Critical comparison of FRET-sensor functionality in the cytosol and endoplasmic reticulum and implications for quantification of ions

Abstract

Genetically-encoded sensors based on fluorescence resonance energy transfer (FRET) are powerful tools for quantifying and visualizing analytes in living cells, and when targeted to organelles have the potential to define distribution of analytes in different parts of the cell. However, quantitative estimates of analyte distribution require rigorous and systematic analysis of sensor functionality in different locations. In this work, we establish methods to critically evaluate sensor performance in different organelles and carry out a side-by-side comparison of three different genetically encoded sensor platforms for quantifying cellular zinc ions (Zn²⁺). Calibration conditions are optimized for high dynamic range and stable FRET signals. Using a combination of single-cell microscopy and a novel microfluidic platform capable of screening thousands of cells in a few hours, we observe differential performance of these sensors in the cytosol compared to the ER of HeLa cells, and identify the formation of oxidative oligomers of the sensors in the ER. Finally, we use new methodology to re-evaluate the binding parameters of these sensors both in the test tube and in living cells. Ultimately, we demonstrate that sensor responses can be affected by different cellular environments, and provide a framework for evaluating future generations of organelle-targeted sensors.

INTRODUCTION

Fluorescent sensors are powerful tools for visualizing and quantifying ions, metabolites, and other species in cells, offering the potential to define the concentration and spatial distribution of such species. But realizing this potential requires systematic and careful characterization of sensor platforms in the complex environment of a living cell and in different subcellular locations. The need for a robust analytical framework for comparing sensor performance and defining sensor functionality in different cellular locations is illustrated by comparing the three families of genetically-encoded FRET-based sensors, Zap, eCALWY, and eZinCh, which have been applied to define Zn²⁺ pools within the cell. Each family is engineered by fusing fluorescent proteins (FP) to Zn²⁺ responsive protein domains. The Zap family links a donor CFP to an acceptor citrine YFP through the zinc-binding domain of the yeast transcription factor Zap1.¹ In the eCALWY family of sensors, cerulean FP and citrine YFP were mutated to recognize each other in the apo state of the sensor, and the metal-binding domains of Atox1 and WD4 were engineered to coordinate Zn²⁺.² In the eZinCh family a Zn²⁺ coordination site was directly engineered into an interface between the β-barrels of cerulean and citrine, which were connected by a flexible linker.³ A notable benefit of genetically-encoded sensors is that they can be targeted to organelles by fusing small peptide signaling sequences to the sensors. Members of these families of sensors have been targeted to different cellular compartments including: cytoplasm, mitochondria, Golgi apparatus, nucleus, and endoplasmic reticulum.^1,3–6 Application of multiple platforms for measuring analytes in cells can increase confidence in the robustness of both the tools and their measurements. These three sensor platforms have been used in the cytosol to estimate labile Zn²⁺ within an order of magnitude, 82–600 pM.^7,8 In contrast, when the same sensors were targeted to the ER of HeLa cells the estimates of labile Zn²⁺ ranged widely from 0.9–1600 pM.^1,3,6 Qualitatively, ZapCY1 is more saturated in the cytosol than in the ER, while eCALWY-4 displays the opposite saturation pattern and eZinCh-2 is about equally saturated in both the ER and the cytosol.^1,3,6 It’s possible the chemical environment of the ER (pH, crowding, or redox state) differentially affects sensors, impacting sensor functionality and accuracy of quantifications. Defining whether the labile Zn²⁺ concentration is higher in the ER than in the cytosol is an important fundamental question in Zn²⁺ biology in order to establish whether the ER might serve as a store of Zn²⁺ that could be mobilized for cellular signaling. More generally, establishing a framework for comparing different sensor platforms and establishing performance metrics in organelles compared to the cytosol may be important for sensors of other analytes such as Ca²⁺, Mg²⁺, or other metal ions.

In this work, we establish a framework for critically comparing ratiometric FRET-based sensors in an organelle of interest by comparing ZapCY1, eCALWY-4, and eZinCh-2 in the ER and cytosol of HeLa cells. We had two main goals in this study: first, outline methods to systematically assess sensor performance, and second, shed light on relative Zn²⁺ levels in the cytosol and ER. We show that calibrations must be optimized to obtain accurate quantification of analytes, and that sensors can perform very differently in the cytosol versus organelles. We applied recently developed microfluidic technology to screen the response of localized sensors to Zn²⁺ over thousands of cells to ensure the trends observed in tens of representative cells hold over large populations⁹. Additionally, we reevaluated the K_d of the sensors in vitro and compared to in situ titrations of the sensors in the ER to better understand how the environment of the ER affects Zn²⁺ binding. Taken together the data indicate that the ER of HeLa cells is depleted of labile Zn²⁺ compared to the cytosol.

RESULTS AND DISCUSSION

Identifying Optimal Calibration Conditions.

Accurate calibration of sensor response is essential for using sensors to estimate analyte levels. Additionally, it is important to critically evaluate and identify optimal response conditions, which may differ in different cellular compartments. In situ calibration of Zn²⁺ sensors involves first measuring the FRET ratio (R) when the cell is at rest (R_rest). Next excess membrane-permeable Zn²⁺-specific chelator, N,N,N′,N′-tetrakis (2-pyridylmethyl) ethylenediamine (TPEN), is added to measure the FRET ratio of the apo sensor (R_apo), followed by addition of excess Zn²⁺, ionophores, and/or permeabilization reagents to measure the FRET ratio of Zn²⁺ replete sensor (R_bound). Stable and maximal sensor responses to the two extrema of Zn²⁺ concentrations are essential for accurately determining the resting saturation of the sensor.

To optimize one in situ calibration protocol that yielded saturating, stable R_bound responses of all three sensors in the ER and cytosol, we tested different concentrations of Zn²⁺, pyrithione, and the inclusion of a membrane-permeabilizing agent, saponin. Our goal was to maximize the dynamic range and stability of the R_bound signal for each sensor (Supplementary Tables S1–2). As shown in Figure 1 and Supplementary Figure S1, under some calibration conditions, addition of saturating concentrations of Zn²⁺ led to an unstable signal characterized by an increase, followed by sharp decrease in the FRET ratio. To define the stability of the R_bound signal, we calculated the percent change in the R_bound signal over time (Supplementary Tables S1 and S2). We determined that addition of 2 nM buffered Zn²⁺, 0.75 μM pyrithione, and 0.001% saponin led to large, stable responses of all three sensor platforms in both the cytosol and ER (Figure 1, Supplementary Figure S1, Supplementary Tables S1 and S2). Increased concentrations of pyrithione and Zn²⁺ or exclusion of saponin resulted in sub-optimal sensor responses that would lead to systematic error in estimated fractional saturation and Zn²⁺ concentrations. As shown in the subsequent section, achieving a stable R_bound was essential for clear estimation of sensor fractional saturation and dynamic range.

Figure 1.

Optimization of zinc conditions for ER (a-c) and NES (d-f) sensors: ZapCY1 (green), eZinCh-2 (red), eCALWY-4 (blue). Arrows indicate washing of cells followed by addition of solutions containing 100 μM ZnCl₂ + 5 μM pyrithione (grey lines), 2 nM buffered Zn²⁺ + 0.75 μM pyrithione (black lines, not done for NES sensors (d-f)), or 2 nM buffered Zn²⁺ + 0.75 μM pyrithione + 0.001% (w/v) saponin (lines in color) to cells treated for 40 minutes with 50 μM TPEN. Raw FRET ratios are provided in Figure S1.

Fractional Saturation of the Sensors.

Using the new calibration conditions, each sensor was calibrated in both the ER and cytosol (Supplementary Figure S2) of HeLa cells. The stability of the FRET ratios and the maximal response of the sensors in each condition allow for clear quantification of the resting fractional saturation and dynamic range (defined as R_bound/R_apo for ZapCY1 and eZinCh-2 and R_apo/R_bound for eCALWY-4) of the sensors in the two cellular locations (Table 1). The fractional saturation describes the percent of sensor that is bound to Zn²⁺ under resting conditions in a particular organelle and is proportional to the concentration of labile Zn²⁺ such that higher fractional saturation suggests higher levels of labile Zn²⁺. The dynamic range is the maximal change in signal upon Zn²⁺ binding and defines the sensitivity of a sensor to small differences in Zn²⁺ concentration.^4,10 The data in Table 1 reveal that two sensor platforms (ZapCY1 and eCALWY-4) show decreased fractional saturation in the ER compared to cytosol, suggesting lower levels of Zn²⁺ in the ER compared to cytosol. One sensor platform (eZinCh-2) showed similar fractional saturation in both locations, but slighter higher saturation, suggesting higher Zn²⁺, in the ER. However, it should be noted that the overall lower dynamic range of eZinCh-2 and eCALWY-4 in the ER compared to the cytosol makes them more prone to error.⁴ Sensor calibrations under previously described conditions generally featured a lower dynamic range and rapidly changing, unstable R_bound signals, which could lead to erroneous estimation of fractional saturation (Supplementary Figure S3, Table S1–2). Given the agreement of ZapCY1 and eCALWY-4 sensors, coupled with the high dynamic range, the stability of the responses, and the consistency of the fractional saturation of ZapCY1, these results suggest that the concentration of labile Zn²⁺ in the ER is lower than in the cytosol.

Table 1.

Fractional saturation and dynamic range of ER- and cytosol-targeted Zn²⁺ sensors.

	ZapCY1		eZinCh-2		eCALWY-4
	Fractional Saturation	Dynamic Range	Fractional Saturation	Dynamic Range	Fractional Saturation	Dynamic Range
ER	4.7 ± 0.4%	2.55 ± 0.02	40 ± 4%	1.47 ± 0.04	21 ± 2%	1.32 ± 0.04
Cytosol	100%	2.62 ± 0.05	33.2 ± 0.8%	2.06 ± 0.06	39.6 ± 0.6%	1.83 ± 0.04

High Throughput Measure of FRET Response.

A weakness of microscopy-based calibrations of FRET responses is that a limited number of cells can be assayed in any given experiment. We recently introduced a microfluidic cytometer capable of measuring FRET ratios for sensor states (R_apo, R_bound), as well as FRET changes in response to systematic addition of ligands within a defined time window from hundreds of ms up to 10 sec, in individual cells expressing sensors in a high throughput manner.^9,11 This cytometer permits measurement of variability in FRET states and FRET responses in thousands of cells to assess variability at the population-level. Here, we use this cytometer to compare cytosolic and ER-targeted sensors for all three sensor platforms. A distinct advantage of this platform is the ability to introduce reagents in a controlled manner that doesn’t vary from cell to cell or experiment to experiment. However, one limitation is that the response time window is limited by device geometry and the pair matching algorithm⁹. Therefore, we only assess a sensor’s ability to respond within a short time window (7.5 sec), which doesn’t correspond to the full sensor response (Supplementary Figure S3 and S4). Given these considerations, we use the platform to define the initial FRET ratios and variability of ratios in the apo state, and examine the responsiveness of sensors upon addition of Zn²⁺, as opposed to the magnitude of response (i.e. dynamic range), within the defined time window.

Figure 2a and 2b show histograms of single cell FRET populations at FRET 1 and FRET 2 for ER- and cytosol-targeted sensors, respectively. Since FRET populations did not exhibit a normal distribution, median based statistics were used to assess the center and heterogeneity of the cell populations. The robust coefficient of variation (RCV) was used to describe the width of the distribution, which serves as an indicator of the heterogeneity of a sensor in the apo state. The median and RCV values for each sensor in each compartment are presented in Supplementary Table S3. Comparison of these values for each sensor allowed us to define whether there were differences in the FRET state in different compartments. ZapCY1 showed similar RCV in the two locations (11.1% and 9.3% in ER and cytosol, respectively) suggesting comparable sensor heterogeneity, and a small, but significant shift in the median FRET 1 ratio when localized to the ER (1.01 versus 0.92 in the ER and cytosol, respectively). The eCALWY-4 sensor showed similarly broad FRET 1 distributions in the ER and cytosol (22.8% and 26.5%, respectively), indicative of increased sensor heterogeneity in both locations compared to ZapCY1, and a significant shift in the median FRET ratio from 0.97 (ER) to 1.55 (cytosol). The eZinCh-2 sensor yielded the tightest FRET 1 distribution in the cytosol (RCV = 5.1%), indicating minimal heterogeneity, but the broadest distribution in the ER (RCV = 32.4%). For eZinCh-2 the FRET 1 median ratio shifted from 1.42 in the ER to 1.03 in the cytosol. Combined, these data reveal that localization to the ER can introduce significant heterogeneity into the apo state of a sensor.

Figure 2:

Microfluidic analysis of FRET sensor response. FRET ratio histograms for HeLa cells expressing ER (a) or cytosol (b) targeted sensors indicating FRET 1 and FRET 2 cell populations. Corresponding pairmatched FRET ratio histograms showing FRET response to pyrithione and Zn²⁺ for ER (c) and cytosol (d) targeted sensors. Inset scattergrams show FRET response on the single cell level. Time delay between FRET 1 and FRET 2 for all data shown is 7.5 seconds. Sample size for each plot is 6500 cells.

This platform also enabled us to measure sensor responses to added analyte in a short time window and identify sub-populations of cells with sensors that are non-responsive sensor or that respond significantly differently than the average. Figure 2c and 2d show the pair-matched single cell response in the form of a histogram (FRET 2/ FRET 1), and the corresponding scattergram that illustrates the FRET change in individual cells (inset). All of the sensors gave rise to responsive populations (Figure 2c–d), but the overall shape of the scattergrams varies from sensor to sensor and is not univariate as the histograms may suggest. This is most exaggerated in ER-eCALWY-4 (Figure 2c, bottom panel), in which the histogram appears broad and unimodal, but the scattergram shows additional sub-populations indicating variability in the kinetics of the response from cell to cell. Comparison of scattergams for ER-targeted versus cytosolic sensors revealed a greater degree of heterogeneity in responsiveness for all ER sensors. While some of this heterogeneity may derive from increased variability in the apo FRET state, some may also arise from variable Zn²⁺ uptake kinetics in different cells. These studies reveal that all three sensor platforms are, to varying degrees, sensitive to some feature of the ER environment in a way that alters their response. Both the in situ calibration experiments and microfluidic analysis indicated that the response of all three sensors was diminished in the ER compared to the cytosol.

Analysis of Sensor Behavior in Cells.

To explore the possible origins of heterogeneity and diminished response in the ER sensor, localization and folding were analyzed by microscopy. Cells expressing ER-ZapCY1 or ER-eCALWY-4 exhibit fluorescence in a tubular network characteristic of the ER (Figure 3a–b). Many cells expressing ER-eZinCh-2 also have proper sensor localization (Figure 3c), but approximately 25% display bright puncta that are not typical of ER structure (Figure 3d), suggesting perturbation of the sensors themselves as well as ER morphology. Perturbation of ER structure by ER-eZinCh-2 occurred regardless of the localization signal used (Supplementary Figure S5).

Figure 3.

Measurement of localization to and aggregation in the ER. ER-ZapCY1 (a) and ER-eCALWY-4 (b) localize properly to the ER. Some cells expressing ER-eZinCH-2 have proper localization (c) but approximately 25% of cells exhibit bright puncta (d) regardless of signal sequence used (Figure S4). Western blot (e) of sensors in the ER under non-reducing and reducing conditions (denoted by – or + DTT) reveals that all sensors form DTT-sensitive oligomers.

Another possible explanation for the issues observed with ER-targeted sensors could be the formation of misfolded sensors or non-functional aggregates in the ER lumen. Fluorescent proteins have been shown to form disulfide-linked oligomers when expressed in the ER.¹² Additionally, all three sensors use Zn²⁺-binding domains with cysteine residues that could potentially form aberrant disulfide bonds. Cells expressing ER-targeted Zn²⁺ sensors were analyzed by immunoblot to determine the degree of sensor oligomerization (Figure 3e). Under non-reducing conditions, all three sensors ran as high-molecular weight species that were not present under reducing conditions. The molecular weight of the higher-running bands suggests the formation of dimers and trimers. These oligomers were not present when the sensors were expressed in the cytosol (Supplementary Figure S6a). While all three sensors oligomerize in the ER, the immunoblot analysis indicated that ER-eZinCh-2 was the most sensitive to oxidation. The extent of oxidation of ER-eZinCh-2 was not dependent on sensor concentration, as even cells with low levels of sensor expression displayed oxidative oligomers (Supplementary Figure S6b).

Taken together with microscopy images that reveal perturbation of ER morphology, these data indicate that eZinCh-2 is poorly suited to the oxidizing environment of the ER. The formation of oligomers may help explain the reduced dynamic range and variable FRET ratios of eZinCh-2 in the ER (Supplementary Table S1). The solvent-exposed, Zn²⁺-binding cysteine residues engineered on the barrel of the fluorescent proteins in eZinCh-2 may be particularly susceptible to oxidation. The formation of intramolecular or intermolecular disulfide bonds between fluorescent proteins could lead to aggregated sensor molecules locked in a variety of FRET states. The oligomers detected by immunoblot may be related to the puncta observed in images of ER-eZinCh-2. However, it is possible that the oligomers formed by these sensors are not always readily detected by imaging.

In vitro Characterization of Sensors.

In order to use sensors to estimate Zn²⁺ concentrations, the apparent dissociation constants, K_d’s, of the sensors must be determined. K_d’s are typically quantified by purifying and titrating the sensors in vitro, and fitting the resulting data using the model developed by Grynkiewicz et al¹³. This model assumes a linear change of the emission intensity at both the donor FP and FRET emission wavelengths across all relevant analyte concentrations. Because this assumption is incorrect for many ratiometric sensors, different K_d’s will be determined when the same dataset is plotted as the ratio of FRET emission over donor FP emission versus the inverse ratio.^8,14 Two models have been developed to determine more precise K_d’s for ratiometric sensors. One method for fitting the in vitro data is to normalize the donor and/or FRET emission data using the emission intensity at the isosbestic point.⁸ However, sensors that bind their analyte with cooperativity will not have an isosbestic point, so this method cannot be universally applied. A second protocol outlined by Pomorski et al.¹⁴ calibrates the emission intensity data at both wavelengths yielding more precise K_d’s, and can be applied to all sensors. This model was applied in this study.

To collect in vitro data, the sensors were titrated with a series of Zn²⁺ buffers at pH 7.4 and the intensities of the fluorescence emission were collected for the donor FP (λ₁ = 481 nm) and the acceptor FP (λ₂ = 529 nm). The ratio of the emission wavelengths (λ₂/λ₁) and inverse ratio (λ₁/λ₂) were plotted as a function of Zn²⁺ concentration (Supplementary Figure S7) and fit to determine the K_d of each sensor. Fits yielded K_d values within 1.7%. ZapCY1 can bind two Zn²⁺ so the data were fit with a model including the Hill coefficient (n) that gave a K_d = 17 pM and n = 0.47. This value indicates a slightly weaker sensor affinity for Zn²⁺ than the previously reported 2.5 pM.¹ Both eCALWY-4 and eZinCh-2 have one Zn²⁺ site and were fit with a model that excluded the Hill coefficient to give K_d = 164 pM and 103 pM, respectively. These values indicate that the sensors bind Zn²⁺ with a higher affinity at pH 7.4 than pH 7.1 (K_d(eCALWY-4) = 630 pM, K_d(eZinCh-2) = 1000 pM), following the trend seen in these sensors’ dissociation constants with increasing pH.^3,6

Because ZapCY1 is fully saturated in the cytosol it cannot be used to estimate the cytosolic concentration of Zn²⁺. Using K_d values and calibration data, the cytosolic Zn²⁺ concentrations were estimated to be 42 ± 2 pM (using eZinCh-2) and 133 ± 3 pM (using eCALWY-4). It should be noted that these numbers are the average and standard error when these equations are applied on a cell by cell basis, and do not include error from the K_d measurements which would increase the uncertainty in these numbers. In the ER, ZapCY1 estimates the concentration of labile Zn²⁺ to be 0.20 ± 0.04 pM, while ER-eCALWY-4 estimates a concentration of 49 ± 6 pM. ER-eZinCh-2 gives an estimate of labile Zn²⁺ in the ER of 70 ± 10 pM.

To better understand how oxidation might affect Zn²⁺ binding to the sensors in the ER, each sensor was purified and the FRET ratios in response to Zn²⁺ in the presence of tris(2-carboxyethyl)phosphine (TCEP), a reducing agent, and diamide, an oxidizing agent, were measured (Supplementary Table S4). All of the sensors responded to Zn²⁺ under reducing conditions. However, under oxidizing conditions the sensors were fluorescent but essentially non-responsive to Zn²⁺. ZapCY1 appeared to be fixed in a high FRET state, with a FRET ratio similar to the R_bound. eZinCh-2 was fixed in a FRET state intermediate between R_bound and R_apo. Finally, eCALWY-4 exhibited an intermediate FRET ratio and responded slightly to Zn²⁺ but in the opposite direction of reduced sensor. In all cases, the presence of fluorescent but non-responsive sensor in cells may explain the decreased dynamic range of sensors in the ER of cells, as only a portion of the signal measured would be due to responsive FRET sensor.

Characterization of Sensor Affinity in Cells.

To determine if the oxidizing environment of the ER affects Zn²⁺ binding to the sensors, we estimated the apparent K_d of the sensors for Zn²⁺ directly in the ER. Previously, in situ titrations such as these were used to determine that the apparent K_d of a Zn²⁺ sensor in the cytosol closely matches the K_d determined with purified protein.¹⁵ However, cysteine residues are known to undergo oxidative modifications that interfere with Zn²⁺ binding.¹⁶ Furthermore, since immunoblots of the sensors indicate the formation of disulfide-linked oligomers and all three sensors feature oxidation-sensitive cysteine residues, it is possible that the environment of the ER might alter the binding properties of the sensors and result in systematic errors in measurements of Zn²⁺ concentration.

In situ titrations of ER-ZapCY1, ER-eZinCh-2, and ER-eCALWY-4 reveal that the apparent K_d in the ER is not significantly different than the K_d measured for purified protein. For the titrations, cells were treated with 50 μM TPEN and 1 μM thapsigargin to reach a stable R_apo, then treated with the optimized reagent conditions to reach a stable R_zinc. The sensors were tested with buffered Zn²⁺ solutions ranging from 2.5 pM to 2 nM labile Zn²⁺. When saponin was excluded from the R_zinc solution, the sensors were unable to respond quickly enough for accurate measurements (Supplementary Figure S8). After addition of pyrithione, saponin, and buffered Zn²⁺, cells were monitored until a stable signal was achieved.

The data were normalized to the TPEN-treated R_apo signal to visualize the magnitude of response to different buffered Zn²⁺ solutions (Figure 4, and raw data for these curves are presented in Supplementary Figures S9–11). Data from in situ titration experiments with all three sensors are plotted together as dynamic range versus Zn²⁺ concentration (Figure 4m). For ZapCY1 the sensor appears halfway saturated at 16 pM labile Zn²⁺, which is consistent with the in vitro K_d. However, the steepness of this binding curve suggests that the cooperativity Zn²⁺ binding to ZapCY1 might be different in cells than in the test tube. Because of the low dynamic range of ER-eZinCh-2 and ER-eCALWY-4 the halfway saturation points of these sensors are more difficult to discern, but both are shifted in the direction of higher Zn²⁺ concentrations (39–72 pM) consistent with their higher in vitro K_d values. While the sensors undergo some oxidative modification and experience reduced dynamic range in the ER, the actual binding interaction between the sensors and Zn²⁺ ions does not appear to be significantly perturbed. Therefore, the in vitro measurements of K_d are still relevant in the ER.

Figure 4.

In situ titrations of ER-targeted zinc sensors ZapCY1 (a, d, g), eZinCh-2 (b, e, h), and eCALWY-4 (c, f, i). Arrows marked (1) indicate addition of 50 μM TPEN + 1 μM thapsigargin to cells. Arrows marked (2) indicate washing out of TPEN and addition of buffered Zn²⁺ + 0.75 μM pyrithione + 0.001% (w/v) saponin to cells. Cells were treated with 6.6 pM (a-c), 16 pM (d-f), or 192 pM (g-i) buffered Zn²⁺ solutions. Plots of dynamic range for ER-ZapCY1, eZinCh-2, and ER-eCALWY-4 over Zn²⁺ concentration suggest that Zn²⁺ binding by the sensors in not drastically altered in the ER (j). Each data point is the average dynamic range and standard error of the mean from a-i. Data for 2 nM Zn²⁺ data point is from Figure 2. Raw FRET ratios and additional data are provided in Figures S8–10.

Important Considerations for Applying FRET sensors to Subcellular Environments.

In this study we systematically examined the performance of three different Zn²⁺ sensor platforms to provide insight into the functionality of sensors in different environments within the cell. We demonstrate the importance of identifying optimal calibration conditions for accurately assessing the fractional saturation and Zn²⁺ concentration. Our finding that inclusion of saponin and use of low concentrations of buffered Zn²⁺ lead to more stable R_bound signals and larger dynamic ranges suggest that flooding cells with high concentrations of ions can yield sub-optimal calibrations. These reagents did not affect the pH or redox state of the ER (Supplementary Figures S12 and S13). This optimization of reagents has broad implications for calibration of other genetically encoded ion sensors, such as those for Ca²⁺, Mg²⁺, and Cu¹⁺ where unstable bound signals are often observed.^17–19

We also demonstrate that sensor performance can be adversely impacted by the oxidizing environment of the ER which can lead to formation of protein oligomers and reduced dynamic range. Although all three sensor platforms showed some degree of perturbation in the ER, as suggested by formation of disulfide linked oligomers and reduced dynamic range, the platforms were differentially affected. eZinCh2 was the most strongly perturbed (dynamic range reduced from 2.06 to 1.47, increase in variability in FRET ratio as measured by RCV from 5.1% to 32.4%, and perturbation of ER morphology in 25% of cells), perhaps due to surface exposed cysteine residues. eCALWY-4 was also perturbed, but to a lesser extent (dynamic range reduced from 1.83 to 1.32, but no change in variability in FRET ratio as measured by RCV: 26.5% vs 22.8% in cytosol and ER, respectively). Finally, ZapCY1 was the least perturbed in the ER environment (dynamic range reduced from 2.62 to 2.55, variability in FRET ratio increased from 9.3% to 11.1% in the cytosol versus ER, respectively). Despite some degree of perturbation in the ER environment, the estimated K_d values for each sensor closely matched the in vitro measurements. Combined, these results suggest that eCALWY-4 and ZapCY1 are more suitable than eZinCh-2 to estimate relative amount of ER Zn²⁺, and of the two sensor platforms ZapCY1 has a higher dynamic range and less evidence of perturbation.

When calibrated using the new conditions identified here, ZapCY1 and eCALWY-4 both have lower fractional saturation in the ER than in the cytosol. Since the apparent K_d of the sensors for Zn²⁺ was not altered in the ER, this decrease in fractional saturation of the sensors supports the conclusion that labile Zn²⁺ levels in the ER are lower than cytosolic levels. However, it is important to note that our study does not exclude the possibility that ER Zn²⁺ could be higher in different types of cells or that changes in cellular state due to signaling could generate a pool of labile ER Zn²⁺. The higher dynamic range of ER-ZapCY1 allows for more accurate quantification and detection of small changes in labile Zn²⁺ in the ER.

This study sought to establish performance metrics for genetically encoded FRET sensors, demonstrating the importance of optimal in situ calibration conditions, measurement of accurate K_d values using the method of Pomorski et al¹⁴, comparison of K_d values in vitro and in organelle-targeted locations, and use of a new microfluidic platform to assess sensor performance and variability in thousands of cells. We reveal that simply targeting a sensor to a particular organelle is not sufficient to guarantee effective performance. Instead, it is important to rigorously examine how performance might be affected by the organelle environment. Our study reveals that different sensor platforms are differentially impacted by organelle environments, suggesting the need for approaches to systematically optimize sensor performance in the desired subcellular location.

METHODS

Details of chemicals, cloning, cell culture, microscope settings, microfluidic experiments, immunoblots, protein purification and in vitro characterization are provided in the Methods section of Supporting Information.

Reagent optimization experiments.

HeLa cells (n ≥ 3 cells for every experiment) were treated with 50 μM TPEN (cytosolic sensors) or 50 μM TPEN with 1 μM thapsigargin (ER sensors) for 40 min prior to imaging. R_apo data was collected, then cells were washed with phosphate, calcium, and magnesium-free HEPES-buffered HBSS, pH = 7.4 to remove the chelate and then treated with pyrithione and Zn²⁺ with or without saponin. Data was normalized to R_apo by dividing the FRET ratio (R) throughout the experiment by the average R_apo value (R/R_apo). Normalized data is presented as the average normalized FRET ratio of at least three cells.

Calibration and in situ titration experiments.

To collect R_resting data for calibration experiments, cells (n ≥ 3 cells for every experiment) were imaged in phosphate-free HEPES-buffered HBSS, pH 7.4 to prevent Zn²⁺ precipitation. To collect R_apo data, cells were treated with 50 μM TPEN (cytosolic sensors) or 50 μM TPEN with 1 μM thapsigargin, (ER sensors, thapsigargin was previously shown to help deplete ER of Zn²⁺¹) until a stable signal was achieved. Cells were then washed with phosphate, calcium, and magnesium-free HEPES-buffered HBSS, pH = 7.4 to remove the chelate and then treated with pyrithione and Zn²⁺ with or without 0.001% (w/v) saponin. For in situ titrations, data was normalized to R_apo for each of the sensors as follows: ER-ZapCY1 (R–R_apo) or ER-eCALWY (1–(R_apo–R)). Normalized data is presented as the average normalized FRET ratio of at least three cells.

Data Analysis.

All imaging data were analyzed in MATLAB (Mathworks). Images were background corrected by drawing a region of interest (ROI) in a blank area of the image and subtracting the average fluorescence intensity of the background ROI from the average intensity of each cell. FRET ratios for each cell were calculated by dividing the background-corrected YFP FRET intensity by the background-corrected CFP intensity ((I_{cellular FRET} – I_{background FRET})/(I_{cellular CFP} – I_{background CFP})). Unless indicated otherwise, error bars are standard error of the mean.