Impedance Responses Reveal β2-Adrenergic Receptor Signaling Pluridimensionality and Allow Classification of Ligands with Distinct Signaling Profiles

The discovery that drugs targeting a single G protein-coupled receptor (GPCR) can differentially modulate distinct subsets of the receptor signaling repertoire has created a challenge for drug discovery at these important therapeutic targets. Here, we demonstrate that a single label-free assay based on cellular impedance provides a real-time integration of multiple signaling events engaged upon GPCR activation. Stimulation of the β2-adrenergic receptor (β2AR) in living cells with the prototypical agonist isoproterenol generated a complex, multi-featured impedance response over time. Selective pharmacological inhibition of specific arms of the β2AR signaling network revealed the differential contribution of Gs-, Gi- and Gβγ-dependent signaling events, including activation of the canonical cAMP and ERK1/2 pathways, to specific components of the impedance response. Further dissection revealed the essential role of intracellular Ca2+ in the impedance response and led to the discovery of a novel β2AR-promoted Ca2+ mobilization event. Recognizing that impedance responses provide an integrative assessment of ligand activity, we screened a collection of β-adrenergic ligands to determine if differences in the signaling repertoire engaged by compounds would lead to distinct impedance signatures. An unsupervised clustering analysis of the impedance responses revealed the existence of 5 distinct compound classes, revealing a richer signaling texture than previously recognized for this receptor. Taken together, these data indicate that the pluridimensionality of GPCR signaling can be captured using integrative approaches to provide a comprehensive readout of drug activity.


Introduction
G protein-coupled receptors (GPCRs) are the most abundant class of cell surface receptors, responding to various types of endogenous stimuli, including hormones, neurotransmitters and odorants, and are the largest family of therapeutic targets [1,2]. Through interactions with various G protein [3] and non-G protein effectors [4], GPCRs elicit a diverse array of signaling events, including production of second messengers, activation of phosphorylation cascades, modulation of ion channel activity and transcriptional regulation. Although it was originally assumed that a given GPCR controlled the activity of a single signaling pathway, it is now recognized that individual receptors can elicit multiple signaling events resulting in global changes in cellular physiology, thus highlighting the pluridimensionality of GPCR signaling [5]. Furthermore, certain GPCR ligands have been shown to differentially modulate distinct subsets of this signaling repertoire, a phenomenon referred to as ''ligand-biased signaling'' or ''functional selectivity'' [5][6][7]. A recent review of the published literature indicates that ligand-biased signaling occurs for a large number of GPCRs involved in a diverse array of physiological functions [8]. For example, several peptide ligands for the angiotensin AT 1A receptor show a clear preference in the activation of b-arrestindependent signaling events, yet possess no intrinsic efficacy towards G protein-dependent signaling [9,10]. Even more striking is the inversion of efficacy observed for some b-adrenergic receptor ligands, which act as inverse agonists for cAMP production yet behave as agonists for the activation of the ERK1/2 MAPK pathways [11,12]. Harnessing such ligand functional selectivity could represent a promising avenue in drug development, as the design of compounds that selectively modulate a pathway involved in a given pathology without collateral effects on other pathways could provide therapeutic benefit with a decreased risk of side effects. Indeed, several recent studies have suggested that functionally selective ligands may provide clinically relevant advantages over unbiased ligands at the same receptor [13][14][15].
Yet, detecting the full extent of such ligand ''functional selectivity'' remains a non-trivial technical challenge and has traditionally involved measuring the relative efficacy of ligands towards distinct pathways engaged by a given receptor using multiple assays. One considerable challenge in obtaining a full description of ligand activity at a given GPCR is the lack of knowledge about the complete signaling repertoire of most receptors. In addition, monitoring multiple signaling pathways can be time and resource consuming and involves the use of different assay formats with different sensitivity and dynamic ranges that can lead to spurious conclusions.
An integrative approach that could capture the global signaling profile of a ligand in a single assay would therefore greatly facilitate the identification of biased ligands and enable their classification into pharmacologically relevant categories. Besides the potential impact on the drug discovery process, such an approach could also provide greater insight into how the pluridimensionality of GPCR signaling is integrated into an overall cellular response. A string of recent studies has explored the use of such integrative assays to study GPCR pharmacology using label-free, cell-based technologies developed to monitor real-time changes in higher-order cellular events such as morphology, viability, adhesion and mass distribution [16][17][18][19][20]. Thus far, however, little effort has been made to understand how the pluridimensionality of GPCR signaling is recapitulated in the responses generated by these assays. One recently developed label-free assay is the measurement of cellular impedance. The use of impedance measurements to monitor global cellular activity is based on the principle that the adhesion of cells directly onto microelectrodes induces changes in the local ionic environment at the electrode/solution interface, conferring an increase in electrode impedance (see Materials and Methods). Any changes in cell morphology and/or adhesion that modulate the physical contact between cell and electrode will be reflected by changes in impedance. Recently, impedance measurements have been used to monitor GPCR signaling [16,18,19] and to assess general mechanisms of drug action [21].
In the present study, we sought to determine if impedance signatures represent an integrative measure of the receptor signaling repertoire that can be used to differentiate among ligands with distinct signaling profiles. Using the b 2 -adrenergic receptor (b 2 AR) as a model, we dissected the impedance responses from individual ligand-receptor pairs and associated the activation of specific signaling events with distinct features of the impedance curves. This dissection of the impedance responses led to the identification of a novel b 2 AR-mediated Ca 2+ signaling event, highlighting the strength of such an integrative readout of cellular response to reveal biological processes even at a well-characterized target. Finally, we performed an unsupervised clustering analysis that permitted a rapid and systematic classification of compounds based on the impedance signatures obtained in both heterologous and primary cell cultures, demonstrating that impedance responses represent signatures that are predictive of the global signaling profiles of b-adrenergic ligands previously shown to display functional selectivity [10,11].
Results b 2 AR activation elicits a multi-featured, time-resolved impedance response We first sought to determine if impedance measurements could provide a composite representation of the multiple signaling events elicited upon receptor activation. To assess this possibility, impedance measurements were made in HEK293S cells stably overexpressing the human b 2 AR (6HisHA-b 2 AR-HEK293S) following treatment with the prototypical b 2 AR-selective agonist isoproterenol (ISO). As shown in Figure 1A, stimulation of cells with ISO led to a concentration-dependent, time-resolved and feature-rich impedance response. Immediately following ISOstimulation, the impedance response briefly decreases below zero before quickly reverting to a positive slope, producing what is referred to hereafter as the ''transient negative phase'' ( Figure 1B). Subsequently, the impedance rises sharply at first (''rapid ascending phase''), followed by a later ''slow ascending phase'' before reaching a maximum, after which it decays slowly. This response was found to be b 2 AR-specific since the response was blocked by the selective b 2 AR antagonist ICI118,551 (ICI) but was not affected by the b 1 AR-selective antagonist CGP20712A ( Figure S1A). In addition, the concentration-response curve describing the maximum impedance response elicited by ISO was progressively right-shifted with increasing concentrations of ICI, indicating a competitive antagonism of the response by b 2 AR blockade ( Figure S1B). Further supporting the b 2 AR specificity of the response, we observed a qualitatively similar ICI-sensitive impedance response in the parental HEK293S cell line, which endogenously express low amounts of b 2 AR ( Figure S2). Although the impedance response maintains the same general shape, a decrease in the magnitude of the positive component of the impedance response and a larger transient negative phase was observed.
Further analyses of the impedance responses were performed in the 6HisHA-b 2 AR-HEK293S cell line to obtain a more robust response to receptor stimulation. The distinct features of the response (i.e. depth and duration of the transient negative phase, slope of the rapid ascending phase and the maximal impedance response) were differentially modulated by increasing concentrations of ISO, suggesting the contribution of distinct signaling events to discrete components of the impedance response. For instance, concentration-response curves describing either the maximum impedance response ( Figure 1C) or the slope of the rapid ascending phase ( Figure 1D), revealed EC 50 values that differ by more than an order of magnitude. Furthermore, two concentrationdependent components can also be distinguished in the transient negative phase. At low concentrations (i.e. under 1 mM), ISO promoted a concentration-dependent increase in the magnitude of the minimum ( Figure 1A). At higher concentrations, however, no further increase in depth was observed, but a quicker recovery from negative impedance values and a shift of the minimum to earlier time points occurred, thus reducing the duration of the transient negative phase. Receptor expression was also found to differentially influence distinct components of the impedance response. As illustrated above, overexpression of the b 2 AR selectively increases the positive components of the impedance response while decreasing the duration and amplitude of the transient negative phase ( Figure S2A). These results further support the above data that the signaling events contributimg to the transient negative phase are fully activated with fewer stimulated receptors. As the stimulus strength (i.e. agonist concentration) or receptor number is increased, the impedance response becomes more positive, likely due to the engagement of another signaling event with a weaker stimulus-coupling. Altogether, these results suggest that the impedance response obtained reflects the integration of multiple signaling events which may be individually represented in the various components of the response.
Impedance responses encode multiple signaling events upon b 2 AR activation b 2 AR activation is known to elicit at least two distinct signaling events in HEK293S cells: cAMP accumulation via adenylyl cyclase (AC) stimulation and activation of the ERK1/2 MAPK  [12]. To directly investigate if these canonical pathways are represented in the impedance response, 6HisHA-b 2 AR-HEK293S cells were pre-treated, prior to ISO stimulation, with the AC inhibitor SQ22536 to inhibit cAMP production or the MEK inhibitor U0126 to block ERK1/2 activation. Interestingly, both of these inhibitors led to similar decreases in the slope of the slow ascending phase and the maximum impedance response without affecting the transient negative phase or the rapid ascending phase (Figure 2A,B). A similar effect on the impedance response was observed using the K97A dominant negative mutant of MEK1 ( Figure S3). SQ22536 and U0126 were found to be pathway-specific since they blocked the ISOstimulated accumulation of cAMP and activation of ERK1/2, respectively, and had no significant effects on the other signaling pathway ( Figure 2C,D). The observation that cAMP and ERK1/2 pathway inhibition led to similar changes in the impedance response suggested that these pathways may converge on a common mechanism underlying the observed changes in impedance. Consistent with this notion, concurrent inhibition of both pathways did not lead to a greater attenuation of the maximal ISO-stimulated impedance response than inhibition of either the cAMP or ERK1/2 pathway alone ( Figure 2E).
From the above data, it is clear that cAMP production and MAPK activation are not sufficient to explain the entire ISO impedance response since a considerable residual signal is observed when both of these pathways are inhibited. To further explore the source of this residual response, we examined the role of the G proteins themselves. The b 2 AR is known to couple to both G s and G i [22]. Chronic treatment with cholera toxin (CTX), which selectively downregulates G s [23], led to a severe attenuation of the impedance response, including a complete loss of the transient negative phase and a significant reduction of the maximum response ( Figure 3A). Interestingly, CTX inhibited ISO-induced cAMP accumulation to a similar extent as the AC inhibitor SQ22536, with no effect on ERK1/2 activation ( Figure 3B,C), yet the effect of CTX on the impedance response is clearly distinct from the one observed following inhibition of the cAMP pathway ( Figure S4A), suggesting the contribution of additional G s -dependent signaling events other than cAMP accumulation to the impedance response. Selective inhibition of G i -dependent signaling with pertussis toxin (PTX) led to changes the first 10 minutes post-stimulation. (C,D) Concentration-response curves describing the maximum impedance response generated (C) and the slope of the rapid ascending phase (D) reveal distinct EC 50 values. Data represent means of at least three independent experiments (6 SEM for C and D). For clarity, error bars are not shown for impedance responses. Typical experimental variability in the impedance responses is demonstrated in Figure S5. doi:10.1371/journal.pone.0029420.g001 in the impedance response that were different from those caused by CTX (Fig. 3A). Specifically, a transient negative phase was still observed following PTX treatment, although it was reduced and delayed compared to vehicle treatment. Although both treatments led to significant reductions in the maximum impedance response, the influence of PTX on this phase of the curve was not as severe as CTX. Taken together, these results indicate that G s and G i signaling have distinguishable contributions to the overall impedance response. PTX treatment significantly inhibited ERK1/2 activation but did not affect cAMP production ( Figure 3B,C). However, the contribution of G i cannot be solely attributed to ERK1/2 activation since complete abolition of this pathway by U0126 (Fig. 2D) affected the ISO-stimulated impedance response to a lesser extent than PTX ( Figure S4B), which only partially blocked ERK1/2 activation. Although the differences observed between acute pharmacological inhibition of cAMP production and ERK1/2 activation versus chronic inhibition of Gs and Gi suggest the contribution of alternative G protein-dependent signaling pathways to the impedance response, we cannot exclude the possibility that the chronic treatment with CTX and PTX may affect other downstream components involved in the response.
Next, we assessed the contribution of Gbc subunits to the impedance response using the pharmacological inhibitor, gallein, which selectively inhibits interactions between Gbc and effector molecules [24]. Pre-treatment with gallein inhibited ERK1/2 activation, with no effect on cAMP accumulation ( Figure 3B,C), and led to an impedance response with reduced kinetics of the slower ascending phase and a decreased maximum response ( Figure 3D), The effect of gallein was similar overall to that observed upon inhibition of the ERK1/2 pathway; however, Gbc inhibition led to an additional loss of the slow decay phase ( Figure  S4C) indicating the existence of additional Gbc-dependent signaling beyond ERK1/2.
Altogether, these data demonstrate that a variety of G proteindependent signaling events are integrated in the ISO-stimulated impedance response, including the activation of canonical second messenger pathways (i.e. cAMP and ERK1/2) and additional G protein-dependent events not yet identified.

Clustering analysis of impedance signatures reveals compound classes with distinct signaling profiles
After demonstrating that impedance responses represent the integration of multiple signaling events elicited upon receptor activation, we hypothesized that compounds with different signaling profiles at the b 2 AR would generate distinct impedance signatures. Impedance responses were obtained for a series of badrenergic ligands ( Figures 4A-I and S5) at concentrations providing .99% receptor occupancy based on their reported K d values [25][26][27]. Differences among ligand signatures were determined by comparing the area between individual impedance curves. Visualization of the pairwise differences using a visual assessment of clustering tendency (VAT; see Materials and Methods) suggested that the ligands fall into 5 distinct clusters ( Figure S6), which we confirmed by complete linkage hierarchical clustering [28] ( Figure 4J). With the exception of Group V compounds,  which elicit impedance responses of completely negative values, differences between compound classes reflect differences in the specific features defined in our previous analysis of the ISOpromoted impedance response. We see great variation, for example, in the maximum impedance responses generated between compound classes. We also observed clear distinctions between groups in the transient negative phase of the response; Group I compounds elicit a relatively short-lived transient negative phase while Group IV compounds completely lack this feature ( Figure 4F-I).
To determine the relationship between ligand signaling profiles with the compound classes obtained in the above analysis of impedance signatures, we assayed each of the ligands for their ability to stimulate cAMP accumulation and ERK1/2 activation in conditions identical to those used for the impedance measurements ( Figure 4K,L). Group I ligands, which include ISO, have high agonist efficacies (full or strong partial agonists) for both cAMP and ERK1/2 responses. Group II ligands are only weak partial agonists toward the ERK1/2 pathway but elicit as strong cAMP responses as Group I. Group III ligands are weak partial agonists for both cAMP and ERK1/2 pathways. Group IV ligands elicit very weak signaling responses for both pathways. Group V ligands are inverse agonists towards both cAMP and ERK1/2 responses. The compound classes defined by impedance signatures are thus highly predictive of the relative efficacies of ligands to elicit cAMP accumulation and ERK1/2 activation.
The role of intracellular Ca 2+ in the b 2 AR-mediated impedance response Since neither cAMP accumulation nor ERK1/2 activation contribute to the transient negative phase observed for most ligand impedance signatures, we reasoned that further analysis of this feature could reveal additional texture among ligand signaling profiles. The shape of the transient negative phase can be defined by two parameters, the duration and minimum value of the negative response. Although no significant correlation between these two parameters could be observed when all of the ligands were considered (r 2 = 0.06, p = 0.79) ( Figure 5), a strong linear relationship emerged when considering Groups III and IV compounds alone (r 2 = 0.97, p = 9610 25 ), reflecting a fundamental difference in the nature of the transient negative phase between Group I/II ligands and Group III/IV ligands. Indeed, Group I and II ligands have a shorter-lived transient negative phase than would be predicted given their amplitudes ( Figure 5 inset), suggesting that these compounds may selectively engage a signaling pathway that accelerates the rapid ascending phase thus shortening the duration of the transient negative phase. Given that many GPCRs are capable of activating pathways leading to an increase in intracellular [Ca 2+ ], we assessed whether a b 2 ARmediated Ca 2+ response could be responsible for accelerating this phase. Consistent with this hypothesis, only Group I and II ligands elicited a significant increase in intracellular [Ca 2+ ], with ISO and epinephrine (Group I) eliciting a much stronger Ca 2+ response than salbutamol and salmeterol (Group II) ( Figure 6A). This response was found to be b 2 AR-selective since receptor blockade with the b 2 AR-selective antagonist ICI completely abolished the ISO-stimulated increase in intracellular [Ca 2+ ] ( Figure S7). This b 2 AR-promoted increase in [Ca 2+ ] was sensitive to the inositol trisphosphate (IP 3 ) receptor antagonist, 2-aminoethoxydiphenyl borate (2-APB) and the intracellular Ca 2+ chelator BAPTA-AM ( Figure 6B), implicating a mobilization from IP 3 -gated intracellular stores in the calcium response. Pre-treatment with either 2-APB or BAPTA-AM strongly inhibited the ascending phases of the impedance response ( Figure 6C), supporting the hypothesis that Ca 2+ mobilization is a key event in the b 2 AR-mediated impedance response.
Interestingly, pre-treatment of cells with either 2-APB or BAPTA-AM significantly modulated the impedance responses not only for ligands in Groups I and II, but also Groups III and IV, which do not elicit a detectable increase in [Ca 2+ ] ( Figure 6A,C-F). Given that both 2-APB and BAPTA-AM not only inhibit ligand-induced Ca 2+ mobilization, but also decrease the basal intracellular [Ca 2+ ] ( Figure 6G), we hypothesized that a minimal concentration of intracellular Ca 2+ is required to initiate the cellular events underlying the ascending phases of the impedance response. To test this hypothesis, we pre-treated cells with the sarco/endoplasmic reticulum Ca 2+ ATPase inhibitor thapsigargin (TG), which will lead to an increase in intracellular Ca 2+ by preventing the reuptake of Ca 2+ , thus emptying the intracellular stores. Since the intracellular Ca 2+ stores are already empty, ISO does not elicit an additional Ca 2+ release following TG inhibition ( Figure S8A). Yet, in contrast with what was observed with 2-APB and BAPTA-AM, and despite the absence of  a de novo Ca 2+ release, the TG treatment does not inhibit the rapid ascending phase of the impedance response ( Figure S8B). In fact, it modestly potentiated the maximal response while slightly reducing the slope of the slow ascending phase. Taken together, these data support the notion that it is the intracellular [Ca 2+ ] itself, more than the de novo Ca 2+ release, that is the prime determinant of the ascending phases. The fundamental role of Ca 2+ in the impedance response is further demonstrated by the effect of the Ca 2+ ionophore A23187 on the impedance responses ( Figure 7). On its own, the ionophore elicits only a weak positive impedance response. However, it greatly potentiates the impedance responses obtained upon stimulation with either b-adrenergic ligands or the direct activator of adenylyl cyclase, forskolin, yielding responses with faster kinetics of the ascending phase and a higher overall maximum response. This effect is particularly evident when considering the forskolin-stimulated impedance response, which consists of a long transient negative phase and a slow rise to a relatively modest maximum response in the absence of the Ca 2+ ionophore. Taken together, these data explicitly demonstrate the importance of Ca 2+ in the impedance response and indicate that, under normal conditions, a minimal [Ca 2+ ] is needed to yield an increase in the b 2 AR-promoted impedance response while additional Ca 2+ mobilization by Group I and II ligands further accelerates the rapid ascending phase. Distinct impedance signatures detected in rat aortic vascular smooth muscle cells To explore the applicability of impedance-based monitoring of the signaling activity of a GPCR in its native cellular context, we assessed the b 2 AR response in rat aortic vascular smooth muscle cells (VSMCs). As was the case in b 2 AR-expressing HEK293S cells, ISO induced a response that was completely abolished upon pre-treatment with the b 2 AR-selective antagonist ICI ( Figure 8A). However, the shape of the impedance response was radically different in VSMCs, indicating that cellular response to receptor activation is cell type-specific. We next assessed the impedance responses induced upon treatment with ligands representing each of the 5 compound classes defined above ( Figure 8B). Using the same clustering criteria as in the 6HisHA-b 2 AR-HEK293S cells, distinct impedance signatures for ISO (Group I), salbutamol (Group II) and labetalol (Group III) were observed. Propranolol (Group IV) and ICI (Group V) generated distinct signatures from these other compounds, but could not be distinguished from each other in VSMCs ( Figure 8C), emphasizing the cell-type specificity of the response. Altogether, these data indicate that despite the fact that the magnitude and direction of the responses are cell-type specific, quantitative analysis of impedance responses can be used to detect distinct ligand signatures in primary cell cultures.

Discussion
In the present study, we demonstrate that impedance responses provide an integrative assessment of the cellular consequence to GPCR stimulation, representing a holistic readout of the various signaling events elicited in real-time. In addition, clustering analyses of the impedance responses provided a means to subclassify ligands in a manner that was predictive of their signaling profiles and revealed a richer signaling texture among badrenergic ligands than previously envisaged.
Analysis of the b 2 AR-promoted impedance response to ISO stimulation revealed the contribution of both G s and G i coupling, Gbc-dependent signaling, as well as cAMP production, ERK1/2 activation and a novel Ca 2+ mobilization response to the overall changes in cellular impedance. The observation that elements of the impedance response that were blocked upon G s or G i inactivation following chronic treatment with CTX and PTX, respectively, could not be fully recapitulated by acute pharmacological inhibition of cAMP or ERK1/2 pathways suggests the contribution of additional G protein-dependent signaling. However, we cannot rule out the possibility that chronic inhibition of G proteins leads to cellular changes (e.g. changes in expression of signaling or adhesion proteins) that indirectly affect the impedance response. Our analysis also provided evidence that both cAMP and ERK1/2 pathways converge on a common downstream effector that mediates their contribution to the impedance response, although the identity of this target remains unknown.
The impedance responses generated following b 2 AR stimulation was found to be extremely sensitive to inhibition of intracellular Ca 2+ mobilization. Although the b 2 AR has previously been shown to increase intracellular [Ca 2+ ] through modulation of L-type voltage-gated Ca 2+ channels [29] and the ryanodine receptor [30], our data implicate a novel IP 3 -gated release of Ca 2+ from intracellular stores. Ca 2+ is a well-established regulator of actin cytoskeleton dynamics [31,32] and plays an important role in controlling cellular adhesion and morphology, through effectors such as cadherins [33] and annexins [34], potentially providing a key mechanistic link between b 2 AR stimulation and the impedance changes observed. Although we found this de novo Ca 2+ release to contribute to the rapid ascending phase of Groups I and II ligands, the impedance responses of ligands that do not themselves induce a Ca 2+ mobilization are also sensitive to a disruption of intracellular Ca 2+ homeostasis, stressing the importance of Ca 2+ in the cellular mechanisms that initiate an impedance response. We propose that the normal resting concentration of intracellular Ca 2+ is necessary and sufficient to permit the cytoskeleton to respond to subsequent signaling events, such as cAMP production or ERK1/2 activation, and that ligands that induce an additional increase in intracellular [Ca 2+ ] further sensitize the cell to these signaling inputs leading to an enhancement of the impedance response. Thus, the overall impedance signature of a given ligand truly represents the integration of the signaling pathways activated.
The contribution of multiple signaling events to the overall impedance response challenges some of the efforts that have been made to reduce such integrated responses to the engagement of a single G protein [17,18]. It is now widely accepted that many GPCRs promiscuously couple to and activate signaling through multiple G protein subtypes [3] as well as G protein-independent pathways. In the current study, we observed that both G s -and G idependent signaling were integrated in the overall b 2 ARpromoted response. Moreover, the fact that the signatures obtained in VSMCs were dramatically different from those observed in 6HisHA-b 2 AR-HEK293S cells highlights the celltype dependency of the impedance response, which likely reflects differences in the relative abundance of signaling molecules and/ or mediators underlying the changes in cell shape and adhesion that give rise to the impedance responses. These observations highlight the inherent complexity of signaling systems and their measurement and demonstrate the diversity of cellular responses that can be generated by the stimulation of a given GPCR in different cell contexts. The ability to detect and dissect these complex cellular responses, as demonstrated in the present study, provides information concerning the full signaling repertoire of a receptor that can be more useful in drug discovery than the attempt to ascribe a single signaling pathway to a receptor.
The fact that multiple signaling events are integrated in the impedance response makes it an appealing platform to classify compounds based on their signaling repertoire. Clustering of the impedance responses generated by a series of b-adrenergic ligands revealed an unprecedented level of ligand texture at the b 2 AR, demonstrating the existence of at least five groups of compounds with distinct impedance signatures. Previous work assessing the signaling profiles of b-adrenergic ligands towards cAMP and ERK1/2 pathways alone using endpoint-specific assays proposed the existence of 3 distinct compound classes [11]. The ability of impedance measurements to detect additional ligand diversity likely reflects the contribution of other pathways not previously assessed. Furthermore, differences in ligand signaling profiles between this study and the former demonstrate the critical influence of assay conditions on the reported signaling properties of a ligand. In the present study, cAMP accumulation and ERK1/ 2 activation were measured in conditions identical to those used for impedance measurements. Without serum starvation, ligands such as propranolol and ICI did not generate a detectable ERK1/ 2 response, as previously reported [11,12]. However, differences in their respective impedance responses indicated distinctions in the overall cellular response induced by ICI and propranolol, which had not previously been observed.
Although cell-based, integrative assays have begun to be employed in some screening applications [35][36][37][38], our results represent a first proof-of-principle of how they can be used not only to identify potential ligands for a given target, but also to differentiate classes of compounds with specific signaling profiles that may correlate with therapeutic or adverse effects. We also demonstrate the unique ability of integrative assays such as impedance measurements to illuminate novel biology not yet revealed by traditional single endpoint assays. The approaches described should be universally applicable to all label-free platforms that provide an integrative cell response, including both impedance-based systems and those based on dynamic mass redistribution. Most certainly, analytical integrative methodologies such as the one presented here, will provide a more in-depth exploration of cellular signaling dynamics and should be transformative in the future of drug discovery.

Materials and reagents
Gallein was obtained from Tocris, U0126 from Cayman, and bucindolol was a generous gift from Dr. Michael Bristow (University of Colorado Health Sciences Center, Denver, CO). Cell culture reagents were from Wisent Incorporated. Other reagents were obtained from Sigma-Aldrich unless stated otherwise.

Stable cell lines and transfection
The pIREShygro_6HisHA-b 2 AR construct was generated as follows: The 6HisHA-b 2 AR gene was amplified by PCR from the previously described vector pCDNA3RSV-6HisHA-b 2 AR [39] using sense and antisense primers containing BsrGI and NheI restriction sites, respectively. The amplified 6HisHA-b 2 AR sequence was digested with BsrGI/NheI and then subcloned in a BsrGI/NheIdigested pIREShygro3 vector (Clontech, Mountain View, CA). The 6HisHA-b 2 AR-HEK392S polyclonal cell line was generated by transfecting the pIREShygro_6HisHA-b 2 AR construct in HEK293S cells [40] and selecting vector-inserted cells by growth in culture media containing 100 mg/ml hygromycin B (WISENT inc, St-Bruno, Quebec, CAN). The 6HisHA-b 2 AR-HEK293S stable cell line was found to express 1.0760.14 pmol of 6HisHA-b 2 AR per mg of membrane protein as determined by a 125 I-CYP radioligand binding assay. Cells were routinely grown at 37uC with 5% CO 2 in Dulbecco's modified Eagle's medium supplemented with 5% fetal bovine serum. For Ca 2+ assays and impedance assays in the presence of the MEK dominant negative mutant, 6HisHA-b 2 AR HEK293S cells were transiently transfected with the obelin biosensor [41] or the MEK1-K97A-Flag mutant [42], respectively, using polyethylenimine and assayed 48 hours later.

Cellular impedance assay
The xCELLigence system (Roche Applied Science) was employed to measure changes in cellular impedance following ligand stimulation [43]. Briefly, this assay is based on the principle that the adhesion of cultured cells directly onto an array of equallydistributed microelectrodes embedded at the bottom of wells of a microtitre plate induces changes in the local ionic environment at the electrode/solution interface, conferring an increase in electrode impedance. As a result, any changes in cell physiological properties that modulate the physical contact between cell and electrode, such as changes in morphology and/or adhesion, will be reflected by changes in the measured impedance, as defined by the cell index variable. A measurement is made in the presence of growth medium prior to cell seeding to determine the background cell index in each well of a 96-well E-plate (Roche Applied Science), which is subtracted from the cell index values generated following cell attachment. Cells were plated at a cell density of 20 000 cells/well and grown for 16-20 hours before ligand addition in the RTCA SP device station (Roche Applied Science). When cells were treated with cholera toxin or pertussis toxin, they were plated at a cell density of 10 000 cells/well, treated with toxins after 24 hours and stimulated with ligand 40 hours after seeding. Experimental variation in cell index values at the time of treatment was limited to 620%, which considerably increased the reproducibility of the data obtained. Cells were treated with compounds as indicated and cell index values were obtained immediately following ligand stimulation every 20 seconds for a total time of at least 100 minutes. Cell index values were normalized by dividing by the cell index at the time of ligand addition and baseline-corrected by subtracting the cell index obtained in vehicle-treated conditions (see Figure S9). All data presented in figures represent means of at least three independent experiments. cAMP accumulation assay Cells were plated, grown in 96-well plates and treated with ligand as described for the cellular impedance assay. cAMP accumulation was measured using the HTFR-cAMP dynamic kit (Cisbio). After 10 minutes of ligand stimulation, cells were lysed on ice in the lysis buffer provided by the manufacturer and immediately snap-froze at 280uC for at least 4 hours. After thawing on ice, 10 ml per well of lysate was transferred to a 384well plate (Lumitrac 200, Greiner Bio-one), and cAMP measurements were performed as per manufacturer's instructions.

ERK1/2 activation assay
Cells were plated, grown in 96-well plates and treated with ligand as described for the cellular impedance assay. ERK1/2 activation was measured using the ERK1/2 AlphaScreen Surefire kit (Perkin Elmer). Cells were lysed on ice after 2 minutes of ligand stimulation in the lysis buffer provided by the manufacturer and placed immediately at 220uC overnight. After thawing on ice, 5 ml were transferred to 384-well ProxiPlates (Perkin Elmer) and ERK1/ 2 activation was measured as per manufacturer's instructions.

Obelin Ca 2+ biosensor assay
Cells were plated, grown in 96-well microtitre plates and transfected as described above. Forty-eight hours after transfection, growth medium was replaced with Tyrode's solution (1 mM CaCl 2 , 140 mM NaCl, 2.7 mM KCl, 900 mM MgCl 2 , 37 mM NaH 2 PO 4 , 5.6 mM d-glucose, 8.3 mM NaHCO 3 and 12.6 mM Hepes, pH 7.5) and cells were incubated with 5 mM coelanterazine H (Nanolight Technology, Pinetop, AZ, USA) for 2 hours. Intracellular Ca 2+ concentration is reflected by the magnitude of bioluminescence emitted by the obelin biosensor and data are expressed as relative luminescence units (RLU). Ligands were injected as indicated and bioluminescence was measured for 1 minute per condition using the SpectraMax L Microplate reader (Molecular Devices). Baseline readings were obtained for 5 seconds prior to ligand addition and subtracted from post-injection readings to obtain a ligand-induced Ca 2+ response. Peak Ca 2+ responses represent the maximum response in the largest peak obtained following ligand addition. Peaks were defined to have a minimum width of ten adjacent points. To quantify the slope of the rapid ascending phase, a line was drawn from two data points after the minimum (to avoid artifacts from shallow minima) to maximum response, and the impedance response was bisected at the point farthest from that line. Then, a second line was drawn from two data points after minimum response to the first bisection point, and the impedance response was bisected again. Slope was calculated as the ratio between the difference in impedance and the difference in time between two data points after the minimum response and the second bisection point.

Data analysis
Differences in ligand impedance signatures were determined by calculating the area between their impedance curves. The area A between two curves i and j was estimated as where I i,t is the impedance response of curve i at time t.
Clustering of the data was assessed using the visual assessment of tendency method (VAT) [44]. Briefly, arrays of the pairwise dissimilarities were sorted by placing the two ligands with the highest pairwise dissimilarity in the first and last rows, respectively. The array was then populated with the remaining ligands by minimizing the dissimilarity among adjacent compounds and maintaining the same order of compounds in rows and columns. To emphasize cluster structure among similar data points, the square root of the sorted dissimilarity maps were displayed.
For complete linkage hierarchical clustering, clustering trees were cut at 0.8 times the median of all pairwise dissimilarities.