Abstract
An amine-containing molecule called Compound A has been reported by a group from Bristol-Myers Squibb to act as a positive allosteric modulator (PAM) at the dopamine D1 receptor. We synthesized the more active enantiomer of Compound A (BMS-A1) and compared it with the D1 PAMs DETQ and MLS6585, which are known to bind to intracellular loop 2 and the extracellular portion of transmembrane helix 7, respectively. Results from D1/D5 chimeras indicated that PAM activity of BMS-A1 tracked with the presence of D1 sequence in the N-terminal/extracellular region of the D1 receptor, a unique location compared with either of the other PAMs. In pairwise combinations, BMS-A1 potentiated the small allo-agonist activity of each of the other PAMs, while the triple PAM combination (in the absence of dopamine) produced a cAMP response about 64% of the maximum produced by dopamine. Each of the pairwise PAM combinations produced a much larger leftward shift of the dopamine EC50 than either single PAM alone. All three PAMs in combination produced a 1000-fold leftward shift of the dopamine curve. These results demonstrate the presence of three non-overlapping allosteric sites that cooperatively stabilize the same activated state of the human D1 receptor.
SIGNIFICANCE STATEMENT Deficiencies in dopamine D1 receptor activation are seen in Parkinson disease and other neuropsychiatric disorders. In this study, three positive allosteric modulators of the dopamine D1 receptor were found to bind to distinct and separate sites, interacting synergistically with each other and dopamine, with the triple combination causing a 1000-fold leftward shift of the response to dopamine. These results showcase multiple opportunities to modulate D1 tone and highlight new pharmacological approaches for allosteric modulation of G-protein–coupled receptors.
Introduction
G-protein–coupled receptors (GPCRs) transduce the biologic signals of numerous hormones, growth factors, and neurotransmitters, thereby representing the largest single class of druggable human proteins (Santos et al., 2017; Alexander et al., 2021). Traditional pharmacological approaches for targeting GPCRs have involved identifying small-molecule drugs that occupy the same pocket as the natural ligand (orthosteric agonists or antagonists). However, accumulating biochemical and structural data indicates that GPCRs may have additional (allosteric) ligand-binding pockets that can modulate receptor function either positively [positive allosteric modulator (PAM)] or negatively [negative allosteric modulator (NAM)] (Stockton et al., 1983; Bruns and Fergus, 1990; May et al., 2007). PAMs affect receptor activation by modulating the binding affinity and/or efficacy of the orthosteric ligand and, at higher concentrations, can directly activate receptors. As therapeutics, PAMs have potential advantages over orthosteric agonists because their actions are largely dependent on the presence of endogenous agonist, which is under physiologic control (May et al., 2007; Kenakin, 2010). Mechanistically, PAMs operate by binding selectively to the activated state of the receptor (Bruns and Fergus, 1990; Ehlert, 2016), driving the thermodynamic equilibrium toward activation, whereas NAMs have the reverse selectivity. In theory, then, any site that changes its configuration between the active and inactive states could host an allosteric modulator, and several such sites have been identified on various GPCRs (Thal et al., 2018).
The D1 receptor is the most highly expressed of the five dopamine receptors (Beaulieu and Gainetdinov, 2011). Although deficiencies in D1 activation are associated with various central nervous system disorders such as Parkinson disease and schizophrenia (Ikemoto et al., 1997; Nakamura et al., 2014; Arnsten et al., 2017), attempts to develop D1 agonists have been hampered by issues including tolerance development, pharmacokinetic limitations, and limited selectivity (Felsing et al., 2019; Svensson et al., 2019). To overcome these problems, we (Svensson et al., 2017; Bruns et al., 2018) and others (Lewis et al., 2015; Shiraki et al., 2015; Luderman et al., 2018) have identified various chemical series with D1 PAM activity (Fig. 1). Among these are two members of the tetrahydroisoquinoline (THIQ) class of D1 PAMs, the pharmacological tool DETQ and the phase II clinical molecule LY3154207 (approved generic name mevidalen) (Hao et al., 2019; Biglan et al., 2022).
Chemical Structures of D1 PAMs. Chemical Abstracts names: DETQ, [2-(2,6-dichlorophenyl)-1-((1S,3R)-3-(hydroxymethyl)-5-(2-hydroxypropan-2-yl)-1-methyl-3,4-dihydroisoquinolin-2(1H)-yl)ethan-1-one]; MLS6585, N-(6-(tert-butyl)-3-carbamoyl-4,5,6,7-tetrahydrobenzo[b]thiophen-2-yl)isonicotinamide; BMS-A1, [1-((rel-1S,3R,6R)-6-(benzo[d][1,3]dioxol-5-yl)bicyclo[4.1.0]heptan-3-yl)-4-(2-bromo-5-chlorobenzyl)piperazine]. The absolute configuration of BMS-A1 is unknown and could be either (1S,3R,6R) or (1R,3S,6S).
To date, two separate binding sites for D1 PAMs have been reported. The most well characterized of these is the inner surface of intracellular loop 2 (ICL2) along with adjacent parts of transmembrane helices 3 and 4. Mutations in this region have marked effects on the affinity of several D1 PAMs, including DETQ as well as compounds from two other structural series (Lewis et al., 2015; Luderman et al., 2018; Wang et al., 2018). LY3154207 has been imaged in the ICL2 site by cryogenic electron microscopy (Xiao et al., 2021; Zhuang et al., 2021; Teng et al., 2022).
In contrast, the D1 PAM activity of MLS6585 is unaffected by mutations to the ICL2 site (Luderman et al., 2018; Wang et al., 2018). MLS6585 was shown to interact supra-additively with DETQ and other ICL2-binders, confirming that it binds to a separate site. In agreement with this observation, the binding site for MLS6585 has recently been reported to include the extracellular portion of transmembrane helix 7 (Luderman et al., 2021).
In 2015, a group from Bristol-Myers Squibb described two D1 PAMs, Compound A and Compound B. Compound B was shown to bind to ICL2, but the binding site for Compound A was not characterized, other than excluding the possibility that it bound to ICL2 (Lewis et al., 2015).
In the current study, we synthesize Compound A and separate the more active enantiomer, which we abbreviate as BMS-A1. Using D1/D5 chimeras, we show that BMS-A1 binds to the N-terminal/extracellular region of the D1 receptor, distinct from ICL2 binders (N-terminal/intracellular) and MLS6585 (C-terminal/extracellular). Pairwise and triple combinations among DETQ, MLS6585, and BMS-A1 in the absence of dopamine show greatly increased allosteric agonist activity, confirming that they occupy distinct binding sites on the D1 receptor that are functionally cooperative. Moreover, dual and triple combinations of the three PAMs synergistically enhance the functional affinity of the orthosteric agonist dopamine, causing a leftward shift of up to 1000-fold in the dopamine EC50 in the triple combination. These results provide the first evidence that a functional four-ligand GPCR complex can be formed, with potential implications for GPCR pharmacology, allostery, and drug discovery.
Materials and Methods
Materials
DETQ was synthesized as previously described (Beadle et al., 2014; Hao et al., 2019). MLS6585, called CID 2886111 in our previous paper (Wang et al., 2018), was purchased from ChemBridge (San Diego, CA). The more active enantiomer of Bristol-Myers Squibb Compound A was synthesized at Eli Lilly and Company (Supplemental Data II: Synthesis of BMS-A1 and Stereoisomers). Dopamine and other pharmacological reagents were purchased from Sigma (St. Louis, MO). Sources of other reagents are provided in individual protocols.
HEK293 Cell Line Expressing the Human D1 Receptor
All experiments except those involving D1/D5 chimeras were carried out in a stable HEK293 cell line expressing the human D1 receptor as previously described (Svensson et al., 2017). This cell line shows a moderate level of D1 receptor expression (Bmax 0.36 pmol/mg protein in 3H-SCH 23390 binding), and allo-agonist activity of DETQ and MLS6585 is relatively low compared with the Jump-In expression system used for the D1/D5 chimeras (Wang et al., 2018).
D1/D5 Chimera Cell Lines
Human D1/D5 chimeras were constructed as previously described (Wang et al., 2018). Human DRD1 and DRD5 cDNA constructs were purchased from Open Biosystems (Huntsville, AL) and ThermoFisher (Waltham, MA), respectively. D1/D5 chimera DNA was synthesized by GenScript (Piscataway, NJ). All constructs were verified by DNA sequencing. Switchover points for all chimeras are described in the supplemental data in Wang et al. (2018). Stable cell lines were established using the Jump-In T-REx HEK293 Retargeting Kit (ThermoFisher). Wild-type and chimera constructs were either directly cloned into pJTI R4 CMV-TO vector or subcloned from pcDNA3.1, then transfected using Fugene HD into Jump-In T-REx HEK293 cells. Transfected cells were selected using 2.5 mg/ml G418 for 10–14 days. Stable cells were induced using 1 µg/ml doxycycline for 24–48 hours, then harvested and suspended in freeze media (FBS with 6% DMSO) at 107 cells/ml, and aliquots were stored in liquid nitrogen. Receptor expression levels in the Jump-In system were typically around 5 pmol/mg protein or roughly 10-fold higher than in the D1 cell line used for α-shift experiments (Wang et al., 2018).
Measurement of cAMP Response
Cyclic AMP experiments were carried out as previously described (Svensson et al., 2017; Wang et al., 2018) with minor modifications. In the D1/D5 chimera studies, BMS-A1 was diluted and dispensed into assay plates in 80 nL DMSO (ProxiPlate-384 Plus, PerkinElmer, Waltham, MA) using acoustic dispensing (Echo, Beckman Coulter, San Jose, CA). Dopamine was dissolved and diluted in DMSO and dispensed in a 10 nL volume (final assay volume 10 μL). In preliminary experiments, a separate dopamine EC20 value was determined for each construct (Wang et al., 2018). An 8 μM final concentration of dopamine was used to define the maximum response to dopamine. The final DMSO concentration was 0.9%.
Curve shift experiments were carried out in the stable HEK293 D1 cell line described above. The first PAM (DETQ or BMS-A1) was diluted and dispensed into assay plates (ProxiPlate-384 Plus, PerkinElmer) using acoustic dispensing with a twofold dilution factor in a 14-point curve in 80 nL DMSO to rows A–N. Dopamine was diluted and dispensed into the assay using acoustic dispensing with a twofold dilution factor and 20-point curve in 80 nL DMSO to columns 3–22. The second and/or third potentiator compound was dispensed into assay plates using acoustic dispensing in 20 nL DMSO to reach final concentration in a 10-μl reaction volume. Diluted compound and dopamine were resuspended with 5 uL STIM buffer, followed by cells (2000 cells/well) in 5 µl STIM. STIM buffer consisted of Hanks’ Balanced Salt Solution supplemented with 0.1% bovine serum albumin, 20 mM HEPES, 500 µM IBMX, and 100 µM ascorbic acid. The final DMSO concentration was 1.8%. To reduce the likelihood of precipitation, test compounds were diluted in 100% DMSO until the penultimate dilution into 50% of the final assay volume.
In both versions of the assay, plates were incubated at room temperature for a total reaction time of 60 minutes. Cyclic AMP production was quantified using homogeneous time-resolved fluorescence detection (Cisbio, Bedford, MA) according to vendor instructions: lysis buffer containing anti-cAMP cryptate (5 µl) and cAMP-D2-conjugate (5 µl) was added to the wells, plates were incubated for an additional 60–90 minutes, and time-resolved fluorescence was detected using an EnVision plate reader (PerkinElmer).
Fluorescence data were converted to cAMP concentrations using a cAMP standard curve. For potentiator-mode concentration-response curves (CRCs), results for each construct were expressed as percent of the window between an EC20 concentration of dopamine alone and the maximum response to dopamine in that construct. For the curve shift assays, concentration-response curves were expressed as percent of window between STIM buffer alone and 8 µM of dopamine.
Curve-Fitting Analysis
Cyclic AMP values were fit to a four-parameter logistic equation with equal weighting using GraphPad version 9.3.1 (San Diego, CA).
The allosteric activity of a PAM is characterized by two experimentally derived parameters: α denotes the leftward shift of the agonist CRC, defined as the ratio between the EC50 for agonist alone and the EC50 for agonist in the presence of one or more PAMs (EC50/EC50’ ratio), and β denotes the increase in maximum efficacy for agonist elicited by one or more PAMs. The overall allosteric boost is defined as γ = α · β. In the present study, β was about 1.1 (Fig. 4), indicating that dopamine was nearly a full agonist in this system. In contrast, α-shifts for the three PAMs separately and in combination spanned up to three orders of magnitude. Given that β contributed only about 1% of the maximum allosteric boost (0.04 log units for β, compared with 3 log units for α), further analysis focused on α.
For α-shift analysis (Figs. 5 and 7), EC50/EC50’ ratios were calculated separately for each day’s experiment, and the experiments were then combined and fit to the equation
where y is α-shift for dopamine, x is the concentration of PAM A varied in CRC, KB is the half-maximal concentration for PAM A, bottom is the α-shift for PAM B (or B + C) at fixed concentration, top is the maximum α-shift, and span is top minus bottom. Weighting was set to the inverse square of the y value (α-shift). For a single PAM, the bottom is fixed to 1, and the equation is mathematically equivalent to equation 25 of Ehlert (1988), except that the variable names have been changed to GraphPad usage. In the case of a CRC for PAM A in the presence of fixed PAM B, bottom is the α-shift for B alone, and top is the maximum α-shift for both PAMs combined. It should be noted that this method of analysis is the same as the Schild analysis customarily used for competitive antagonists and NAMs, except that here α denotes a leftward shift.
Standard error values for fitted parameters were computed in GraphPad as previously described (Wang et al., 2018). The S.E. of the log EC50 provided by GraphPad was converted to S.E. of the untransformed (linear) EC50 by the equation
Results
Characterization of the Enantiomers of Compound A
In the initial publication (Lewis et al., 2015), Compound A was a racemic mixture of the two enantiomers with anti relative stereochemistry between the cyclopropyl and piperazine groups. We resynthesized Compound A (Supplemental Data II: Synthesis of BMS-A1 and Stereoisomers), separated the two enantiomers, and found that one enantiomer was roughly threefold more potent than the other (EC50 values 1240 ± 130 nM versus 3500 ± 300, Supplemental Table 1) [the two syn isomers were also synthesized and were found to be less potent (EC50 values of 6100 and 7300 nM)]. The fairly similar activity of the four isomers suggests a rather loose fit with the D1 receptor, which is also consistent with their modest affinity values relative to the size of the molecule. The more active enantiomer, henceforth referred to as BMS-A1, was used for all subsequent studies.
Identification of the General Binding Region (Quadrant) for BMS-A1 Using D1/D5 Chimeras
BMS-A1 was tested in four D1/D5 chimeras that were previously used to characterize the binding sites for DETQ and MLS6585. In each of these, half of the D1 receptor (N-terminal versus C-terminal or extracellular versus intracellular) was replaced with its D5 counterpart (Wang et al., 2018). The combined results for the four chimeras identify which quadrant of the receptor is required for PAM activity; for instance, DETQ is only active when the N-terminal and intracellular halves are D1, implying that it binds to the N-terminal/intracellular quadrant. In wild-type D1 receptor, BMS-A1 potentiated the dopamine response with an EC50 of 3.4 μM and a maximum response that was 117% of the dopamine maximum (Fig. 2; Table 1). This is in reasonable agreement with published results for the racemic mixture (Lewis et al., 2015). In contrast, BMS-A1 antagonized the dopamine response in the wild-type D5 receptor (we did not further characterize whether the antagonism was orthosteric, allosteric, or nonspecific). Replacing either the extracellular or N-terminal half of the D1 receptor with its D5 counterpart abolished PAM activity, whereas replacing the cytoplasmic or C-terminal half with D5 left PAM activity unchanged. These results indicate that BMS-A1 binds to the N-terminal/extracellular quadrant of the D1 receptor (“upper left” quadrant in the standard orientation for depiction of GPCRs). BMS-A1 thus occupies a unique third allosteric site, setting it apart from DETQ (lower left quadrant) and MLS6585 (upper right quadrant) (Wang et al., 2018; Luderman et al., 2021).
Potentiation by BMS-A1 of the cAMP accumulation elicited by an EC20 concentration of dopamine in D1/D5 chimeras. The dopamine window is defined as the range between the cAMP response to an EC20 concentration of dopamine and the response to a maximum (8 μM) concentration of dopamine. The receptor diagram is taken (CC-BY) from Wang et al. (2018). The NT/CT breakpoint was at the extracellular end of TMH4, and the out/in breakpoints were at the midpoints of each transmembrane helix; see the supplemental data of Wang et al. (2018) for the exact breakpoints. Values are means ± S.E., n = 3. In the D5, D5out/D1in, and D5NT/D1CT constructs, BMS-A1 inhibited cAMP accumulation. Although GraphPad was able to find best-fit values for the inhibitory curves, it was unable to compute confidence intervals for several fit parameters, and the curves are therefore included only as visual aids. CT, C-terminal; NT, N-terminal.
Best-fit values for BMS-A1 CRCs in D1/D5 chimeras (data from Fig. 2).
Bottom and top values are expressed as percent of the dopamine window, defined as the difference between the maximum response to dopamine and the response to an EC20 concentration of dopamine. S.E. values of best-fit parameters were calculated as described in Materials and Methods.
Mutual Synergy between Three D1 PAMs in the Absence of Dopamine
If two molecules stabilize the same protein conformation but act at different sites, their free energies of binding should be additive, implying that their effects on the conformational equilibrium should be multiplicative (synergistic). Since agonists and PAMs both work by selectively binding to the active state of a given receptor (Bruns and Fergus, 1990; Ehlert and Griffin, 2014), a PAM + PAM combination should show the same qualitative synergy as a PAM + agonist combination as long as the two PAMs bind to independent sites. In agreement with this prediction, DETQ and MLS6585 have been shown to interact multiplicatively in the absence of dopamine (Wang et al., 2018). In the case of three independent PAM sites, each pairwise combination as well as the triple combination should show synergy. This prediction provides a confirmatory test of the independence of the three sites.
In the current series of experiments, the limit of detection for an increase in cAMP was about 1% of the maximum response to dopamine. MLS6585 and BMS-A1 showed very small but detectable allo-agonist activity: 1.2% ± 0.1 for 80 μM MLS6585 and 2.0% ± 0.5 for 80 μM BMS-A1 (see bottom values for DETQ curves in Supplemental Table 2). The allo-agonist response (if any) to DETQ was below the detection limit. Previously (Svensson et al., 2017), DETQ was reported to show variable allo-agonist activity, ranging from 2.7% to 13% (n = 4). The prior study used the same cell line but was conducted about 8 years before the current experiments; the difference may be due to minor variations in experimental protocol.
Each pairwise PAM combination showed much greater cAMP accumulation than the sum of each PAM separately (Fig. 3; Supplemental Table 2). For DETQ plus MLS6585, maximum cAMP was 4.7% of the dopamine maximum, compared with <1% and 1.2% separately; for DETQ plus BMS-A1, 15.2% compared with <1% and 2.0%; and for MLS6585 plus BMS-A1, 26.6% compared with 1.2% and 2.0%. All three pairwise combinations showed synergy, implying that the three PAMs bind to different sites but stabilize the same active conformation or conformations. Further supporting this conclusion, the triple combination showed an additional supra-additive interaction, reaching 63.8% of the dopamine maximum.
Mutual potentiation of the allosteric agonist effects of DETQ, MLS6585, and BMS-A1 in the absence of dopamine. Values are means ± S.E. [n = 4 for (A–C), n = 5 for (D)]. Best-fit parameters for the curves are provided in Supplemental Table 2.
Hill coefficients for BMS-A1 in this experiment were noticeably higher than unity, typically about 2 (Fig. 3, B and C; Supplemental Table 2).
Multiplicative leftward Shift of the Dopamine CRC by Combinations of Two PAMs
From receptor theory, the mutual synergy between two PAMs in the absence of dopamine should also cause a larger leftward shift in the dopamine CRC compared with either PAM separately. In other words, the three free energies (PAM A, PAM B, and agonist) should be additive, resulting in multiplicative effects on affinity. This has been confirmed for the combination of MLS6585 with ICL2 binders (Luderman et al., 2018). Assuming that BMS-A1 activates the same receptor conformation, it should show a similar supra-additive response with either DETQ or MLS6585 separately and possibly also in a triple combination with both PAMs.
In agreement with this prediction, BMS-A1 caused a marked increase in the ability of DETQ to elicit a leftward shift (α-shift) in the dopamine CRC (Fig. 4, middle column), increasing the α-shift from 18-fold for DETQ alone to about 200-fold for the combination (Fig. 5B; see Table 2 for best-fit parameters). A similar multiplicative interaction was observed between BMS-A1 and MLS6585 (Fig. 4, right column), in which MLS6585 increased the α-shift for BMS-A1 from about 100-fold to roughly 500-fold (Fig. 5C) [the CRCs for BMS-A1 began to curve down at the highest concentrations but did not achieve well-defined plateau values; because of this, the best-fit values for KB and maximum α-shift were poorly characterized due to high correlation between the two parameters, but the initial slopes of the curves were still well defined (Table 2)]. Finally, the expected synergy between DETQ and MLS6585 was confirmed (Fig. 4, left column; Fig. 5A). Each of these interactions was concentration dependent (Fig. 5).
Potentiation of the cAMP response to dopamine by pairwise combinations of PAMs. In the left and middle columns, each curve represents a different concentration of DETQ; in the right column, each curve represents a different concentration of BMS-A1. Values are means of two replicates from a single experiment. The EC50 for dopamine alone was 55 ± 5 nM (n = 2). Composite best-fit α-shift values from this experiment and one other are given in Fig. 5.
Leftward shift of concentration-response curves for dopamine in the presence of pairwise combinations of PAMs. EC50/EC50’ (α-shift) is the control EC50 for dopamine divided by its EC50 in the presence of one or more PAMs. Values are means ± S.E. of α-shift values derived from two experiments. Curves are from a model in which top, bottom, and KB are allowed to vary, and the Hill coefficient is set to 1. See “Curve-Fitting Analysis” in Materials and Methods for a more detailed description. Best-fit parameters are provided in Table 2.
Best-fit parameters for leftward shifts of the dopamine CRC in the presence of pairwise combinations of PAMs (Fig. 5)
The maximum response to dopamine was barely increased by the PAMs, either singly or in combination (Fig. 4), with β (fold increase in the maximum response to agonist) usually 1.1 or less. Dopamine is therefore nearly a completely full agonist (efficacy >90%), and overall allosteric enhancement, defined as γ = α · β (Ehlert and Griffin, 2014), is almost completely dominated by α.
It should be noted that the 10 and 20 μM concentrations of BMS-A1 lowered the maximum response to dopamine (Fig. 4, right column) while still affording a small additional leftward shift of the dopamine EC50 (Fig. 5C), suggesting nonspecific inhibition of cAMP accumulation at these higher concentrations.
Triple Combination
The combination of all three PAMs caused an additional leftward shift in the CRC for dopamine (Fig. 6), with a maximum shift of 996-fold (Fig. 7; Table 3). The interactions were separately concentration-dependent for each of the three PAMs (Fig. 7).
Potentiation of the cAMP response to dopamine in the presence of two or three PAMs. Values are means of two replicates from a single experiment. Composite best-fit α-shift values from this experiment and two others are given in Fig. 7.
Additional leftward shift for DETQ in the presence of one or both of the other two PAMs. EC50/EC50’ (α-shift) is the control EC50 for dopamine divided by its EC50 in the presence of one or more PAMs. Values are means ± S.E. of α-shift values derived from three experiments, except for DETQ control curves (n = 2). Curves are from a model in which top, bottom, and KB are allowed to vary, and the Hill coefficient is set to 1. See “Curve-Fitting Analysis” in Materials and Methods for a more detailed description. Best-fit parameters are provided in Table 3.
Best-fit parameters for leftward shifts of the dopamine CRC in the presence of pairwise and triple combinations of PAMs (Fig. 7)
Leftward shifts from combinations of PAMs were much greater than additive but were consistently slightly less than would be expected from a strictly multiplicative interaction (Fig. 8). For instance, DETQ caused an 18-fold leftward shift by itself, but its incremental shift was 10-fold in the presence of 5 μM MLS6585, eightfold in the presence of 5 μM BMS-A1, and fivefold in the presence of both PAMs. A similar pattern was seen for MLS6585 and BMS-A1. In each case, the lowest incremental α-shift was seen when both of the other two PAMs were present.
Diagram summarizing α-shift interactions. Absolute α-shift values are shown in blue at the vertices of the cube and are taken from fitted bottoms and tops in Fig. 7A. Incremental shifts (in red) are shown at the edges of the cube. PAM concentrations are 5 μM for MLS6585 and BMS-A1; DETQ data are for CRCs up to 2 μM.
Discussion
The most important findings of this study are 1) the identification of a third allosteric site on the dopamine D1 receptor and 2) the observation that D1 PAMs that act at different sites interact synergistically with each other as well as with dopamine.
Previously, various D1 PAMs, exemplified by DETQ, were shown to bind to intracellular loop 2 (Lewis et al., 2015; Luderman et al., 2018; Wang et al., 2018). A structurally different series of D1 PAMs, exemplified by MLS6585, was shown to bind to another site and to interact cooperatively with the ICL2 binders (Luderman et al., 2018; Wang et al., 2018). This site, which has not yet been characterized in detail, appears to be located in the C-terminal/extracellular quadrant of the D1 receptor (Luderman et al., 2021). Another D1 PAM, Compound A, was shown not to bind to ICL2, but its site of action was not further delineated (Lewis et al., 2015). The unique structure of Compound A led us to speculate that it could bind to a different site from other published D1 PAMs. We therefore synthesized BMS-A1, the more active enantiomer of Compound A, and characterized its binding location and its interaction with other D1 PAMs. We found that BMS-A1 binds to the N-terminal/extracellular quadrant of the dopamine D1 receptor, a location that is distinct from the two previously described D1 PAM sites. In addition, we found that BMS-A1 interacts multiplicatively with PAMs that bind to the other two sites, indicating that all three classes of PAMs stabilize the same active conformation (or population of conformations) of the D1 receptor.
A Simple Thermodynamic Model of Allosterism
According to the theory of linked equilibria (Monod et al., 1965), proteins that are under allosteric control can transition between two conformations, an inactive and an active state. Any ligand that binds selectively to one of the two conformations will drive the equilibrium in the direction of the corresponding state. PAMs bind to the active state, favoring activation, and NAMs bind to the inactive state, suppressing activity (Ehlert and Griffin, 2014). An interesting consequence occurs when two ligands bind selectively to the same active conformation but at different sites: their free energies will be additive, which implies that their functional effects will be multiplicative, up to some saturation point. PAMs potentiate agonists (and vice versa) in this way. Importantly, in the simple two-state model, the active-state conformation is the same regardless of how the receptor is occupied, whether by agonist, allosteric modulator or modulators, or any combination of agonist and modulators (Ehlert, 2016). Output is determined solely by the fraction of receptors in the active state. The current study provides experimental results that can be used to test this model. The D1 PAMs appear to be particularly useful for this purpose because they appear to lack more complicated types of activity such as effector bias, probe dependence, or a required order of binding (Svensson et al., 2017; Luderman et al., 2018).
Diversity of GPCR Allosteric Sites
The identification of a third allosteric site on the D1 receptor brings home the point that any location that changes its conformation between the active and inactive conformations has the potential to host a ligand that distinguishes between these conformations. Such a ligand will be a PAM or NAM, depending on the direction of its selectivity. Given that much of the surface of a typical GPCR will shift in the transition between active and inactive states (Rasmussen et al., 2011), this provides considerable scope for discovery of different allosteric sites, and multiple such sites have been reported for GPCRs (Thal et al., 2018; Wang et al., 2021). The exact location of the binding site for BMS-A1 requires further delineation, but the N-terminal/extracellular quadrant is known to host allosteric modulators for GPR40, P2Y1, and PAR2 (Thal et al., 2018). This general location is also the site of a cleft or tunnel through which several lipid mediators enter the orthosteric sites of their respective GPCRs (Hanson et al., 2012). The adenosine A1 PAM series has been localized by cryogenic electron microscopy to the extracellular sections of transmembrane helices 1, 6, and 7 (Draper-Joyce et al., 2021) and thus could overlap with the binding sites for BMS-A1 or MLS6585, or both. The A1 PAM series bears some structural resemblance in its core scaffold to MLS6585. Somewhat oddly, none of the three D1 PAMs binds to the vestibule region [extracellular loop (ECL)-2, ECL3, and nearby transmembrane segments], which is a well known hotspot for muscarinic NAMs and PAMs (Kruse et al., 2013). However, unlike the five muscarinic receptors, the D1 and D5 receptors show poor conservation of ECL2, suggesting disordered structure (https://www.uniprot.org).
Reciprocity of PAM and Agonist
Although PAMs and orthosteric agonists at first glance would appear to possess qualitatively different types of activity, the two-state model indicates that both operate by selectively binding to the activated state, the only difference being the binding location. The apparent qualitative difference is undoubtedly due to the shaping of the agonist/receptor interaction by millions of years of evolution, resulting in much higher active-state affinity and selectivity for endogenous agonists compared with PAMs, which are generally discovered by screening and optimized by (at most) a few years of medicinal chemistry. Although PAMs are thought of as potentiators for agonists, the reverse interpretation is equally valid, as seen by the ability of agonists to enhance the binding of PAMs (Gavish and Snyder, 1980). The parallelism between agonists and PAMs leads to the prediction that PAMs should have some degree of efficacy on their own (allo-agonism), which was observed already in the first GPCR PAMs (Bruns and Fergus, 1990). Furthermore, provided they act at different sites, PAMs should potentiate each other, and the current study confirms this prediction in each of the possible 2- and 3-way combinations (Fig. 3).
Saturation of Multiplicative leftward Shift
The two-state model predicts that multiplicative interactions should begin to wane when a large fraction of the receptor population is already in the active state. Such a ceiling effect is seen for the triple PAM combination (Fig. 8). Can we use these results to estimate the fraction of D1 receptors in the active state under basal conditions? The 1000-fold α-shift observed with the triple combination (Fig. 7) indicates that this PAM combination increases the active-state population by 1000-fold, which, in turn, implies that less than 1/1000th of the receptor is constitutively active. At the same time, the increase in dopamine affinity for the triple combination lags behind the roughly 4000-fold shift expected from multiplying the three single α-shifts, indicating that this combination is beginning to approach a limit that presumably represents full occupation of the active state. The maximum α-shift is consistent with an inactive:active-state ratio (ε0) of around 1600 under basal conditions, although this estimated value has a wide confidence interval and is almost entirely dependent on a single data point (Fig. 9; Table 4). Nevertheless, the same combination (2 μM DETQ + 5 μM MLS6585 + 5 μM BMS-A1) caused a 63.8% allo-agonist response in the absence of dopamine (Fig. 3), again suggesting that this combination approaches but does not reach full occupation of the active state. Overall, the results provide a tentative demonstration that PAM combinations can be used to probe relative levels of active and inactive receptor states.
α-Shifts predicted from multiplying single-PAM shifts compared with experimentally observed α-shifts. Data are from Fig. 7/Table 3, with additional data for 5 μM MLS6585 + 0.2 μM BMS-A1 (± DETQ) and the converse 0.2 μM + 5 μM combination. Data were fit to a three-parameter CRC model with the bottom fixed to 1. The best-fit value for maximum observed α-shift was 1610 (95% confidence interval, 900–7400); half-maximal observed shift occurred at a predicted shift of 1710 (95% confidence interval 830–7300).
Comparison of values predicted from multiplying α-shifts from single PAMs with experimentally observed α-shifts for double and triple combinations
We also constructed a chemical equilibrium-based model for the four-ligand interaction with the D1 receptor (Supplemental Data I: Four-Site Allosteric Model), and predictions from this model were qualitatively consistent with the experimental data.
Limitations and Advantages of D1 PAMs as Pharmacological Tools
The CRC for BMS-A1 shows a downward slope at higher concentrations beginning around 10–20 μM (Fig. 4, right column). This phenomenon is also seen with DETQ and MLS6585 at about the same concentrations (data not shown). It does not appear to be due to activity at the D1 receptor because DETQ is much more potent than the other two compounds in PAM activity but about the same as the others in this inhibitory effect. It may involve inhibition of cAMP generation by a nonspecific mechanism associated with hydrophobic compounds, such as formation of aggregates (Irwin et al., 2015). Because of this inhibition, it is difficult to measure the true maximum α-shifts for MLS6585 and BMS-A1. On the other hand, DETQ reaches a plateau before the onset of this downward effect, so its roughly 20-fold maximum α-shift is probably a true measure of its selectivity for the active form over the inactive form of the D1 receptor. More potent analogs of MLS6585 and BMS-A1 would be useful.
Conversely, the D1 PAMs may be useful tools for studying the two-state model because they lack more complicated pharmacology seen with other PAMs. In particular, DETQ and MLS6585 appear to lack biased signaling and probe dependence (Svensson et al., 2017; Luderman et al., 2018). In addition, the three D1 PAMs do not bind to the vestibule region and therefore are unlikely to impede access of ligands to the orthosteric binding pocket. Finally, the availability of PAMs acting at three independent sites on the D1 receptor affords a unique opportunity to investigate various phenomena predicted by the two-state model, such as multiplicative effects between PAMs and between PAMs and agonists.
Therapeutic Possibilities
PAMs amplify physiologic control circuits, and the potential therapeutic advantages of this approach have long been known (Bruns and Fergus, 1990; Nemeth et al., 1998; Christopoulos and Kenakin, 2002). In the D1 system in particular, the PAM mevidalen, a close analog of DETQ, has shown efficacy versus both motor and nonmotor symptoms of patients with Lewy body dementia and Parkinson disease (Biglan et al., 2022).
Going further, combinations of allosteric modulators may have advantages over single agents. For instance, two PAMs that acted at different sites could be effective at much lower doses than either dosed separately, potentially circumventing pharmacokinetic limitations or non-mechanism-related toxicity. Greater efficacy might also be obtained from a greater maximum α-shift of the combination or from greater allo-agonism in situations where the endogenous agonist is not present. Conversely, our results are a reminder that synergistic combinations (agonist plus PAM or PAM plus PAM) carry the risk of therapeutic overshoot.
Biased signaling involves the presence of two different active conformations, each related to a different downstream effector pathway (Kenakin and Christopoulos, 2013). In theory, compounds showing biased signaling could have a superior margin of safety, but designing drugs with a clinically significant degree of signal bias has proven to be difficult. The thermodynamic model, when extended to two active conformations, predicts that signaling bias should be multiplicative, which could allow two drugs, each with a modest degree of bias (agonist plus PAM or two PAMs acting at independent sites), to together produce a more robust degree of bias.
Combinations of NAMs acting at independent sites could also have advantages. A single NAM has a maximum rightward shift determined by its selectivity for the inactive state. With two NAMs acting at different sites, rightward shifts would multiply, allowing lower dosing and/or a larger maximum effect. In addition, if two NAMs acting at different sites each separately showed therapeutic activity, treatment with the combination might make it less likely for a single mutation to cause resistance, which could be advantageous in treating disorders such as cancer (Hany et al., 2022) and infectious disease (Naidu et al., 2022) in which the appearance of resistance can abrogate the effectiveness of a treatment.
Basis for Discovery of Allosteric Drugs
In conclusion, our results re-emphasize two points, both known from theory but not so often exemplified experimentally. First, whenever a protein is under allosteric control, it must exist in at least two conformations. Any part of the protein that changes its shape between these conformations can potentially host a molecule that binds selectively to one of these conformations, allosterically modifying the protein’s function. This presents opportunities for discovery of allosteric drugs. In the current study, we describe four such sites on the dopamine D1 receptor: the orthosteric site for dopamine and three independent sites for allosteric modulators. Second, whenever two or more ligands bind selectively to the same protein conformation, their free energies are additive, hence their functional effects are multiplicative. This presents opportunities for combinatorial effects on physiology. In the current study, we describe up to 1000-fold amplification of the potency of dopamine by combinations of three allosteric modulators.
Authorship Contributions
Participated in research design: Wang, Goldsmith, Svensson, Willard, Bruns.
Conducted experiments: Wang, Goldsmith.
Contributed new reagents or analytic tools: Goldsmith, Beck.
Performed data analysis: Wang, Hembre, Goldsmith, Willard, Bruns.
Wrote or contributed to the writing of the manuscript: Wang, Hembre, Goldsmith, Svensson, Willard, Bruns.
Footnotes
- Received August 5, 2022.
- Accepted December 1, 2022.
This work was funded by Eli Lilly & Co.
All authors are current or former employees of Eli Lilly & Co. The authors have no conflicts of interest to declare.
↵
This article has supplemental material available at molpharm.aspetjournals.org.
Abbreviations
- CRC
- concentration-response curve
- ECL
- extracellular loop
- GPCR
- G-protein–coupled receptor
- ICL2
- intracellular loop 2
- NAM
- negative allosteric modulator
- PAM
- positive allosteric modulator
- Copyright © 2023 by The Author(s)
This is an open access article distributed under the CC BY Attribution 4.0 International license.