Research ArticleGPCR SIGNALING

Preassembled GPCR signaling complexes mediate distinct cellular responses to ultralow ligand concentrations

See allHide authors and affiliations

Science Signaling  09 Oct 2018:
Vol. 11, Issue 551, eaan1188
DOI: 10.1126/scisignal.aan1188

Ultrasensitivity of GPCRs

Most analyses of signaling through G protein–coupled receptors (GPCRs) are performed using nanomolar or micromolar concentrations of ligand. Civciristov et al. found that femtomolar concentrations of ligand activated signaling by the endogenous β2-adrenergic receptor (β2AR) and muscarinic acetylcholine receptor M3 (M3R) in several cell types. Such ultralow concentrations of ligand stimulated signaling that was qualitatively distinct from that elicited by high concentrations and depended on activation of preassembled GPCR complexes. In contrast, high concentrations of ligand elicited signaling through GPCRs that were not part of complexes in addition to those in preassembled complexes. Such qualitative differences in signaling elicited by different ligand concentrations suggest that low doses of GPCR-targeting drugs could have therapeutic effects and may have implications for the mechanisms of action and side effects of these drugs.

Abstract

G protein–coupled receptors (GPCRs) are the largest class of cell surface signaling proteins, participate in nearly all physiological processes, and are the targets of 30% of marketed drugs. Typically, nanomolar to micromolar concentrations of ligand are used to activate GPCRs in experimental systems. We detected GPCR responses to a wide range of ligand concentrations, from attomolar to millimolar, by measuring GPCR-stimulated production of cyclic adenosine monophosphate (cAMP) with high spatial and temporal resolution. Mathematical modeling showed that femtomolar concentrations of ligand activated, on average, 40% of the cells in a population provided that a cell was activated by one to two binding events. Furthermore, activation of the endogenous β2-adrenergic receptor (β2AR) and muscarinic acetylcholine M3 receptor (M3R) by femtomolar concentrations of ligand in cell lines and human cardiac fibroblasts caused sustained increases in nuclear translocation of extracellular signal–regulated kinase (ERK) and cytosolic protein kinase C (PKC) activity, respectively. These responses were spatially and temporally distinct from those that occurred in response to higher concentrations of ligand and resulted in a distinct cellular proteomic profile. This highly sensitive signaling depended on the GPCRs forming preassembled, higher-order signaling complexes at the plasma membrane. Recognizing that GPCRs respond to ultralow concentrations of neurotransmitters and hormones challenges established paradigms of drug action and provides a previously unappreciated aspect of GPCR activation that is quite distinct from that typically observed with higher ligand concentrations.

INTRODUCTION

G protein–coupled receptors (GPCRs) are the largest class of signaling proteins at the cell surface. These receptors can sense a diverse range of stimuli—from photons and odors to hormones and large peptides—to induce intracellular signal transduction cascades that mediate specific cellular responses. GPCRs are ubiquitously distributed across all cell types, are involved in many diseases, and are the targets of 50% of marketed drugs (1). The intracellular domains of GPCRs interact with heterotrimeric G proteins, and agonist binding to GPCRs stabilizes an active receptor conformation that promotes the dissociation of the heterotrimeric G proteins into Gα and Gβγ subunits. The activated G proteins then interact with other intracellular effectors to induce downstream signaling. One of the downstream targets of activated Gα subunits is adenylyl cyclase (AC), which converts adenosine triphosphate (ATP) into cyclic adenosine monophosphate (cAMP). However, it is increasingly clear that GPCRs do not exist in isolation. Instead, GPCR activity is closely coordinated by the assembly of receptors into higher-order protein complexes [e.g., (28)] that can restrict GPCR signaling to highly organized compartments within the cell, to activate receptor- and location-specific responses (2, 4, 9, 10). The spatial and temporal properties of these intracellular signals are very important for the control of distinct physiological outcomes (2, 4, 917).

Although the assembly of GPCRs into protein complexes enables precise spatiotemporal control over signaling, the physical interactions between the receptor and other proteins in the complex are likely to alter the pharmacological properties of the GPCR itself. We previously reported that the relaxin receptor, RXFP1, preassembles into a large signaling complex that facilitates activation of the receptor by attomolar concentrations of relaxin (8). Whereas responses to such “ultralow” concentrations of biologically active compounds are well documented in cytokine signaling (18), such high ligand sensitivity for GPCRs is not widely reported. Typically, nanomolar-micromolar concentrations of ligand are used to activate GPCRs in experimental systems using global cellular measurements, such as calcium mobilization or cAMP accumulation assays, as readouts for GPCR activity. Nevertheless, there are reports that some GPCRs, including the β2-adrenergic receptor (β2AR), opioid receptors, and angiotensin receptors, can respond to femtomolar (10−15 M) concentrations of ligand in endogenous, physiological systems (1926). Despite these observations, there is little mechanistic insight to explain these nonconventional responses, which are typically measured as changes in cell biology that occur far downstream of the receptor (such as cell adhesion or glucose uptake for the β2AR, analgesia or neuroprotection for opioid receptors, and blood vessel contraction for angiotensin receptors). Without a detailed characterization of the putative extreme sensitivity of these important and ubiquitous receptors, it is unclear whether this sensitivity is a widespread fundamental property of GPCRs and whether ultralow concentrations of ligands have a unique and physiologically relevant role in the cell.

Here, by measuring endogenous GPCR activity with high spatial and temporal resolution, we detected responses from various GPCRs across an extraordinarily wide range of ligand concentrations from attomolar to millimolar. We found that two prototypical GPCRs, the β2AR and the muscarinic acetylcholine receptor M3 (M3R), were activated by femtomolar concentrations of ligand. Mathematical modeling predicted that femtomolar concentrations of ligand can feasibly activate, on average, 40% of cells in a population over a period of 5 min, as observed in our assays, provided that individual cells are capable of responding to one to two binding events. Signaling in response to femtomolar concentrations of ligand depended on the preassembly of a higher-order signaling complex at the plasma membrane. Compared to higher concentrations of ligand, receptor activation by femtomolar concentrations resulted in both a spatially and temporally distinct intracellular signal and a distinct response at the cellular level. The physical interaction between the GPCR and other proteins in the signaling complex appeared to allosterically alter the pharmacological properties of the receptor to enhance the sensitivity to ligand. The ability of many prototypical GPCRs to respond to ultralow concentrations of ligand suggests that a better understanding of this sensitivity is necessary for future research and drug discovery.

RESULTS

Ultralow concentrations of ligand activate endogenous GPCRs

Typically, GPCR ligands within the nanomolar-micromolar concentration range are reported to activate receptors in experimental systems; however, there have been reports of GPCRs responding to femtomolar concentrations of ligand, which are well below conventionally defined pEC50 (negative logarithm of the half maximal effective concentration) values, in endogenous physiological systems [e.g., (1926)]. We have previously shown that RXFP1 induces a biphasic increase in intracellular cAMP that is characterized by a remarkably wide range of pEC50 values [10.9 aM (attomolar, 10−18 M) versus 0.3 nM] (8). This differs from typical biphasic response profiles, wherein each pEC50 value is closely clustered within the nanomolar-micromolar concentration range (27). To determine whether this sensitivity to femtomolar (and lower) concentrations of ligand is a widespread property of GPCRs, we measured cAMP after activation of members of eight different GPCR families, six of which are present endogenously in human embryonic kidney (HEK) 293 cells and, as negative controls, two for which we could not detect any mRNA (Fig. 1, A to C, and fig. S1, A to H) (28, 29). HEK293 cells endogenously express transcripts encoding the A2B adenosine receptor (activated by adenosine), the β1AR and β2AR [activated by isoproterenol (Iso)], α-adrenergic receptor subtypes 2B and 2C, the EP1-4 prostanoid receptors [activated by prostaglandin E1 (PGE1)], the M3R [activated by carbachol (CCh)], the δ-opioid receptor (DOP, activated by SNC80), and the dopamine receptors D2R and D4R (activated by dopamine). We found no transcripts encoding any of the receptors for relaxin (RXFP1 to RXFP4) or the receptors for glucagon-like peptide 1 (GLP-1 and GLP-2). The endogenous receptors canonically couple to Gαs (adenosine A2B, β1AR, β2AR, and the prostanoid, relaxin, and glucagon-like peptide receptors) to stimulate AC activity, Gαi/o2B-AR, α2C-AR, DOP, D2R, and D4R) to inhibit AC activity, or Gαq/11 (M3R) to stimulate Ca2+ mobilization. Subnanomolar concentrations of adenosine, Iso, PGE1 (Fig. 1A), CCh, SNC80, or dopamine (Fig. 1B) increased cAMP. As expected, there was no change in baseline cAMP in response to relaxin or GLP-1, both of which activate receptors that are not produced in HEK293 cells (Fig. 1C). For the endogenous receptors, we observed biphasic concentration-response curves ranging from attomolar to millimolar, in which the two response phases were separated by a very wide concentration range. All ligands caused an increase in cAMP at femtomolar concentrations (table S1); when the ligand reached nanomolar concentrations, ligands that activated Gαs-coupled GPCRs caused a further increase in cAMP (Fig. 1A), whereas ligands that activated Gαi/o- or Gαq/11-coupled GPCRs decreased cAMP back to baseline (Fig. 1B). To determine whether this characteristic biphasic response was cell type specific, and as a further control, we repeated the same experiment in CHO-K1 (Chinese hamster ovary K1 cells). These cells do not endogenously produce adrenergic or muscarinic receptors [National Center for Biotechnology Information (NCBI) Gene Expression Omnibus accession GSE75521; (30)], and accordingly, we observed no change in cAMP from baseline upon activation with Iso or CCh over a wide range of concentrations (fig. S2A). In contrast, we detected changes in cAMP after activation of members of four GPCR families that are produced endogenously [NCBI Gene Expression Omnibus accession GSE75521; (30)] in CHO-K1 cells (Fig. 1, D and E): the adenosine (A2A and A2B), prostanoid (EP1 and EP4), 5-hydroxytryptamine (5-HT; 5-HT1B, 5-HT6, and 5-HT7), and proteinase-activated (PAR1 and PAR2) receptors. Again, all ligands caused a biphasic change in cAMP from baseline: an initial increase in cAMP at femtomolar concentrations, followed by a further increase (adenosine and PGE1; Fig. 1D) or a decrease back to baseline (5-HT and thrombin; Fig. 1E) when the ligand reached nanomolar concentrations. These data suggest that sensitivity to ultralow concentrations of ligand is a potentially fundamental property of many endogenous GPCRs, irrespective of cell type and canonical G protein–coupling profile.

Fig. 1 GPCRs respond to subnanomolar concentrations of ligand.

(A to C) Quantification of cAMP in native HEK293 cells stimulated with increasing concentrations of adenosine, the βAR agonist Iso, or PGE1 (A); the muscarinic acetylcholine receptor agonist CCh, the δ-opioid receptor agonist SNC80, or dopamine (B); and relaxin or GLP-1 (C) in the absence of the phosphodiesterase (PDE) inhibitor 3-isobutyl-1-methylxanthine (IBMX) (n = 6 to 9 independent experiments; see also table S1). (D and E) Quantification of cAMP in native CHO-K1 cells stimulated with increasing concentrations of adenosine or PGE1 (D) and 5-HT or thrombin (E) in the presence of IBMX (n = 6 independent experiments). (F) Quantification of cAMP in native HEK293 cells stimulated with increasing concentrations of adrenaline, noradrenaline, or acetylcholine in the absence of IBMX (n = 6 to 8 independent experiments; see also table S1). (G) Quantification of cAMP in primary human cardiac fibroblasts (CFs) stimulated with increasing concentrations of Iso or CCh in the absence of IBMX (n = 5 to 6 independent experiments). (H) Quantification of cAMP in native HEK293 cells or HEK293 cells transiently expressing scrambled or β2AR small interfering RNA (siRNA), stimulated with increasing concentrations of Iso in the absence of IBMX (n = 6 independent experiments). (I) Expression of β2AR mRNA in native HEK293 cells or HEK293 cells transiently expressing scrambled (scram.) or β2AR siRNA as determined by qRT-PCR (n = 3 independent experiments). (J) Quantification of cAMP in native HEK293 cells or HEK293 cells transiently expressing scrambled or M3R siRNA, stimulated with increasing concentrations of CCh in the absence of IBMX (n = 6 independent experiments). (K) Expression of M3R mRNA in native HEK293 cells or HEK293 cells transiently expressing scrambled or M3R siRNA as determined by qRT-PCR (n = 3 independent experiments). (L) Quantification of cAMP in HEK293 cells transiently expressing the β2AR or M3R and stimulated with increasing concentrations of Iso or CCh, respectively, in the absence of IBMX (n = 3 to 4 independent experiments). All data are expressed as the means ± SEM of n independent experiments. *P < 0.05 and **P < 0.01 versus HEK293 controls, one-way analysis of variance (ANOVA) with Tukey’s multiple comparison test.

To further understand this highly sensitive signaling, we selected two prototypical GPCRs for detailed examination: the β2AR, a classical Gαs-coupled receptor that responds to Iso, and the M3R, a classical Gαq/11-coupled receptor that responds to CCh. To our knowledge, there are no reports of muscarinic receptors responding to femtomolar concentrations of ligand; however, there are previous reports that activation of the β2AR by picomolar concentrations of ligand (well below the EC50 values) leads to increased cell adhesion (19) and glucose uptake (20). Transcripts encoding both the β2AR and M3R are endogenously produced in HEK293 cells (fig. S1, C and E), and we confirmed localization of both proteins to the plasma membrane of HEK293 cells using fluorescent ligand binding (fig. S2, B and C). Subnanomolar concentrations of the endogenous β2AR or M3R ligands adrenaline or noradrenaline and acetylcholine, respectively, elicited similar increases in cAMP in HEK293 cells as did the synthetic ligands Iso and CCh (Fig. 1F). We observed the same biphasic response after addition of the β2AR-selective agonists salbutamol and formoterol (fig. S2D); no selective M3R agonists are available. Further, similar highly sensitive responses to Iso and CCh were observed in primary cultures of human cardiac fibroblasts that endogenously produce β2AR and M3R (Fig. 1G and fig. S2E). This highlights that activation of endogenous GPCRs by ultralow concentrations of ligand is a general feature of at least some endogenous systems. To confirm that responses to ultralow concentrations of ligand were receptor dependent, we knocked down the endogenous β2AR or M3R in HEK293 cells; this abolished cAMP responses to subnanomolar concentrations of Iso or CCh, respectively (Fig. 1, H to K). Knockdown of β2AR had no effect on the cAMP response to CCh, and knockdown of M3R had no effect on the cAMP response to Iso (fig. S2, F and G). This confirms that receptor knockdown did not merely reduce baseline cAMP so that responses to subnanomolar Iso or CCh were undetectable but that cAMP responses to ultralow concentrations of Iso or CCh required β2AR or M3R, respectively. Because responses to subnanomolar concentrations of ligand were undetectable by the cAMP assay after exogenous expression of the β2AR or M3R (Fig. 1L), we suggest that receptor overexpression may mask the responses to subnanomolar concentrations of ligand typically observed in endogenous systems. This could be because overexpressed receptors cause increased constitutive activity and therefore increase the baseline cAMP concentration within the cell (compare vehicle responses in Fig. 1, A and B, to those in Fig. 1L). Alternatively, the overexpressed receptors may alter the composition of the signaling complexes that are required to respond to ultralow concentrations of ligand (31), thus allowing the prototypical signaling response to dominate.

We next wanted to determine whether ultralow and high concentrations of ligand activated qualitatively different signaling pathways or only a quantitative difference in signaling. To address this, we used a sensitive plasma membrane–targeted cAMP Förster resonance energy transfer (FRET) biosensor (32) that allowed us to gain a higher-resolution measure of cAMP produced at the plasma membrane in real time and in single live cells. Activation of the endogenous β2AR in HEK293 cells with 1 fM Iso caused a relatively slow, gradual increase of cAMP at the plasma membrane (1.898 min−1) over 5 min (Fig. 2, A and B). In contrast, a high concentration of Iso (100 nM) caused a more rapid increase in cAMP at the plasma membrane (0.666 min−1, threefold faster than responses to 1 fM Iso), which then declined (Fig. 2, A and B). Preincubation of the cells with 100 nM ICI-118,551, an adrenergic receptor antagonist, blocked the sustained plasma membrane cAMP response to 1 fM Iso (Fig. 2C and fig. S2H), further demonstrating the receptor dependence of this signal. Whereas activation of the endogenous M3R by 1 fM CCh also caused a relatively slow, gradual increase in plasma membrane–associated cAMP over 5 min, there was no response to a high concentration of CCh (1 μM; Fig. 2, D and E). The absence of a cAMP signal in response to a high concentration of CCh and the distinct temporal profiles of cAMP generated by ultralow versus high concentrations of Iso demonstrate that the signaling outcomes of high versus ultralow concentrations are qualitatively different and not merely due to changes in the amount of signaling (33). Preincubation of the cells with 10 nM N-methyl scopolamine (NMS; a muscarinic receptor antagonist) blocked the sustained plasma membrane cAMP response to 1 fM CCh (Fig. 2F and fig. S2H), confirming the receptor dependence of this signal. Inhibiting Gαi/o proteins with NF023 had no effect on the cAMP response to Iso or CCh (fig. S2, I and J), suggesting that differences in signaling at high concentrations are not due to the activation of additional G proteins that inhibit cAMP production. Thus, endogenous β2AR and M3R induce sustained increases in cAMP at the plasma membrane in response to remarkably low concentrations of ligand. Critically, stimulating either the ultralow or high concentration phases resulted in different temporal signaling profiles.

Fig. 2 Femtomolar concentrations of ligand cause sustained increases in plasma membrane–localized cAMP and require an intact orthosteric binding site and only one binding event per cell.

(A) Measurement of cAMP at the plasma membrane in single native HEK293 cells using the FRET biosensor pmEpac2, which reversibly binds cAMP. Cells were stimulated with vehicle, 1 fM Iso, or 100 nM Iso (n = 47 to 79 cells). (B) Representative ratiometric pseudocolor images of cells from (A) at the indicated time points after stimulation. Scale bars, 10 μm. (C) Measurement of cAMP at the plasma membrane in single native HEK293 cells preincubated with the β2AR antagonist ICI-118,551 before stimulation with vehicle or 1 fM Iso (n = 51 to 97 cells). (D) Measurement of cAMP at the plasma membrane in single native HEK293 cells stimulated with vehicle, 1 fM CCh, or 1 μM CCh (n = 29 to 53 cells). (E) Representative ratiometric pseudocolor images of cells from (D) at the indicated time points after stimulation. Scale bars, 10 μm. (F) Measurement of cAMP at the plasma membrane in single native HEK293 cells preincubated with the M3R antagonist NMS before stimulation with vehicle or 1 fM CCh (n = 56 to 95 cells). (G) Measurement of cAMP at the plasma membrane in single HEK293 cells transiently expressing wild-type (WT) FLAG-β2AR or the orthosteric binding site D3.32A mutant FLAG-β2AR and stimulated with vehicle, 1 fM Iso, or 1 pM Iso (n = 43 to 151 cells). (H) Measurement of cAMP at the plasma membrane in single HEK293 cells transiently expressing WT or D3.32A mutant 3HA-M3R and stimulated with vehicle, 1 fM CCh, or 1 pM CCh (n = 119 to 186 cells). (I) Measurement of cAMP at the plasma membrane in single HEK293 cells transiently expressing M3R-DREADD and stimulated with vehicle, 1 fM CCh, or 1 fM CNO (n = 57 to 89 cells). All cells were stimulated at 0 min, and a maximal cAMP response (Max) was induced after 5 min by stimulating the cells with forskolin, IBMX, and PGE1. Individual cells were analyzed from experiments performed on three independent occasions. Data are expressed as the means ± SEM of n cells, normalized to the maximal cAMP response induced after 5 min (F/FMax). (J) Fraction of HEK293 cells within the field of view that increased cAMP at the plasma membrane after a 5-min exposure to 1 fM or 100 nM Iso. Data were analyzed from experiments in Fig. 3 (A and B) with an area under the curve (AUC) greater than 0.697 considered statistically significantly increased compared to vehicle control. Data are expressed as the means ± SEM of six independent experiments. (K) The 95% credible interval for responses to 1 fM Iso over 5 min, using 1000 randomly subsampled parameter sets from the Markov chain Monte Carlo (MCMC) sampling procedure. The red line shows the time course with parameters consistent with the maximum a posteriori probability (MAP) estimate. The solid gray line shows the median, and the dashed gray lines show the 95% credible interval for the subsampled parameter sets. The 1 fM Iso data from (J) is shown as crosses; for two of these, only a small region (~2%) of sampled parameter space allows the model to reach these points. (L) Normalized frequency of binding for 1 fM Iso from 100 independent model simulations with the MAP estimate parameter set. The average number of binding events is 1.13 per cell.

Activation of GPCRs by femtomolar concentrations of ligand requires an intact orthosteric binding site

In addition to the primary orthosteric binding site, many GPCRs have allosteric binding sites within the extracellular vestibule (a surface-exposed area above the binding pocket), which can fine-tune receptor activity (34). All-atom molecular dynamic simulations have demonstrated that β2AR and M3R ligands make initial contact with this extracellular vestibule before achieving the final pose in the orthosteric binding pocket (35, 36). We therefore wondered whether this highly responsive state of the β2AR and M3R was due to ligand binding to an allosteric, high-affinity binding site or, alternatively, to the canonical orthosteric site.

In cAMP assays, the response to femtomolar concentrations of ligand was masked when the receptors were exogenously expressed (Fig. 1L). However, the plasma membrane–localized cAMP FRET biosensor is more sensitive than the cAMP accumulation assay and has high spatial resolution, which allowed us to detect changes in cAMP abundance in single HEK293 cells in response to activation of exogenously expressed receptors by femtomolar concentrations of ligand (fig. S3, A to D). We used this approach to measure cAMP at the plasma membrane of single cells after transient expression of receptors bearing mutations in the orthosteric binding site. Mutation to alanine of a conserved orthosteric binding site residue within transmembrane domain 3 [D3.32A using Ballesteros-Weinstein numbering (37), essential for ligand binding to aminergic receptors (38, 39)] of β2AR (D113A) and M3R (D148A) abolished plasma membrane cAMP in response to 1 fM or 1 pM ligand (Fig. 2, G and H, and fig. S4, A to D). This mutation also inhibited canonical signaling in response to high concentrations of Iso and CCh (fig. S4, A and D). To confirm that the orthosteric site was necessary for responses to ultralow ligand concentrations, we used a well-characterized mutant form of M3R called M3R-DREADD (M3R designer receptor exclusively activated by designer drugs), which is selectively activated by clozapine-N-oxide (CNO) but not by other ligands (fig. S4E) (40, 41). After expression of M3R-DREADD in cells, 1 fM CNO, but not CCh, increased plasma membrane cAMP (Fig. 2I and fig. S4E). Together, this confirms that activation of β2AR, M3R, and M3R-DREADD by subnanomolar concentrations of ligand requires an intact orthosteric binding site.

Mathematical modeling supports GPCR responses to femtomolar concentrations of ligand

Cellular responses to such ultralow concentrations of GPCR ligands are not commonly reported. However, we have clearly shown that these responses can occur in different cell lines, are observed using distinct cell assays, are receptor dependent, and can be eliminated by mutation of the orthosteric binding pocket. To further explore the biophysics of receptor activation at such ultralow ligand concentrations, we developed a mathematical model based on chemical kinetics and used it to determine whether the observed cell activation by ultralow concentrations of ligands can be explained by a simple ligand-receptor interaction.

We considered a model wherein the activation of a cell is proportional to the number of occupied receptors. We also took into account the fraction of cells in the population that are competent to be activated by ligand (71.1%, determined from single-cell FRET experiments using the high concentration of Iso; Fig. 2J). To simulate stochastic ligand-receptor binding kinetics in response to 1 fM Iso, we used Gillespie’s algorithm (42). We used an MCMC algorithm to sample potential parameter sets and used Bayesian statistics to estimate the probability distributions of the following parameters in our model: kr and kact (dissociation and activation rate constants, respectively), Kd (equilibrium dissociation constant), and fc (fraction of cells competent for activation) (see Materials and Methods for model details; fig. S5, A to C). A detailed description of our procedure can be found in (43). MCMC sampling allowed us to calculate credible intervals for the time course of ligand binding in response to 1 fM Iso (Fig. 2K) and the number of binding events per cell (Fig. 2L). From this procedure, we determined the MAP parameter estimates (analogous to best-fit parameter estimates from nonlinear regression). For the MAP parameter estimates, we found that more than 70% of the cell population had less than two binding events and less than 10% had more than two binding events in the allotted time (Fig. 2L). The average number of binding events was slightly more than one per cell. Our model therefore suggests that it is feasible for cells to respond to femtomolar concentrations of ligand and predicts that the cells must be sufficiently sensitive (which means that kact must be sufficiently large) to respond to just one or two binding events per cell. Such highly efficient and amplified signaling is commonly observed in response to cytokines (18). We then input the fastest published on-rate constant (1.2 × 1010 M−1 min−1 for the μ-opioid receptor ligand carfentanil) and slowest published off-rate constant (4.8 × 10−4 min−1 for the M3R ligand tiotropium) for a GPCR ligand (44) to evaluate the capabilities of a “super ligand.” The model revealed that one binding event per cell would occur in response to concentrations of the super ligand as low as 25 aM.

Responses to femtomolar concentrations of ligand depend on a preassembled signaling complex

We hypothesized that the signal amplification required to cause cell activation in response to one to two ligand binding events per cell may be achieved by the formation of highly specialized signaling complexes to allow rapid and more efficient coupling to intracellular pathways. We therefore sought to identify the signaling proteins involved in the cAMP response to femtomolar concentrations of Iso. The plasma membrane cAMP response was abolished after pharmacological inhibition of Gαs with NF449, of Gβγ with the peptide mSIRK, or of AC with 2′,5′-dideoxyadenosine (ddA), suggesting that femtomolar concentrations of Iso lead to activation of AC through Gαs and Gβγ to increase plasma membrane–associated cAMP (Fig. 3A and fig. S6, A and B). Consistent with our hypothesis, complexes formed by the β2AR and large scaffolding proteins such as A kinase anchoring protein 79 (AKAP79), AKAP250, PDEs, and β-arrestins are important for many responses to nanomolar concentrations of ligand (3, 5, 6). We found that the plasma membrane cAMP response to femtomolar concentrations of Iso depended on the scaffolding proteins AKAP250 and β-arrestins (Fig. 3A and fig. S6, C to F).

Fig. 3 A preassembled β2AR signaling complex controls the response to femtomolar concentrations of ligand.

(A) Measurement of cAMP at the plasma membrane in response to 5 min of stimulation with vehicle or 1 fM Iso in single native HEK293 cells that were pretreated with the Gαs antagonist NF449, the Gβγ inhibitor mSIRK, the negative control peptide mSIRK L9A, or the AC inhibitor ddA or transient expression of scrambled, AKAP250, β-arrestin 1, or β-arrestin 2 siRNA (n = 36 to 254 cells). (B) Measurement of cAMP at the plasma membrane in response to 5 min of stimulation with vehicle or 1 fM Iso in single native HEK293 cells was pretreated with the Gαi/o antagonist NF023, the PDE inhibitor IBMX, or the PKA inhibitor KT5720 or transient expression of PDE4D3 dominant negative (dn), PDE4D5 dn, pSilencer control, or AKAP79 short hairpin RNA (shRNA) (n = 22 to 254 cells). (C) Measurement of cAMP at the plasma membrane after 5 min of stimulation with vehicle or 1 fM Iso in HEK293 cells transiently expressing the β2AR. Cells were pretreated with the Gαi/o antagonist NF023, the PDE inhibitor IBMX, or the PKA inhibitor KT5720 or transient coexpression of PDE4D3 dn, PDE4D5 dn, pSilencer control, or AKAP79 shRNA (n = 22 to 153 cells). All cells (A to C) were stimulated at 0 min, and a maximal cAMP response was induced after 5 min by the addition of forskolin, IBMX, and PGE1. Individual cells were analyzed from experiments performed on three independent occasions. Data are expressed as the means ± SEM of n cells and represented as the 5-min AUC. **P < 0.01 and ***P < 0.001 versus vehicle control, two-way ANOVA with Sidak’s multiple comparison test; ^P < 0.05, ^^P < 0.01, and ^^^P < 0.001 versus untreated control, two-way ANOVA with Dunnett’s multiple comparison test. (D) Cartoon showing the regions of the β2AR C-terminal tail (CT) that were tagged with glutathione S-transferase (GST). (E) Quantification of proteins identified as required for activation of cAMP in response to 1 fM Iso in GST pulldowns from lysates of unstimulated native HEK293 cells using the indicated immobilized CT-GST fusions. GST pulldowns were assayed for endogenous Gαs (short and long forms), transgenically expressed AC2-HA, endogenous β-arrestin 1, and endogenous β-arrestin 2 (n = 5 to 6). (F) Quantification of proteins identified as required for regulation of constitutive activity of the preassembled β2AR complex in GST pulldowns from lysates of unstimulated native HEK293 cells using the indicated CT-GST fusions. GST pulldowns were assayed for endogenous Gαi in cells transgenically expressing AKAP79-HA, endogenous PKA, transgenically expressed PDE4D5 dn, and transgenically expressed AKAP79-HA (n = 3 to 4). For GST pulldown assays (E to F), band densities were normalized for equivalent amounts of GST and expressed relative to GST alone. Data are means ± SEM of n independent experiments. *P < 0.05, **P < 0.01, and ***P < 0.001 versus GST alone, two-way ANOVA with Dunnett’s multiple comparison test. (G) Representative immunoblots (IB) showing Gαs, β-arrestin 1, β-arrestin 2, PKA, PDE4D, HA, and Gαi in GST pulldown assays of lysates from cells using GST alone or the indicated CT-GST fusions. (H) Representative immunoblots showing β2AR and HA after HA immunoprecipitation (IP) of lysates from HEK293 cells transiently expressing HA-AKAP250. (I) Representative images of cells coexpressing β2AR-CFP and a YFP-tagged component of the β2AR-associated complex or the positive control pmEpac2, after acceptor photobleaching of a region of the plasma membrane (dotted box). Gray solid boxes indicate areas of the plasma membrane that were photobleached previously. Scale bars, 10 μm. (J) FRET efficiency at the plasma membrane between β2AR-CFP and YFP-tagged components of the protein complex, calculated from acceptor photobleaching FRET experiments from two regions of interest (ROIs) per cell with four cells analyzed per biological replicate (n = 24 ROIs). Data are expressed as the means ± SEM of n ROIs. *P < 0.05 and ***P < 0.001 versus β2AR-CFP/Gαq-YFP FRET efficiency, Kruskal-Wallis with Dunn’s multiple comparison test; ^^P < 0.01 and ^^^P < 0.001 versus β2AR-CFP/Gαq-YFP FRET after conversion to binary values (1 = FRET, 0 = no FRET) and then chi-square test. (K) Cartoon of the preassembled β2AR signaling complex required for responses to femtomolar concentrations of Iso. Stimulation of cells with 1 fM Iso activates a Gαs- and Gβγ-mediated stimulation of AC2 that depends on AKAP250 and β-arrestins 1 and 2. This increase in cAMP causes the sequential activation of PKA and PDE4D5, which cooperates with Gαi/o to oppose the increase in cAMP. This tonic opposition depends on AKAP79. Hierarchy of proteins within the cartoon is based on whether proteins mediate activation or inhibition and reported protein-protein interactions (3, 5, 53, 59, 84, 105).

The plateau in the cAMP response to ultralow ligand concentrations (Fig. 1, A, B, and D to G) indicates that the balance between production and breakdown of the second messenger is tightly controlled. Whereas the proteins that are required for increased cAMP in response to activation of endogenous receptors are readily identified using inhibitors or genetic targeting, complications may arise when using the same approach to reveal proteins important for cAMP breakdown because any observed increase in basal cAMP activity could be due to the inhibitors affecting any of the multiple endogenous receptor systems. However, by performing experiments in parallel in cells transiently expressing the β2AR, we can be more confident that any observed changes in baseline cAMP are due to a specific effect of the inhibitor on β2AR activity. The efficacy of this approach is illustrated by the identification of distinct proteins involved in the regulation of β2AR versus M3R basal activity.

Because the β2AR can also couple to inhibitory Gαi/o proteins, we first assessed the effect of the Gαi/o antagonist, NF023. Inhibition of Gαi/o increased vehicle-stimulated plasma membrane cAMP in native HEK293 cells (Fig. 3B and fig. S6G) and in HEK293 cells transiently expressing the β2AR (Fig. 3C and fig. S6H). The same effect was observed after adenosine diphosphate (ADP) ribosylation of Gαi/o proteins by pertussis toxin (PTx; fig. S6, I and J). This suggests that there is constitutive activity of the endogenous β2AR in these cells, which is normally tonically opposed by the activity of Gαi/o. There was no additional increase in plasma membrane cAMP after stimulation with 1 fM Iso, suggesting that there is an upper limit for the induction of cAMP by the putative preassembled β2AR complex. Because cAMP can only be degraded by PDE activity, we next examined the effect of a PDE inhibitor, IBMX. In both native cells (Fig. 3B and fig. S6K) and in cells transiently expressing the β2AR (Fig. 3C and fig. S6L), IBMX pretreatment increased vehicle-stimulated plasma membrane cAMP, with no additional increase after stimulation with 1 fM Iso. We observed the same increase in constitutive plasma membrane cAMP activity after pharmacological inhibition of protein kinase A (PKA), which is activated by cAMP and often controls feedback inhibition pathways, with KT5720 (Fig. 3, B and C, and fig. S6, K and L). PDE4D contributes a high proportion of PDE activity in HEK293 cells (45), and PKA activates the long isoforms PDE4D3 and PDE4D5 (46). Overexpression of dominant negative (dn) forms of PDE4D3 (PDE4D3 dn) and PDE4D5 (PDE4D5 dn) caused an increase in vehicle-stimulated plasma membrane cAMP in native HEK293 cells (Fig. 3B and fig. S6, M and N). Whereas 1 fM Iso stimulated an additional increase in plasma membrane cAMP in cells expressing PDE4D3 dn, there was no further increase compared to vehicle in cells expressing PDE4D5 dn. This suggested that although PDE4D5 may repress the constitutive activity of the putative preassembled β2AR complex, PDE4D3 merely decreases basal cAMP globally in the cell. When we performed the same experiment in cells transiently expressing the β2AR, only coexpression of PDE4D5 dn, but not of PDE4D3 dn, caused the same increase in vehicle-stimulated plasma membrane cAMP with no further increase in response to 1 fM Iso (Fig. 3C and fig. S6O). Because PKA is tethered in close proximity to the β2AR under resting conditions by the scaffolding protein AKAP79 (3), we assessed the effect of AKAP79 knockdown on cAMP production. Knockdown of AKAP79 (fig. S6P) statistically significantly increased vehicle-stimulated plasma membrane cAMP, and there was no further increase in plasma membrane cAMP after the addition of 1 fM Iso in native HEK293 cells (Fig. 3B and fig. S6Q) or in cells exogenously expressing the β2AR (Fig. 3C and fig. S6R). This suggests that AKAP79 plays an important role in repressing responses to 1 fM Iso.

That the inhibition of proteins that repress cAMP production causes an increase in signaling under nonstimulated conditions (Fig. 3, B and C) suggests both an inherent constitutive activity of the β2AR signaling complex and that it may be preassembled under nonstimulated conditions. To confirm this, and to also identify the region of the receptor that interacts with other proteins in the complex, we performed GST pulldowns using the intracellular regions of the β2AR (Fig. 3D and fig. S7A). Under nonstimulated conditions, proteins required for activation (Gαs, AC, and β-arrestins 1 and 2) and inhibition (Gαi, PKA, PDE4D5, and AKAP79) of the β2AR interacted with C-terminal helix 8 (CT1, residues 330 to 357) (Fig. 3, E to G, and fig. S7, A to D). Although we could not readily detect interactions with some proteins encoded by transcripts that occur at very low abundance in HEK293 cells (AC, AKAP79, and PDE4D; fig. S7E), exogenous expression of the protein of interest enabled detection of interactions with GST-CT1. This also revealed the involvement of AC2 in the production of cAMP downstream of β2AR: Gαs and Gβγ coincidently activate AC2, AC4, and AC7 (47), and β2AR GST-CT1 pulled down exogenously expressed AC2–human influenza hemagglutinin tag (HA) from cell lysates. Further, although we were unable to pull down Gαi from native HEK293 cell lysates, β2AR GST-CT1 pulled down endogenous Gαi from HEK293 cell lysates transiently expressing AC2-HA, PDE4D5 dn, or AKAP79-HA (Fig. 3, F and G, and fig S7B). The propensity of AKAP250 to oligomerize (48) prevented pulldown of endogenous or exogenously expressed AKAP250; however, exogenously expressed HA-AKAP250 coimmunoprecipitated with the endogenous β2AR under nonstimulated conditions (Fig. 3H). To confirm that the β2AR signaling complex was preassembled at the plasma membrane in intact cells, we used acceptor photobleaching FRET to monitor interactions between cyan fluorescent protein (CFP)–tagged β2AR (β2AR-CFP) and some yellow fluorescent protein (YFP)–tagged components of the complex (Gαs, AKAP79, β-arrestins 1 and 2, and PKA) identified in signaling and GST pulldown experiments (Fig. 3I). We measured FRET within two regions of the plasma membrane for each cell analyzed. Despite colocalization of proteins, FRET was not always detected in both regions of the plasma membrane (table S2), suggesting that the β2AR signaling complex is only formed in discrete membrane domains. Because of this nonuniform formation of the β2AR signaling complex, the data are not normally distributed. Analysis of the FRET efficiency revealed statistically significant interactions at the plasma membrane under basal conditions between β2AR-CFP and Gαs-YFP and PKA-YFP versus the negative control Gαq-YFP (Fig. 3J). Conversion of the data to binary values (0 = no FRET, 1 = FRET) revealed statistically significant FRET between β2AR-CFP and all components tested: Gαs-YFP, AKAP79-YFP, YFP–β-arrestin 1, YFP–β-arrestin 2, and PKA-YFP (Fig. 3J and fig. S7F). Therefore, a preassembled β2AR signaling complex responded to 1 fM Iso by stimulating Gαs-Gβγ activation of AC2 to increase cAMP in a manner that depended on AKAP250 and β-arrestins. This cAMP production was tonically opposed by Gαi/o inhibition of AC2, and PKA stimulated PDE4D5 activity in a manner that depended on AKAP79 (Fig. 3K).

The cAMP produced in response to activation of the M3R by 1 fM CCh required a set of proteins distinct from those required for cAMP production downstream of the β2AR. There was no effect of Gαs inhibition on the plasma membrane cAMP response to 1 fM CCh (Fig. 4A and fig. S8A), suggesting that an alternate pathway can activate AC in this context. Activation of the M3R by micromolar concentrations of CCh induces a cAMP response that depends on a signaling complex comprising AKAP79, AC2, PKC, PKA, and Gαq/11 (7). Similarly, we found that the plasma membrane cAMP response to femtomolar concentrations of CCh was abolished after pharmacological inhibition of Gαq/11, Gβγ, PKC, and AC (Fig. 4A and fig. S8, A to C). Thus, for the M3R, ultralow concentrations of ligand lead to Gαq/11-Gβγ activation of PKC, which stimulates AC to increase cAMP. In contrast to the β2AR complex, there was no effect of knockdown of AKAP250; however, knockdown of either β-arrestin 1 or β-arrestin 2 abolished the plasma membrane cAMP response to 1 fM CCh (Fig. 4A and fig. S8, D and E).

Fig. 4 A preassembled M3R signaling complex controls the response to femtomolar concentrations of ligand.

(A) Measurement of cAMP at the plasma membrane in response to 5 min of stimulation with vehicle or 1 fM CCh in single native HEK293 cells that were pretreated with the Gαs antagonist NF449, the Gαq/11 inhibitor UBO-QIC, the Gβγ inhibitor mSIRK, the negative control peptide mSIRK L9A, the PKC inhibitor GF109203X, or the AC inhibitor ddA or transiently transfected with scrambled, AKAP250, β-arrestin 1 siRNA, or β-arrestin 2 siRNA (n = 39 to 316 cells). (B) Measurement of cAMP at the plasma membrane in response to 5 min of stimulation with vehicle or 1 fM CCh in single native HEK293 cells that were pretreated with the Gαi/o antagonist NF023, the PDE inhibitor IBMX, or the PKA inhibitor KT5720 or transiently transfected with PDE4D3 dn, PDE4D5 dn, pSilencer control, or AKAP79 shRNA (n = 31 to 316 cells). (C) Measurement of cAMP at the plasma membrane after 5 min of stimulation with vehicle or 1 fM CCh in HEK293 cells transiently expressing the M3R. Cells were pretreated with the Gαi/o antagonist NF023, the PDE inhibitor IBMX, or the PKA inhibitor KT5720 or transiently cotransfected with PDE4D3 dn, PDE4D5 dn, pSilencer control, or AKAP79 shRNA (n = 65 to 193 cells). All cells (A to C) were stimulated at 0 min, and a maximal cAMP response was induced after 5 min with forskolin, IBMX, and PGE1. Individual cells were analyzed from experiments performed on three independent occasions. Data are expressed as the means ± SEM of n cells, and represented as the 5 min AUC. **P < 0.01 and ***P < 0.001 versus vehicle control, two-way ANOVA with Sidak’s multiple comparison test; ^^P < 0.01 and ^^^P < 0.001 versus untreated control, two-way ANOVA with Dunnett’s multiple comparison test. (D) Cartoon showing the regions of the M3R third intracellular loop (ICL3) that were tagged with GST. (E) Quantification of proteins required for activation of cAMP in response to 1 fM CCh in GST pulldowns from unstimulated native HEK293 cells using the indicated GST-ICL3 fusions. GST pulldowns were assayed for endogenous Gαq/11, endogenous PKC (from cells transgenically expressing with AKAP79-HA), transgenically expressed AC2-HA, endogenous β-arrestin 1, endogenous β-arrestin 2, and transgenically expressed AKAP79-HA (n = 3 and 4). (F) Quantification of GST pulldowns from unstimulated native HEK293 cell lysates of proteins required for regulation of constitutive activity of the preassembled M3R complex: endogenous PKA and transgenically expressed PDE4D3 dn (n = 3 to 4). For GST pulldown assays, band densities were normalized for equivalent amounts of GST and expressed relative to GST alone. Data are means ± SEM of n independent experiments. *P < 0.05, **P < 0.01, and ***P < 0.001 versus GST alone, two-way ANOVA with Dunnett’s multiple comparison test. (G) Representative immunoblots showing Gαq/11, PKC, HA, β-arrestin 1, β-arrestin 2, PKA, and PDE4D in GST pulldown assays of lysates using GST alone or the indicated ICL3-GST fusions. (H) Representative images of cells coexpressing M3R-CFP and a YFP-tagged component of the M3R protein complex or the positive control pmEpac2, after acceptor photobleaching of a region of the plasma membrane (dotted box). Scale bars, 10 μm. (I) FRET efficiency at the plasma membrane between M3R-CFP and YFP-tagged components of the protein complex, calculated from acceptor photobleaching FRET experiments from two ROIs per cell with four cells analyzed per biological replicate (n = 24 ROIs). Data are expressed as the means ± SEM of n ROIs. *P < 0.05 and ***P < 0.001 versus M3R-CFP/Gαs-YFP FRET efficiency, Kruskal-Wallis with Dunn’s multiple comparison test; ^P < 0.05 and ^^^P < 0.001 versus M3R-CFP/Gαs-YFP FRET after conversion to binary values (1 = FRET, 0 = no FRET) and then chi-square test. (J) Cartoon of the preassembled M3R signaling complex required for responses to femtomolar concentrations of CCh. Stimulation of cells with 1 fM CCh activates a Gαq/11-Gβγ-PKC–mediated stimulation of AC2 that depends on AKAP79 and β-arrestins 1 and 2. This increase in cAMP causes the sequential activation of PKA and PDE4D3, which opposes the increase in cAMP. Hierarchy of proteins within the cartoon is based on reported protein-protein interactions (5, 55, 64).

As observed for the β2AR (Fig. 3, B and C), inhibition of Gαi/o increased vehicle-stimulated plasma membrane cAMP in native HEK293 cells; however, 1 fM CCh stimulated a further increase in plasma membrane cAMP compared to the vehicle control (Fig. 4B and fig. S8F). This suggests that Gαi/o does not inhibit the preassembled M3R signaling complex. In HEK293 cells transiently expressing the M3R, there was no effect of the Gαi/o antagonist NF023 on the plasma membrane cAMP produced in response to vehicle or 1 fM CCh (Fig. 4C and fig. S8G). The same effect was observed after ADP ribosylation of Gαi/o proteins by PTx (fig. S8, H and I). In contrast, inhibition of PDEs or PKA increased vehicle-stimulated plasma membrane cAMP in both native HEK293 cells (Fig. 4B and fig. S8J) and after transient expression of the M3R (Fig. 4C and fig. S8K), with no further increase in plasma membrane cAMP after stimulation with 1 fM CCh. This confirmed that the M3R also displays an inherent constitutive activity that is likely due to preassembly of a signaling complex, as identified for the β2AR. Expression of both PDE4D3 dn and PDE4D5 dn in native HEK293 cells caused a statistically significant increase in vehicle-stimulated plasma membrane cAMP, with no further increase in plasma membrane cAMP in response to 1 fM CCh (Fig. 4B and fig. S8L). However, after coexpression of the M3R, only PDE4D3 dn caused an increase in vehicle-treated plasma membrane cAMP with no further increase in response to 1 fM CCh (Fig. 4C and fig. S8M). Therefore, as for responses to high concentrations of CCh (7), PDE4D3 represses cAMP activity of the M3R. AKAP79 was required for negative regulation of the β2AR complex. Although knockdown of AKAP79 increased vehicle-stimulated plasma membrane cAMP in native HEK293 cells (Fig. 4B and fig. S8N), it did not affect plasma membrane cAMP in response to vehicle treatment but did abolish the response to 1 fM CCh in cells transiently coexpressing the M3R (Fig. 4C and fig. S8O). Thus, as for cAMP responses to micromolar concentrations of CCh (7), an increase in cAMP in response to 1 fM CCh depended on AKAP79.

To confirm that these proteins can preassemble with the M3R, we performed GST pulldowns from unstimulated HEK293 cell lysates and showed that proteins required for activation (Gαq/11, PKC, AC, β-arrestins 1 and 2, and AKAP79) and repression (PKA and PDE4D3) of ultrasensitive M3R signaling required residues 305 to 457 of the third intracellular loop (ICL3) of M3R for assembly into a complex with the receptor (Fig. 4, D to G, and fig. S9, A to D). As we observed with the β2AR, we could not detect endogenous interactions with some proteins encoded by transcripts that occurred at very low abundance in HEK293 cells (AC, AKAP79, and PDE4D; fig. S7E), but exogenous expression of the protein of interest enabled detection of interactions between these proteins and the GST-tagged regions of M3R (Fig. 4, D to G, and fig. S9, A to D). Again, as with β2AR, this also revealed the involvement of AC2 in the stimulation of cAMP downstream of M3R. PKC and Gβγ can activate AC2 (47), and M3R GST–ICL3-2 pulled down exogenously expressed AC2-HA from cell lysates. We were unable to pull down PKC from native HEK293 cell lysates; however, endogenous PKC was pulled down by GST–ICL3-2 from cell lysates transiently expressing AC2-HA, AKAP79-HA, or PDE4D3 dn (Fig. 4, D and G). As with the β2AR, to confirm preassembly of the M3R signaling complex at the plasma membrane of intact cells, we used acceptor photobleaching FRET between M3R-CFP and YFP-tagged components (Gαq, AKAP79, β-arrestins 1 and 2, and PKA and PKC) of the signaling complex (Fig. 4H). Formation of the M3R complex did not always occur in regions of protein colocalization (table S2), and the data were nonnormally distributed, suggesting that the M3R signaling complex forms in discrete regions of the plasma membrane. Analysis of the FRET efficiency revealed statistically significant interactions between M3R-CFP and Gαq-YFP, YFP-β-arrestins 1 and 2, and YFP-PKC versus the negative control Gαs-YFP (Fig. 4I). After conversion of the data to binary values (0 = no FRET, 1 = FRET), we observed statistically significant FRET between the M3R-CFP and all components tested: Gαq-YFP, AKAP79-YFP, YFP–β-arrestin 1, YFP–β-arrestin 2, PKA-YFP, and YFP-PKC (fig. S9E). Therefore, a preassembled M3R signaling complex responds to 1 fM CCh by stimulating Gαq/11-Gβγ-PKC–mediated activation of AC2 to increase cAMP in a manner that depends on AKAP79 and β-arrestins, and this cAMP is tonically opposed by PKA stimulated PDE4D3 (Fig. 4J).

Together, these data reveal that although activation of the β2AR and M3R by femtomolar concentrations of ligand produces the same sustained increase in cAMP, the responses require preassembly of signaling complexes comprising a distinct subset of proteins that associate with different regions of the receptors (Figs. 3K and 4J).

GPCRs activate sustained, compartmentalized signals in response to femtomolar concentrations of ligand

Next, we investigated whether signaling in response to femtomolar concentrations of ligand extends to downstream pathways other than cAMP, whether this signaling differs from that induced by high concentrations of ligand, and whether this also occurs in human cardiac fibroblasts. We measured changes in extracellular signal–regulated kinase (ERK) and PKC activity in different subcellular domains using FRET biosensors (that contain phosphorylation target sequences for ERK or PKC, respectively) targeted to different areas of the cell (4951). Activation of the endogenous β2AR in HEK293 cells and human cardiac fibroblasts did not affect the activity of cytosolic ERK but increased nuclear ERK activity in individual cells (Fig. 5, A to D, and fig. S10A). Mimicking the temporal dynamics of the cAMP response (Fig. 5, E and F), 1 fM Iso caused a sustained increase in nuclear ERK, whereas 100 nM Iso resulted in a transient increase (Fig. 5, A and D). There was no effect of 1 fM CCh on ERK activity in HEK293 cells (Fig. 5C) or in the cardiac fibroblasts (fig. S10B). In contrast, 1 fM CCh caused a sustained increase in cytosolic, but not in plasma membrane–localized, PKC activity in both cell types (Fig. 5, G to J, and fig. S10C), whereas a high concentration (1 μM) generated a transient increase in cytosolic PKC activity in both cell types and an increase in plasma membrane PKC activity in the cardiac fibroblasts (Fig. 5, G to J, and fig. S10C). This again mimicked the temporal dynamics of the M3R cAMP response: 1 fM CCh caused a sustained increase in plasma membrane–localized cAMP, whereas 1 μM CCh induced a delayed and transient increase in plasma membrane cAMP that peaked at 15 min in HEK293 cells and at 5 min in the cardiac fibroblasts (Fig. 5, K and L). There was no effect of 1 fM Iso on PKC activity in the two cell types (Fig. 5I and fig. S10D). Therefore, activation of GPCRs by ultralow concentrations of ligand also affects other intracellular signaling pathways in addition to cAMP production. In contrast to responses to high concentrations of ligand, this signaling is sustained and restricted to defined subcellular compartments. This demonstrates that activation of GPCRs by ultralow concentrations of ligand induces signaling that is qualitatively different compared to the canonical responses activated by concentrations in the nanomolar to micromolar range.

Fig. 5 Stimulation of the β2AR and M3R by femtomolar concentrations of ligand activates sustained and compartmentalized kinase signaling.

(A to F) Single native cells were stimulated with vehicle or the indicated concentration of Iso for 20 min. (A) ERK activity detected in the nucleus of HEK293 cells using the FRET biosensor, EKAR, which is reversibly phosphorylated by ERK and targeted to the nucleus (nucEKAR) (n = 118 to 133 cells). Data are normalized to the maximal ERK response (F/FMax). (B) Representative ratiometric pseudocolor images of cells from (A) at the indicated time points after stimulation. Scale bars, 10 μm. (C) ERK activity detected in the cytosol using the cytoEKAR FRET biosensor or nucleus (nucEKAR) of HEK293 cells. Some cells were stimulated with 1 fM CCh instead of Iso for 20 min (n = 13 to 130 cells). Data are represented as the 20-min AUC. (D) ERK activity detected in the nucleus of human CFs (n = 38 to 61 cells). Data are normalized to the baseline ERK response (F/F0). (E) cAMP detected at the plasma membrane in HEK293 cells (n = 31 to 44 cells). Data are normalized to the maximal cAMP response induced after 20 min (F/FMax). (F) cAMP detected at the plasma membrane of human CFs (n = 22 to 53 cells). Data are normalized to the baseline cAMP response (F/F0). (G to L) Single native cells were stimulated with vehicle or the indicated concentration of CCh for 20 min. (G) PKC activity detected in the cytosol of HEK293 cells using the FRET biosensor, CKAR, which is reversibly phosphorylated by PKC (n = 185 to 226 cells). Data are normalized to the maximal PKC response induced after 20 min (F/FMax). (H) Representative ratiometric pseudocolor images of cells from (G) at the indicated time points after stimulation. Scale bars, 10 μm. (I) PKC activity detected at the plasma membrane (pmCKAR) or in the cytosol (cytoCKAR) of HEK293 cells. Some cells were stimulated with 1 fM Iso instead of CCh (n = 10 to 175 cells). Data are represented as the 20-min AUC. (J) PKC activity detected in the cytosol of human CFs (n = 69 to 124 cells). Data are normalized to the baseline PKC response (F/F0). (K) cAMP detected at the plasma membrane of HEK293 cells (n = 32 to 44 cells). Data are normalized to the maximal cAMP response induced after 20 min (F/FMax). (L) cAMP detected at the plasma membrane of human CFs (n = 31 to 50 cells). Data are normalized to the baseline cAMP response (F/F0). All cells were stimulated at 0 min, and a maximal ERK, PKC, or cAMP response was induced after 20 min with PDBu (ERK), PDBu plus phosphatase inhibitors (PKC), or forskolin plus IBMX and PGE1 (cAMP). Individual cells were analyzed from experiments performed on three independent occasions. Data are expressed as the means ± SEM of n cells. ***P < 0.001 versus vehicle control, two-way ANOVA with Sidak’s multiple comparison test.

Activation of GPCRs by femtomolar concentrations of ligand causes a unique cellular response

Both the location and duration of intracellular signals are extremely important for generating appropriate and distinct cellular responses (2, 4, 9). Because GPCR activation by femtomolar concentrations of ligand causes sustained signals in defined cellular compartments, this suggests that each femtomolar GPCR response may orchestrate a distinct cellular signal compared to both higher ligand concentrations and other ligands at femtomolar concentrations. Here, we used proteomic analysis by liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS) as a sensitive and global assessment of the consequences of activation of endogenous GPCRs by femtomolar concentrations of ligand in HEK293 cells. Activation of endogenous GPCRs induced a proteomic pattern that was unique to both the receptor that was stimulated and the ligand concentration (Fig. 6, A and B, and table S3). For the β2AR, the abundances of 56 proteins were uniquely affected by 1 fM Iso compared to vehicle or 100 nM Iso. From these, we identified proteins that were exclusively increased in response to 1 fM Iso but not in response to 100 nM Iso or either concentration of CCh. These included five proteins that have roles in RNA processing and protein synthesis (Fig. 6C): SF3B5 (splicing factor 3B subunit 5), a component of the spliceosome important for pre-mRNA splicing; TXNL4A (thioredoxin-like protein 4A), part of the machinery involved in spliceosome assembly; RPS21 (40S ribosomal protein 21), a component of the 40S ribosomal subunit; GUF1 (translation factor GUF1), which promotes protein synthesis and acts as a fidelity factor during translation; and TXNDC9 (thioredoxin domain-containing protein 9), which negatively affects protein folding by inhibiting the adenosine triphosphatase activity of the chaperonin TCP1 (T-complex protein 1) complex. These results, in addition to the sustained increase in nuclear ERK activity (Fig. 5, A to C), suggested that ultralow concentrations of Iso may affect gene expression. In agreement with the proteomic data, only 1 fM Iso, but not 100 nM Iso or CCh, increased gene transcription over a period of 4 hours (Fig. 6, D and E) as assessed by a green fluorescent protein (GFP) reporter under the control of the constitutive EF1α promoter. We observed a similar increase in gene transcription in response to 1 fM Iso, but not to 100 nM Iso or CCh, in human cardiac fibroblasts (Fig. 6, F and G). In HEK293 cells, we observed no effect of inhibition of Gαi/o by NF023 on the lack of response to 100 nM Iso (fig. S10E). This shows that the absence of a signal in response to 100 nM Iso was not due to activation of inhibitory pathways and that, therefore, the responses to 1 fM and 100 nM Iso are qualitatively different. Together, these data demonstrate a unique role for increased gene transcription in cellular responses to the activation of the β2AR by femtomolar concentrations of ligand that is not triggered by higher concentrations of ligand.

Fig. 6 Activation of the β2AR and M3R by femtomolar concentrations of ligand causes distinct whole-cell responses.

(A) Representative hierarchical clustering of proteins with increased (blue) or decreased (red) abundance in native HEK293 cell populations after stimulation with vehicle, 1 fM or 100 nM Iso, or 1 fM or 10 μM CCh for 4 hours. Data are expressed as Z-scores of the ligand-stimulated log2 change in protein abundance compared to vehicle (see also table S3) (n = 3 independent experiments). (B) Proteins with a statistically significant increase or decrease in abundance in native HEK293 cells after stimulation with Iso or CCh were classified by Gene Ontology (GO) terms and grouped into the indicated categories. A Biological Process GO term was included if it occurred in at least two of the three independent experiments. (C) Log2 change in protein abundance in the indicated native HEK293 treatment groups versus vehicle control for SF3B5, TXNL4A, RPS21, GUF1, and TXNDC9, all of which are involved in RNA processing and protein synthesis (n = 3). (D and E) GFP fluorescence in single native HEK293 cells expressing the pEF1α-GFP reporter after stimulation with vehicle, 1 fM or 100 nM Iso (D; n = 196 to 204 cells), or 1 fM or 10 μM CCh (E; 177 to 194 cells) for 4 hours. Individual cells were analyzed from three independent experiments. Data are expressed relative to baseline fluorescence (F/F0). (F) GFP fluorescence in single human CFs expressing the pEF1α-GFP reporter after stimulation with vehicle, 1 fM Iso, or 100 nM Iso for 4 hours (n = 64 to 107 cells). Individual cells were analyzed from four independent experiments. Data are expressed as relative fluorescence units (RFU) per cell. (G) GFP fluorescence in single human CFs expressing the pEF1α-GFP reporter after stimulation with vehicle, 1 fM CCh, or 10 μM CCh (n = 109 to 121 cells) and activation of Cdc42, as measured by the Raichu-Cdc42 FRET biosensor, which detects GDP/GTP binding, in single human CFs after stimulation with vehicle, 1 fM Iso, or 100 nM Iso (n = 133 to 178 cells) expressed as the 4-hour AUC. (H) Log2 change in protein abundance in the indicated native HEK293 treatment groups versus vehicle control for GGA1, PDE6D, ILK, VPS52, and GPSM1, all of which are involved in protein trafficking and cytoskeletal networks (n = 3). (I and J) Activation of Cdc42 in single native HEK293 cells after stimulation with vehicle, 1 fM or 10 μM CCh (I; n = 305 to 323 cells), or 1 fM or 100 nM Iso (J; n = 304 to 401 cells) for 4 hours. Individual cells were analyzed from three independent experiments. Data are expressed relative to baseline FRET (F/F0). (K) Activation of Cdc42 in single human CFs after stimulation with vehicle, 1 fM CCh, or 10 μM CCh for 4 hours (n = 150 to 159 cells). Individual cells were analyzed from three independent experiments. Data are expressed relative to baseline FRET (F/F0). All data are expressed as the means ± SEM of n cells or independent experiments. **P < 0.01 versus vehicle control, two-way ANOVA with Dunnett’s multiple comparison test (C and H). ***P < 0.001 versus vehicle control, two-way ANOVA (D, F, I, and K).

Similarly for the M3R, 1 fM CCh affected the abundances of 35 proteins in HEK293 cells compared to vehicle or 10 μM CCh. From these, we identified proteins that were exclusively increased in response to 1 fM CCh but unaffected by 10 μM CCh or either concentration of Iso. These included five proteins that affect trafficking, cytoskeletal networks, and small G protein signaling (Fig. 6H): GGA1 (ADP-ribosylation factor-binding protein), which plays a role in protein sorting and trafficking between the trans-Golgi network and endosomes; PDE6D (a cGMP PDE), which regulates the subcellular targeting of Ras small guanosine triphosphate (GTP)–binding proteins; ILK (integrin-linked protein kinase), which is implicated in cell architecture, adhesion, and anchorage-independent growth; VPS52 (vacuolar protein sorting-associated protein 52), which is a component of the retrograde transport and endocytic recycling machinery; and GPSM1 (G protein signaling modulator 1), a guanine nucleotide dissociation inhibitor that uncouples G protein signaling from GPCRs. These results suggested that ultralow concentrations of CCh might be important for the regulation of cellular trafficking, cytoskeletal organization, and signaling by small G proteins. To test this hypothesis, we used Raichu-Cdc42, a FRET biosensor that reports on activation of the Rho GTPase Cdc42 (52); Rho guanosine triphosphatases are important regulators of cytoskeletal organization and trafficking (53). In agreement with the proteomic data, over 4 hours, only 1 fM CCh caused an increase in Cdc42 activity; 10 μM CCh, 1 fM Iso, and 100 nM Iso did not (Fig. 6, I and J). We observed the same increase in Cdc42 activity in response to 1 fM CCh, but not to 10 μM CCh or Iso, in the human cardiac fibroblasts (Fig. 6, G and K). As seen for cAMP, inhibition of Gαi/o by NF023 did not alter the lack of response to 10 μM CCh (fig. S10F) in HEK293 cells. This shows that 10 μM CCh did not affect Cdc42 activity and that the responses to 1 fM and 10 μM CCh are qualitatively different. Therefore, activation of the M3R by femtomolar concentrations of CCh causes an increase in Cdc42 activity, which can affect many basic cellular processes including cell morphology, migration, endocytosis, and cell cycle progression (49). As seen for the β2AR, these data demonstrate that activation of the M3R by ultralow ligand concentrations generates a unique cellular response compared to high ligand concentrations.

DISCUSSION

The current findings uncover a previously unappreciated dimension of GPCR signaling, with several prototypical GPCRs initiating cellular responses to subnanomolar concentrations of ligand that are distinct from responses elicited by higher ligand concentrations. This extremely high sensitivity of GPCRs to ligand was observed in multiple cell types, was receptor dependent, and required an intact orthosteric binding site in the receptor. Mathematical modeling suggested that these responses were triggered in an individual cell by one to two binding events, which would necessitate signal amplification. The preassembled signaling complexes we identified may play an important role in amplifying the response to individual receptor binding events by allowing highly efficient coupling to the signaling machinery. Activation of GPCRs by ultralow concentrations of ligand caused sustained signals within defined subcellular compartments. In contrast, higher concentrations of ligand enabled many more binding events to receptors both within and outside of complexes to generate qualitatively different responses at the whole-cell level (Fig. 7).

Fig. 7 GPCR signaling complexes respond to femtomolar concentrations of ligand.

GPCRs exist in preassembled protein complexes at the plasma membrane. (1) Simulation of stochastic ligand-receptor binding kinetics reveals that the addition of a 1 fM solution of ligand under our assay conditions would result in an average of one to two binding events per cell within 5 min. (2) One to two binding events stimulate strong signal amplification, which depends on a preassembled protein complex at the plasma membrane and results in (3) a relatively slow and gradual increase in the signal over time. (4) Addition of a high concentration solution of ligand (100 nM Iso or 1 μM CCh) results in a much greater number of binding events and activates receptors that are present in preassembled complexes and in any uncomplexed receptors. (5) The resulting activation stimulates a signal that is qualitatively different from that elicited by ultralow ligand concentrations, such as (6) no signal (CCh-stimulated cAMP, EF1α gene transcription, or Cdc42 activity) or (7) a more rapid increase in the signal that then declines (Iso-stimulated cAMP, nuclear ERK, or cytosolic PKC).

Although a sensitivity to femtomolar concentrations of biological compounds is well below the accepted binding affinity of GPCRs, we were able to simulate stochastic ligand binding kinetics to reveal that the addition of femtomolar solutions of ligand under our assay conditions would result in, on average, roughly one binding event per cell over 5 min. First, this suggests that responses to ultralow concentrations of ligand are triggered by only a few GPCR molecules at the cell surface and, second, that activation of one to two receptors results in highly efficient signal amplification. Such signal amplification resulting from activation of only a few receptors at the cell surface is commonly observed for cytokine receptors (18). There are several ways in which such a high degree of signal amplification could occur. Our studies using inhibitors, GST pulldowns, and acceptor photobleaching FRET suggest that a preassembled, functional, higher-order signaling complex is essential for responses to ultralow concentrations of ligand and that the inherent activity of the GPCR is tightly controlled and limited. The close proximity of receptor, G proteins, and effectors to one another would allow a small number of activated receptors to cause a very rapid increase in signaling. Moreover, an assembled signaling complex may alter the local environment of a ligand near a receptor in such a way that the ligand spends more time in close proximity to the receptor, perhaps allowing a ligand to rebind to the receptor multiple times or to bind to the receptor for a longer time, thereby increasing the apparent sensitivity of the receptor to the ligand (54, 55). The mere presence of β2AR at the plasma membrane of cells can more than double the local concentration of ligand (56). In addition, if these signaling complexes cluster owing to oligomerization of AKAPs (48, 57), this would result in a high local concentration of receptors at the plasma membrane, with the clustered receptors effectively acting as a “ligand sink” to again increase the apparent receptor affinity. Last, the protein-protein interactions within the complex may allosterically alter the properties of other associated proteins. This could conceivably result in higher affinity binding by the receptor, by locking the transmembrane helices in an open conformation or reducing the dynamic fluctuations of the ligand binding site, to increase ligand accessibility to the binding pocket or to stabilize the ligand-receptor interaction to generate a signal robust enough to elicit a cellular response. Binding of a positive allosteric nanobody to the intracellular regions of the β2AR can increase the affinity of β2AR for Iso by up to 15,000-fold (58); this demonstrates that intracellular allosteric modulation of a subset of receptors could create two defined receptor populations with widely different ligand sensitivity. Allosteric interactions within the signaling complex may also lower the activation threshold of G proteins and other downstream effectors. Previous studies suggest that the association of PKC with AKAP79 locks the kinase into an active conformation, and PKC becomes insensitive to inhibitors that compete with ATP for binding to the kinase (59, 60). For the M3R, this heightened PKC activity could be very important for facilitating the efficient activation of AC2 by the kinase in response to ultralow concentrations of CCh.

The production, activity, and degradation of cAMP after stimulation of both the β2AR and M3R by femtomolar concentrations of ligand involve many proteins that are also required for responses to high concentrations of ligand (3, 57, 61). Although high-sensitivity responses are associated with many familiar components of GPCR signaling, the dynamics of the interacting proteins within the signaling complex must differ depending on the abundance of ligand to produce unique signaling outcomes. We found that the proteins of the preassembled β2AR complex interacted with the CT1 region of the C-terminal tail of β2AR. This is consistent with previous reports of interactions between the C-terminal tail of the β2AR and proteins such as AKAP79, AKAP250, PKA, G protein receptor kinase 2 (GRK2), and Src (3, 62, 63). All proteins within the preassembled M3R complex interacted with the ICL3 domain of M3R. This is also consistent with previous reports of interactions between the M3R ICL3 and proteins such as Gαq/11, Gβγ, phospholipase Cβ, GRKs, β-arrestins, and casein kinase 2 (6467). Moreover, conformational changes within this loop region are important for the formation of M3R dimers (68). For both the β2AR and M3R, a large number of proteins interact with the receptors through the same intracellular regions. However, crystal structures of the β2AR in complex with Gαs (69) and electron microscopy reconstruction of the β2AR in complex with β-arrestin (70) or a β2AR/V2 vasopressin receptor chimera in complex with both Gαs and β-arrestin (71) suggest that there is little available space for any additional proteins to interact with a monomeric receptor. Nevertheless, these sorts of interactions may be feasible because of the highly flexible structure of AKAPs and the tendency for both AKAP250 and AKAP79 to form higher-order homo- and hetero-oligomeric structures (48, 57). AKAPs may therefore play an important role in supporting the efficient scaffolding of a large number of proteins. Consequently, we may envisage a higher-order assembly of a signaling complex that, by scaffolding a large number of effector proteins, generates a high amount of signal amplification in close proximity to the receptor.

Responses to very subtle environmental cues have been described from bacteria to mammals. Some metalloregulatory proteins have femtomolar sensitivity to control zinc homeostasis in bacteria (72, 73), and it is proposed that Escherichia coli uses subfemtomolar zinc sensing to gain information about the host niche and forms biofilms only in certain environments (74). Similarly, bacteria sense host iron as an environmental cue to express virulence factors (75); free iron is kept at ultralow levels (10 yoctomolar; 10−24 M) in vertebrates (76), so bacterial siderophore proteins bind iron with extremely high affinity (enterobactin binds iron with a Kd of 10−35 M) (77, 78). Here, we show that mammalian cells can generate qualitatively unique responses to ultralow concentrations of GPCR ligands. It is therefore tempting to speculate that the purpose of this high sensitivity is similar: to assess or sample the niche and tailor cellular phenotypes accordingly. Thus, we could anticipate that cells exposed to ultralow concentrations of adrenaline may develop a phenotype distinct from cells that are exposed to ultralow concentrations of acetylcholine. We suggest that this extremely large dynamic range of GPCR signaling is widespread throughout this receptor superfamily and that a low amount of continuous receptor activation may play a critical role in maintaining cell phenotypes in response to subtle environmental cues. The realization that many prototypical GPCRs respond to ultralow concentrations of ligand has important implications for the current understanding of GPCR signaling and the drug design process. For example, current treatment regimens do not consider such low dosages or whether drugs can discriminate between isolated receptors or receptors assembled into protein complexes as mechanisms for response specificity. This may go some way toward explaining the high attrition rates in GPCR drug development. Translation of these concepts could allow “repurposing” of existing drugs using tailored dosage or by targeting the drugs to specific protein complexes.

MATERIALS AND METHODS

Complementary DNAs

AC2-HA (8) and AKAP79-HA (79) were described previously. The pEF1α-AcGFP-C1 vector was from Clontech, and human M3R and 3HA-M3R were from the Bloomsburg University cDNA Resource Center (www.cdna.org). Human β2AR (80) was a gift from R. Summers, and FLAG-β2AR (81) was a gift from R. Lefkowitz. M3R-DREADD (Y149C and A239G) (40) was a gift from B. Roth. PDE4D3 dn D484A (82) and PDE4D5 dn D556A (83) were gifts from M. Houslay. HA-AKAP250 (84) was a gift from C. Malbon, and pSilencer and AKAP79 shRNA (59) were gifts from J. Scott.

pmEpac2 (32) was a gift from D. Cooper. nucEKAR EGFP-mRFP (Addgene plasmid 18682) and cytoEKAR EGFP-mRFP (Addgene plasmid 18680) were gifts from K. Svoboda (49). CytoCKAR (Addgene plasmid 14870) and pmCKAR (MyrPalm-CKAR; Addgene plasmid 14862) were gifts from A. Newton (50, 51). Raichu-Cdc42 (Raichu-Cdc42/Cdc42CT) was a gift from M. Matsuda (52) and was contained within the pCAGGS vector (85), which was a gift from J. Miyazaki.

s-YFP (Addgene plasmid 55781) and Gαq-YFP (Addgene plasmid 55782) were gifts from C. Berlot (86). AKAP79-YFP (AKAP79 in a pEYFP-N1 vector) was a gift from M. Dell’Acqua (87). YFP-β-arrestin 1 and YFP-β-arrestin 2 were gifts from M. Caron (88). PKA catalytic subunit-YFP (PKA-YFP) was a gift from M. Zaccolo (89). YFP-PKC-βII-YFP (YFP-PKC; Addgene plasmid 14866) was a gift from A. Newton (50).

FLAG-β2AR D3.32A (D113A) and 3HA-M3R D3.32A (D148A) were generated using the QuikChange II Kit (Agilent Technologies). The D3.32A annotation uses the Ballesteros-Weinstein numbering system (37). FLAG-β2AR-CFP and 3HA-M3R-CFP were generated by subcloning FLAG-β2AR and 3HA-M3R into pECFP-N1. GST-tagged fragments of the β2AR and M3R intracellular regions were generated by amplifying the required region from the full-length complementary DNA (cDNA) using polymerase chain reaction (PCR) and cloning into pGEX-4T1. Shorter regions (β2AR-ICL1, β2AR-ICL2, M3R-ICL1, and M3R-CT1) were generated by annealing complementary primers and cloning into pGEX-4T1. The following GST-tagged fragments of the β2AR were generated: ICL1 (residues 59 to 71), ICL2 (residues 134 to 150), ICL3 (residues 221 to 274), CT (residues 330 to 413), CT1 (residues 330 to 357), CT2 (residues 358 to 386), and CT3 (residues 387 to 413). The following GST-tagged fragments of the M3R were generated: ICL1 (residues 91 to 104), ICL2 (residues 165 to 185), ICL3 (residues 256 to 489), ICL3-1 (residues 256 to 304), ICL3-2 (residues 305 to 457), ICL3-3 (residues 458 to 489), CT (residues 546 to 590), CT1 (residues 546 to 560), and CT2 (residues 561 to 590).

Drugs

The vehicle for Iso, adrenaline, noradrenaline, dopamine, and serotonin was 0.1% (w/v) ascorbic acid, present in experiments at a final concentration of 0.0001% (w/v). The vehicle for CCh, adenosine, salbutamol, acetylcholine, and thrombin was ultrapure (Milli-Q) water and that for CNO, PGE1, SNC80, and formoterol was dimethyl sulfoxide, both present in experiments at a final concentration of 0.01% (v/v). The vehicle for relaxin was 0.1% (v/v) trifluoroacetic acid, present in experiments at a final concentration of 0.0001% (v/v), and that for GLP-1 was 0.025% (v/v) acetic acid, present in experiments at a final concentration of 0.000025% (v/v).

Cell culture

HEK293 and CHO-K1 cells (American Type Culture Collection; negative for mycoplasma contamination) were used as well-characterized generic cell lines with endogenous expression of GPCRs. The cells were grown in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 5% (v/v) fetal bovine serum (FBS). For HEK293 cells, all assay dishes and plates were precoated with poly-d-lysine (5 μg/cm2). Primary cultures of human cardiac fibroblasts (ScienCell Research Laboratories) were grown in poly-l-lysine–coated culture flasks (2 μg/cm2) in DMEM supplemented with 5% (v/v) FBS, fibroblast growth supplement 2 (ScienCell Research Laboratories), penicillin (100 U/ml), and streptomycin (100 μg/ml).

HEK293 cells were transfected using linear polyethyleneimine (PEI) (90). For experiments using single transfection of siRNA (AlphaScreen cAMP assay), cells were transfected with 25 nM scrambled, β2AR, or M3R SMARTpool ON-TARGETplus siRNA (GE Healthcare Dharmacon) using Lipofectamine 2000 (Invitrogen). Human cardiac fibroblasts were transfected using X-tremeGENE 9 (Roche) at a 1:3 DNA/transfection reagent ratio.

RNA sequencing

RNA was extracted from two passages of HEK293 cells (P0 and P37) using the RNeasy Mini Kit (Qiagen), and transcriptome sequencing was performed by the Beijing Genomics Institute.

cAMP quantification assay

cAMP from cell populations was measured in duplicate using the AlphaScreen cAMP accumulation assay (PerkinElmer) as described previously (91) with the following modifications to ensure the maximum dynamic range and sensitivity. Cells were seeded into 96-well plates and grown to confluency. On the day of the experiment, cells were preincubated with stimulation buffer [Hanks’ balanced salt solution (HBSS) with 5 mM HEPES, 5.6 mM glucose, 1.3 mM CaCl2, and 0.1% (w/v) bovine serum albumin (BSA) (pH 7.4)] for 45 min at 37°C, before addition of ligands, vehicle, or positive control (50 μM forskolin and 100 μM IBMX) diluted in stimulation buffer for 30 min at 37°C. For HEK293 cells and human cardiac fibroblasts, the experiment was performed in the absence of PDE inhibition; for CHO-K1 cells, the experiment was performed in the presence of 500 μM IBMX. To terminate the reaction, the buffer was aspirated, and 50 μl of ice-cold ethanol was added per well. After ethanol evaporation at 37°C, the cell precipitate was resuspended in 30 μl of detection buffer [5 mM HEPES, 0.3% Tween 20, and 0.1% (w/v) BSA (pH 7.4); 130 μl for positive control samples], and then, 10 μl was transferred to a 384-well white OptiPlate (PerkinElmer) on ice. After addition of anti-cAMP acceptor beads (in the presence of 500 μM IBMX) and donor beads with biotinylated cAMP for 1 hour, the plate was read using an EnVision Multilabel Plate Reader (PerkinElmer), and data were analyzed against a standard curve using GraphPad Prism from n biological repeats as stated.

Quantitative real-time polymerase chain reaction

RNA was extracted from HEK293 cells and primary human cardiac fibroblasts using the RNeasy Mini Kit (Qiagen). Quantitative real-time polymerase chain reaction (qRT-PCR) was performed in triplicate from 100 ng of RNA using the iScript One-Step RT-PCR Kit (Bio-Rad) and CFX96 Touch Real Time PCR Detection System (Bio-Rad) according to the manufacturer’s instructions. TaqMan probes (Applied Biosystems) used in this study were as follows: ADRB2, Hs00240532_s1; CHRM3, Hs00265216_s1; and ACTB, Hs99999903_m1. The 2−ΔCt method (92) was used to analyze results, and data are expressed as 2−ΔCt (difference in Ct value of the gene of interest relative to the housekeeping gene, ACTB) from n biological repeats as stated.

Fluorescent ligand binding

HEK293 cells were seeded into black, optically clear 96-well plates and grown to 80% confluency. Cells were washed in phosphate-buffered saline (PBS) and then incubated with a nuclear stain (Hoescht 33342, Pierce) and a saturating concentration of antagonist (1 μM ICI-118,551 for β2AR binding or 100 μM N-methyl scopolamine for M3R binding) or vehicle control for 1 hour at room temperature. The fluorescent ligands (1 μM BODIPY-propranolol for β2AR binding or 100 nM BODIPY-pirenzepine for M3R binding, both from CellAura Technologies) were added for 10 min at room temperature. Buffer was removed from the cells and replaced with PBS before fluorescence imaging using a high-content PerkinElmer Operetta with an Olympus LUCPlanFLN 20× [numerical aperture (NA), 0.45] objective. Nuclei were visualized using the Hoescht 33342 filter set (excitation, 360 to 400 nm; emission, 410 to 480 nm), and BODIPY fluorescence was visualized using the Cy5 filter set (excitation, 620 to 640 nm; emission, 640 to 680 nm). Four fields of view were captured per well, and data were automatically analyzed by determining the mean BODIPY fluorescence per well using Harmony High Content Imaging and Analysis software (version 3.5.2). BODIPY fluorescence was expressed relative to the vehicle-treated control in triplicate from n biological repeats, as stated.

High-content ratiometric FRET imaging

Ratiometric FRET imaging was performed as described previously (9, 90, 93). We detected changes in cAMP levels using Epac2-camps (94) targeted to the plasma membrane (32), which undergoes a conformational change after cAMP binding to the cAMP-binding domain of Epac2. Changes in ERK or PKC activity were detected using EKAR or CKAR, respectively, which undergo conformational change after ERK or PKC phosphorylation of a target sequence. We used EKAR targeted to the cytosol or nucleus (49) and CKAR targeted to the plasma membrane or cytosol (50, 51). Changes in Cdc42 activity were detected using Raichu-Cdc42, which undergoes a conformational change after GTP displaces guanosine diphosphate (GDP) within residues 2 to 176 of Cdc42 (52).

HEK293 cells were seeded in black, optically clear 96-well plates and grown to 70% confluency before transfection with PEI. Human cardiac fibroblasts were transfected using X-tremeGENE 9 in suspension and seeded in half-area black, optically clear 96-well plates at 90% confluency. To measure activation of endogenously expressed receptors, HEK293 cells were transfected with a FRET biosensor (90 ng per well), and human cardiac fibroblasts were transfected with a FRET biosensor (100 ng per well). For overexpression of mutant receptors, HEK293 cells were cotransfected with a receptor (55 ng per well) and a FRET biosensor (40 ng per well). For experiments with siRNA, HEK293 cells were cotransfected with an additional 25 nM scrambled, β-arrestin 1, β-arrestin 2, or AKAP250 SMARTpool ON-TARGETplus siRNA (GE Dharmacon) for 72 hours. For experiments involving dominant negative constructs or shRNA, HEK293 cells were cotransfected with an additional plasmid (50 ng per well) for 72 hours. Before the experiment, HEK293 cells were partially serum restricted overnight in 0.5% FBS (v/v) DMEM.

Cells were pretreated with inhibitors for 30 min at 37°C in HBSS, and inhibitors were used at the following concentrations: 100 μM ddA, 1 μM GF109203X, 100 μM IBMX, 1 μM KT5720, 5 μM mSIRK or mSIRK L9A, 10 μM NF023, 10 μM NF449, and 100 nM UBO-QIC. Antagonists were preincubated with the cells for 10 min, and were used at 100× the Ki (100 nM ICI-118,551 and 10 nM N-methyl scopolamine). Cells were pretreated with PTx (100 ng/ml) at 37°C and 5% CO2 in 0.5% FBS (v/v) DMEM for 16 hours.

Fluorescence imaging was performed using a high-content GE Healthcare INCell 2000 Analyzer with a Nikon Plan Fluor ELWD 40× (NA, 0.6) objective and FRET module as previously described (90). For CFP/YFP (pmEpac2, cytoCKAR, pmCKAR, and Raichu-Cdc42) emission ratio analysis, cells were sequentially excited using a CFP filter (430 nm/24 nm) with emission measured using YFP (535 nm/30 nm) and CFP (470 nm/24 nm) filters and a polychroic filter optimized for the CFP/YFP filter pair (Quad3). For GFP/RFP (cytoEKAR and nucEKAR) emission ratio analysis, cells were sequentially excited using a fluorescein isothiocyanate (FITC) filter (490 nm/20 nm) with emission measured using dsRed (605 nm/52 nm) and FITC (525 nm/36 nm) filters and a polychroic filter optimized for the FITC/dsRed filter pair (Quad4). Cells were either imaged every 20 s for 5 min (image capture of 5 wells per 20 s) or every 1 min for 20 min (image capture of 14 wells per min). At the end of each experiment, the same cells were stimulated with the following positive controls to maximally activate the biosensor: 10 μM forskolin, 100 μM IBMX with 100 nM PGE1 for Epac2, 200 nM phorbol 12,13-dibutyrate (PDBu) for EKAR, or 200 nM PDBu with phosphatase inhibitor cocktail (Sigma-Aldrich) for CKAR. Only HEK293 cells with >5% change in F/F0 (FRET ratio relative to baseline for each cell) after stimulation with positive controls were selected for analysis, and the data were expressed relative to the positive control (F/FMax). For human cardiac fibroblasts, only cells with >3% change in F/F0 after stimulation with positive controls were selected for analysis, and data were expressed as the F/F0 due to the variation in responses to the positive controls. Data were analyzed using in-house scripts written for the FIJI distribution of ImageJ (95), as previously described (90).

Ratiometric pseudocolor images were generated as previously described (96). A multiplication factor of 10 was applied using the Ratio Plus plugin, the Green Fire Blue LUT was applied, and the brightness and contrast range was set to the minimum and maximum FRET ratios within the image stack. The rate of cAMP increase over 5 min (Fig. 2A) was determined by fitting the plateau of the response using an exponential equation (plateau followed by one phase association) in GraphPad Prism.

Enzyme-linked immunosorbent assay

HEK293 cells in 10-cm dishes were transfected with 3 μg of pcDNA3.1, FLAG-β2AR, FLAG-β2AR D3.32A, 3HA-M3R, or 3HA-M3R D3.32A and then seeded into 48-well plates 24 hours after transfection. Forty-eight hours after transfection, cells were washed with tris-buffered saline [TBS; 50 mM tris (pH 7.5) and 150 mM NaCl] and fixed (4% paraformaldehyde in TBS for 30 min). Cells were washed in TBS, blocked [1% (w/v) skim milk and 0.1 M NaHCO3; 4 hours at room temperature with shaking] and then incubated with primary antibodies overnight at 4°C [mouse anti-HA or anti-FLAG, both 1:2000 in 0.1% (w/v) BSA in TBS]. Cells were washed three times with TBS and then incubated with secondary goat anti-mouse–horseradish peroxidase (HRP) antibody solution [1:2000, 0.1% (w/v) BSA in TBS for 2 hours at room temperature]. SIGMAFAST OPD substrate solution (Sigma-Aldrich) was added, and the reaction was terminated with 3 M HCl. The samples were transferred to a 96-well plate, and optical density at 492 nm was measured using an EnVision Multilabel Reader (PerkinElmer). Data are expressed as the fold change in receptor expression compared to pcDNA3.1-transfected cells from n biological repeats as stated.

Mathematical modeling

The kinetics of ligand-receptor binding for a population of cells is defined byL+RikrkfBi(1)where i is an index denoting a particular cell, L represents free ligand, R represents the unbound receptor, B represents the occupied receptor, and kf and kr are association and dissociation rate constants, respectively. Activation of a cell is taken to be proportional to the number of occupied receptorsBi+CikactBi+Ci*(2)where kact is the activation rate constant, C represents an inactive cell, and C*represents an active cell. Note that Ci has a value of 1 until activation and 0 thereafter. In addition to the kinetic parameters, we introduce fc, the fraction of cells competent to be activated by ligand. This parameter is introduced to account for any intracellular conditions (e.g., gene expression, cell cycle state, etc.) that may prevent a cell from responding to ligand.

For 1 fM Iso, we simulated the stochastic ligand-receptor binding kinetics using Gillespie’s algorithm (42). This approach is not computationally feasible when considering the high ligand concentration (100 nM), because the number of reaction events per unit time scales linearly with the number of molecules in the system (120,440 per well for 1 fM versus 1.2 × 1013 per well for 100 nM). As we use molecule copy numbers in these simulations, the concentrations of biochemical species and the association rate constant, kf, must be converted to the appropriate units#M=[M]NAV(3)kf,#=kfNAV(4)where V is the extracellular volume (200 μl), M is a biochemical species, and NA is the Avogadro constant. To estimate the concentration of occupied receptors ([B]) at high ligand concentration, we make a quasi–steady-state approximation for the ligand-receptor interaction because the total ligand concentration, [LT], is much greater than the total receptor concentration, [RT][B]=[RT]kf[LT]kr+kf[LT](5)

We can also calculate the average concentration of occupied receptors per cell[Bi]=[B]Ncells(6)

The fraction of cells, FA, that are active after a time, t, isFA=1eλt(7)with λ = kact ⋅ 〈[Bi]〉 as the average rate of activation for each cell. For kact > 10−4s−1, all cells are activated in less than 1 min when [B] ≈ [RT].

We used a Bayesian approach to estimate the following parameters in our model: kr and kact, which are rate constants in the model defined above with units of s−1; KD, which is the equilibrium dissociation constant in molar units (M) for ligand-receptor binding and can be used to calculate kf, given kr; and fc, which is the fraction of cells competent for activation (dimensionless).

Our procedure uses an MCMC algorithm to estimate the probability distribution of the parameters’ values similar to the procedure outlined in (43). In Bayesian statistics, this estimated distribution is called a parameter’s posterior. For each parameter set sampled during the MCMC run, estimating the posterior requires calculating both the probability of observing the experimental data given a particular set of parameters (the likelihood) and the probability of the parameters given an assumed probability distribution (the parameter’s prior distribution).

Two parameters’ means and SDs have already been characterized in the literature, log10 KD (58) and kr (97). We assign log10 KD to have a normal distribution as its prior, with mean, μ, and SD, σP(log10KD)= Normal(μ=9.768,σ=0.612)(8)

Assuming normality for kr results in statistically significant probability density for values below zero. We therefore assign kr to have a gamma distribution as its prior, with the gamma distribution’s parameters α and β calculated such that the distribution’s mean, α/β, and SD, α/β2, correspond to the mean and SD reported in the literature, 0.05 and 0.0255, respectivelyαβ=0.05(9)αβ2=0.0255(10)P(kr)= Gamma(α=3.845,β=76.894)(11)

The prior for the fraction of competent cells, fc, can be specified based on our data as follows. We assume that 100 nM Iso is a saturating concentration that should activate all competent cells, and so we calculate the mean and SD of the cells that are activated in response to 100 nM Iso (Fig. 2J) and assign fc to have the normal distributionP(fc)= Normal(μ=0.711,σ=0.092)(12)with μ and σ calculated from the data in Fig. 2J. The rate of receptor-dependent cell activation relies on incomplete knowledge of the relevant signaling pathways. However, we can still constrain this parameter with a uniformly distributed prior over a finite range. We assume that the activation rate must be sufficiently fast to activate cells given potential values of kr and that excessively fast activation rates are not physically realizable. Thus, we setP(log10kact)= Uniform(4,2)(13)

Other fixed parameters used in the model are volume of medium (200 μl per well), number of cells (30,000 per well), and number of receptors [18,000 per cell; (98, 99)]. Our MCMC sampling was performed for 1,000,000 iterations with a constant jump size of 0.2 (in log space), and we discarded the first 10,000 points as the burn-in period. Parameter updates were accepted using the Metropolis-Hastings criterion, with about 37% of the attempted updates being rejected. The sampling trace for log10 KD appears to have reached stationarity (fig. S5A). From this, we can characterize the posterior distribution of each parameter; the posteriors for three of the four free parameters strongly reflect their priors (fig. S5B). The exception, kact, reveals a posterior that is shifted toward larger values, with near uniformity for parameters larger than 0.01. We can further characterize the correlations between the free parameters by looking at their pairwise scatterplots (fig. S5C). All pairwise relationships result in a Spearman’s rank correlation coefficient, ρ, of less than 0.05, meaning that dependency between any pair of parameters is unlikely.

Immunoblotting

Proteins were resolved by SDS–polyacrylamide gel electrophoresis using 10% tris-glycine or precast 4 to 15% Mini-PROTEAN TGX gels (for AKAP250 co-IP only; Bio-Rad) and transferred to 0.45-μm LF PVDF (low fluorescence polyvinylidene difluoride) membranes (Bio-Rad) using a Trans-Blot SD Semi-Dry Transfer Cell (for 75 min at 10 V; Bio-Rad). Membranes were blocked for 1 hour at room temperature [5% (w/v) BSA for GST pulldowns or 5% (w/v) skim milk powder in PBS with 0.1% (v/v) Tween 20 (PBS-T) for confirmation of protein knockdown and overexpression or co-IP] and incubated with primary antibody overnight at 4°C [diluted in 1% (w/v) BSA for GST pulldowns or 1% (w/v) skim milk powder in PBS-T for confirmation of protein knockdown and overexpression or co-IP]. Membranes were washed, incubated with secondary antibody [diluted in PBS-T for fluorescent secondary antibodies for GST pulldowns or 1% (w/v) skim milk powder in PBS-T for HRP-conjugated secondary antibodies for confirmation of protein knockdown and overexpression or co-IP] for 1 hour at room temperature, and washed. Immunoreactivity was detected by fluorescence for GST pulldowns (fluorescently conjugated secondary antibodies) or ECL (enhanced chemiluminescence) for confirmation of protein knockdown and overexpression or co-IP (Millipore, HRP-conjugated secondary antibodies). Fluorescence was detected using the Odyssey Classic Infrared Imager (LI-COR Biosciences), with resolution set at 169 μm and the intensity adjusted to be in the linear range for infrared fluorescence detection. ECL was detected using the ChemiDoc Touch Imaging System (Bio-Rad), with exposures adjusted to be in the linear range for chemiluminescence.

Antibodies for immunoblotting

Immunoblotting was performed using primary antibodies recognizing AKAP79 (Millipore, ABS102; rabbit, 1:1000), β-actin (Abcam, ab36956; rabbit, 1:1000), β-arrestin 1/2 (Cell Signaling Technology, 46745; rabbit, 1:1000), β-arrestin 1 (Abcam, ab31868; rabbit, 1:1000), β-arrestin 2 (Millipore, AB6022; rabbit, 1:1000), β-tubulin (Santa Cruz Biotechnology, sc-9104; rabbit, 1:5000), Gαi3 (Santa Cruz Biotechnology, sc-262; rabbit, 1:1000), Gαq/11 (Santa Cruz Biotechnology, sc-392; rabbit, 1:1000), Gαs (Millipore, 06-237; rabbit, 1:1000), gravin (AKAP250; Sigma-Aldrich, G3795; mouse, 1:1000), GST (Sigma-Aldrich, G1660; mouse, 1:25,000), HA (Abcam, ab9110; rabbit, 1:5000), PDE4D [Abcam, ab14613 (for GST pulldowns) or Santa Cruz Biotechnology, sc-25814 (for confirmation of protein overexpression); rabbit, 1:1000], PKA (Santa Cruz Biotechnology, sc-903; rabbit, 1:1000), or PKC (Millipore, 05-83; mouse, 1:1000). Immunoblotting was detected using the following fluorescent or HRP-conjugated secondary antibodies: goat anti-mouse 680 (LI-COR, 926-68070; 1:10,000), goat anti–mouse-HRP (Abcam, ab97023; 1:2000), goat anti-rabbit 800 (LI-COR, 926-32211; 1:10,000), and goat anti-rabbit-HRP (Cell Signaling Technology, 70745; 1:2000 to detect β-arrestin 2, 1:5000 to detect AKAP79, β-arrestin 1/2, and PDE4D).

Confirmation of protein knockdown and dominant negative overexpression

HEK293 cells were seeded into six-well plates and grown to 70% confluency. Cells were transfected with 25 nM scrambled or targeted siRNA or 1.5 μg of pcDNA/pSilencer, targeted shRNA, or dominant negative cDNA for 72 hours using PEI. After transfection, cells were lysed in 100 μl of modified radioimmunoprecipitation assay lysis buffer [50 mM tris (pH 7.4), 375 mM NaCl, 1 mM EDTA, 1% (v/v) Triton X-100, 0.5% (w/v) sodium deoxycholate, and 0.1% (w/v) SDS] for 30 min on ice. Lysates were centrifuged (at 10,000g for 15 min at 4°C), and protein concentration in the supernatant was determined using the Bradford Ultra reagent (Expedeon). Laemmli sample buffer was added to the supernatants, and samples were incubated at 37°C for 30 min before immunoblotting.

GST pulldowns

GST-tagged fragments were expressed in BL21(DE3)pLys cells at 37°C after induction with 0.1 mM isopropyl-β-D-thiogalactopyranoside. Cells were lysed by sonication (three pulses for 30 s at 70% amplitude; Qsonica Q125) in lysis buffer [50 mM tris (pH 8), 300 mM NaCl, 10 mM MgCl2, 1 mM EDTA, 1 mM dithiothreitol (DTT), lysozyme (0.25 mg/ml), protease inhibitor cocktail, and 100 U deoxyribonuclease I (DNase I)]. The homogenates were centrifuged (at 15,000g for 20 min at 4°C), and the supernatants were incubated with glutathione Sepharose 4B resin (GE Healthcare; for 1 hour at 4°C). The resin was washed [50 mM tris (pH 8), 150 mM NaCl, 1 mM EDTA, and 1 mM DTT] until no protein remained in the eluate, and then, an equal volume of PBS [with protease inhibitors and 0.02% (w/v) NaN3] was added to the resin.

HEK293 cells were seeded into 175-cm2 flasks and grown to confluency. For overexpression pulldowns, cells were transfected with 20 μg of AKAP79-HA, PDE4D3 dn, PDE4D5 dn, AC2-HA, or HA-AKAP250 using PEI. Cells were lysed in lysis buffer [50 mM tris (pH 7.4), 100 mM NaCl, 10% (v/v) glycerol, 0.3% (v/v) NP-40, 2 mM DTT, 1 mM PMSF, 1 mM benzamidine, 10 mM β-glycerophosphate, 2 mM Na3VO4, protease inhibitor cocktail, and 100 U DNase I] by rotating for 30 min at 4°C and then passing 10 times through a 21-gauge needle. The cell homogenates were centrifuged (at 500g for 3 min at 4°C) and then incubated with the GST-β2AR or GST-M3R fragment resin for 4 hours at 4°C with rotation. The GST-β2AR or GST-M3R fragment resin was washed twice in lysis buffer [with 0.03% (v/v) NP-40], before the bound proteins were eluted in Laemmli buffer and incubated at 37°C for 30 min before immunoblotting. Immunoreactive bands were quantified by densitometry using Image Studio Lite 4.0 software (LI-COR Biosciences). Data for each fragment are normalized for equivalent amounts of GST and expressed relative to GST-alone control from n biological repeats as stated.

Co-immunoprecipitation

HEK293 cells in 175-cm2 flasks were transfected with 20 μg of HA-AKAP250 using PEI. Forty-eight hours after transfection, cells from eight flasks were lysed in modified lysis buffer [50 mM tris (pH 7.4), 100 mM NaCl, 10% (v/v) glycerol, 1% (v/v) Triton X-100, 0.5% (v/v) NP-40, 2 mM DTT, 1 mM PMSF, 1 mM benzamidine, 10 mM β-glycerophosphate, 2 mM Na3VO4, protease inhibitor cocktail, and 100 U DNase I] by rotating at 4°C for 20 min. The cell homogenates were centrifuged (at 16,000g for 10 min at 4°C) and then halved and incubated with anti-HA affinity matrix (Roche Life Science) overnight at 4°C with rotation. The anti-HA affinity matrix was washed three times in lysis buffer, before the bound proteins were eluted in Laemmli buffer and incubated at 37°C for 30 min before immunoblotting.

Acceptor photobleaching FRET

HEK293 cells were seeded in six-well plates and grown to 70% confluency before cotransfection with FLAG-β2AR-CFP or 3HA-M3R-CFP (0.6 μg per well) and one of the following YFP-tagged proteins (0.6 μg per well): Gαs-YFP, Gαq-YFP, AKAP79-YFP, YFP-β-arrestin 1, YFP-β-arrestin 2, YFP-PKC, or PKA-YFP. The FRET biosensor, pmEpac2, was used as a positive control. Four hours after transfection, cells were reseeded (40,000 cells per well) into an eight-well μ-slide (iBidi). Twenty-four hours after transfection, cells were rinsed in PBS, fixed [with 4% (w/v) paraformaldehyde for 30 min at room temperature], rinsed three times in PBS, and then stored at 4°C.

Acceptor photobleaching FRET was performed using a Leica SP8 confocal microscope with HCX PL APO 63× CS2 (NA, 1.40) oil objective using the FRET Acceptor Photobleaching wizard in the LAS X software suite. A ROI was selected, and the acceptor channel was bleached at 70% laser intensity (514 nm) until the YFP signal was reduced by at least 90%. CFP (ultraviolet laser excitation, 405 nm; emission, 465 to 511 nm) and YFP (laser excitation, 514 nm; emission, 532 to 603 nm) emissions were then measured. For each biological replicate, three cells and two ROIs per cell were analyzed (a total of 24 ROIs from four biological replicates). FRET efficiency was calculated by the LAS X software suite using the following equation: FRETeff = (Donorpost − Donorpre)/Donorpost. Gαq-YFP and Gαs-YFP were used as negative controls for FLAG-β2AR-CFP and 3HA-M3R-CFP FRET, respectively. Because of nonuniform distribution of protein complexes at the plasma membrane, and a large number of “0” FRET values, the data were not normally distributed and were therefore statistically analyzed using a nonparametric Kruskal-Wallis test. For further analysis, data were converted to binary values (0 = no FRET, 1 = FRET) and analyzed using a chi-square test with two-sided P values and 95% confidence interval.

LC-MS/MS: FASP protein digestion and dimethyl labeling

HEK293 cells were seeded in six-well plates and grown to confluency. Cells were treated with vehicle or ligand for 4 hours and then incubated in lysis buffer [100 mM tris (pH 7.6), 4% w/v SDS, and 100 mM DTT] at 95°C for 3 min, before sonication (for 30 s at 30% amplitude; Qsonica Q125) and centrifugation (at 16,000g for 5 min at room temperature). The supernatant (100 μg) was digested using the FASP Protein Digestion Kit (Expedeon), with trypsin digestion overnight at 37°C. Digested peptides were labeled as previously described (100) using 40 mM 12C3 light or 13C3 heavy formaldehyde with 20 mM NaCNBOH for 1 hour at 37°C, before the reaction was quenched with formic acid (to pH 2.5). The light- and heavy-labeled samples were mixed at a 1:1 ratio and desalted using C-18 desalting columns and three washes with 0.1% (v/v) formic acid. Samples were eluted in 70% (v/v) acetonitrile and 0.1% (v/v) formic acid and then dried by SpeedVac (Labconco). Peptides were fractionated after resuspension in strong anion exchange (SAX) buffer (20 mM acetic acid, 20 mM phosphoric acid, and 20 mM boric acid) (pH 11) and loading onto stage tips containing five layers of anion exchange discs. The first fraction was collected after centrifugation (at 1000g for 3 min at room temperature). A total of seven fractions were collected by sequentially eluting fractions from the stage tips in SAX buffer at pH 8, 6, 5, 4, and 3 and SAX buffer 7 [10% (v/v) formic acid (pH 1)]. Fractions were dried and then resuspended in 2% (v/v) acetonitrile with 1% (v/v) formic acid by sonication at 37°C for 10 min before LC-MS/MS.

Proteomic analysis by LC-MS/MS: Data collection and analysis

Samples were analyzed by LC-MS/MS using a Q Exactive or Q Exactive Plus Orbitrap mass spectrometer (Thermo Scientific) coupled online with an UltiMate 3000 RSLC nano-UHPLC (Thermo Scientific). Samples were injected onto an Acclaim PepMap100 RSLC C18 analytical column [pore size, 100 Å; 75 μm (inner diameter) × 50 cm reversed-phase nanoViper column; Thermo Scientific] with 95% buffer A [0.1% (v/v) formic acid] at a flow rate of 250 or 300 nl/min. The peptides were eluted over 60 min using a gradient to 42.5% buffer B [80% (v/v) acetonitrile and 0.1% (v/v) formic acid]. The eluate was nebulized and ionized using a Nano ElectroSpray Ion Source (Thermo Scientific) with a coated borosilicate emitter and a capillary voltage of 1700 V. Peptides were selected for MS/MS analysis using Xcalibur software (Thermo Scientific) in full MS/dd-MS2 (TopN) mode with the following parameter settings: MS AGC target, 3 × 106; MS maximum injection time, 120 ms; MS/MS TopN = 10 or 12; MS/MS AGC target, 1 × 105, MS/MS maximum injection time, 120 ms; normalized collision energy, 27; and isolation window, 2 or 1.8 mass/charge ratio. Dynamic exclusion was set to 15 s. Protein identification and quantification were performed using MaxQuant software (version 1.5.3.17) (101). Searches were performed against human sequences downloaded from UniProt (August 2015 version) (102) using the following parameters: specific digestion with trypsin with up to two missed cleavages, protein N-terminal acetylation and methionine oxidation were set as variable modifications, and cysteine alkylation was set as a fixed modification.

Data were analyzed using Perseus software (version 1.5.0.15). Common contaminants, reverse peptides, and proteins identified only by a modification site were removed. All data were expressed relative to vehicle-treated controls (heavy/light or transformed 1/[heavy/light], as appropriate), ratios were log2 normalized to allow quantitative analysis, and any nonvalid values were removed. Only proteins that differed statistically significantly from vehicle controls (t test with P < 0.05) were retained and Z-scored to prepare the data for clustering. Hierarchical clustering was performed using default settings. Data for individual proteins are expressed as the log2 change relative to vehicle control from n biological repeats as stated.

The proteins used for hierarchical clustering were further classified by their Biological Process GO term, using the Database for Annotation, Visualization and Integrated Discovery (DAVID, version 6.7) (103, 104) to generate pie charts. Classifications with P < 0.05 were used to group proteins according to biological function, synonymous classifications were removed, and the number of proteins classified within these groups was counted. Only classifications that were identified in at least two biological replicates were included within the final count.

High-content EF1α-GFP imaging

HEK293 cells were seeded in black, optically clear 96-well plates and grown to 70% confluency before cotransfection with pEF1α-AcGFP-C1 (50 ng per well) and pDsRed-N1 (50 ng per well; transfection efficiency control) using PEI. Twenty-four hours after transfection, cells were washed with PBS and partially serum restricted in phenol red–free DMEM supplemented with 0.5% FBS (v/v) overnight. Human cardiac fibroblasts were transfected with pEF1α-AcGFP-C1 (50 ng per well) and pDsRed-N1 (50 ng per well; transfection efficiency control) using X-tremeGENE 9 in suspension and then seeded in black, optically clear 96-well plates at 90% confluency. Experiments in human cardiac fibroblasts used HBSS.

Fluorescence imaging was performed using a high-content GE Healthcare INCell 2000 Analyzer with a Nikon Plan Fluor ELWD 40× (NA, 0.6) objective. Sequential GFP/dsRed imaging used FITC (excitation, 490 nm/20 nm; emission, 525 nm/36 nm) and dsRed (excitation, 555 nm/25 nm; emission, 605 nm/52 nm) filters and the Quad4 polychroic filter. Baseline images were taken every 10 min for 40 min, cells were stimulated with ligand, and images were taken every 10 min for 4 hours. Data were analyzed by selecting 70 cells per well using FIJI, and the GFP fluorescence intensity was expressed relative to the average baseline GFP fluorescence intensity for each cell (F/F0). For human cardiac fibroblasts, all transfected cells were selected, and data are expressed as RFU due to variation in transfection efficiency.

Statistics

All data points are the means ± SEM of at least three independent experiments unless otherwise stated. All data were analyzed using GraphPad Prism with statistically significant differences (P < 0.05) determined using Kruskal-Wallis or chi-square analysis (acceptor photobleaching FRET) or one- or two-way ANOVAs (all other experiments) with appropriate post tests, as stated.

SUPPLEMENTARY MATERIALS

www.sciencesignaling.org/cgi/content/full/11/551/eaan1188/DC1

Fig. S1. Endogenous expression of transcripts encoding GPCRs in HEK293 cells.

Fig. S2. Biphasic changes in cAMP are due to activation of endogenous β2AR and M3R.

Fig. S3. FRET biosensors can detect responses of endogenous or exogenous receptors to femtomolar concentrations of ligand.

Fig. S4. Plasma membrane localization and activity of mutant β2AR and M3R.

Fig. S5. Modeling responses to femtomolar concentrations of Iso.

Fig. S6. Identification of proteins involved in mediating responses to 1 fM Iso.

Fig. S7. The β2AR forms a preassembled signaling complex.

Fig. S8. Identification of proteins involved in mediating responses to 1 fM CCh.

Fig. S9. The M3R forms a preassembled signaling complex.

Fig. S10. Femtomolar ligand concentrations activate compartmentalized signaling and unique cell responses.

Table S1. Femtomolar concentrations of ligand increase cAMP.

Table S2. The β2AR and M3R constitutively form complexes at the plasma membrane.

Table S3. Proteins from hierarchical clustering analysis.

REFERENCES AND NOTES

Acknowledgments: We thank R. J. Summers, D. M. F. Cooper, P. M. Sexton, and A. Christopoulos for critical discussion and C. J. Nowell (Monash Institute of Pharmaceutical Sciences Imaging, Flow Cytometry and Analysis Core Facility), C. Huang (Monash Biomedical Proteomics Facility), and C. Choo for technical assistance. Funding: This work was supported by a National Health and Medical Research Council of Australia R.D. Wright Fellowship (1061687) and a Project Grant (1047633) (to M.L.H.), a Monash Institute of Pharmaceutical Sciences Strategic Grant (to M.L.H.), and a NIH/NIGMS grant (P50GM085273) (to W.S.H.). Author contributions: Conceptualization: M.L.H. Methodology: R.S., B.A.E., and M.L.H. Formal analysis: S.C., R.S., and M.L.H. Investigation: S.C., A.M.E., R.S., C.K.P., O.K., S.J.C., W.S.H., M.C., and M.L.H. Writing (original draft): M.L.H. Writing (review and editing): S.C., A.M.E., R.S., B.A.E., W.S.H., M.C., and M.L.H. Supervision: A.M.E., O.K., W.S.H., M.C., and M.L.H. Funding acquisition: M.L.H. Competing interests: The authors declare that they have no competing interests. Data and materials availability: The MS proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE (106) partner repository with the dataset identifier PXD011033. All other data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials.
View Abstract

Navigate This Article