Development of Functionally Selective, Small Molecule Agonists at Kappa Opioid Receptors*

Background: Kappa opioid receptor (KOR) signaling may produce antinociception through G protein or dysphoria through βarrestin pathways. Results: Two highly selective, brain penetrant agonist scaffolds bias KOR signaling toward G protein coupling and produce antinociception in mice. Conclusion: Described are first-in-class small molecule agonists that bias KOR signaling through G proteins. Significance: Functionally selective KOR agonists can now be used in vivo. The kappa opioid receptor (KOR) is widely expressed in the CNS and can serve as a means to modulate pain perception, stress responses, and affective reward states. Therefore, the KOR has become a prominent drug discovery target toward treating pain, depression, and drug addiction. Agonists at KOR can promote G protein coupling and βarrestin2 recruitment as well as multiple downstream signaling pathways, including ERK1/2 MAPK activation. It has been suggested that the physiological effects of KOR activation result from different signaling cascades, with analgesia being G protein-mediated and dysphoria being mediated through βarrestin2 recruitment. Dysphoria associated with KOR activation limits the therapeutic potential in the use of KOR agonists as analgesics; therefore, it may be beneficial to develop KOR agonists that are biased toward G protein coupling and away from βarrestin2 recruitment. Here, we describe two classes of biased KOR agonists that potently activate G protein coupling but weakly recruit βarrestin2. These potent and functionally selective small molecule compounds may prove to be useful tools for refining the therapeutic potential of KOR-directed signaling in vivo.

The kappa opioid receptor (KOR) 5 is a seven-transmembrane G protein-coupled receptor (GPCR) (1). The cognate neuropeptides for KOR are endogenous opioids, including the dynorphin peptides. Dynorphins and KOR are widely expressed throughout the central nervous system (2)(3)(4). In a canonical sense, KOR activation is defined by agonist-induced coupling of the G␣ subunit of the heterotrimeric G proteins in the pertussis toxinsensitive G i/o family (5), subsequent inhibition of adenylyl cyclase (6), activation of inward-rectifying potassium channels (7), and blockade of calcium channels (8).
There is considerable evidence that selective KOR agonists produce antinociception in animal models (9 -12), and mice lacking KOR expression are no longer responsive to the antinociceptive effects of a selective KOR agonist U50,488 (13). The G protein-mediated signaling events are believed to contribute to the analgesic properties of KOR agonists (14,15). Unlike MOR agonists, KOR agonists do not cause physical dependence nor do they produce respiratory failure; thus, they are attractive as potent analgesics (16). However, KOR activation has also been implicated in producing an array of undesirable side effects, including dysphoria (17,18), sedation (10), diuresis (11), hallucination (19), and depression (20). These adverse effects limit the therapeutic potential of KOR agonists as analgesics.
It is becoming widely evident that activation of GPCRs by chemically distinct agonists can promote receptor coupling to distinct pathways in different tissues (21)(22)(23). The KOR also has the potential to signal through multiple downstream signaling cascades, and there is increasing evidence that signaling to alternative cascades may promote side effects associated with KOR activation. In addition to G protein coupling, another proximal event following GPCR activation is the agonist-promoted recruitment of ␤arrestins. As regulatory and scaffolding proteins, ␤arrestins can lead to desensitization of the receptor by impeding further G protein coupling; in some cases, ␤arrestins facilitate the assembly of protein scaffolds, thereby acting as signal transducers (24 -26). For example, ␤arrestin2 recruitment has been shown to promote recruitment and acti-vation of MAPKs, including ERK1/2 (27,28) and p38 (29). Interestingly, studies in AtT-20 cells and mouse striatal neurons suggest that the KOR may signal via a GPCR kinase 3 (GRK3)-and a ␤arrestin2-dependent mechanism to activate the stress MAPK p38 and that this signaling cascade may mediate the dysphoric effects of KOR agonists (29,30). Moreover, KOR activation has been shown to lead to ERK1/2 phosphorylation in both neurons and astrocytes through both G protein and ␤arrestin-dependent mechanisms (28,29,31,32). Endogenously elevated dynorphin levels produced by repeated swim stress tests lead to KOR-dependent ERK1/2 phosphorylation in the mouse brain that was not dependent on GRK3 expression (33). Therefore, ERK activation may be a useful indicator of both ␤arrestin-dependent and G protein-dependent signaling states downstream of KOR (34).
Because G protein-mediated signaling is implicated in KORinduced antinociception and ␤arrestin2-mediated signaling has been implicated in the aversive and dysphoric properties of KOR agonists, the development of G protein-biased agonists may provide a means to optimally tune KOR therapeutics. In two different screening campaigns, we discovered two new classes of KOR agonists, triazole probe 1 and isoquinolinone probe 2 (Fig. 1). The triazole scaffold was identified under the auspices of the Molecular Libraries Probe Production Centers Network via a high throughput screening campaign. In this work, which was done in collaboration with colleagues at Sanford-Burnham Research Institute and Duke University, the National Institutes of Health Small Molecule Repository was screened to identify KOR ligands based on activation of ␤arrestin2 recruitment (35,36). The isoquinolinone scaffold arose from a 72-member library prepared by a tandem Diels-Alder acylation reaction that was screened for binding at potential GPCR targets by the NIMH Psychoactive Drug Screening Program (37,38). The lead isoquinolinone was reported as a KOR agonist showing selectivity and high binding affinity for KOR over MOR, DOR, and other screened GPCRs (37,38); moreover, the triazole lead also displayed high selectivity for KOR over MOR and DOR ( Fig. 1) (35,36). The isoquinolinone compounds are particularly interesting in that they lack the basic nitrogen center common in small molecule KOR ligands (10, 39 -41); the best known ligand lacking this feature is salvinorin A, a natural neoclerdane diterpene found to be a highly selective, potent KOR agonist (19).
Following their initial disclosure, these scaffolds were subjected to iterative rounds of medicinal chemistry and structureactivity relationship studies with the goal of developing KOR agonists that are biased toward G protein coupling. Here, we report five triazole analogues and two isoquinolinone analogues ( Fig. 1) that activate KOR in a manner that is preferentially biased toward G protein signaling with minimal effects on ␤arrestin2 recruitment and downstream ERK1/2 activation.
Animals-C57BL/6J male mice between 5 and 7 months old from The Jackson Laboratory (Bar Harbor, ME) were used in accordance with the National Institutes of Health Guidelines for the Care and Use of Laboratory Animals and with approval by The Scripps Research Institute Animal Care; the Scripps vivarium is fully AAALAC-accredited.
Membrane G Protein Signaling-Membranes were prepared according to a modified procedure of Schmid et al. (42). Briefly, CHO-hKOR cells were serum-starved for 1 h in DMEM/F-12 media, collected with 5 mM EDTA, washed with PBS, and stored at Ϫ80°C. For each assay, cell pellets were homogenized via Teflon-on-glass homogenizer in buffer (10 mM Tris-HCl, pH 7.4, 100 mM NaCl, 1 mM EDTA), passed through a 26-gauge needle eight times, centrifuged twice at 20,000 ϫ g for 30 min at 4°C, and resuspended in assay buffer (50 mM Tris-HCl, pH 7.4, 100 mM NaCl, 5 mM MgCl 2 , 1 mM EDTA, and 3 M GDP). For each reaction, 15 g of membrane protein were incubated in assay buffer containing ϳ0.1 nM [ 35 S]GTP␥S and increasing concentrations of compounds in a total volume of 200 l for 1 h at room temperature. The reactions were terminated by rapid filtration over GF/B filters using a 96-well plate harvester (Brandel Inc., Gaithersburg, MD). Filters were dried overnight, and radioactivity was determined with a TopCount NXT high throughput screening microplate scintillation and luminescence counter (PerkinElmer Life Sciences).
Whole Cell G Protein Signaling-CHO-hKOR cells were seeded in a 96-well tissue culture plate (BD Biosciences) at 60,000 cells/well. On the 2nd day, cells were serum-starved for 1 h followed by treatment with saponin (50 g/ml, Sigma) for 4 min to briefly permeabilize cells (44). Cells were then incubated with [ 35 S]GTP␥S and increasing concentrations of compounds in assay buffer in a 200-l volume for 1 h at room temperature as described for membrane G protein signaling above. The [ 35 S]GTP␥S binding was measured using the same methods described above.
Cellular Impedance-CHO-hKOR cells were plated at 60,000 cells per well in CellKey TM microplates (Molecular Device, Sunnyvale, CA) in complete DMEM/F-12 media. The 2nd day, the cells were equilibrated at room temperature in Hanks' balanced salt solution (Invitrogen) containing 0.1% BSA (fatty acid-free, Sigma) and 20 mM HEPES for 30 min. Changes in impedance due to changes in extracellular current (Z iec ) were recorded using CellKey TM cellular analysis system (Molecular Device) for 35 min after treatment with increasing doses of compounds (45,46). Maximal changes in impedance were extracted and plotted in dose-response curves.
␤Arrestin2 Recruitment (EFC)-The ␤arrestin2 EFC assays were performed according to the manufacturer's (DiscoveRx) protocol with slight modification as described previously (42). Briefly, the U2OS-hKOR-␤arrestin2-EFC cells were plated at 5000 cells/well in 20 l of Opti-MEM media containing 1% FBS for overnight in 384-well white plates. The 2nd day, cells were treated with compounds for 90 min followed by 1-h incubation of detection reagent. Luminescence values generated following the substrate addition were obtained using a Synergy HT luminometer (BioTek, Winooski, VT). All compounds were run in duplicate per assay and normalized to vehicle-treated cells.
␤Arrestin2 Imaging-For concentration-response studies, U2OS-hKOR-␤arrestin2-GFP cells were plated in a 384-well Aurora Black walled, clear bottom plate at a density of 5000 cells per well and incubated overnight. Cells were serumstarved for 30 min prior to treatment (20 min) and then fixed with pre-warmed 4% paraformaldehyde and stained with nuclear dye Hoechst (1:1000) for 30 min. Cells were then washed three times with PBS and stored in PBS until imaged. Images were acquired with a ϫ20 objective on the CellInsight (Thermo Scientific), and the number of spots per cell was determined using the Cellomics Spot Detector BioApplication algorithm (version 6.0). For imaging of live cells, an Olympus FluoView IX81 confocal microscope (Olympus, Center Valley, PA) was used. Live cell imaging was assessed in CHO-hKOR cells transfected with ␤arrestin2-YFP (4 g) via electroporation plated on collagen-coated glass-bottom dishes (MatTek, Ashland, MA) (42). Single focal plane images were obtained using ϫ100 objective between 5 and 20 min after drug treatment. Cells that did not respond to compound stimulation were treated with 10 M U69,593 to validate the cell's potential to respond prior to concluding a negative drug effect (data not shown). Individual images were adjusted for brightness/contrast and size (scale bars are indicated).
Radioligand Binding-Receptor binding assays were performed as described previously (42,47). Cell pellets (prepared as for G protein coupling) were homogenized in homogenization buffer (10 mM Tris-HCl, pH 7.4, with 1 mM EDTA) and centrifuged twice at 20,000 ϫ g for 30 min at 4°C. The resulting membrane pellet was homogenized in assay buffer ( H]diprenorphine) and increasing concentrations of the test compounds. Nonspecific binding was determined for each radioligand in the presence of 10 M U69,593 (KOR), 10 M DAMGO (MOR), or 10 M naltrindole (DOR). All binding assays were terminated by filtration through GF/B glass fiber filters (pretreated with 0.1% polyethyleneimine) on a Brandel cell harvester (Brandel Inc., Gaithersburg, MD). Filters were dried and counted with Microscint on a TopCount NXT high throughput screening microplate scintillation and luminescence counter (PerkinElmer Life Sciences).
ERK1/2 Western Blot Analysis-CHO-hKOR cells in 6-well plates were serum-starved for 1 h prior to treatment (10 min) as described previously (42). Antibodies to phosphorylated ERK1/2 and total ERK1/2 (Cell Signaling) were used, and the ratio of phosphorylated ERK1/2 to total ERK1/2 was calculated and normalized to vehicle treatment. Percentage of response was calculated according to the maximal responses by 10 M U69,593 stimulation.
Tissue (Brain) Distribution-Test compounds were formulated as 1 mg/ml solution in 10% DMSO, 10% Tween 80, and sterile distilled H 2 O and were dosed at 10 mg/kg intraperitoneally in C57Bl-6 mice. Brain and plasma samples were taken at 30 and 60 min. Plasma and brain were mixed with acetonitrile (1:5 v/v or 1:5 w/v, respectively). The brain sample was disrupted with a probe tip sonicator. Samples were centrifuged at 16,000 ϫ g, and the compound concentration in the supernatant was determined using liquid chromatography (Shimadzu, Japan)/tandem mass spectrometry (AB Sciex, Framingham, MA) operated in positive ion mode using multiple reaction monitoring methods. Separate standard curves were prepared in blank plasma and brain matrix. Brain concentration was calculated as compound per mg of tissue and converted to a concentration assuming a density of 1, where 1 g of tissue equals 1 ml.
Pharmacokinetics-Pharmacokinetic parameters were determined in C57Bl-6 mice by intravenous dosing in the tail vein. Mice (n ϭ 3) were dosed with drug at 2 mg/kg, and ϳ20 l of blood was collected at 0, 5, 15, 30, 60, 120, 240, and 480 min post-dose. Plasma was generated by standard centrifugation techniques resulting in ϳ10 l of plasma that was immediately frozen. Drug levels were determined using an ABSciex 5500 mass spectrometer using multiple reaction monitoring and mass transitions of 431.3381.1 for compound 1.1 and 461.33286.1 for compound 2.1. Pharmacokinetic parameters were calculated using a noncompartmental model (Phoenix WinNonlin, Pharsight Inc.).
Plasma and Brain Binding-Plasma protein binding and nonspecific brain binding were determined using equilibrium dialysis. All samples were tested in triplicate. A 96-well equilibrium dialysis chamber (HTDialysis, Gales Ferry, CT) with 12,000 molecular weight cutoff dialysis membranes were used. Human plasma and rat plasma were provided by a commercial vendor (Pel-Freez Biologicals, Rogers, AR). Equal volumes of buffer and plasma containing 2.5 M drug were added to opposite sides of the dialysis membrane. The plate was covered and allowed to shake in a 37°C incubator for 16 h. Similar methods were used for nonspecific brain binding except a brain homogenate (1 part rat brain, 3 parts buffer) was used in place of the plasma. After 16 h, the concentration of drug in the plasma/brain versus plasma compartments was determined by LC-MS/MS. The fraction bound was calculated as Determination of logP-Estimation of octanol/water partition coefficient (logP) was determined using a fast gradient reverse phase HPLC method as described by Valko et al. (48) without modification. Briefly, chromatographic retention times for test compounds and 10 standards were determined using a linear acetonitrile gradient on a Luna C18(2) 50 ϫ 4.6 mm 5-mm column. Retention time was used to calculate the chromatographic hydrophobicity indices in acetonitrile (CHI ACN ). CHI ACN and hydrogen bond count were used to calculate the logP according to the formula logP ϭ 0.047 ϫ CHI ACN ϩ 0.36 ϫ hydrogen bond count Ϫ 1.10. Refer to Valko et al. (48) for details and a full description of the method validation.
Warm Water Tail Immersion-Antinociception was evaluated in the warm water tail-immersion (tail flick) assay (49°C) as described previously (47,49,50). In brief, the tip (ϳ2 cm) of the mouse's tail was submerged in warm water, and the time until it was withdrawn from the water was recorded. Response latencies were measured prior to (basal) and at the indicated times following drug administration (30 mg/kg, intraperitoneally given as 10 l/g). U50,488H and the test compounds were prepared in a vehicle containing 10% Tween 80 and 10% DMSO in sterile distilled H 2 O. A cutoff latency of 30 s was imposed to prevent tissue damage. Vehicle alone had no effect on response latencies (data not shown).
Data Analysis and Statistics-Sigmoidal dose-response curves were generated using three-parameter nonlinear regression analysis in GraphPad Prism 6.01 software (GraphPad, La Jolla, CA). All compounds were run in parallel assays in 2-4 replicates per n. The EC 50 values and maximal responses (E max ) of drugs were obtained from the average of each individual experiment following nonlinear regression analyses and are reported as the means Ϯ S.E.
To express the results of each response assay in a format suitable for comparison, each data set was fit, using GraphPad Prism Version 6.01, to the operational model (51) expressed as Equation 1 (52), where E max is the maximal response of the system; A is the molar concentration of the drug, and K A is the equilibrium dissociation constant. For each assay, the maximal response (E max ) is constrained to be a shared value. The parameter is defined, in the operational model, as the agonist efficacy, and the log(/K A ) ratio (transduction coefficient) provides a single parameter value that is useful for the comparison of agonist activity (51). For all studies, U69,593 is used as the reference compound and is assayed in parallel with the test compounds. To produce an appropriate estimate of log(/K A ) for the full reference agonist (U69,593), in each assay, the log K A for the full agonist was set constant at 0 (i.e. K A ϭ 1 M). The resulting log(/K A ) ratio for the reference agonist is used to produce the "normalized" transduction coefficient (⌬log(/K A ) ratio) for each of the test compounds shown in Equation 2, For a limited number of test compounds in certain assays, the logK A value in the equation was constrained to be less than 0 (i.e. K A Ͻ1 M) to permit convergence to the model. An asterisk in the tables indicates these few individual cases.
Bias factors (52)(53)(54) are derived by subtracting the ⌬log(/ K A ) test-U69 for each assay (⌬⌬log(/K A ) assay1 Ϫ assay2 ) and are calculated as shown in Equation 3, The statistical tests used are noted in the figure legends, where Student's t test indicates an unpaired two-tailed analysis. All studies were performed n Ն 3 independent experiments performed in multiple replicates.

RESULTS
KOR-mediated G protein signaling was evaluated using a [ 35 S]GTP␥S binding assay in cell membranes from CHO-K1 cells stably expressing the human KOR (CHO-hKOR) (42). Concentration-response curves were run in parallel with U69,593, a well characterized potent, full agonist at KOR, as a reference agonist for the assay; EC 50 values and maximal   50 values exceeding 3000 nM are generally not considered a good fit, the values are presented here for comparison. EFC refers to the KOR-␤arrestin2 enzyme fragment complementation assay; mG refers to membrane G protein signaling assays, and imaging refers to automated high content imaging of ␤arrestin2-GFP translocation to KOR. NC indicates not converged. n Ն 3 individual assays; means are presented Ϯ S.E. Maximal stimulation over vehicle treatment with U69,593 was ϳ4-fold for ͓ 35 S͔GTP␥S binding, ϳ8-fold for the EFC assay, and ϳ20-fold for the imaging assay. To determine the relative potencies for recruiting ␤arrestin2 to the agonist-stimulated KOR, two cell-based assays, as described in the Molecular Libraries Probe Production Centers Network probe report, were used (35). The commercial EFC assay consists of U2OS cells expressing human KOR and ␤arrestin2 (U2OS-hKOR-␤arrestin2-EFC), each tagged with a fragment of ␤-galactosidase. The degree of ␤arrestin2 recruitment is measured as increases in luminescence intensities triggered by enzyme complementation occurring upon receptor and ␤arrestin2 engagement (55). The second assay utilizes U2OS cells expressing human KOR and ␤arrestin2 tagged with green fluorescent protein (U2OS-hKOR-␤arrestin2-GFP) (56,57). The translocation of ␤arrestin2-GFP is measured by automated high content imaging analysis that detects a difference in diffuse cellular GFP distribution to the agonist-induced formation of fluorescent punctae (56). In both cases, U69,593 is used as the full agonist reference ligand.

Entry
While maintaining potent agonism in the G protein signaling assays, the triazole and isoquinolinone compounds only weakly recruit ␤arrestin2, with potencies exceeding 2 M in both assays, whereas the potency of U69,593 is ϳ130 nM in the ␤arrestin2 EFC assay and ϳ205 nM in the imaging assay ( Fig. 3; Table 1). Examination of the concentration-response curves in both assay platforms (Fig. 3, A and C) reveals that each of the triazole compounds shows minimal ␤arrestin2 recruitment at lower concentrations (Ͻ1 M). However, stimulation dramatically increases at higher concentrations (Ͼ1 M); this is also true for compound 2.1 of the isoquinolinones (Fig. 3, B and D).
This phenomenon was verified for representative compounds using confocal microscopy to visualize ␤arrestin2-GFP translocation in the U2OS cells (data not shown) that confirmed the results obtained in the high content imaging platform (Fig. 3, C  and D).
G protein signaling and ␤arrestin2 recruitment were, by design, performed in different cellular backgrounds (CHO-hKOR and U2OS-hKOR-␤arrestin2-EFC or U2OS-hKOR-␤arrestin2-GFP). The U2OS cell line was used for the commercial enzyme fragment complementation assay because it gave a larger assay window than that observed for the CHO-hKOR-␤arrestin2-EFC commercial assay that is also available from the manufacturer. This was preferred to bias the system toward detecting even weak ␤arrestin2-KOR interactions. Furthermore, even though CHO-hKOR-␤arrestin2-EFC cells could be used, it is important to realize that the KOR and ␤arrestin2 are linked to fragments of ␤-galactosidase and therefore, although the parental cell line is of the same origin, the stably transfected cell lines are not identical.
To determine whether the cell line (U2OS versus CHO-K1) contributed to the apparent bias, we also evaluated ␤arrestin2 (tagged with YFP) recruitment in the CHO-hKOR cell line by confocal microscopy for select agonists of each class. In the confocal microscopy studies of CHO-hKOR cells, 1.1 does not promote ␤arrestin2 recruitment at low concentrations (100 nM and 1 M) but induces punctae formation at 10 M (Fig. 3E), consistent with the studies in U2OS cell imaging assays and the EFC ␤arrestin2 assay. The ability to visualize the agonist-induced ␤arrestin2 recruitment also suggests that the positive signal at high concentrations is not an artifact of the assay. In contrast to the observations in the other ␤arrestin2 assay platforms, 2.1 revealed no visible recruitment of ␤arrestin2-YFP even at 10 M, in the CHO-hKOR cells (Fig. 3E). The lack of response to 2.1 in the CHO-hKOR cells transfected with ␤arrestin2-YFP is likely due to a low detection window with the low degree of stimulation falling below detectable thresholds. Finally, when tested in the MOR EFC assay, no stimulation is

Analysis of bias comparing G protein signaling and ␤arrestin2 recruitment in reference to U69,593 activity
Bias factors (⌬⌬log(/K A ) assay1 Ϫ assay2 ) were calculated as described under "Experimental Procedures." The asterisk indicates log(/K A ) values were constrained to Ͻ0 (less than 1 M relative affinity) to permit convergence to the operational model. mG indicates membrane G protein signaling; imaging indicates automated imaging analysis. n Ն 3 individual assays; means are presented Ϯ S.E. ␤arr2 is ␤arrestin2.   DECEMBER 20, 2013 • VOLUME 288 • NUMBER 51

JOURNAL OF BIOLOGICAL CHEMISTRY 36709
seen at any dose (1, 10, and 100 M), which also argues against the effect being an artifact of the assay (data not shown). The studies presented in Tables 1 and in Figs. 2 and 3 suggest that the triazole and isoquinolinone analogues presented here bias KOR function toward G protein signaling over the recruitment of ␤arrestin2. To compare the relative differences within and between assays, curves were fit to the operational model (51) to derive transduction coefficients (log(/K A )) for the ref-erence ligand (U69,593) and the "test" ligands using GraphPad Prism Version 6.01, as described previously (42,52,54). The transduction coefficient represents the relative propensity of the ligand to generate a signal proportional to the agonist's calculated relative affinity for engaging the receptor based on its performance in the assay. Subtracting the log(/K A ) for U69,593 from the values obtained for the test ligands generates a "normalized" transduction coefficient for the test ligand within the assay, or the ⌬log(/K A ). To determine preference between assays, bias factors were calculated as described under "Experimental Procedures" ( Table 2) (54). Although such modeling may be imperfect for extreme cases of bias, we have presented the analysis here for qualitative comparison; each of the compounds displays bias for activation of KOR toward G protein signaling over ␤arrestin2 recruitment.
Because the [ 35 S]GTP␥S binding assay was performed in membrane preparations whereas the ␤arrestin2 recruitment studies were performed in whole cells, we considered that these contextual differences could confound our interpretations of bias. Therefore, we selected a representative triazole and isoquinolinone and performed two additional G protein signaling assays in whole cell preparations (Fig. 4). Using a low concentration of saponin to permeabilize the CHO-hKOR cells allowing uptake of the radionucleotide, [ 35 S]GTP␥S binding was performed in whole cells plated on 96-well plates (Fig. 4A). U69,593, 1.1, and 2.1 potently stimulate G protein signaling in this whole cell assay. Furthermore, in the absence of the brief permeabilization step, U69,593 does not induce [ 35 S]GTP␥S binding due to the lack of available radionucleotide inside the cell.
A label-free, cellular impedance assay was also utilized to assess relative potencies in the living CHO-hKOR cells (Fig.  4B). This assay measures changes in impedance resulting from cytoskeletal re-organization leading to changes in cell shape (45). GPCRs promote actin reassembly patterns reflective of the coupling of particular G␣ proteins. In our analysis, the cellular response signature resembles that of a typical G␣ i -coupled GPCR (Fig. 4C) (58). Importantly, no response was observed in   Table 3. Data are presented as the mean Ϯ S.E. (n Ն 5).

TABLE 3
Representative triazole and isoquinolinone compounds in whole cell G protein signaling assays EC 50 and E max values were calculated as described in Table 1. Bias factors (⌬⌬log(/ K a ) assay1 Ϫ assay2 ) were calculated as described under "Experimental Procedures." ␤Arrestin2 EFC and ␤arrestin2 imaging transduction coefficients are found in Table 2. The asterisk indicates that log K a values were constrained to Ͻ0 (less than 1 M relative affinity) to permit convergence to the operational model. n Ն 5 individual assays; means are presented Ϯ S.E. Maximal stimulation over vehicle treatment with U69,593 was ϳ2-fold for whole cell [ 35 S]GTP␥S binding (wcG) and ϳ8-fold for the cellular impedance assay. the parental CHO cell line (data not shown). In this live cell assay format, 1.1 and 2.1 are fully efficacious and nearly as potent as U69,593 in promoting KOR-dependent responses indicative of G protein signaling (Table 3).
Further analysis was undertaken to compare all G protein signaling assays to all ␤arrestin2 recruitment assays. Data from Tables 2 and 3 were used to determine bias factors that are presented graphically in Fig. 5. In summary, regardless of the assay format and the cell type, the triazole and isoquinolinone maintain bias toward KOR-induced G protein signaling over ␤arrestin2 recruitment.
To further characterize the compounds at a downstream signaling pathway, ERK1/2 phosphorylation was investigated. GPCRs can utilize both G protein-dependent and ␤arrestinmediated signaling pathways to activate ERK1/2 MAPKs (34). Interestingly, both classes of compounds activate ERK with potencies between 300 and 6000 nM as compared with ϳ5 nM for U69,593. Although this appears to correlate with their relative potency for recruiting ␤arrestin2 (Fig. 6, A and B, and Table 4), further experiments must be undertaken to further define these pathways.
It is noteworthy that although these agonists are not very potent in this assay, their efficacy for activating ERK exceeds that of the reference compound U69,593. Because the assay utilized here involves an immunohistochemistry approach, it is possible that the increase in fluorescence could be due to the induction of other kinases that may be recognized by the phospho-specific ERK1/2 antibodies used. Therefore, Western blot analysis was performed to determine whether the increase in phosphorylation detected could be attributed to the 42-and 44-kDa bands, p-ERK1 and p-ERK2, respectively. Western blot analysis confirmed that the intensity of 42-and 44-kDa bands are elevated over that observed for U69,593-induced phosphorylation suggesting that this effect is not an artifact of the 384well plate immunohistochemistry approach (Fig. 6C). Furthermore, none of the compounds activate ERK1/2 in the CHO-K1 parental cell line when tested at 1, 10, or 20 M suggesting that these effects are KOR-dependent (data not shown). Because the efficacy of the test compounds exceeded that of the reference compound in this assay, U69,593 was insufficient to define the maximum potential of the system, and therefore, these data were not fit to the operational model. Additional studies, investigating differences in ERK1/2 populations (cytosol and nuclear) as well as the temporal regulation of ERK1/2 phosphorylation, will be needed to make accurate comparisons between the agonists.  Tables 2 and 3 for G protein signaling assays and ␤arrestin2 recruitment assays were used to calculate bias factors. Bias factors are presented in Tables 2 and 3 and are plotted here on a logarithmic scale (base 10). As the reference agonist, the bias of U69,593 conforms to unity in all assays. The pathways represented are as follows: membrane [ 35 S]GTP␥S binding (mG protein); ␤arr2 EFC, ␤arr2 imaging, cellular impedance, and whole cell [ 35 S]GTP␥S binding (wcG protein). Independent of the platform used to assess G protein signaling or ␤arrestin2 recruitment, both 1.1 and 2.1 display bias for G protein signaling.
It is possible that, given the different receptor expression levels and cellular context, the ligands may not have comparable opportunities to interact with the receptor. To address this question, radioligand competition binding assays were used to determine whether the test ligands maintained comparable affinity and ability to displace   (Table 5). Further studies were performed to determine selectivity profiles of binding to MOR and DOR expressed in CHO-K1 cells. isoquinolinone analogues (B) for 10 min, and phosphorylated (p-ERK) and total ERK (t-ERK) were detected by fluorescence intensity in 96-well plate "In-cell Western" assays. The ratio of p-ERK1/2 to total ERK1/2 was calculated and reported normalized to the percentage of maximal response induced by U69,593. C and D, Western blot analysis confirms that the p42 and p44 bands increase with drug treatment when examined with the p-ERK1/2 antibody. The high degree of ERK1/2 phosphorylation in CHO-hKOR cells induced by the triazole (C) or isoquinolinone (D) stimulation exceeds that observed for U69,593 (U69) (10 M, 10 min), confirming the observations made in the In-cell Western format. Ratios of p-ERK1/2 over total ERK1/2 were normalized to the maximal U69,593 stimulation observed (n Ն 3; *, p Ͻ 0.05; **, p Ͻ 0.01, Student's t test); molecular masses are indicated in kDa. Veh, vehicle. The triazole and isoquinolinone derivatives, like the initial probes, display high selectivity for KOR binding over mu and delta opioid receptors (Table 5) (35,36,38). Ultimately, these compounds, of which there is a growing collection, may become important tools for ascertaining the relative contributions of G protein-dependent, ␤arrestin2-independent signaling at KOR in vivo. As an early step in characterization, the pharmacokinetic properties of 1.1 and 2.1 were assessed in mice. Brain and plasma compound contents were determined following a single systemic, intraperitoneal administration (10 mg/kg, intraperitoneal), and both compounds were detected at low micromolar concentrations in brain at 30 and 60 min (Table 6). Pharmacokinetic distribution and clearance rates were determined following a 2 mg/kg (i.v.) tail vein injection and timed blood collections. Plasma protein binding and nonspecific brain binding were also determined for these compounds. These parameters, along with the determined "logP" values are presented in Table 6 and indicate good brain exposure and reasonable pharmacokinetic properties. Together these findings suggest that these scaffolds may represent favorable leads for development of biased KOR agonists with CNS penetration.
Because G protein-mediated KOR signaling has been proposed as the mechanism underlying KOR-mediated antinociception, the antinociceptive effects of 1.1 and 2.1 were then assessed using the warm water tail-flick assay (Fig. 6) (10,59). Mice were injected with 30 mg/kg, intraperitoneal doses of either 1.1, 2.1, or the selective KOR agonist U50,488H, as this dose of U50,488H has been shown to induce antinociception in mice in this assay (60). The triazole and isoquinolinone compounds significantly increased the latency to tail withdrawal similar to that seen with U50,488H, with the effects peaking at 20 min post drug treatment (Fig. 7) demonstrating that these agonists are capable of inducing antinociception in mice.

DISCUSSION
The triazole and isoquinolinone analogues described herein are highly selective KOR agonists that induce receptor signaling biased toward G protein signaling over ␤arrestin2 recruitment. When tested in vivo, 1.1 and 2.1, a triazole and an isoquinolinone, respectively, prove to be brain-penetrant. Subsequently, 1.1 and 2.1 produce antinociception in the mouse tail flick test, which correlates with the ability of these compounds to potently activate G protein signaling in cell-based assays.
Prior studies suggest that KOR agonists that do not engage ␤arrestin2-mediated signaling pathways may prove to have less aversive properties such as dysphoria (30,61,62) (Fig. 8). It is hopeful that the generation of compounds, such as those in the series described here, will provide pharmacological tools that will aid in the elucidation of KOR-mediated signaling cascades in cellular systems and, importantly, in vivo. The fact that these series possess very high affinity and selectivity for KOR over other opioid receptors, are brain penetrant, and have favorable pharmacokinetic parameters will certainly aid in these ventures.

affinities for KOR agonists in the U2OS-hKOR-␤arrestin2-EFC, CHO-hKOR, CHO-hDOR, and CHO-hMOR cell lines
Binding parameters were determined for KOR using ͓ 3 H͔U69,593 and 10 M unlabeled U69,593 to determine the nonspecific binding. Saturation binding assays revealed 11.45 Ϯ 2.26 pmol/mg receptors in the U2OS-hKOR-␤arrestin2-EFC cell line and 2.07 Ϯ 0.48 pmol/mg receptor in the CHO-hKOR cell line. All of the compounds fully displaced ͓ 3 H͔U69,593 binding in a manner that did not significantly differ from displacement achieved with U69,593 (extra sum of squares F test, p Ͼ 0.05). To determine selectivity, ͓ 3 H͔diprenorphine was used for DOR with 10 M unlabeled naltrindole to determine nonspecific binding; ͓ 3 H͔DAMGO was used for MOR binding with 10 M naloxone used to determine nonspecific binding. Saturation binding assays revealed an affinity of 0.342 Ϯ 0.07 nM and 5.51 Ϯ 1.0 pmol/mg receptors in the CHO-hDOR cell line and an affinity of 0.443 Ϯ 0.02 nM and 5.36 Ϯ 0.76 pmol/mg receptor in the CHO-hMOR cell line. None of the compounds fully displaced radioligand binding to DOR or MOR at the highest concentrations tested (10 M); K i values were calculated for curves that displaced more than 50% of radioligand binding; those that did not displace more than 50% are listed as Ͼ10,000 nM. n Ն 3 individual assays; means are presented Ϯ S.E. K i ratios were calculated based on the K i values indicated (10,000 is used where Ͼ10,000 is indicated to define the upper limit of affinity).   (42). The triazole and isoquinolinone compounds investigated herein differ from 6Ј-GNTI in that they are full agonists in G protein signaling and they are able to activate ␤arrestin2 recruitment in the cell-based assays, although at a greatly diminished potency.

Entry
Furthermore, the triazoles and isoquinolinones do not potently activate ERK1/2, further distinguishing them from 6Ј-GNTI. The increasing number of compounds identified with diverse signaling profiles downstream of KOR activation may be useful for docking studies to the continually refined crystal structure of these GPCRs (63), which may provide insight into the chemical signatures driving specific receptor confirmations and signaling events. The triazoles and isoquinolinones consistently activate ERK1/2 with less potency than that observed for U69,593. Therefore, it is attractive to speculate that the triazoles and isoquinolinones are biased toward G protein signaling over ERK1/2 phosphorylation. We hesitate, however, to make this claim based on two confounding variables. In this particular pathway, U69,593 does not adequately serve as a reference compound. The second issue, and perhaps the most important, is that ERK1/2 activation is a downstream effector that represents the consolidation of multiple upstream cascades (potentially originating from affecting different G proteins, ␤arrestins, calcium influx, etc.). Although our negative controls demonstrate that the activation of ERK by the test compounds is due to KOR (no response in parental CHO-K1 cells), it is difficult to interpret what component of cellular signaling is participating more (overshoot in efficacy) or less (rightward shift in potency) in generating the response profile elicited by these compounds downstream of KOR. Regardless, the compounds consistently reveal more potency in the G protein signaling assay than in the ERK1/2 and ␤arrestin assays. Overall, these results are intriguing and represent an area for further explorations into the compartmental (nuclear versus cytoplasmic (64)) and temporal (transient or sustained (28,64)) aspects of GPCR-stimulated ERK activation in respect to ␤arrestin and G protein-dependent mechanisms.
It will be of great interest to utilize these compounds in additional behavioral and physiological assays to determine the contribution of ␤arrestin2 to KOR signaling in vivo. However, it will first be important to determine that the signaling bias observed across the various cell-based assay platforms is maintained in vivo and whether these pathways are associated with behaviors. If indeed these newly discovered KOR-biased agonists, which are brain penetrant following systemic injections, prove to maintain signaling bias in the endogenous setting, these compounds may serve as important tools for investigating the contributions of ERK activation and/or ␤arrestin2 recruitment to KOR-mediated effects in vivo, and it is hopeful that these new pharmacological tools will serve the community for exploring relevant KOR biology in an endogenous setting. FIGURE 8. Schematic of isoquinolinone and triazole compound signaling at KOR compared with U69,593. Activation at KOR mediates multiple signaling cascades leading to G protein coupling, ␤arrestin recruitment, and ERK kinase phosphorylation. The isoquinolinones and triazoles act as biased agonists, preferentially inducing G protein coupling over ␤arrestin2 recruitment. This lack of preference for ␤arrestin2 recruitment appears to correlate with a loss in potency for ERK1/2 activation. Because KOR-mediated antinociception has been attributed to G protein signaling mechanisms and because KORinduced interactions with ␤arrestin2 are proposed to induce dysphoria, biasing KOR activation toward G protein signaling and away from ␤arrestin2 pathways may be the key to induce antinociception and avoid dysphoria (30,57,58).