Minireview: More Than Just a Hammer: Ligand “Bias” and Pharmaceutical Discovery

Abstract

Conventional orthosteric drug development programs targeting G protein-coupled receptors (GPCRs) have focused on the concepts of agonism and antagonism, in which receptor structure determines the nature of the downstream signal and ligand efficacy determines its intensity. Over the past decade, the emerging paradigms of “pluridimensional efficacy” and “functional selectivity” have revealed that GPCR signaling is not monolithic, and that ligand structure can “bias” signal output by stabilizing active receptor states in different proportions than the native ligand. Biased ligands are novel pharmacologic entities that possess the unique ability to qualitatively change GPCR signaling, in effect creating “new receptors” with distinct efficacy profiles driven by ligand structure. The promise of biased agonism lies in this ability to engender “mixed” effects not attainable using conventional agonists or antagonists, promoting therapeutically beneficial signals while antagonizing deleterious ones. Indeed, arrestin pathway-selective agonists for the type 1 parathyroid hormone and angiotensin AT₁ receptors, and G protein pathway-selective agonists for the GPR109A nicotinic acid and μ-opioid receptors, have demonstrated unique, and potentially therapeutic, efficacy in cell-based assays and preclinical animal models. Conversely, activating GPCRs in “unnatural” ways may lead to downstream biological consequences that cannot be predicted from prior knowledge of the actions of the native ligand, especially in the case of ligands that selectively activate as-yet poorly characterized G protein-independent signaling networks mediated via arrestins. Although much needs to be done to realize the clinical potential of functional selectivity, biased GPCR ligands nonetheless appear to be important new additions to the pharmacologic toolbox.

Despite the fact that heptahelical G protein-coupled receptors (GPCRs) are by far the most successfully exploited class of drug targets, accounting for nearly half of all pharmaceuticals in current use (1), the conceptual framework guiding GPCR drug discovery programs for decades has been remarkably simple. Dating back to the original application of allosteric models to membrane receptor function in the 1960s (2, 3), the basic concepts are that GPCRs exist in equilibrium between conformationally discrete “off” and “on” states that are distinguished by their ability to trigger downstream responses, and that ligands act by perturbing this equilibrium (4, 5). Within this framework, the actions of a ligand can be fully described by only 2 terms; the equilibrium dissociation constant of the ligand-receptor complex (K_d), and the maximal observed change in receptor activity (V_max). Hence, GPCR ligands are classified as agonists if they can elicit a maximal response, partial agonists if they only generate a submaximal response at saturating ligand concentration, and antagonists if they lack intrinsic efficacy but competitively inhibit agonist responses. Later refinements of this “2-state” model, such as the “extended ternary complex” (6) and “cubic ternary complex” (7) models that were developed to explain the capacity of “inverse agonists” to reduce the basal activity of constitutively active mutated GPCRs, simply added terms accounting for the probability that the receptor might spontaneously transition to the active state in the absence of ligand. They did not consider the possibility of multiple active states.

According to the American psychologist Abraham Maslow, “if all you have is a hammer, everything looks like a nail” (8). The pharmacologic equivalent of Maslow's hammer is shown in Figure 1A. If GPCRs can only be off or on, then all ligands can do is change the conformational equilibrium, increasing the proportion of receptors in the on state in settings in which receptor activity is insufficient and decreasing it in the presence of excess endogenous agonist. Thus, conventional agonists and antagonists change the quantity of receptor activity, but only the receptor determines what signals are transmitted by the on state. Partial agonists, by virtue of their inability to completely shift the receptor equilibrium at saturating concentration, may exert “protean” effects (9) in systems with differing levels of constitutive basal receptor activity, but even they do not qualitatively change signaling.

Figure 1.

Evolving concepts of orthosteric GPCR ligand action. A, The conventional view of ligand efficacy assumes that all downstream GPCR signaling arises from a single “on” state. In this case, agonists (Ag) can increase receptor activity (R*) when levels of the endogenous ligand (H) are insufficient, and antagonists (Ant) can decrease receptor activity (R) in the face of endogenous ligand excess, but only the intensity of signaling is changed, not its character. B, Schematic depicting a hypothetical GPCR with 5 conformationally distinct “active” states (R*1–R*5), each of which couples the receptor to downstream G protein (Gs; Gq/11; G12/13) and non-G protein (arrestin2 [Arr2]; arrestin3 [Arr3]) effectors with different efficiency. Note that the 1:1 coupling between active state and effector depicted is an oversimplification. In such a system, a full agonist (A) will produce a full system response in all downstream effectors, just as in the conventional model. In contrast, “biased” agonists (B) engage different active receptor conformations with variable intrinsic efficacy, a property that permits them to activate some downstream pathways, eg, arrestin-dependent signals, while antagonizing others. The ability to engender “mixed” effects permits biased agonists to qualitatively change GPCR signaling. AC, adenylyl cyclase; GEF, guanine nucleotide exchange factor; LIMK, lim domain-containing kinase; PKA, protein kinase A; PKC, protein kinase C; PLC, phospholipase C; MEK, MAPK kinase.

If all you have is a hammer, then the only way forward is to find new nails to drive. By the mid-1990s, innovative GPCR drug discovery was limited to finding new targets for conventional agonist and antagonist ligands, or focusing on subtype selectivity for known drug targets like the muscarinic cholinergic and serotonin receptors. With the knowledge that more than half of the nearly 1000 GPCRs present in the human genome were orphan receptors with unknown ligands and physiological functions, “deorphanizing” GPCRs became the major hope for finding new drugs (10, 11), and many promising drug targets, like ghrelin, orexins/hypocretins, and endocannabinoid receptors (12, 13) have emerged from these efforts. But what have been missing are new tools. More than just hammers, we have needed pharmacologic tools to expand the scope of efficacy beyond conventional agonism and change GPCR signaling qualitatively, not just quantitatively.

The Basis of Bias

Two concepts that together form the foundation for a novel class of GPCR-targeted therapeutics have gained attention over the last decade. The first of these, termed “pluridimensional efficacy” by Michel Bouvier in 2006 (14), recognizes that GPCRs signal by engaging multiple G protein and non-G protein effectors. Although it had long been recognized that GPCRs can couple to 2 or more unrelated G protein families at physiological levels of expression, enabling a single receptor to engage multiple signaling pathways simultaneously or activate them differentially in different tissues (15–17), it was the recognition that a robust G protein-independent signaling network lay beneath the surface of heterotrimeric G protein signaling that revealed the true breadth of GPCR efficacy (18, 19). The best-characterized non-G protein effectors are the arrestins, a family of 4 cytosolic proteins first described for their role in GPCR desensitization and the modulation of phototransduction (20). Arrestins bind tightly to agonist-occupied receptors that have been phosphorylated by GPCR kinases (GRKs) (21). Arrestin binding sterically precludes further G protein activation, producing homologous desensitization, and in the case of the nonvisual arrestins, arrestin2 and 3 (β-arrestin1 and 2), the arrestin C terminus binds elements of the clathrin endocytic machinery, leading to the clustering and sequestration of desensitized receptors (22, 23).

But the end is also the beginning, because arrestins also function as ligand-regulated scaffolds, recruiting a host of arrestin-bound proteins to agonist-occupied GPCRs and initiating a “second wave” of signaling that begins as G protein signaling wanes and persists as receptors coalesce and transit the intracellular compartment (19). Beginning in 1999 with the report that arrestin2 bound the nonreceptor tyrosine kinase, c-Src, and recruited it to activated β₂-adrenergic receptors, studies linking arrestins to GPCR activation of novel enzymatic effectors began appearing with regularity (24). Among others, putative arrestin-regulated effectors now include Src family kinases (24–26), components of the ERK1/2 and c-Jun N-terminal kinase 3 MAPK cascades (27–29), the E3 ubiquitin ligase, Mdm2 (30), the cAMP phosphodiesterases, PDE4D3/5 (31), diacylglycerol kinase (32), the inhibitor of nuclear factor-κB, IκBα (33), the Ral-GDP dissociation stimulator, Ral-GDS (34), the actin filament-severing protein, cofilin (35), and the Ser/Thr protein phosphatase 2A (PP2A) (36, 37). Although the full extent of physiologically relevant arrestin signaling remains inadequately characterized, it is increasingly apparent that arrestin-dependent GPCR signaling performs numerous functions, among them enhancing second messenger degradation, regulating cytoskeletal dynamics controlling vesicle trafficking, exocytosis and cell migration, and promoting cell survival, growth, and hyperplasia (19, 38).

The second critical thread in the bias story is the formulation of a general allosteric model of GPCR signaling by Kenakin and colleagues beginning in the 1990s (39–41, 43). As more techniques to measure GPCR efficacy were developed, it became apparent that GPCRs do not behave as though they have a single “active” conformation that transmits all downstream responses with equal efficiency (44, 45). Reports of “reversal of potency,” in which 2 ligands exhibit opposite rank order of potency for 2 downstream responses measured in the same system (46–49); or “reversal of efficacy,” in which a single ligand exhibits opposing efficacy toward 2 different downstream responses (50–52), offered definitive proof that structurally distinct ligands can activate the same GPCR in different ways, something that can only occur if there is more than one “active” receptor state. Biochemical (53–55) and biophysical (56–58) approaches provided additional evidence that GPCRs adopt multiple active states that couple the receptor to different downstream effectors with different efficiency.

Kenakin's allosteric model (40, 44, 59) envisions GPCRs not as off-on switches, but as “ensembles” of tertiary conformations, some subset of which elicit measurable outcomes that operationally define them as active states. At any instant, the biological activity of the system thus reflects the distribution of the receptor population across each of the component microstates that make up the conformational ensemble. Any molecular interaction that changes the energy landscape of the system has the potential to affect the conformational ensemble in a way that affects signaling. Considered as allosteric proteins, GPCRs are thus susceptible to numerous inputs that modify their signaling properties. In addition to orthosteric effects exerted via the ligand-binding site, lateral allosteric effects arising from protein-protein interactions within the plasma membrane or cytosol, and allosteric modulation arising from the interaction of small molecules with sites outside the ligand-binding pocket, all affect downstream signaling. In effect, orthosteric ligands and allosteric modulators work the same way and can be described in the same mathematical terms (43).

The concepts of “agonist trafficking,” “functional selectivity,” and “biased agonism” merge the principles of pluridimensional efficacy with allosteric models of GPCR efficacy to produce a comprehensive description of orthosteric ligand behavior (39, 59–64). If receptors are conformationally flexible, there is no a priori reason that the conformational ensemble stabilized by one ligand should duplicate either the distribution of spontaneously formed active states or those generated by a structurally distinct ligand. Figure 1B illustrates the differences between conventional and biased agonism using a hypothetical GPCR with 5 conformationally distinct “active” states (R*1–R*5), each of which couples the receptor to downstream G protein and non-G protein effectors with different efficiency. In such a system, a conventional full agonist (A) can be viewed as possessing equivalent intrinsic efficacy for R*1–R*5, ie, it does not discriminate between them. Although there are exceptions (65), native ligands are mostly full agonists, because they coevolved with the receptor to stabilize active receptor conformations in the correct proportions to elicit the most physiologically adaptive set of downstream signals. Like full agonists, conventional partial agonists would not discriminate between active states; they simply possess lower intrinsic efficacy such that only a fraction of the receptors will be in an active state at full receptor occupancy. In contrast, the intrinsic efficacy of a biased agonist (B) differs between active states, such that at full receptor occupancy some active states are favored (R*4–R*5) over others (R*1–R*3).

It is important to recognize that under the right circumstances both partial and biased agonists can activate only a subset of the receptor's full signaling repertoire. In the case of a partial agonist, responses that are tightly coupled or highly amplified will be preserved because fractional receptor activation is sufficient to generate a maximal response, whereas responses that are weakly coupled or that have an activation threshold, in which full receptor activation is necessary for a maximal response, will be disproportionally decreased or absent. Similarly, unamplified responses, such as many of those transmitted via stoichiometric receptor-arrestin complexes, may be undetectable due to the reduced fraction of receptors in an active conformation. This form of variability, arising from differences in “signal strength,” does not require the existence of multiple active states, and the rank order of potency of a series of conventional full and partial agonists will not vary when multiple responses are compared, because the intrinsic efficacy of each ligand is the same across all active states (59).

On the other hand, biased agonists, because they only stabilize a subset of active states, possess the ability to activate some pathways while antagonizing others, leading to qualitative differences in signaling. Soon after the discovery of arrestin-dependent signaling, it became apparent that the G protein- and arrestin-coupled modes of GPCR signaling were pharmacologically dissociable, ie, ligands could discriminate between receptor conformations responsible for these 2 distinct forms of GPCR signaling (66). Ligand bias can range from relatively modest deviations in rank order of potency, to frank reversal of efficacy, wherein the characterization of a ligand as agonist, antagonist, or inverse agonist is completely assay dependent. Among the more dramatic examples of functional selectivity are ligands like ICI118551 and [D-Trp¹²,Tyr³⁴]-PTH(7–34), that function as inverse agonists for heterotrimeric G protein signaling while acting as partial agonists for arrestin-dependent GPCR sequestration and arrestin signaling (67, 68). Such ligands are clearly activating the same receptor in different ways than the native agonist; thus it is the ligand-receptor complex, not the receptor alone, that is specifying the cellular response. In effect, the receptor bound to a biased ligand is a different functional entity, with different signaling characteristics, than the same receptor bound to a conventional agonist or partial agonist.

The Faces of Bias

The goal of any pharmaceutical agent is to modify a biological response in target cells. An orthosteric ligand, conventional or biased, can only effect such change via the ligand-binding pocket of its cognate receptor. Everything that follows, from activation of effectors and second- messenger generation to integrated biological responses like cell proliferation, differentiation, survival, and death, is encoded at that one point.

Biased efficacy in vitro

What range of effects can be obtained through functional selectivity? Allosteric models of GPCR function are based on the concept of conformational selection, which posits that structurally distinct ligands may stabilize active receptor conformations in different proportions than the native ligand, but do not force receptors into sterically unfavorable “novel” conformations or couple them to novel effectors. If the native ligand is truly pluripotent, ie, couples the receptor to all possible effectors, then the actions of a biased ligand must comprise a subset of the whole. Experimental data examining the actions of biased ligands in vitro using a wide range of readouts tend to bear this out.

The most immediate consequence of ligand binding is a change in receptor conformation, which can be monitored at different points within the receptor using intramolecular fluorescence probes (56, 57). In contrast to rhodopsin, in which covalently bound trans-retinal is sufficient to stabilize the active state of the receptor even in the absence of a G protein (69, 70), ligand-induced activation of the β₂-adrenergic receptor involves generation of a number of intermediate conformations. Using ¹³CH₃-ϵ-Met nuclear magnetic resonance, Nygaard et al (71) found that whereas agonist binding destabilizes the inactive state, it does not stabilize the fully active conformation observed in the β₂-adrenergic receptor-G_s complex (72). This higher degree of conformational heterogeneity may permit structurally distinct ligands to direct receptor coupling to different downstream effectors. When measured using site-specific one-dimensional ¹⁹F-nuclear magnetic resonance spectroscopy, agonist, inverse agonist, and arrestin-selective β₂-adrenergic receptor agonists were found to differentially affect the movement of transmembrane helices VI and VII (73), such that G protein activation correlated with movement of helix VI, whereas coupling to arrestins depended upon movement of helix VII. Importantly, helix VI and helix VII were shown to move independently, providing a physical basis for biased agonism and a conformational “signature” predictive of arrestin selectivity. Dynamic studies performed using atomic force microscopy-based single-molecule force spectroscopy similarly demonstrated that ligands differentially perturb the conformational equilibrium of the receptor (74), presumably thereby biasing the probability of coupling to different effectors.

The first step in signal transduction occurs when the activated GPCR touches its proximal effectors, events that can be monitored in live cells using bioluminescence resonance energy transfer-based technologies (75, 76). A recent study by Saulière et al (77) using a comprehensive panel of bioluminescence resonance energy transfer-based sensors encompassing all heterotrimeric G protein families and arrestins to compare the efficacy of the native angiotensin AT_1A receptor agonist, angiotensin II (AngII), with that of the arrestin-biased agonist, [Sar¹,Ile⁴ Ile⁸]-AngII, offers insights into physical basis of signal bias. In this study, [Sar¹,Ile⁴,Ile⁸]-AngII was found to behave as a weak partial agonist for heterotrimeric G proteins, proportionally activating the same G protein species as AngII, albeit with much lower efficacy than the native ligand. Reflective of its arrestin signal bias, [Sar¹,Ile⁴,Ile⁸]-AngII recruited arrestin3 to the AT_1A receptor with nearly comparable efficacy. Consistent with the theory of conformational selection, the data suggested that biased ligands select receptor conformations that couple the receptor to downstream effectors with different efficiency than the native ligand, but do not drive the receptor to engage otherwise unrecognized effectors.

The next level of signal propagation occurs as G proteins, arrestins, and potentially other non-G protein effectors, activate second messenger pathways and protein phosphorylation cascades. This is the level at which most in vitro studies of ligand bias have been performed, and, to date, most have failed to find unique signaling events produced by biased agonists. Rather, biased ligands appear to activate a subset of the downstream second messenger and protein kinase cascades regulated by their conventional counterparts (19, 78). In their systematic comparison of AngII and [Sar¹,Ile⁴,Ile⁸]-AngII, Saulière et al (77) were able to discern relatively subtle and cell-type specific differences in the production of second messengers, but these probably arose from “signal strength” effects that led to differential activation of graded and threshold responses. Nonetheless, their results at least admit the possibility that biased agonists can qualitatively change information flow by coupling the receptor to effectors with different efficiency.

Unbiased surveys of conventional and arrestin-selective agonism to date have likewise found little in the way of “novel” signaling events triggered by biased agonists. A side-by-side global phosphoproteomic comparison of AngII and [Sar¹,Ile⁴,Ile⁸]-AngII performed using LTQ-Orbitrap high-resolution mass spectrometry was able to detect more than 1183 regulated protein phosphorylation sites of 10 000 surveyed, but of these 756 (64%) were unique to AngII, 369 (34%) were regulated by both AngII and [Sar¹,Ile⁴,Ile⁸]-AngII, and only 58 (5%) were unique to [Sar¹,Ile⁴,Ile⁸]-AngII (79). A less sensitive gel-based phosphoproteomic comparison of the same 2 ligands uncovered novel arrestin-dependent AT_1A receptor pathways leading to regulation of PP2A-Akt-glycogen synthase kinase 3β (GSK3β) signaling and stimulation of prostaglandin E2 synthesis by prostaglandin E synthase type 3, but still found that both AngII and [Sar¹,Ile⁴,Ile⁸]-AngII activated them (80).

The same story appears to hold when examining integrated biological responses such as cell proliferation or migration. For example, both AngII and [Sar¹,Ile⁴,Ile⁸]-AngII have been reported to produce arrestin-dependent inotropic and lusitropic effects in isolated murine cardiomyocytes (81). Likewise, the ability of [Sar¹,Ile⁴,Ile⁸]-AngII to stimulate proliferation of neonatal rat cardiomyocytes is a property shared with AngII; its lack of a hypertrophic response reflects its inability to activate G_q/11-dependent pathways, not its ability to engender novel responses (82). Similarly, the capacity to promote arrestin-dependent cell survival and stimulate migration in primary murine osteoblasts is a property shared by the arrestin-selective type 1 PTH receptor agonist, [D-Trp¹²,Tyr³⁴]-PTH(7–34), and its conventional counterpart PTH(1–34) (83).

Spatial and temporal bias

Most discussions of ligand bias focus on the capacity of ligands to change the efficiency with which GPCRs activate different downstream effectors. But cellular responses are also affected by the kinetics and spatial compartmentalization of pathway activation (84, 85). Another facet of bias relates to the ability of structurally distinct ligands to change the duration and/or location of signaling, ie, not changing the effector per se, but rather the dynamics of its activation.

Variation in ligand structure can affect both the duration and subcellular location of heterotrimeric G protein signaling. For example, both PTH(1–34) and a corresponding fragment of PTHrp, PTHrp(1–36), robustly activate G_s-cAMP signaling via the PTH₁ receptor, but the 2 ligands differ markedly in the kinetics of receptor association and dissociation, with PTHrp(1–36) exhibiting a slower on rate and faster off rate. Consequently, the cAMP response to PTH(1–34) is prolonged, and live cell imaging demonstrates that although cAMP generation in response to PTHrp(1–36) is limited to the plasma membrane, PTH(1–34) continues to stimulate cAMP production from within the endosomal compartment after the receptor has internalized (86–88). These differences appear to arise from conformational selection, in that ligands that exhibit high affinity for the G_s-uncoupled PTH₁ receptor, including PTH(1–34) and a series of synthetic N-terminal PTH analogs, possess higher binding affinity, increased potency, and prolonged G_s activation kinetics, than ligands, like PTHrp(1–36), that bind with high affinity only to the precoupled receptor-G protein complex (89, 90).

Similarly, ligand structure can change the stability of the GPCR-arrestin complex, leading to alterations in desensitization, trafficking, and signaling kinetics, and even admitting the possibility that arrestin-dependent signaling can be dissociated from arrestin-dependent desensitization. Zimmerman et al (91) found that a series of angiotensin peptide analogs that all supported arrestin2 recruitment and caused AT_1A receptor-arrestin complexes to traffic to endosomes differed widely in their ability to promote arrestin-dependent signaling. Using fluorescence recovery after photobleaching, they demonstrated ligand-specific differences in the stability of the receptor-arrestin complex, such that ligands that promoted tighter arrestin binding were effective activators of ERK1/2, whereas ligands with faster off rates were severely compromised in their ability to generate arrestin-dependent signals. Interestingly, they found ligand-specific differences in the degree to which arrestin recruitment, activation, and signaling were sensitive to knockdown of GRK2 and GRK6, suggesting that ligand-induced changes in receptor conformation affected arrestin conformation and function by eliciting different patterns of receptor phosphorylation, consistent with the “phosphorylation code” hypothesis proposed by Tobin and colleagues (92).

Bias In Vivo

If biased GPCR agonists are to find a prominent place in the pharmacopoeia of the future, it will be because they can produce biological effects in vivo that are not attainable using conventional agonist or antagonist ligands. Although some drugs approved for clinical use, eg, the β-adrenergic receptor antagonists propranolol and carvedilol, have been shown in retrospect to exhibit a degree of bias (67, 93), no current pharmaceuticals are known to possess unique clinical efficacy based on their ability to bias GPCR signaling. Nonetheless, data from several preclinical studies using arrestin- or G protein-biased ligands offer encouragement.

Arrestin pathway-selective bias

In bone, the conventional PTH₁ receptor agonist, PTH(1–34), stimulates bone-forming osteoblasts, increasing both osteoblast number and activity, while at the same time accelerating bone turnover by causing osteoblasts to secrete soluble factors that increase the number and activity of bone-resorbing osteoclasts (94). When administered intermittently, PTH(1–34) produces a net increase in bone mass, a property that is exploited clinically for the treatment of severe osteoporosis. Traditionally, the actions of PTH in bone have been attributed to G_s-cAMP signaling (95), so the expectation would be that a ligand like (d-Trp¹²,Tyr³⁴)-PTH(7–34), which antagonizes PTH₁ receptor-G_s coupling, would at best have no effect on bone mass. Paradoxically, mice treated with (d-Trp¹²,Tyr³⁴)-PTH(7–34) also exhibit increased bone formation, with greater trabecular bone volume, increased osteoblast number, and accelerated mineral apposition, but with no effect on osteoclast number or markers of bone turnover (96). The key to this unexpected phenotype apparently lies in the ability of (d-Trp¹²,Tyr³⁴)-PTH(7–34) to expand the osteoblast pool through arrestin-dependent cell cycle regulation and antiapoptotic signaling, while “uncoupling” the PTH₁ receptor from G_s-cAMP-dependent activation of osteoclasts (83, 96).

In the cardiovascular realm, the positive inotropic and lusitropic responses of isolated murine cardiomyocytes to [Sar¹,Ile⁴,Ile⁸]-AngII (81) appear to translate into improved cardiac contractility in vivo. An analogous arrestin pathway-selective AT_1A receptor agonist, TRV120027 [Sar-Arg-Val-Tyr-Ile-His-Pro-(D)-Ala-OH], that also stimulates arrestin-dependent activation of Src, ERK1/2, and endothelial nitric-oxide synthase in vitro, similarly improves cardiomyocyte contractility (97). Administration of TRV120027 to rats reduced mean arterial blood pressure, as did the nonpeptide AT_1A receptor antagonists losartan and telmisartin, but unlike the neutral antagonists, which decrease cardiac performance, TRV120027 increased cardiac performance and preserved cardiac stroke volume. In a canine rapid-pacing model of heart failure, TRV120027 produced cardiac unloading while preserving renal function, suggesting that it may have utility in heart failure treatment (98, 99).

G protein pathway-selective bias

Bias goes both ways, and G protein pathway-selective ligands have also shown promise. The most obvious benefit of G protein selectivity is that such ligands would escape arrestin-dependent desensitization. Prolongation of opiate-induced analgesia was the first phenotype of arrestin3-knockout mice to be described, demonstrating the critical role of arrestin3 in desensitization of central nervous system opioid receptors (100, 101). Recently, TRV130, a biased μ-opioid receptor agonist that demonstrates G protein signaling potency and efficacy similar to morphine, but with reduced arrestin recruitment and receptor sequestration, was shown in mice to produce comparable analgesia with less gastrointestinal dysfunction and respiratory depression than equianalgesic doses of morphine (102). Such results suggest that circumventing opioid receptor desensitization may lead to longer lasting or safer analgesics.

The other application of G protein bias would be avoiding unwanted arrestin-dependent side effects. For example, the niacin receptor, GPR109A, has been shown to decrease serum-free fatty acids by activating G_i/o while at the same time producing cutaneous flushing through arrestin-dependent activation of phospholipase A₂ (103). Predictably, a G protein pathway-selective GPR109A agonist, MK-0354, has been shown to decrease serum-free fatty acids in vivo without causing flushing, making it a potentially useful agent for treating hyperlipidemia (104). Another highly promising area relates to the mechanism of action of antipsychotics. β-Catenin and Akt signaling in the D₂ dopamine receptor-rich striatum of mice is regulated by an arrestin3 scaffolded PP2A-Akt-GSK3β complex (36, 37). Within the complex, PP2A dephosphorylates Akt Thr³⁰⁸, keeping it tonically inactive. Because phosphorylation by Akt inhibits GSK3α/β, keeping arrestin-bound Akt quiet increases GSK3α/β activity. GSK3β, in turn, phosphorylates β-catenin, accelerating its proteosomal degradation and thereby reducing its transcriptional activity. Predictably, striatal extracts from arrestin3-null mice show higher levels of β-catenin expression. The mood stabilizer, lithium, disrupts the interaction between arrestin, Akt, and PP2A, relieving PP2A-mediated inhibition of Akt allowing it to phosphorylate and inactivate GSK3β (105). Interestingly, the clinical efficacy of all classes of antipsychotic drug correlates directly to their D₂ receptor binding affinity and ability to antagonize the receptor, and it was found that although different classes of antipsychotics exhibit complex efficacy profiles with respect to D₂ receptor-G protein coupling, they share the property of antagonizing the D₂ receptor-arrestin3 interaction (106), suggesting that antagonism of arrestin signaling contributes to the activity of all antipsychotics.

Predicting Biased Outcomes

When dealing with conventional agonism or antagonism, a reasonable prediction of drug effects can be derived from study of the native ligand. Biased agonists, on the other hand, activate receptors in “unnatural” ways, and their biological effects may be less predictable. Whereas a concern with allosteric modulators, which bear no structural relationship to the orthosteric ligand, is unexpected, “off-target” effects (107), developers of biased therapeutics must be aware of the possibility of unpredicted “on-target” effects; not side effects arising from unexpected drug-protein interactions, but rather from the drug affecting its intended target in unintended ways. This is particularly true for arrestin pathway-selective agonists, because our appreciation of the scope physiologically relevant GPCR signals transmitted via arrestins remains incomplete (19, 38, 108).

By way of illustration, parametric gene set enrichment analysis of cDNA microarray data from bone of mice treated with PTH(1–34) or the arrestin-biased ligand, [d-Trp¹²,Tyr³⁴]-PTH(7–34), found a striking degree of nonoverlap in the genome level response (83). As expected of a conventional PTH₁ receptor agonist, PTH(1–34) affected gene clusters classically associated with embryologic skeletal patterning and PTH actions in bone, including Wnt/β-catenin signaling, bone morphogenic protein signaling, TGF-β signaling, phosphatidylinositol 3-kinase/Akt signaling, and ERK/MAPK signaling, that corresponded to regulation of processes linked to the control of skeletal morphogenesis, osteoblast differentiation, matrix biosynthesis and mineralization, and bone turnover (109–111). In contrast, [d-Trp¹²,Tyr³⁴]-PTH(7–34) primarily affected pathways regulating apoptosis, eg, ataxia telangectasia-mutated signaling and p53 signaling; cell survival, eg, phosphatidylinositol 3-kinase/Akt signaling; and cell cycle checkpoint regulation. Thus, although both agents increased trabecular bone mass, they appeared to do so in largely different ways.

Unexpected actions of [Sar¹)-Ile⁴-Ile⁸)]-AngII have been reported as well. In a murine model of heart failure, [Sar¹-Ile⁴-Ile⁸]-AngII was found to promote aldosterone production via arrestin2- and ERK1/2-dependent up-regulation of the steroidogenic acute regulatory protein, StAR, a mitochondrial transport protein that is the rate-limiting step in the biosynthesis of steroid hormones. In this model, the resultant elevation in circulating aldosterone levels led to salt and volume retention, adverse cardiac remodeling, and heart failure progression (112). Arrestin-mediated signaling also appears to promote vascular smooth muscle cell hyperplasia and an exaggerated vascular injury response. In primary vascular smooth muscle, [Sar¹-Ile⁴-Ile⁸]-AngII stimulates proliferation by promoting AT_1A receptor-dependent epidermal growth factor receptor transactivation (113–115) and activating antiapoptotic signaling pathways (116).

Expanding the Pharmacologic Tool Kit

Rather than an occasional anomaly, it is now clear that ligand bias is a general property in GPCR signaling, even being found in nature, as for example between CCL19 and CCL21, the 2 endogenous ligands for the CC chemokine receptor 7 (65). Table 1 lists several GPCRs for which biased ligands have been described and the therapeutic arenas in which ligand bias may prove useful. Conditions caused by mutated constitutively active GPCRs, eg, familial male precocious puberty (LH receptor; Ref. 131), hyperthyroidism due to toxic follicular adenomas (TSH receptor; Ref. 132), and some cases of familial nephrogenic diabetes insipidus (V2 vasopressin receptor; Ref. 133), offer additional settings in which ligands that favor G protein- or arrestin coupling might offer therapeutic advantage. Even more promising, arrestin-mediated signaling pathways have been implicated in the pathogenesis of many common clinical conditions; eg, asthma (134, 135), inflammatory bowel disease (136, 137), certain malignancies (138), and autoimmune disorders (139), suggesting that ligands that bias GPCR signaling to favor or oppose arrestin coupling may have widespread application.

Table 1.

Potential Targets for Biased GPCR Therapeutics

Receptor	Ligand(s)	Observed Bias	Therapeutic Arena	References
β1/2-Adrenergic	Carvedilol; ICI118551	Arrestin-selective agonism	Congestive heart failure	14, 67, 93, 117
AT1 angiotensin	Sar1,Ile4,Ile8-AngII; TRV120027	Arrestin-selective agonism	Congestive heart failure	81, 97–99
α2-Adrenergic	Desipramine	Arrestin-selective agonism	Hypertension	118
PTH1 parathyroid hormone	[d-Trp12,Tyr34]-bPTH (7–34)	Arrestin-selective agonism	Osteoporosis	83, 96
D1 dopamine	A-77636	Arrestin-selective agonism	Parkinson's disease	119
CCR5 chemokine	AOP-RANTES	Arrestin-selective agonism	HIV infection	120
CB1 cannabinoid	ORG27969	Arrestin-biased allosteric modulator	Anxiety, depression, pain, obesity	121
GPR109A nicotinic acid	MK-0354	Gi/o-selective agonism	Hyperlipidemia	103, 104
μ-Opioid	TRV130	Gi/o-selective agonism	Analgesia	102
κ-Opioid	6′-Guanidinonaltrindole	Gi/o-selective agonism	Analgesia, dysphoria	122–124
D2 dopamine	Aripiprazole	Gi/o-selective agonism	Schizophrenia, psychosis	106, 125
5-HT2A serotonin	2,5-Dimethoxy-4 iodoamphetamine	Gi/o-selective agonism	Schizophrenia, hallucinations	126, 127
GLP-1 glucagon-like peptide	Oxyntomodulin	Gs-selective agonism	Diabetes, β-cell mass/function	128–130

Beyond the challenges that pluridimensional efficacy poses in terms of understanding the molecular pathogenesis of disease and defining the ligand efficacy profile most likely to prove therapeutic, the phenomenon of ligand bias has implications for the design of assays to characterize GPCR signaling in the research or drug discovery settings (44, 140–142). The traditional approach of using unidimensional high-throughput assays for ligand classification is inadequate in an environment in which ligand classification is assay dependent (44, 45). Instead, drug discovery approaches need to be flexible. If the goal is to define efficacy in a limited number of known dimensions, eg, G protein and arrestin coupling, then multiplexing assays may be adequate to identify functional selectivity. Beyond assays based on ligand binding, G protein activation, or second-messenger generation, high-content assays based on imaging techniques that use fluorescent signals can provide both temporal and spatial information about signaling (143, 144). Considering just one such event, the interaction of GPCRs with arrestins, responses can be monitored through direct visualization of GPCR/arrestin-green fluorescent protein complexes (145, 146), with bioluminescence resonance energy transfer (147), with enzyme fragment complementation (148), or with protease-activated transcriptional reporter genes (149). Greater efficiency can be obtained by multiplexing green fluorescent protein- and immunofluorescence-based assays to provide simultaneous readouts of multiple signaling pathways within the same cell (150). Once empiric data on the potency and efficacy of a series of ligands across 2 or more dimensions of efficacy are generated, the extent of ligand bias, eg, ΔΔLog(τ/KA) or ΔΔLog (Relative Activity) (43, 151), can be quantified relative to a reference agonist based on adaptations of the Back-Leff operational model (4, 5).

If, on the other hand, the objective is to detect whether compounds exert any form of activity against a receptor, then technology designed to detect integrated whole-cell responses may be necessary. Resonant wave guide grating technology has led to the development of optical biosensors that can measure dynamic mass redistribution signals in living cells (152, 153). This technology can detect interactions of GPCRs with cytosolic signaling molecules at a depth of 150–200 nM and detect receptor internalization. The resulting dynamic mass redistribution signal is a noninvasive cell-based technology that can measure virtually any receptor activation in any cell type in real time. The approach has been applied to the detection and quantification of functional selectivity in intact cells (154). Another technology that can be used for the same purpose involves measuring changes in the electrical impedance of cell monolayers caused by receptor-mediated changes in cell mass redistribution (155, 156). The principal limitation of these “label-free” systems is that because whole-cell responses represent the integration of numerous pathways, it may be difficult or impossible to deconvolute the kinetic pattern of the response in a manner that allows identification of the specific signaling pathway(s) being affected. On the other hand, label-free systems can be used with primary cells, circumventing the potential for signaling artifacts in engineered cell systems and allowing ligands to be screened in therapeutically relevant cell types.

Finally, the recent emergence of high-resolution crystal structures of GPCRs in their agonist-bound, antagonist-bound, and effector-coupled states has begun to enable model-based or virtual in silico approaches to drug discovery (157, 158). Virtual screens have identified novel orthosteric and allosteric ligands for the D₁, D₂ (159), and D3 (160) dopamine, CB₁ (161) and CB₂ (162) cannabinoid, and CXCR7 (163) receptors, among others. Importantly, a large library virtual screen against the activated β₂-adrenergic receptor crystal structure identified “hits” that exhibited arrestin-selective signaling bias (164), suggesting that virtual structure-based approaches might be used to fashion ligands with specific efficacy.

Conclusions

In the native context, GPCRs and their natural ligands coevolved to control signaling in the most physiologically adaptive manner. Signaling, reflected in the distribution of GPCR-active states and the efficiency with which these states engage downstream effectors, is “balanced” to meet the needs of the organism. Biased agonists, on the other hand, alter the distribution of active states. The resulting signals are not qualitatively different, but are “unbalanced” compared with the native ligand. As these unbalanced signals propagate, differences in the tissue response may arise from relatively subtle differences in signal strength or the kinetics of pathway activation and deactivation. Certainly, the presence of disease can cause adaptive responses to become maladaptive, as in the case of endogenous catecholamine signaling in the setting of congestive heart failure (165), and this is one arena in which biased agonists hold promise as therapeutics. The ability to “recalibrate” GPCR signaling to adapt to the needs of a perturbed system is something that cannot be achieved using conventional agonist or antagonists (166). But we must remain cognizant of the converse. The same properties that make them valuable as potential therapeutics may also make them more prone to produce unanticipated effects.

It is clear that the concepts of pluridimensional efficacy and GPCR allosterism hold the keys to new drugs. Small molecule allosteric modulators have already found their way into the clinic for the treatment of secondary hyperparathyroidism (167) and advanced HIV disease (168). Biased agonists, tools that can qualitatively change GPCR signaling for therapeutic benefit, seem to be poised on the horizon. The future will tell whether they prove to be the utilitarian wrenches, saws, and screwdrivers of the pharmacologist's tool kit or a collection of left-handed monkey wrenches with limited utility or specialized application.