A BRET Ca²⁺ sensor enables high-throughput screening in the presence of background fluorescence

Abstract

The intrinsic fluorescence of samples confounds the use of fluorescence-based sensors. This is of particular concern in high-throughput screening (HTS) applications using large chemical libraries containing intrinsically fluorescent compounds. To overcome this problem, we developed a bioluminescence resonance energy transfer (BRET) Ca²⁺ sensor, CalfluxCTN. We demonstrated that it reliably reported changes in intracellular Ca²⁺ concentrations evoked by an agonist and an antagonist of the human muscarinic acetylcholine receptor M₁ (hM₁R) even in the presence of the fluorescent compound fluorescein, which interfered with a standard fluorescent HTS sensor (Fluo-8). In an HTS using a chemical library containing fluorescent compounds, CalfluxCTN accurately identified agonists and antagonists that were missed or miscategorized using Fluo-8. Moreover, we showed that a luciferase substrate that becomes activated only when inside cells generated long-lasting BRET signals in HTS, enabling results to be reliably compared among replicate samples for hours. Thus, the use of a self-luminescent sensor instead of a fluorescent sensor could facilitate the complete screening of chemical libraries in a high-throughput context and enable analysis of autofluorescent samples in many different applications.

INTRODUCTION

Monitoring cellular and molecular processes with fluorescent sensors is an important tool for researchers (1, 2). Some of the sensors commonly used for visualizing and quantifying cytosolic calcium (Ca²⁺) signaling are small-molecule probes such as Fura-2, Fluo-4, and Fluo-8 (3–5), whereas others are genetically encoded, such as GCaMP6 and cameleon (6, 7). Genetic engineering by structure-based mutagenesis and directed evolution has optimized the properties of fluorescent proteins to adapt them to myriad applications. However, all these sensors depend on fluorescent excitation that can trigger autofluorescence, phototoxicity, photoresponses (in photosensitive tissues), and photobleaching of the sensor. All living tissues have background autofluorescence due to photoexcitable molecules, such as reduced nicotinamide adenine dinucleotide phosphate and flavins, and subcellular structures, such as lysosomes. Many tissues also have especially high autofluorescence due to the presence of characteristic pigments like chlorophyll or hemoglobin or extracellular matrix components like collagen and elastin or due to fixation (8, 9).

Although background fluorescence is sometimes merely an annoyance that can be circumvented by using a fluorescent sensor with different excitation/emission wavelengths and/or by using very narrow bandwidth filters, in some applications, the background fluorescence creates uncorrectable problems that have deleterious biomedical implications. One such application is the discovery of new pharmacological agents by high-throughput screening (HTS) of cells expressing targets of interest. HTS often relies on fluorescent sensors of a physiological response to report the effect of test compounds on the activity of a target of interest. However, when the compounds being tested (for example, agonists, antagonists, and/or allosteric modulators) are themselves fluorescent, this background fluorescence interferes with the signal from the sensor (1). Many compounds within small-molecule libraries that are used for HTS, especially xanthines (10), curcumins (11), and coumarins (12), exhibit intrinsic fluorescence emission. These very same molecules may have beneficial biomedical properties: For example, xanthine derivatives are well-characterized neural stimulants, and both curcuminoids and coumarin-like molecules may be effective in treating a variety of cancers (13–16). Moreover, the fluorescence of compounds in adjacent wells can often “bleed over,” thereby obscuring the signal from an efficacious compound in a nearby well. As we report here, potentially effective compounds that are intrinsically fluorescent will cause a background that might cause many useful compounds to be ignored in HTS applications or, at best, necessitate retesting by a nonfluorescent assay. In some HTS situations, an entire plate of compounds might be discarded because of the global interference by one highly fluorescent compound in the plate.

An optimal HTS sensor would be impervious to fluorescence interference and enable an assay of every compound in the HTS library in the initial screen. Luciferase-based sensors are an option that avoids the problems associated with fluorescence excitation and have been previously used in HTS contexts with outputs ranging from cell viability to gene expression changes (1, 17). A luminescent sensor does not require excitation because the light signal is a result of an enzyme-catalyzed chemical reaction rather than fluorescence. Therefore, the optical measurement of the response is made in a dark background, and the luminescent signal itself does not excite detectable intrinsic fluorescence of any compound in the chemical library. However, previous HTS experiments using luminescent sensors relied on older-generation luciferases that are dim, resulting in noisy readouts, and attempts to demonstrate the advantages of luminescent sensors in screens of libraries that include fluorescent compounds have been limited to assessing sensor responsiveness in the presence of autofluorescence of the sample (18).

We have shown the usefulness of bioluminescence resonance energy transfer (BRET) (19, 20) sensors in another context in which overlapping fluorescence spectra is deleterious, namely, in optimal coupling of BRET sensors with optogenetic actuators (21). BRET sensors are ratiometric, so they monitor a targeted activity independently of the intracellular concentration of the sensor, which may vary from cell to cell for genetically encoded sensors, because the measured quantity is the ratio of intensities at two wavelengths and not a single intensity alone. Moreover, ratiometric recording is less sensitive than intensity recording to the presence of fluorescent or light-absorbing compounds in the library because the change that is monitored is not the signal intensity of one wavelength but the ratio of two separate wavelengths.

Here, we demonstrated that BRET sensors will detect efficacious compounds that are invisible to fluorescent sensors and avoid the deleterious effects of fluorescence in adjacent wells when a fluorescent sensor is used. We previously demonstrated the advantages of Calflux over previous luminescent sensors of Ca²⁺ fluxes and applied it to cell culture and brain slice preparations (21). Our BRET Ca²⁺ sensor is genetically encoded to allow targeting to specific cell types and/or subcellular loci, and it uses a luciferase (NanoLuc) that is 100 to 150 times brighter than other luciferases (22) and has an excellent signal-to-noise ratio (SNR). We showed that background fluorescence did not interfere with screens for agonists or antagonists of the human muscarinic receptor 1 (hM₁R), a G_q-coupled G protein–coupled receptor that stimulates intracellular Ca²⁺ signaling, when using the BRET sensor under conditions in which fluorescent sensors are blind. We also introduced a luciferase substrate not previously used for HTS that extends the lifetime of the BRET signal for hours such that stable and reliable HTS results are directly comparable among replicate plates. Last, we undertook a side-by-side comparison of Calflux with Fluo-8 in an HTS of a typical chemical library that included a few fluorescent compounds and found that Calflux accurately identified agonists and antagonists that were missed or miscategorized using Fluo-8.

RESULTS

Development of the CalfluxCTN sensor

We developed a BRET sensor for free Ca²⁺ consisting of NanoLuc luciferase (22) and the fluorescent protein Clover (23) connected by the same linkers and Ca²⁺-sensitive troponin C (TnC) sequence (from Opsanus tau) as the fluorescent Ca²⁺ indicator Twitch-2B (Fig. 1A and fig. S1A) (24). We call this Ca²⁺ sensor CalfluxCTN for “calcium flux composed of Clover, troponin, and NanoLuc.” In vitro experiments showed that the Ca²⁺-sensitive troponin sequence of purified CalfluxCTN underwent a conformational change as the [Ca²⁺] was increased from 0.017 to 39 μM, bringing NanoLuc closer to Clover so that resonance energy transfer could occur with a concomitant spectral shift from 450 to 515 nm (Fig. 1B and fig. S1B) (21, 24, 25). Several configurations of Clover, NanoLuc, and the troponin sequence were tested (fig. S2, A to E), and the CloverΔC9-T-N configuration (hereinafter called CalfluxCTN) was determined to be optimal in terms of the BRET ratio response to Ca²⁺ changes (fig. S2E). The BRET response of CalfluxCTN occurred within a range of Ca²⁺ concentrations that span from typical cytosolic [Ca²⁺] to typical Ca²⁺ activation concentrations (~100 nM to ~1 μM).

Fig. 1. BRET response of CalfluxCTN to Ca²⁺ and characteristics of emission with different substrates.

(A) Schematic of CalfluxCTN. In the presence of its enzymatic substrate, NanoLuc (N) bioluminescence is at 450 nm. When the troponin (T) moiety binds Ca²⁺, it undergoes a conformational change that brings NanoLuc closer to Clover (C). This increases the probability of BRET, leading to the activation of Clover and the emergence of a mix of 450- and 515-nm signals. (B) In vitro spectral measurements of CalfluxCTN emission in response to buffers of different [Ca²⁺] using furimazine as the NanoLuc substrate. Spectra were normalized to the emission intensity at 444 nm (n = 3 independent experiments). R², coefficient of determination. (C) Unstimulated CHO cells expressing hM₁R and CalfluxCTN were tested with furimazine and different coelenterazine (coel) analogs as substrates for Calflux to assess emission spectra and relative intensity. Data are representative of n = 3 independent experiments. a.u., arbitrary units. (D) The total brightness of CalfluxCTN light emission from unstimulated CHO cells expressing hM₁R and CalfluxCTN over the course of 3 hours was measured using chemically blocked (extracellularly inert) coelenterazine analogs ViviRen and blocked coelenterazine 400a. Data points represent means ± SEM of three independent experiments (n = 3).

As compared with our previously reported CalfluxVTN (where Venus served as the fluorescent moiety) (21), purified CalfluxCTN was more responsive to very low Ca²⁺ concentrations, but CalfluxVTN exhibited a larger dynamic range to changes in [Ca²⁺] (fig. S3, A and B). We chose to use CalfluxCTN in HTS applications rather than CalfluxVTN because of its enhanced sensitivity to low [Ca²⁺] and because its Clover emission peak (515 nm) most closely matched that of fluorescein (fig. S3C), which we use here to model the interfering effect of endogenously fluorescent compounds in a screening library. CalfluxCTN’s NanoLuc emission at 450 nm matches the excitation for fluorescein. These overlapping wavelengths therefore constitute the worst-case scenario for fluorescence interference and would therefore provide a rigorous proof-of-principle test for whether a luminescent sensor is superior to a fluorescent sensor for HTS using libraries that include a substantial number of fluorescent compounds.

Use of a blocked luciferase substrate to stabilize signals

A primary reason that luminescent sensors have not been as popular as fluorescent sensors for HTS is that the substrates for coelenterazine-based luciferases tend to be unstable, causing the signal intensity to decay over 30 min to 1 hour, thus complicating the screening of hundreds of test plates at a time. The substrate designed by Promega to optimally work with NanoLuc is furimazine (22), but other analogs of coelenterazine are effective substrates for NanoLuc and Calflux (figs. S4, A to C, and S5). Using Chinese hamster ovary (CHO) cells stably expressing CalfluxCTN and human muscarinic acetylcholine receptor M₁ (hM₁R) and stimulated with the hM₁R agonist carbachol to induce Ca²⁺ release from intracellular stores (26, 27), we tested different coelenterazine analogs with Calflux to determine the optimal compromise between brightness and sustained signal intensity for HTS applications. The bimodal spectrum of CalfluxCTN was essentially identical between all tested substrates (Fig. 1C). The luminescence of Calflux was brightest with furimazine, but coelenterazine 400a (1-bisdeoxycoelenterazine) (28) was equivalent (Fig. 1C). Whereas the emission spectrum of Renilla luciferase is blue-shifted with coelenterazine 400a (28), with NanoLuc, all substrates showed a consistent spectrum. The luminescence signal with coelenterazine h (2-deoxycoelenterazine) (29) was about one-half of that with furimazine or coelenterazine 400a, and the signal with native coelenterazine (29) was ~30× lower than with the brighter substrates.

Although furimazine and coelenterazine 400a allowed the brightest signal with Calflux, the signal was short lived, peaking about 10 min after substrate addition to cells transfected with CalfluxCTN and decaying over the next several hours (Fig. 1D and fig. S4, A and B). Because Calflux is a ratiometric reporter (Fig. 1, B and C, and fig. S1B), the signal decay was not a problem for accurate estimation of [Ca²⁺], as shown by the stability of the BRET ratio for at least 3 hours (fig. S4C). However, as the signal intensity of a BRET sensor declines toward background, the SNR also decays. We therefore tested two versions of protected substrates that are inactive extracellularly but activated by intracellular enzymes (such as esterases) that remove the chemical blocking groups after they permeate cells. One such protected substrate is ViviRen, a commercially available protected version of coelenterazine h. The luminescence from Calflux-transfected cells with ViviRen began at a 10-fold lower amount than with furimazine or coelenterazine 400a and declined further but stabilized at a low luminescence value after 1 hour (Fig. 1D). We also tested acetoxymethyl bisdeoxycoelenterazine (30), a protected version of coelenterazine 400a that we refer to here as “blocked 400a.” The luminescence of Calflux-transfected cells treated with this blocked 400a started at a 10-fold lower amount than in cells treated with furimazine or coelenterazine 400a, but the luminescence signal from cells was very stable for at least 3 hours thereafter (Fig. 1B). Because Calflux is so much brighter than previously available luminescent sensors (for example, those not based on NanoLuc), the 10-fold lower initial intensity was not a problem for SNR, and the stable signal allowed more time flexibility for HTS assays and other assays that continue for an extended time. Therefore, we consider that the use of blocked 400a allows an optimal trade-off between signal intensity and longevity.

Fluorescent compounds can broadly affect signal readout across HTS plates and wells

Many compounds in small-molecule libraries used for HTS have chromophoric or fluorophoric properties that obscure signals from fluorescent sensors. For example, a study of fluorescence profiles of more than 70,000 compounds found that more than 5% of library members had substantial fluorescence (31). A potential consequence of this fluorescence is the perturbation of the signal from wells adjacent to those containing a fluorescent compound, so the signals from multiple wells may be missed during initial screens with fluorescent sensors. To investigate this phenomenon, we placed CHO cells expressing hM₁R and loaded with Fluo-8 into 96-well plates, stimulated them with carbachol, and imaged them with a typical HTS system. The background from a well also containing 1 μM fluorescein (D3) obscured the Fluo-8 fluorescence signal of that same well (Fig. 2, A and B). It is also possible that the fluorescence of some compounds may be so bright that their emission conceals the signal emanating from adjacent wells, thereby preventing evaluation of the activity of the compounds of the adjacent wells, although they are not intrinsically fluorescent. This “flare-out” phenomenon is illustrated by 10 μM fluorescein in one well (I7) generating an intense fluorescence capable of obscuring signals from nearby wells, causing signal saturation in digital imaging (Fig. 2B). This concentration (10 μM) corresponds to the standard concentration at which compounds are typically tested in initial high-throughput screens, and we performed the imaging with an industry standard HTS system equipped with an interline charge-coupled device (CCD) camera (Hamamatsu FDSS 7000). Other, more sensitive HTS cameras such as the WaveFront Panoptic may be affected even more by the flare-out and spillover problem such that up to half of the entire plate cannot be accurately imaged (Fig. 2, C and D). With the Panoptic detector, not only was the scanning of the column perturbed but also the data from entire rows of wells were lost as a result of the strong fluorescence of a single well (Fig. 2, C and D). It is not unreasonable to expect the presence of fluorescent compounds in screening libraries to result in collateral loss of hundreds to thousands of potential hits in a large HTS.

Fig. 2. Background fluorescence in a single well interferes with HTS.

(A) Fluorescence image of a 384-well plate in which the left half of the plate was loaded with hM₁R-expressing CHO cells treated with 5 μM Fluo-8 AM, and the right half of the plate was free of cells. The image was taken with using a Hamamatsu FDSS 7000, and the acquisition settings were set for a typical fluorescence screen. (B) Image taken after fluorescein was added to two wells of the plate shown in (A). One micromolar fluorescein was added to well D3, and 10 μM fluorescein was added to well I7. The red color indicates saturation of the signal. (C and D) A different HTS detector (WaveFront Panoptic) was used to image plates containing 1 (C) or 10 μM (D) fluorescein in well K10. Green and red colors indicate saturation of the signal. Data in (A) to (D) are representative of n = 3 independent tests. (E) Ca²⁺ measurements in CHO cells expressing hM₁R and loaded with Fluo-8 before stimulation with 30 μM carbachol at t = 3 s in the presence of various concentrations of fluorescein. (F) Ca²⁺ measurements in CHO cells coexpressing hM₁R and CalfluxCTN and stimulated with carbachol at t = 3 s. Data in (E) and (F) were acquired using a POLARstar OPTIMA plate reader, are means ± SD of measurements from 12 wells, and are representative of three independent experiments.

The background problem minimizes SNR so that fluorescent sensors become blind to efficacious compounds. The ability of Fluo-8 to detect a robust Ca²⁺ flux was obviated by even a modest background fluorescence (Fig. 2E). In contrast, the BRET assay using CalfluxCTN was unperturbed by the same background conditions and retained an excellent signal (Fig. 2F). This result means that the BRET assay can detect signals that are obscured in the fluorescence assay.

Application of CalfluxCTN for HTS

Having established that CalfluxCTN was a sensitive reporter of changes in Ca²⁺ in vitro and in cells (Figs. 1, C and D, and 2F), we sought to determine its applicability in identifying compounds that stimulate or inhibit intracellular Ca²⁺ signaling by HTS. We first tested whether the presence of a fluorescent compound affected the measurement of BRET signals in plates of unstimulated CHO cells stably expressing CalfluxCTN and hM₁R. We used fluorescein for this test because its emission peak coincides with that of CalfluxCTN (fig. S3C), constituting a worst-case scenario for potential interference. The BRET ratio of CalfluxCTN in unstimulated CHO cells was essentially unaffected by the presence of fluorescein at concentrations up to ~100 μM (Fig. 3, A and B), which is more than 100 times the concentration that overwhelmed Fluo-8 (Fig. 2E).

Fig. 3. BRET response of CalfluxCTN is insensitive to background fluorescence.

(A) Unstimulated CHO cells stably expressing CalfluxCTN and hM₁R were plated across an entire 384-well plate, and a range of concentrations of fluorescein was added across the wells (0.02 μM to 1 mM). The fluorescence emission of the plate (excitation at 482/35 nm) was captured by the WaveFront Panoptic II to show the gradient of fluorescence intensities generated by fluorescein. The left and right halves of the plate are duplicates of each other. (B) Luminescent BRET ratios of the CHO cells in the fluorescein gradient plate in (A). Data are means ± SD of 32 wells for each fluorescein concentration and are representative of three independent experiments.

The BRET response of these cells stimulated by carbachol was dose dependent and reproducible (Fig. 4, A and B). In a typical HTS 384-well plate format using a WaveFront Panoptic that permits imaging of every well simultaneously, we monitored the response of the CalfluxCTN-transfected hM₁R cells in real time with an image capture rate of every 100 to 250 ms per image. The prestimulation signals in the green (520 nm emission) and blue (485 nm emission) ranges were consistent among the wells, yielding a reproducible BRET ratio before carbachol treatment of about 1.0 (Fig. 4C). At 0.6 min after the prestimulation data were collected, either 30 μM carbachol or vehicle control was added in a “checkerboard” pattern (Fig. 4C). Calflux reported the [Ca²⁺] rise in carbachol-treated cells as a simultaneous increase in green luminescence and a decrease in blue luminescence (Fig. 4, C and D), as expected from the in vitro characterization of Calflux response to Ca²⁺ (Fig. 1B). These data were quantified as BRET ratios and arranged as time courses of the control- versus carbachol-treated groups (Fig. 4E). There was a clear separation between the two groups and precise quantification of the Ca²⁺ fluxes elicited by hM₁R stimulation with low well-to-well variability. The peak response to carbachol at 1 min for each well was separated from the vehicle-treated wells (Fig. 4F), and the resulting Z′ factor was calculated to be 0.52, which is well within the commonly accepted range for excellent separation of response versus no response for HTS (0.4 to 1.0) (32).

Fig. 4. Response of CalfluxCTN to hM₁R stimulation.

CHO cells stably coexpressing hM₁R and CalfluxCTN were imaged using the WaveFront Panoptic system in a 384-well plate. (A) Time course of the cells responding to different concentrations of the cholinergic agonist carbachol (Cbcl). Error bars at each time point are means ± SD for 16 wells. (B) Distribution of concentration-response curve ratios of five independent experiments as in (A). Center line, median; box limits, upper and lower quartiles; whiskers, 1.5× interquartile range; and points, outliers. (C) Luminescence intensity photograph of an entire plate showing each well from which luminescence was measured at ~485 and ~520 nm before (pre) and after (post) carbachol stimulation. Insets show the magnified image of nine wells after addition of 30 μM carbachol in a checkerboard pattern. EM, emission. (D) Intensity measurements taken from the 485- and 520-nm channels from well A1 in the plate in (C) illustrating the reciprocal 485/520-nm intensity changes elicited by carbachol addition at 0.6 min, from which the BRET ratios were calculated. (E) Time courses for the wells shown in (C). The lines represent the mean response, and the error bars are ±SD for 192 wells for each treatment. (F) Scatter plot showing the distribution of the peak responses to either carbachol or vehicle measured at time = 1 min. Dashed lines indicate the mean of each group. The Z′ factor for this representative plate was 0.52. Data in (C) to (F) are representative of n = 3 independent experiments.

As predicted from other experiments (Figs. 2F and 3, A and B), the BRET assay of hM₁R stimulation by carbachol was unperturbed by background fluorescence from fluorescein (Fig. 5, A to C). The response was essentially identical in the presence (+ F) or absence (− F) of 10 μM fluorescein (± F), which was also true with the blocked 400a substrate (Fig. 5D), which has the additional advantage of much longer-lived high SNR signals (Fig. 1D). Therefore, a strong fluorescence background of compounds in an HTS library should have little or no effect on the kinetics, well-to-well variability, or Z′ factors of a BRET sensor such as Calflux (Fig. 5D).

Fig. 5. Fidelity of CalfluxCTN BRET signal changes elicited by agonists in highly fluorescent backgrounds.

(A) Images of a 384-well plate containing CHO cells stably coexpressing hM₁R and CalfluxCTN, with half of the wells (columns 1 to 12) also containing 10 μM fluorescein. Bright-field and fluorescence images are shown as well as the luminescence at 520 nm before (pre) and after (post) treatment with 30 μM carbachol in a checkerboard pattern. (B) Time course of BRET ratio changes in the plate shown in (A) in response to 30 μM carbachol (Cbcl) versus vehicle (Veh) in the presence (+ F) or absence (− F) of 10 μM fluorescein. Mean ± SD of each treatment is shown. n = 96 wells per treatment for this experiment, which is representative of three independent experiments. (C) Peak response BRET ratio for each well to the checkerboard carbachol stimulation of the plate in (A), in which wells A1 to P12 contained fluorescein and wells A13 to P24 did not. (D) Summary of Z′ factors of CalfluxCTN with blocked (BC400a) and nonblocked (C400a) coelenterazine 400a substrates in the presence or absence of fluorescein. Mean ± SEM is shown for n = 3 independent experiments.

CalfluxCTN reveals antagonists that escape detection using fluorescent sensors

As a further demonstration of the value of a luminescent Ca²⁺ sensor, we compared the ability to detect a muscarinic receptor antagonist in a mock HTS screen using Calflux versus a fluorescence-based Ca²⁺ sensor. Scopolamine is a well-characterized antagonist of the carbachol stimulation of M₁R (33). We sought to determine whether an intrinsically fluorescent antagonist with the properties of scopolamine (represented here by scopolamine + fluorescein) could be detected using a sensor with a fluorescence emission profile similar to that of the antagonist (Fluo-8). We found that even a strong fluorescence background had no significant effect on the Calflux BRET signal for carbachol-induced stimulation of M₁R or the antagonism of this response by scopolamine (Fig. 6A). On the other hand, modest concentrations of fluorescein obscured the Fluo-8 signal to the extent that the difference between the presence versus the absence of scopolamine became invisible (Fig. 6, B to D, and fig. S6). In particular, for Fluo-8 measurements, the relatively limited fluorescence background caused by 0.156 μM fluorescein (and higher concentrations) created an environment in which there was no differential between the presence versus the absence of scopolamine due to the high baseline (fig. S6). In the context of an HTS, this means that a modest fluorescence background will completely degrade the ability to detect a potentially valuable antagonist when using a fluorescent sensor, whereas a luminescent BRET sensor will readily detect the antagonist.

Fig. 6. BRET sensor reveals antagonist action undetected by fluorescent sensors.

(A) CHO cells coexpressing hM₁R and CalfluxCTN and (B) CHO cells coexpressing hM₁R and CalfluxCTN and additionally loaded with Fluo-8 were stimulated with 30 μM carbachol at t = 3 s in the presence or absence of 10 μM scopolamine and in the presence of the indicated concentrations of fluorescein. The BRET ratio of CalfluxCTN (A) or the intensity of Fluo-8 fluorescence (B) were measured from t = 0 to 15 s. Data are means ± SD of 16 wells from representative 96-well plates (n = 3 independent experiments). (C) Quantification of the BRET ratio percent difference between carbachol stimulation with and without (±) scopolamine in the presence of the indicated concentrations of fluorescein. (D) Quantification of the percent difference in the Fluo-8 signal between carbachol stimulation with and without (±) scopolamine in the presence of the indicated concentrations of fluorescein. Data in (C) and (D) are means ± SEM of 48 wells from three independent experiments.

Comparison of CalfluxCTN with Fluo-8 in a test library screen

To test whether the BRET sensor is superior to a fluorescent sensor in a realistic HTS application, we performed a side-by-side comparison of CalfluxCTN and Fluo-8 in a screen of a test library. A selection of 320 compounds from the Microsource Spectrum Collection, including two compounds that are obviously fluorescent (fig. S7), was dispensed into a 384-well plate containing hM₁R-expressing CHO cells either coexpressing CalfluxCTN or loaded with Fluo-8. This library contains a wide range of biologically active and structurally diverse compounds, consisting of about 50% drug components, 30% natural products, and 20% other bioactive components, all of which were screened at a concentration of 10 μM. We also included the well-characterized Ca²⁺ ionophore ionomycin at 10 μM and a fluorophore-conjugated selective antagonist of the M₁R, telenzepine at 0.15 μM (A488-telenzepine) (34). The protocol was to establish a stable baseline of sensor signal and then inject an aliquot of the library compound into each well at t ~ 15 s to detect whether 10 μM of the small-molecule compound had an agonist effect on the M₁R. At t ~ 160 s, 10 μl of 200 μM carbachol was injected into each well to detect compounds with M₁R antagonist activity.

As expected, there were a variety of response patterns (Fig. 7A and data file S1). For example, both sensors detected the response in well F16 (VU0243313) as indicative of a compound that had neither agonist nor antagonist action; the injection of the compound at t ~ 15 s did not stimulate a Ca²⁺ response, nor did it inhibit the carbachol-elicited response at t ~ 160 s. Well E23 contained the fluorescent A488-telenzepine antagonist, and the result with CalfluxCTN showed an inhibition of the carbachol stimulus as expected. However, the intrinsic fluorescence of A488-telenzepine saturated the Fluo-8 signal, muddling the experimental interpretation so that it could be construed to elicit an agonist response. Well M15 contained a strongly fluorescent compound (VU0244360) for which the CalfluxCTN signal showed no action on the M₁R, but the Fluo-8 signal could be misinterpreted as indicating agonist activity. Well O1 contained ionomycin, and both CalfluxCTN and Fluo-8 responded immediately upon its addition at t ~ 15 s. Both CalfluxCTN and Fluo-8 accurately reported the nonfluorescent compounds in wells B4 (VU0239848) and D15 (VU0239836) as an M₁R antagonist and agonist, respectively.

Fig. 7. Comparison of CalfluxCTN with Fluo-8 in a test library screen.

CHO cells coexpressing hM₁R and CalfluxCTN were either preloaded with Fluo-8 (for fluorescence detection) or treated with furimazine substrate (for BRET detection) before addition of compounds from the test library and subsequent treatment with carbachol. (A) Responses of six different wells as monitored by CalfluxCTN and Fluo-8 to the addition of the test compound at t ~ 15 s followed by the addition of 40 μM carbachol at t ~ 160 s. BRET ratio (red) and Fluo-8 intensity (blue) are presented as normalized values in arbitrary units (a.u.). Labels refer to individual wells: F16, compound no. VU0243313 from the Vanderbilt Library (10 μM streptozotocin); E23, A488-telenzepine (0.15 μM) (34); M15, VU0244360 from the Vanderbilt Library (10 μM 3,6-diamino-10-methylacridinium); O1, ionomycin (10 μM ionomycin); B4, VU0239848 from the Vanderbilt Library (10 μM propantheline); and D15, VU0239836 from the Vanderbilt Library (10 μM spiperone). The test library experiment was performed four times with equivalent results. (B) The response data from four independent screens of the test library (data file S1) were plotted as a function of Fluo-8 response (abscissa) versus CalfluxCTN BRET ratio (ordinate). Cyan points plot the response or nonresponse to injection at t ~ 15 s of each compound from the library (plotted points are signals integrated from 15 to 160 s); magenta points plot the response or nonresponse to the subsequent addition of 40 μM carbachol at t ~ 160 s (plotted points are signals integrated from 160 to 305 s). The points for the well numbers containing some key compounds are labeled. (C) Higher-magnification view of a portion of (B).

We compared the responses measured by fluorescence (Fluo-8) and BRET (CalfluxCTN) to all test compounds as well as the subsequent response to carbachol (Fig. 7, B and C, and data file S1). For compounds that did not elicit an agonist response nor inhibit the carbachol-induced response, such as that in F16, the responses to the compounds clustered in a cloud of points where fluorescence/BRET was 1.0/1.0, and the responses to the subsequent addition of carbachol were distributed along the ~45° trajectory between 1.0/1.0 and 2.2/2.5 (Fig. 7C). Agonists such as the compound in well D15 and the ionophore ionomycin evoked initial responses along the ~45° distribution. In general, the data obtained from CalfluxCTN versus Fluo-8 were equivalent, but there were a few important exceptions. The fluorescent telenzepine antagonist in well E23 appeared to be an agonist from the Fluo-8 signal, but CalfluxCTN correctly reported it as an antagonist. Moreover, the library compounds in wells K6 and M15 also would be initially categorized as agonists by the Fluo-8 response, but CalfluxCTN accurately reported them as no-action compounds (Fig. 7, B and C).

DISCUSSION

The use of fluorescent sensors of physiological and molecular processes is a common and standardized methodology, but it suffers from technical problems. One of these liabilities is interference due to the excitation of unwanted and interfering autofluorescence of the sample or fluorescent reagents. In many cases using a fluorescent sensor, the problem can be ameliorated by using a sensor with a fluorescence spectrum that is distinct from the background fluorescence and/or by using very narrow bandwidth filters. This problem ceases to exist when using a self-luminescent sensor. We demonstrated that luminescence sensors—here exemplified by a BRET luminescence Ca²⁺ sensor—are optimally suited to monitor cellular processes when samples include discernible background fluorescence. We demonstrate its utility for HTS of chemical libraries in which a fraction of the compounds is intrinsically fluorescent. Although some of these fluorescent compounds might provide a valuable starting place for the development of a beneficial therapeutic agent, they may remain effectively untested in an HTS, because their fluorescence overwhelms the efficiency and SNR of fluorescent sensors (Figs. 2E; 6, B and D; and 7, A to C; and fig. S6). Even worse, these fluorescent compounds can even prevent accurate measurement of other compounds’ activity in adjacent wells (Fig. 2, B to D). The presence of as few as 1000 fluorescent compounds in a screening collection could obscure the activity of 10 times as many compounds.

Here, we used the fluorescein-based Ca²⁺ sensor Fluo-8 as a prototypical fluorescent sensor to compare to our BRET-based Ca²⁺ sensor, CalfluxCTN. Fluorescence sensors with excitation and emission wavelengths different from fluorescein-based sensors have been developed and can be used to mitigate potential interference from fluorescent compounds with fluorescein-like spectral properties, but these will have the same limitations as any other fluorescence-based sensor for fluorescent compounds whose spectra do appreciably overlap. In a library of thousands of chemical compounds, there will always be some compounds with fluorescent properties that interfere with any given fluorescence-based sensor. There are several factors that contribute to promising compounds being “lost” in the screening process, including false negatives due to fluorescence as discussed here. A technique that can reduce the negative effect of any of these factors is valuable even if it helps to identify only a few additional compounds that would otherwise not be detected, because this may reveal important structural characteristics necessary for pharmacological activity. Many historical drugs were discovered and/or derived from the dye industry, in which many of the compounds were developed precisely because they absorbed light and many were fluorescent (35).

Therefore, unlike fluorescence-based approaches, a BRET or other luminescent sensor enables more comprehensive library coverage. Use of such a sensor would also avoid the disruption of signals from adjacent wells by bright fluorescent wells during plate reads (Figs. 2F; 6, A and C; and 7, A to C; and fig. S6). However, no technique is perfect. In the case of a BRET or other luminescent sensor, the signal depends on an enzymatic reaction by luciferase, and it is conceivable that some of the compounds in a library could inhibit the luciferase reaction, causing an apparently false-negative result. In general, compounds that directly inhibit luciferases are rare; in particular, because NanoLuc (the luciferase that is active within Calflux) is an adenosine 5′-triphosphate–independent luciferase, it is rarely inhibited by compounds from screening libraries, and those compounds that do inhibit NanoLuc have weak potency (36). If, however, a compound in a library directly inhibits (or activates) the luciferase reaction, the virtue of a ratiometric BRET reporter is that the luminescence signal in both the blue and green channels will be depressed (or enhanced) as compared with adjacent wells, and therefore, it will be obvious that the compound affected the luciferase activity rather than cellular process being monitored (for example, changes in Ca²⁺ concentration in the case of Calflux).

Our test library experiment (Fig. 7, A to C, fig. S7, and data file S1) validated the use of Calflux to avoid false negatives or false positives. In general, the results using Calflux tracked those obtained with Fluo-8. However, both false negatives and false positives marred the interpretation of results obtained with Fluo-8. The Fluo-8 results would have identified the compounds in wells E23, K6, and M15 as acting as positives in the agonist category and negatives in the antagonist category. However, CalfluxCTN correctly identified K6 and M15 as having no activity in the assay (false positives with the Fluo-8 sensor) and E23 as an antagonist in the assay (a false negative with the Fluo-8 sensor). In the practice of HTS, the luxury of high replicates is usually not possible; many resources depend on using an assay that has the sensitivity and robustness to detect the relatively few active compounds that are hopefully contained within the vast library that is screened. The concealment of one or more active compounds due to an artifact such as fluorescence background wastes the time invested to avoid false negatives and ultimately deprives the project of potentially rich structure-activity knowledge that could inform downstream discovery efforts.

An optimal strategy for an HTS facility that reuses the same small-molecule libraries and is therefore familiar with which compounds are fluorescent would be to group all fluorescent compounds on plates to be initially screened with a luminescence (preferably BRET) reporter. The remaining plates that are devoid of fluorescent compounds could be screened with a traditional fluorescence sensor without concern of interferences such as those that we have characterized in this study. The plates containing fluorescent compounds would be screened with a BRET or other luminescent sensor to generate an accurate initial response profile. A BRET luminescence sensor is preferable to a non-BRET, intensity-only luminescence sensor, because any compound that is directly interfering with the luciferase’s activity would be obvious, as described above; for a BRET sensor, both the blue and green channels will be depressed or enhanced by compounds that interfere with the activity of luciferase, whereas a true effect on the sensor sequence (in the case of this study, the troponin moiety) will be expressed as a change of BRET ratio.

The advantages of Calflux that we demonstrate here go beyond HTS applications. Any live cell imaging in which cell or tissue autofluorescence is above background will benefit from a luminescent sensor with a readout of BRET ratio or simply luminescence intensity (BRET/luminescence). This principle of BRET/luminescence sensor superiority in applications compromised by autofluorescence extends not only to fusion construct sensors as used here but also to the original bimolecular BRET interaction assay paradigm (18–20). Moreover, the imperviousness of BRET/luminescence sensors to autofluorescence has a consequence that can be applied to an innovative multiplex screening strategy to reduce the number of wells or plates needed in the initial screen. This strategy involves the placement of multiple (~10) compounds in individual single wells, where each compound is screened in several different wells in different combinations with other compounds. By comparing which wells show positive signals with the particular combination of compounds, it is possible to reduce by fivefold or more the number of wells or plates in the initial screen, thereby reducing costs of cells, reagents, media, and other resources. Multiplex screening is not feasible with fluorescent sensors applied to libraries that include intrinsically fluorescent compounds, because the distribution of fluorescent compounds will interfere with true signals and nullify the pattern of wells exhibiting hits as correlated with the distribution of specific compounds. A self-luminescent sensor that is impervious to autofluorescence will enable such an innovative multiplex screening strategy.

MATERIALS AND METHODS

DNA plasmid construction

The CalfluxCTN DNA sequence was assembled from 5′ to 3′: (i) Clover (with a nine–amino acid deletion of the C terminus: “CΔ9”), (ii) TnC from O. tau, and (iii) NanoLuc. The template for the Ca²⁺-sensitive TnC domain for CalfluxCTN was derived from the Twitch fluorescent Ca²⁺ indicator version 2B (24), whereas the plasmids pcDNA3.1 Clover (Thermo Fisher Scientific) and pNL1.1 [Promega identification (ID) no.: N1001] were used as DNA templates for the Clover and NanoLuc, respectively. The Förster resonance transfer inherent in BRET is very sensitive to the orientation and the distance between donor and acceptor molecules (19), and therefore the use of different linkers can alter both the orientation and the distance between donor and acceptor. On the basis of our experience with various linkers in the creation of CalfluxVTN (21), we tested various linkers in the process of constructing CalfluxCTN. Of the various forms that we tested, we found that the linker between donor and acceptor achieved by truncation of nine amino acids of Clover produced the best BRET transfer in our tests (figs. S1, A to C, and S2, A to E).

To create Escherichia coli expression plasmids, we inserted DNA into the pRSETB plasmid using the restriction enzyme sites Eco RI and Hind III to create an N-terminal six-histidine–tagged (His6) protein. For mammalian expression, pcDNA3.1/Puro-CAG-VSFP-CR (Addgene ID no.: 40257) was used as the backbone, where voltage-sensitive fluorescent protein (VSFP) portions were removed and CalfluxCTN was inserted at the Nhe I and Bam HI restriction enzyme sites.

In vitro Ca²⁺ assays

BL21 E. coli was used for expression of CalfluxCTN under the control of the pRSETB plasmid. Bacteria were grown in shaking culture for 48 hours at 37°C, and cells were lysed by sonication. Using the His6 tag fused to the construct, we purified the protein with TALON Co⁺⁺ metal affinity resin (Clontech Laboratories Inc.). To assess the emission spectrum of the purified proteins in response to changing Ca²⁺ concentrations, we used a QuantaMaster fluorescence spectrophotometer. Furimazine (Promega ID no.: N1110) solution (10 μM final concentration for these experiments) was added to each tube containing purified protein combined with a variety of Ca²⁺ buffers (Invitrogen), and the light emitted at 400 and 600 nm was measured.

Mammalian cell expression and selection of stably expressing cell lines

CHO cells stably expressing hM₁R were a gift from C. Niswender, Vanderbilt University. The CHO cells were grown in Gibco Dulbecco’s modified Eagle’s medium (DMEM)/F-12 (Ham) + GlutaMAX supplemented with 10% fetal bovine serum (FBS), 10 mM Hepes, and 1% antibiotic-antimycotic (Gibco) in a humidified 37°C incubator, and they tested negatively for contamination. To generate cells that stably expressed both CalfluxCTN and hM₁R, we transfected the hM₁R-expressing CHO cells with the CalfluxCTN expression plasmid using the transfection reagent GeneCellin (Bulldog Bio). Because the hM₁R-expressing CHO cells used G418 selection, both puromycin (10 μg/ml) and G418 sulfate [800 μg/ml; Research Products International (RPI) Corp.] were added after 72 hours to the selection media. Cells that survived the selection process were counted in a hemacytometer, diluted (0.8 cells per 200 μl), and plated in clear-bottomed plates for clonal selection. Afterward, cells were again grown in growth media free of selection agents for the remainder of experiments.

Optical data acquisition for cell culture experiments

Cells were imaged in either Opti-MEM media (Gibco) or in a simple Hepes-buffered salt solution [1.26 mM CaCl₂, 0.49 mM MgCl₂, 0.41 mM MgSO₄, 5 mM KCl, 0.44 mM KH₂PO₄, 4.16 mM NaHCO₃, 150 mM NaCl, 0.34 mM Na₂HPO₄, and 10 mM Hepes (pH 7.2); filter-sterilized d-glucose was added fresh at 0.6% w/v]. For the photomultiplier tube–based plate reader experiments, a POLARstar OPTIMA plate reader (BMG Labtech Inc.) was used to collect the light emitted through filters of 470/10-nm emission and either 530/10-nm emission or 520-nm long pass (LP). Cells were imaged either in the previously mentioned Hepes-buffered solution or in Opti-MEM media without phenol red. For the 384-well plate experiments, where the image of the entire plate of wells is captured simultaneously, either a custom WaveFront Panoptic plate imager (WaveFront Biosciences) or a Hamamatsu FDSS 7000 was used. The FDSS 7000 used an Orca-ER CCD camera (Hamamatsu) to capture images. The custom Panoptic used an electron-multiplying CCD camera (Andor iXon). For BRET recordings, the filters used were custom 460/70-nm emission and 520-nm LP 50-mm diameter filters (Semrock); each filter was rotated in front of the emission path using a filter turret to monitor the desired wavelength output. The images were binned 4 × 4, with a preamp gain of 2 and an exposure time of 100 to 500 ms (exposure time depended on the initial brightness of each particular plate). For fluorescence recordings using either the Panoptic or the FDSS 7000, a green fluorescent protein filter set with a 482/35-nm excitation and a 536/40-nm emission was used.

Coelenterazine analogs

Furimazine was obtained from Promega, whereas coelenterazine 400a/1-bisdeoxycoelenterazine (C400a) and propionated coelenterazine 400a/acetoxymethyl bisdeoxycoelenterazine (blocked C400a, also known as “BC400a”) were synthesized by the Vanderbilt Chemical Core (Nashville, TN) using the method described by Levi and co-workers (30). The other coelenterazine analogs were purchased from Nanolight Technology (Prolume Ltd.). Analogs were dissolved in 200-proof ethanol, and the concentrations used in each experiment are stated in each figure legend. For long-term storage, each analog was stored under argon gas at −80°C.

Ca²⁺ flux imaging with Fluo-8

Fluo-8 acetomethyl ester (AM) (Abcam plc, CAS no.: 1345980-40-6) was used at a final concentration of 5 μM and incubated with the cells for 50 min to 1 hour before washout with phosphate-buffered saline (Gibco) and then imaging. Because CHO cells actively pump molecules like Fluo-8 out of the cells to the extracellular medium, 1 mM of the organic anion transporter inhibitor probenecid (Sigma-Aldrich) was used to inhibit those pumps and maintain a stable concentration of cytosolic deesterified Fluo-8 for imaging. Probenecid was not required for imaging with Calflux.

Carbachol concentration response curves

To measure the sensitivity of the Ca²⁺ response of CalfluxCTN to carbachol stimulation in CHO-hM1R cells, we aliquoted serial dilutions of carbachol (Sigma-Aldrich) to a 384-well plate. The maximum final concentration of carbachol per well was 20 μM, and then it was diluted threefold serially until 12 concentrations were achieved, ranging between 0.17 nM and 20 μM carbachol. For Fluo-8 AM carbachol concentration response curves (CRCs), an identical protocol with the same concentrations was used, and the half-maximal effective concentration (EC₅₀) was determined by curve fitting in R.

Carbachol checkerboard

A 30 μM carbachol (saturating, final concentration) versus vehicle checkerboard was used to stimulate Ca²⁺ fluxes to determine the effectiveness of CalfluxVTN as an HTS sensor (“vehicle” was cell culture medium). Using the robotic arm and 384-well head of the Panoptic imager, we added carbachol versus vehicle to a black, clear-bottomed 384-well plate containing cells from another 384-well plate with wells containing either 60 μM (2×) of carbachol or vehicle distributed in a checkerboard pattern. For experiments with C400a, ~10 μM final concentration was added immediately before imaging the plate. However, when BC400a was used as the substrate, cells were incubated with 60 μM BC400a for 35 min before imaging to allow some time for intracellular conversion of BC400a to C400a. Carbachol and vehicle solutions included a substrate so that there was not a dilution of substrate concentration when solutions were added to the cells in the well.

Ca²⁺ measurements in the presence of fluorescein-induced backgrounds

For tests of Ca²⁺ sensor performance in autofluorescent backgrounds, 0.002 to 10 μM fluorescein (final concentrations) was added to wells by a threefold serial dilution as described above for carbachol CRCs. The BRET ratio (for CalfluxCTN) or intensity (for Fluo-8) measurements were compared between cells in fluorescein-containing medium with cells residing in nonfluoresceinated medium.

Scopolamine CRCs

To determine the effective working concentration of scopolamine (muscarinic receptor antagonist) that would antagonize a 30 μM stimulation by carbachol, we plated CHO cells incubated in 5 μM Fluo-8 AM in 96-well plates in Opti-MEM with serial 10-fold dilutions of scopolamine (50 μM to 0.05 nM and 0 nM). After recording the unstimulated signal in the presence of scopolamine, 30 μM carbachol was injected into each well, and the fluorescence change was recorded.

Effect of fluorescein on recordings of antagonist activity

In the experiments in which the response to scopolamine was determined in the presence of different concentrations of fluorescein (Fig. 6 and fig. S6), CHO cells were plated in 96-well plates in which half of the wells contained 10 μM scopolamine. Then, 30 μM carbachol was injected into each well using the POLARstar OPTIMA plate reader, and the difference between the responses after carbachol addition of ±10 μM scopolamine was recorded. (The 10 μM concentration of scopolamine was determined to be close to the minimum concentration that could produce maximal antagonism of the 30 μM carbachol stimulation. In addition, 10 μM is also the standard concentration of test compounds in initial HTS assays.) Whereas half of wells contained 10 μM scopolamine, all wells also contained serial dilutions of fluorescein (10, 2.5, 0.625, 0.156, 0.039, 0.010, 0.002, and 0 μM). This multifactor paradigm allowed a comparison of fluorescein’s effect on inhibited and noninhibited cells. Identical plate configurations were used for both fluorescent Fluo-8 and BRET CalfluxCTN measurements.

Screening of small-molecule test library with CalfluxCTN

The test library was derived from the Microsource Spectrum Collection (purchased from Microsource Discovery Systems Inc., Gaylordsville, USA), including two compounds that are detectably fluorescent (fig. S7). This library contains a wide range of biologically active and structurally diverse compounds consisting of about 50% drug components, 30% natural products, and 20% other bioactive components, all of which were screened at a concentration of 10 μM. We also included two compounds that are not part of the Microsource Spectrum Collection: the well-characterized Ca²⁺ ionophore ionomycin at 10 μM and a fluorophore-conjugated selective antagonist of the M1 muscarinic receptor, telenzepine, at 0.15 μM (Alexa488-telenzepine) (34).

About 2750 cells in 30 μl of growth medium [DMEM/F-12 medium supplemented with 10% FBS, 1× antibiotic-antimycotic, puromycin (10 μg/ml), and G418 disulfate (800 μg/ml)] were seeded into each well of PureCoat Amine 384-well black/clear flat-bottom plates (Corning no. 356719) and incubated at 37°C for 36 to 40 hours. After reaching 70 to 80% confluency, the cells were rinsed with 80 μl of assay medium (Opti-MEM without phenol red, Gibco no. 11058021, and without FBS) three times, and the growth medium was replaced by 20 μl of assay medium. The cells were incubated at 37°C for an additional 3 to 4 hours.

For the fluorescence assay with Fluo-8, 20 μl of assay medium containing 2.4 μM Fluo-8 AM (AAT Bioquest no. 21083), 0.025% Pluronic F-127 (Invitrogen no. P6867), 0.5% dimethyl sulfoxide (DMSO), and 2.5 mM probenecid (Sigma-Aldrich no. P8761) was added to each well, and the plate was incubated for 1 hour to let the cells take up the Fluo-8 AM and convert it to Fluo-8. The cells were rinsed with 80 μl of assay medium plus 2.5 mM probenecid 3×, and each well was lastly filled with 20 μl of assay medium plus 2.5 mM probenecid. Fluorescence images were acquired in the WaveFront Panoptic plate imager at 37°C as described in a previous section. After an initial baseline fluorescence measurement (~15 s), 20 μl of 20 μM chemical compound from the library dissolved in assay medium plus 2.5 mM probenecid and 0.2% DMSO was robotically injected into each well. Ten microliters of 200 μM carbachol in assay medium plus 2.5 mM probenecid was subsequently injected into each well ~145 s later (final carbachol concentration was 40 μM).

For the BRET (luminescence) assay, 20 μl of assay medium containing 20 μM furimazine and 0.2% ethanol was added to each well. After a 15-min incubation, the assay medium was aspirated down to 20 μl, and luminescence images were acquired by the Panoptic plate imager as described above. Twenty microliters of 20 μM library compound in assay medium plus 10 μM furimazine, 0.1% ethanol, and 0.2% DMSO was injected into each well at ~15 s in the assay, and subsequently, 10 μl of 200 μM carbachol in assay medium containing 10 μM furimazine and 0.1% ethanol was injected into each well ~145 s later. We used furimazine as the substrate in this screening of the test library rather than bc400a because we had already established by the prior experiments that bc400a enables a stable signal over time. Therefore, we used the commercially available furimazine as the substrate in the test library experiment so that we were testing the HTS under conditions and with reagents that are available to all laboratories.

Data analyses

Data from the POLARstar OPTIMA and the WaveFront Panoptic apparatuses were collected as text files. The software program RStudio Integrated Development Environment (IDE), based on the language R (including the packages tidyverse, stringr, scales, and RColorBrewer), was used to load, extract, and visualize all data. For time series experiments, emission counts were normalized by dividing all data points by the baseline emissions before stimulation, when the signals collected from the “blue” and “green” channels were stable. For checkerboard experiments, the time point of peak data was defined as mean of the 5 s when the highest BRET ratio was reached in carbachol-stimulated wells. Z′ factor calculations (32) were performed using the average peak ratio after high-carbachol stimulation compared with the same time point in the vehicle-treated wells in a checkerboard-formatted 384-well plate. The Z′ factor for each plate was calculated as follows:

$$Z\prime =1-\frac{3\left(\text{SDpos}+\text{SDneg}\right)}{\mid \text{MEANpos}-\text{MEANneg}\mid }$$

where SDpos represents the SD of the carbachol-stimulated wells, SDneg represents the SD of the vehicle-treated wells, and MEANpos and MEANneg are the means of the carbachol- and vehicle-treated wells, respectively.

To fit the BRET ratio versus Ca²⁺ response in purified CalfluxCTN in vitro and for carbachol CRCs, we used a nonlinear least squares function in R with the following formula:

$$y=a+\frac{b-a}{1+{\left(\frac{d}{x}\right)}^c}$$

where y is the BRET ratio, x is the Ca²⁺ concentration, a is the minimum value, b is the maximum value, c is the Hill coefficient, and d is EC₅₀. Starting parameters for each model were set by visual inspection of the data and then allowed to converge on the most accurate coefficients.

Data and materials availability:

All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials. The DNA construct encoding Calflux has been deposited with Addgene to make it freely available to other researchers.