_2ARs, suggesting that G238^5.42 also serves as a structural determinant of lipid-mediated G protein signaling through 5-HT_2ARs. It remains unknown how the lipid environment impacts antagonist affinity or selectivity at 5-HT_2ARs.

Taken together, published experimental and computational work supports the conclusion that G^5.42 contributes to the observed 5-HT_2AR selectivity of structurally distinct antagonists. However, this model incompletely describes 5-HT_2AR selectivity, as pimavanserin demonstrates high selectivity over 5-HT_2B and modest selectivity over 5-HT_2CRs despite the presence of G^5.42 in both receptors. To better understand the molecular and structural features governing selective binding at 5-HT_2ARs over other 5-HT₂-type receptors, a more detailed understanding of GPCR structure is required.

3.2.3. Amino acid ensembles governing 5-HT_2AR selectivity

Apart from single amino acid differences, amino acid ensembles can change the 3-dimensional (3D) architecture of binding pockets. A sophisticated model of this was recently put forth by Kimura et al. in detailing the first 5-HT_2AR crystal structures (79). Structural comparisons of a risperidone-bound 5-HT_2AR, ergotamine-bound 5-HT_2BR, and ritanserin-bound 5-HT_2CR indicated that the length and conformation of extracellular loop 2 (ECL2) was nearly identical between 5-HT_2A and 5-HT_2CRs yet was 4–6 residues longer in the 5-HT_2BR due to a shortened TM4. In addition, divergent amino acid residues near the extracellular ends of TM4 and TM5 within the 5-HT₂-type receptors were proposed to contribute to the formation of a unique side chain rotamer of F^5.38, located one helical turn closer to ECL2 than G^5.42, in the 5-HT_2AR compared to that in 5-HT_2B and 5-HT_2CRs.

In the risperidone-bound 5-HT_2AR crystal structure, a hydrophobic interaction between F213^4.63 and F234^5.38 appeared to orient the side chain rotamer of F234^5.38 toward the extracellular end of TM4, leading Kimura et al. (79) to propose that a 5-HT_2AR-specific rotamer of F234^5.38 increases the volume of the side pocket formed by G238^5.42 to realize a “side-extended cavity” (Figure 5). In contrast, the side chains of F217^5.38 and F214^5.38 in 5-HT_2B and 5-HT_2CRs, respectively, oriented toward the cytosol, effectively reducing the side-cavity’s volume. The orientation of F^5.38 within 5-HT_2B and 5-HT_2CRs appeared to originate from steric restrictions with K193^4.63 in 5-HT_2BRs, and I192^4.63 or D211^5.35 in 5-HT_2CRs (79), potentially explaining the observed selectivity of pimavanserin to bind 5-HT_2ARs over 5-HT_2B and 5-HT_2CRs. Notably, although this work represents the most extensive mutagenesis investigation of the 5-HT_2AR to date, no mutations of 5-HT_2AR were made at positions X^4.63 or X^5.35 to suggest that a side chain rotamer of F^5.38 could be perturbed to impact the affinity of antagonists with subtype-selectivity at 5-HT_2ARs.

Figure 5: Putative molecular determinants of subtype-selective binding at 5-HT₂-type receptors.

Superimposition of 5-HT₂-type receptors suggests that a “side extended cavity” is formed in the 5-HT_2AR (light blue color) by a unique rotamer of F^5.38, making hydrophobic interactions with the non-conserved residue F^4.63 (purple circle, top left). Structures of the 5-HT_2AR (PDB: 6A93, top right), 5-HT_2BR (PDB: 6DS0, bottom left), and 5-HT_2CR (PDB: 6BQH, bottom right) indicate that ECL2 of the 5-HT_2BR exhibits greater conformational freedom compared 5-HT_2A and 5-HT_2CRs, wherein K^ECL2 is electrostatically constrained (orange dashed lines). The side chain of G^5.42 is shown for reference to the side cavity.

To test the role of interactions between F213^4.63 and F234^5.38 in mediating subtype-selective antagonist binding at 5-HT_2ARs, our lab point-mutated F213^4.63 within the 5-HT_2AR to the structurally equivalent residue in 5-HT_2BRs (K^4.63) to mimic the side chain rotamer of F^5.38 found in 5-HT_2BR crystal structures. We then assessed the affinity of several 5-HT_2AR antagonists with varying selectivity, including risperidone and pimavanserin, at point-mutated F213K^4.63 5-HT_2ARs (80). However, no difference in antagonist potency (pK_b) was observed for any ligand compared to WT 5-HT_2ARs. These results suggest that F213^4.63 alone does not meaningfully contribute to facilitating selective antagonist binding to 5-HT_2ARs.

Ligand binding is a dynamic process wherein ligands and amino acid residues sample discreet interaction profiles and conformational states across a probability distribution (see ref. (112) for review). Accordingly, receptor crystal structures are static representations of a dynamic system that may not reliably indicate structurally or pharmacologically meaningful interactions. Therefore, we used molecular modeling to identify residues that could dynamically modulate the rotamer of F234^5.38 in 5-HT_2ARs. Our simulations identified the side chain of D231^5.35 (F214^5.35 in 5-HT_2B and D211^5.35 in 5-HT_2CRs), located one helical turn above F234^5.38, as a candidate for influencing the F234^5.38 rotamer (80). However, D231F^5.35 5-HT_2ARs were functionally null in response to 5-HT, and specific binding was not detected with [³H]spiperone, [³H]ketanserin, or [³H]mesulergine (80). Notably, the absence of a fluorescent tag precluded us from confirming proper membrane localization of D231F^5.35 5-HT_2ARs.

Careful inspection of published 5-HT₂-type receptor crystal structures indicates that the D231F^5.35 mutation could have destabilized the conformation of ECL2, potentially leading to a misfolded receptor and, subsequently, inadequate membrane-trafficking. For example, the crystal structure of a risperidone-bound 5-HT_2AR shows the formation of an electrostatic cage around K220^ECL2, formed by the negatively charged side chains of D217^4.67 and D231^5.35, which may limit the conformational freedom of, and therefore stabilize, ECL2 (Figure 5). Similar interactions are apparent in the crystal structure of a ritanserin-bound 5-HT_2CR, where the positively charged side chain of K199^ECL2 is enclosed by the negatively charged side chains of D196^4.67 and D211^5.35 (Figure 5). Thus, the absence of K^ECL2 near TM4 in the 5-HT_2BR, and the presence of F214^5.35, may promote the conformational freedom observed in 5-HT_2BR ECL2 (127, 128), and contribute to the ability of some antagonists to bind 5-HT_2A and 5-HT_2CRs with higher affinity. Moreover, residues on the extracellular side of D^3.32 are less conserved across 5-HT₂-type receptors than those on the intracellular side of D^3.32, suggesting that antagonist selectivity could involve interactions with residues near ECL2. Although, steric interactions with M218^5.39 in the 5-HT_2BR do not appear to sufficiently explain the low affinity of pimavanserin since it exhibits ~2-fold higher affinity at V235M^5.39 5-HT_2ARs (80).

Concerning the modest selectivity of pimavanserin to bind 5-HT_2A over 5-HT_2CRs, careful consideration of its affinity at, and selectivity over, 5-HT_2CRs is warranted. Using molecular modeling techniques identical to those described elsewhere (80), we show that the binding pose of pimavanserin in 5-HT_2CRs is nearly identical to that reported for 5-HT_2ARs, even insofar as F^5.38 adopts a raised rotamer conformation (Figure 6). Notably, the only non-conserved residues present in the binding pockets of 5-HT_2A and 5-HT_2CRs are S/A^5.46 and I/V^4.56 (80). Although mutagenesis studies do not support a role for S/A^5.46 in mediating subtype selectivity of pimavanserin (79, 80), pimavanserin does exhibit 4-fold increase in potency at I206V^4.56 5-HT_2ARs (79), suggesting that steric tolerance at this position may enhance affinity at 5-HT_2CRs.

Figure 6: Proposed binding mode of pimavanserin at 5-HT_2A (light green) and 5-HT_2CRs (dark green).

Molecular docking and molecular dynamics simulations indicate that pimavanserin engages nearly identical amino acid residues within the binding pocket of 5-HT_2A and 5-HT_2CRs. The primary determinant of selectivity, the extension of the isobutoxybenzyl moiety into the side-extended cavity between TM4 and TM5 (purple circle), as allowed by the small side chain of G^5.42, is conserved in both receptors and is accompanied by a raised rotamer conformation of F^5.38 in both receptor subtypes.

Insight into the ability of ligands to discriminate between 5-HT_2A and 5-HT_2CRs comes from a recent report detailing the crystal structure of a spiperone-bound D₂R that proposes a mechanism for the ability of spiperone to bind D₂Rs and 5-HT_2ARs selectively over 5-HT_2CRs (129). Although the D₂R lacks a side-extended cavity, it does possess a spatially distinct extended binding pocket. Located between TM2 and TM3, the extended binding pocket—comprised of V87^2.57, W90^2.60, V91^2.61, L94^2.64, W100^23.50(ECL1), F110^3.28, V111^3.29, and C182^45.50(ECL2)—accommodates the phenyl ring of spiperone. Notably, the side chains of W90^2.60 in D₂Rs and V130^2.60 in 5-HT_2ARs allow the side chain rotamers of F110^3.28 and W151^3.28 to flip and expose the extended binding pocket. In 5-HT_2CRs, however, the bulkier side chain of L109^2.60 may restrict W130^3.28 from flipping, thereby impeding the formation of an extended binding pocket (129). These findings were supported by the observation that the affinity of spiperone at W90L^2.60 D₂Rs was more severely attenuated (~20-fold) than at F110W^3.28 D₂Rs (~11-fold) (129), although the affinity of spiperone at W90V^2.60 D₂Rs was not reported. These findings are supported by medicinal chemistry work noting that replacing the phenyl ring on spiperone with smaller lipophilic substituents (e.g., −H, −CH₃, −CH₂CH₃, or − CH[CH₃]₂) negatively impacted affinity at 5-HT_2A, 5-HT_2C, and D₂Rs (130). In conclusion, these structural differences may contribute, in part, to the observed selectivity of spiperone to bind D₂ and 5-HT_2ARs over 5-HT_2CRs.

3.3. Alternative approaches for designing 5-HT_2AR selective antagonists

Most aminergic ligands incorporate a basic amine moiety that, upon protonation, can form a strong ionic bond with D^3.32 in the orthosteric binding pocket (105, 131). Such an anchoring strategy can be helpful in generating high affinity ligands, although protonated amines are also significant contributors to ligand promiscuity. In fact, screening studies have found that 20–50% of positively charged novel chemical entities bind to aminergic receptors with an IC₅₀ < 10 μM, and serotonin receptors are among the most common targets (132, 133). Therefore, the design of non-basic ligands can be an effective strategy for developing novel, selective ligands.

Non-basic ligands comprise the most subtype selective 5-HT₂-type receptor antagonists available. For example, RS-127445 is a 5-HT_2BR antagonist that does not possess a basic amine moiety and is not positively charged at physiological pH. Yet, it displays >1,000-fold selectivity over 5-HT_1A, 5-HT_1B, 5-HT_1D, 5-HT_2A, 5-HT_2C, 5-HT₃, 5-HT₅, 5-HT₆Rs, and 100 other receptors and ion channels, determined by Cerep screening (134). Successfully generating a selective 5-HT_2BR antagonist is an impressive accomplishment since ligand affinity at 5-HT_2BRs may predict pharmacological promiscuity, with some surveys finding that ~50% of positively charged compounds interact with 5-HT_2BRs (132, 133). Similarly, indoline urea derivatives such as SB 242084, SB 243213, and SB 206553 lack basic amines and are among the most selective antagonists targeting 5-HT_2B and/or 5-HT_2CRs (135–137). Thus, rational strategies to identify non-basic 5-HT_2AR ligands hold promise for discovering novel subtype-selective ligands. One such approach was taken to discover 6-(1-ethyl-3-(quinolin-8-yl)-1H-pyrazol-5-yl)pyridazin-3-amine, a non-basic subtype-selective 5-HT_2CR agonist, identified through screening a compound library against D134A^3.32 5-HT_2CRs—an active site receptor variant that is functionally responsive to neutral, but not positively charged, agonists (138). Analogous point-mutated 5-HT_2ARs might also be useful in identifying selective non-basic inverse agonists in platforms like R-SAT^™ (68).

Results from recent molecular docking studies provide an alternative approach to achieving subtype-selective binding. Docking studies with SB 242084 suggest that it may interact with the same conserved amino acid residues in each of the 5-HT₂-type receptors, but that its selectivity stems from its ability to make closer, more numerous hydrophilic contacts in the 5-HT_2CR, e.g., with D134^3.32, S138^3.36, and Y358^7.43 (117). Moreover, ligand affinity is defined by the relationship between the association rate (k_on) of a free ligand to an unbound receptor, forming a binary complex, and the dissociation rate (k_off) of the bound ligand from the binary complex (K_d = k_off/k_on). Conceptually, it is helpful to consider k_off as reflecting the stability of a ligand-receptor complex, with more stable complexes having lower k_off values, and hence, higher affinity. Given this, it is interesting to speculate that closer hydrophilic contacts between SB 242084 and 5-HT_2CR residues may promote a uniquely stable ligand-receptor complex with a low k_off value. Further computational and experimental investigations to this end would provide a more textured understanding of how ligands are stabilized in 5-HT₂-type receptors.

4. Computational approaches to assist in drug design targeting 5-HT_2ARs

4.1. Ligand-based drug discovery: QSAR

In the pre-structural biology era, 3D quantitative structure-activity relationships (3D-QSAR) were commonly used in drug discovery programs targeting 5-HT receptors. Among the most popular methods is comparative molecular field analysis (CoMFA), which correlates the biological activity (e.g., binding affinity) of a series of compounds with molecular features governing non-covalent ligand-receptor interactions, such as electrostatic and steric properties. Compared to conventional QSAR models based on the Hansch analysis, CoMFA models are similarly capable of predicting biological activity yet can incorporate structurally diverse analogs typically omitted from traditional QSAR analyses (139).

The CoMFA method to predict the activity of novel ligands is based on a training set of experimental data. For example, 3D-QSAR studies on a series of 3-(1,2,5,6-tetrahydropyridin-4-yl)indole derivatives targeting 5-HT_2ARs, yielded a model that accurately predicted the affinity of six novel ligands based on a training set composed of 45-compounds (139). Moreover, CoMFA models modified to account for hydrogen bonding and hydrophobic interactions have been generated to predict electrostatic, steric, and hydrophobic determinants of affinity for a series of conformationally restricted butyrophenones at rat 5-HT_2A and D₂Rs (140).

4.2. Use of homology modeling and molecular docking to delineate mechanisms of receptor inactivation and ligand affinity

Several GPCR crystal structures were solved in the early 2000s, including rhodopsin (141) and human β₂-adrenergic receptors (142). They were widely used to build homology models of 5-HT receptors to better understand the molecular mechanisms governing receptor function and ligand affinity and selectivity. Using a 5-HT_2AR model based on the crystal structure of bovine rhodopsin in combination with molecular modeling, site-directed mutagenesis, and functional assays, Shapiro et al. (143) investigated the molecular mechanism of 5-HT-mediated 5-HT_2AR activation. Their findings were consistent with the hypothesis that an ionic bond between R173^3.50 and E318^6.30, on the intracellular face of the receptor, serves as an essential interaction that is disrupted upon receptor activation (143). The side chain of R173^3.50 has also been implicated in resolving the inverse agonist activity of constitutively activated C322K^6.34 5-HT_2AR variants, supported by molecular docking and molecular dynamics simulations with a model of the 5-HT_2AR based on metarhodopsin (144, 145).

A homology model of the 5-HT_2AR based on rhodopsin helped predict the relative contribution of S239^5.43 and S242^5.46 in the binding of tryptamine and phenylalkylamine-based agonists. Using a combination of site-directed mutagenesis and radioligand binding to confirm model predictions, Braden and Nichols demonstrated that a hydrogen bond could form between S239^5.43 and tryptamines substituted at the 4- or 5-position with hydrogen-bond acceptors, as well as phenylalkylamines substituted at the 4-position (121). Moreover, phenylalkylamine binding was predicted to involve a hydrogen bond between the 2- and 5-position substituents and S159^3.36 and S239^5.43, respectively. This prediction was supported over a decade later by the structure of a 5-HT_2AR bound to the phenylalkylamine derivative N-(2-hydroxybenzyl)-2,5-dimethoxy-4-cyanophenylethylamine, demonstrating the formation of a unique hydrogen bond between the C(2)-OH moiety and the side chain of S159^3.36 (106). In contrast, the side chain of S159^3.36 does not appear to be involved in the binding of LSD or ketanserin at 5-HT_2ARs (106, 120).

4.3. Virtual high-throughput screening

Traditional drug discovery programs often rely on experimental HTS with large compound libraries in biological assays, and known active ligands, to identify leads for SAR and guide drug development (146). However, while these methods have proved fruitful, they barely scratch the surface of exploring the chemical space of drug-like molecules—routinely cited to be >10²⁰—thus limiting the pace of medication development (147). Compared to experimental HTS, virtual HTS, pairing large computational libraries of compounds and molecular docking, is a more economical screening approach to drug discovery (148–151). Several studies have used virtual HTS with models of 5-HT₂-type receptors to identify possible novel ligands (152, 153), with some predicted to demonstrate subtype-selective binding (154). Importantly, these investigations have led to the discovery of experimentally validated novel ligands with high affinity at (155), or selectivity for (156, 157), particular 5-HT₂-type receptors.

With the increasing abundance of experimental 5-HT_2AR structures, 13 at the time of writing (79, 106, 111, 126), coupled with the availability of free ultra-large ligand libraries, including the ZINC database (http://zinc20.docking.org)—an expanding library of over a billion compounds—the discovery of novel ligands with new scaffolds is increasingly accessible. Virtual HTS and ultra-large library docking have proven fruitful in identifying novel and potent receptor modulators (158–161). It seems increasingly likely that these methods will assume a more prominent role in drug discovery campaigns searching for novel 5-HT_2AR ligands as we advance. To predict the functional activity of novel ligands as agonists or antagonists, and especially functionally selective (biased) agonists, however, requires additional computational methods capable of tracking global changes in receptor structure and receptor interactions with signaling proteins.

4.4. Molecular dynamics as a method to predict functionally relevant receptor conformations

Both structural and dynamic factors determine protein function. To better understand the dynamic and temporal aspects of protein function, molecular dynamic (MD) simulations are an integral and powerful computational approach. For example, one microsecond-long MD simulations using a 5-HT_2AR homology model bound to psychedelic or non-psychedelic 5-HT_2AR agonists identified distinct conformations of intracellular loop 2 (ICL2) characteristic of psychedelic agonists (162). These findings suggest that psychedelic 5-HT_2AR agonists could stabilize discreet receptor conformational states relevant to intracellular signaling cascades, as suggested by experimental work (163). Moreover, MD simulations have been used to predict that the side chains of S242^5.46 and N343^6.55 in the 5-HT_2AR contribute to biased agonism, a prediction later supported by experiment (164).

When designed to encompass a sufficient temporal window, MD simulations may also distinguish agonist from antagonist ligands at the 5-HT_2AR. This was shown by Shan et al., using 350 nanosecond MD simulations of a 5-HT_2AR model, based on rhodopsin and β₂AR crystal structures, in complex with a full agonist, a partial agonist, and an antagonist of canonical 5-HT_2AR signaling (i.e., 5-HT, LSD, and ketanserin, respectively) (165). The simulations indicated that ligands of varying functional activity differentially influenced well-documented microswitches of GPCR activation and stabilized conformational states of the receptor that lead to distinct deformations in the surrounding plasma membrane. Using a complementary approach, we have aimed to expand the work described elsewhere (143) to characterize 5-HT_2AR dynamics with one microsecond MD simulations of an apo-5-HT_2AR, methiothepin-bound 5-HT_2AR (antagonist-state), and an ergotamine-bound 5-HT_2AR (agonist-state). We confirmed that a stable ionic bond (~2 Å) between R173^3.50 and E318^6.30 locks the apo-5-HT_2AR in an inactive state. This bond is also present in the inactive methiothepin-bound 5-HT_2AR, and broken (~4 Å) upon conformational transitions to the active state in the ergotamine-bound 5-HT_2AR between 300–700 nanoseconds (Figure 7). Similar findings have been presented independently for the 5-HT_2BR (166).

Figure 7:

Molecular dynamics simulation results of an unbound, antagonist-, and agonist-bound 5-HT_2AR (APO, methiothepin, and ergotamine, respectively). Ionic lock (E318^6.30—R173^3.50) distances are shown as a function of simulation time.

5. Considerations and conclusions

The development of selective antagonists for clinical and preclinical purposes will likely continue, although the methods used to discover them are destined to evolve. Historically, the discovery of putatively selective 5-HT_2AR antagonists relied on experimental screening and lead optimization. Yet, sparingly few compounds have withstood the test of rigorous receptor profiling in vitro, and the number of chemically diverse antagonists with robust selectivity for 5-HT_2ARs remains low. Moving forward, it would be incumbent for medicinal chemists and pharmacologists reporting on selective ligands to clearly describe the context used to define a ligand as “selective”. As an example, a ligand may be defined as selective if it has >100-fold higher affinity at its target GPCR, based on in vitro affinity tests at a pre-defined set of off-target GPCRs within the target’s receptor family, e.g., “compound X is selective for 5-HT_2A relative to 5-HT_2B, and 5-HT_2CRs based on in vitro affinity assays using antagonist radiolabels.” Moreover, novel compounds should be characterized alongside reference ligands in orthogonal assays to account for affinity differences due to assay conditions and facilitate fair comparison across literature reports (80, 167). Authors should then note that the compound may only be useful—without further characterization—as an in vitro tool.

Discerning ligand selectivity with confidence in vivo poses a distinct challenge, due to a need to quantify ligand occupancy at on- and off-target binding sites across multiple tissues, doses, and time points following administration. In addition, the pharmacokinetic parameters used to determine occupancy will likely vary between routes of administration (i.e., intraperitoneal, subcutaneous, oral, or intravenous), and thus would require further experimentation to validate selectivity following dosing across routes. These experiments may be seen as cost-prohibitive, especially for academic labs developing medication candidates, but these data ultimately would help prevent misinterpretation of results and provide the broader scientific community (e.g., behavioral pharmacologists, neurophysiologists, etc.) critical information to guide the selection, and dosing, of tool compounds for in vivo use.

In the absence of detailed pharmacokinetic profiling, there is the modus operandi of titrating antagonist dose to prevent behaviorally disruptive or off-target effects while achieving a specific behavioral outcome. However, there can be problems with this approach if the behavioral outcome under evaluation is affected by multiple targets. For example, the head-twitch response in mice is associated with activation of the 5-HT_2AR, yet is also modulated through numerous other mechanisms (163, 168, 169). Without the pharmacodynamic and pharmacokinetic information described above, we cannot make firm conclusions about selective receptor engagement in vivo or about the role of the GPCR target in the measured outcome. For example, we have shown that M100,907—which has ~160-fold in vitro selectivity for mouse 5-HT_2A over 5-HT_2CRs—significantly suppresses the head-twitch response in adult, male C57BL/6J mice elicited by 1 mg/kg (±)-2,5-dimethoxy-4-iodo-amphetamine (s.c.), when administered at 0.0025 mg/kg (s.c.) (71), though, at this dose, only ~35% of the head-twitch response is blocked. Meanwhile at 0.025 mg/kg and 0.25 mg/kg M100,907, ~85% and 100% of the head-twitch response is blocked, respectively. Without evaluating receptor occupancy at these doses, it is not possible to conclude what dose (if any) of M100,907 selectively blocks central 5-HT_2ARs that elicit the head-twitch response.

Selective knockdown of 5-HT_2AR with genetic tools in combination with autoradiography would also clarify the in vivo selectivity of a ligand following administration at a defined dose, route, timestep. For example, [³H]M100,907 could be administered at various doses to 5-HT_2AR knockout mice together with another, structurally unique 5-HT_2AR antagonist, and receptor occupancy could be used to determine the dose at which it binds off-targets, considering potential issues of non-specific binding.

There is also the option of combining ligands in an experiment to interrogate mechanism. For example, SB 242084, a putatively selective neutral antagonist of 5-HT_2CRs might be co-administered with pimavanserin to mask the potent inverse agonist activity of pimavanserin at 5-HT_2CRs. However, SB 242084 would block activity-dependent signaling through 5-HT_2CRs, like pimavanserin, without preserving its native signaling capacity in response to synaptic release of 5-HT. Moreover, ligand combinations increase the possibility of confounding pharmacokinetic and pharmacodynamic interactions in vivo, e.g., competition for common metabolic enzymes might increase the plasma concentration of each ligand, and parent compound metabolites, and their pharmacological activities, are often unknown.

Here, we conclude that M100,907 and pimavanserin, on balance, are the most authenticated putatively selective 5-HT_2AR antagonists available. Nevertheless, M100,907 and pimavanserin, as well as all other 5-HT_2AR-selective antagonists, have off-target activities that need to be considered when used to elucidate biological mechanisms. Thus, we echo a recommendation made nearly forty-years ago (8) that mechanistic investigation of 5-HT_2ARs by pharmacological antagonism should utilize two structurally distinct antagonists with different off-target activities, albeit, feasibility of conducting the same experiments twice with different antagonists is limited by increased costs.

For the reasons detailed above, we define “selective” empirically within a context, meaning that if a ligand can be administered such that it acts on its intended target without acting on pre-defined off-targets, then it is selective. We acknowledge the shortcomings of this definition, because of the “unknown unknowns” phenomenon, i.e., there are tens of thousands of proteins, and ~800 GPCRs, expressed in mammalian tissue, and a ligand could have activity at any one of them. Because this is a critical issue, we believe that medicinal chemists and pharmacologists should convene to establish parameters for defining a ligand as “selective” with the goal of providing the general scientific community information to choose dose ranges for in vivo studies in particular species.

The design of novel unambiguously selective 5-HT_2AR antagonists will require expertise spanning various fields. Structural biology studies could help by delineating the structure of 5-HT₂-type receptors bound to subtype-selective antagonists from multiple chemotypes to identify points of contact reliably involved in subtype-selective binding. Notably, structural biology studies often detail a single receptor conformational state, leaving unanswered questions about the dynamic nature of the receptor. Therefore, a library of diverse ligand-receptor structures could improve the efficiency of ultra-large library docking and prediction of ligand function in silico, which will likely lead to the next generation of selective 5-HT_2AR antagonists. Nevertheless, the proper implementation of MD simulations in the context of drug discovery is still a challenge, and computational approaches supported by experimental evidence will provide the most compelling models for ligand-receptor interactions.

Abbreviations

LSD

lysergic acid diethylamide

R-SAT^™

receptor selection and amplification technology^™

TM

transmembrane domain

ECL

extracellular loop

ICL

intracellular loop

Q-SAR

Quantitative structure-activity relationships

CoMFA

comparative molecular field analysis

MD

molecular dynamics

HTS

high throughput screening