Robust calibration and quantification of FRET signals using multiplexed biosensor barcoding

Summary

Förster resonance energy transfer (FRET) between fluorescent proteins (FPs) underpins many genetically encoded fluorescent biosensors for monitoring biochemical activities in live cells. However, the FRET ratio (acceptor-to-donor signal ratio), commonly used as a proxy for FRET efficiency, is highly sensitive to imaging parameters, complicating data interpretation. Using FP-based barcodes, we introduced calibration standards into subsets of cells for normalization of fluorescence signals. Theoretical analysis indicated the need for both high- and low-FRET standards for calibration under different excitation intensities. We validated this prediction using engineered “FRET-ON” and “FRET-OFF” standards, demonstrating that calibrated FRET ratios are independent of imaging conditions. Including donor-only and acceptor-only cells enabled simultaneous determination of FRET efficiency for multiple biosensors. Calibration also restored expected reciprocal changes in donor and acceptor signals, often obscured by imaging fluctuations and photobleaching. Together, our studies introduce a simple strategy for robust, multiplexed FRET biosensor imaging, facilitating cross-experimental and long-term studies.

Graphical abstract

Highlights

• •FRET biosensor signals can be calibrated using high and low FRET standards
• •Imaging standards and sensors in barcoded cells streamline calibration and multiplexing
• •The approach simplifies the determination of actual FRET efficiency
• •Calibration restores reciprocal donor and acceptor trends obscured by imaging drifts

Sensor; Biotechnology; Biophysical Chemistry; Methodology in biological sciences

Introduction

Cellular functions depend on precisely coordinated activities of large numbers of biomolecules in space and time. To understand physiological and pathological processes in cells, it is often necessary to track the spatiotemporal dynamics of molecular activities as well as their responses to perturbations. Genetically encoded fluorescent biosensors, which incorporate fluorescent proteins (FPs) that alter their emission properties or subcellular localization in response to specific biomolecular activities, are powerful tools for monitoring these dynamic processes.^1,2,3,4 Compared to traditional biochemical methods, genetically encoded fluorescent biosensors offer several key advantages. For example, they enable continuous monitoring of molecular activities in live cells and can be directed to specific subcellular compartments, providing precise localized readouts. Biosensors also reveal cell-to-cell variability that might be masked in ensemble measurements. Over the past three decades, a large number of genetically encoded fluorescent biosensors have been developed to detect a wide range of cellular targets ranging from receptors, signaling proteins, metabolites, metals, and physical properties of cells,^5,6 highlighting their versatile use in cell biology. These tools not only advanced our understanding of fundamental cellular processes but also enabled the development of innovative approaches for studying disease mechanisms, drug responses, and the effects of genetic or environmental perturbations in living systems. Biosensor data have also been extensively used in modeling to quantitatively understand the kinetics, regulatory networks, and spatiotemporal dynamics of molecular activities within cells.^{7,8,9,10,11,12}

One of the most commonly used strategy for constructing genetically encoded fluorescent biosensors is based on Förster resonance energy transfer (FRET), a process involving non-radiative energy transfer from a donor fluorophore to a nearby acceptor fluorophore.^13,14 FRET biosensors detect specific molecular activities by undergoing conformational changes that alter the distance or orientation between a donor and acceptor FP, leading to measurable changes in energy transfer efficiency.^15,16,17 For example, many kinase biosensors consist of a substrate specific to the kinase of interest and a binding domain for the phosphorylated form of the substrate, flanked by donor and acceptor FPs. Upon kinase activation, phosphorylation triggers a binding event that induces a conformational change, altering the FRET efficiency, i.e., the fraction of energy transferred from the donor to the acceptor. This change in FRET efficiency serves as a quantitative readout of kinase activity.^{2,18,19,20,21,22,23} In addition, some biochemical events such as the activation of G-proteins involve subtle conformational changes that are cumbersome to detect with other approaches but can be readily monitored in live cells using FRET biosensors.^{24,25,26,27,28,29,30,31}

FRET efficiency can be determined by various means, including spectral imaging, acceptor photobleaching, and donor fluorescence lifetime imaging.³² The most commonly used method, sometimes referred to as sensitized emission FRET, involves measuring the emission of both the donor and acceptor under donor excitation. Ideally, the signal in the acceptor channel should be entirely due to FRET. However, the acceptor can also be directly excited by the excitation beam, albeit at a lower efficiency, and donor fluorescence can bleed through into the acceptor channel. To correct these issues, parameters for signal crosstalk must be determined using donor- and acceptor-only samples imaged at different wavelengths.^33,34 This process is time-consuming, and multiple imaging steps can introduce additional variability. In practice, the acceptor-to-donor signal ratio, or FRET ratio, is often used as a convenient but imperfect surrogate for FRET efficiency. This ratio is influenced by imaging conditions, such as laser intensity and detector sensitivity settings, making it difficult to directly compare results across different imaging sessions. Consequently, most FRET biosensor experiments are confined to a single imaging session of limited duration, typically lasting up to several hours.

Another challenge of genetically encoded fluorescent biosensors is their difficulty in multiplexing due to the limited availability of spectral space.^{35,36,37,38,39} Specifically, many FRET biosensors employ cyan fluorescent protein (CFP) and yellow fluorescent protein (YFP) (or their variants) as the FRET donor and acceptor due to their favorable spectral properties.^32,40 However, the overlap in their emission spectra prevents the simultaneous imaging of multiple FRET biosensors in the same sample. To address this issue, we recently developed a biosensor barcoding method that enables highly multiplexed imaging of biosensors. In this method, cells expressing different biosensors are labeled with distinct pairs of barcoding proteins, which are blue or red FPs targeted to different subcellular locations. These barcoding proteins have spectra separable from those of commonly used biosensors based on GFP, CFP, and YFP. Barcoded cells expressing various biosensors are mixed for simultaneous imaging, and the identity of each biosensor in each cell is established by analyzing the barcode through machine learning models.^41,42,43

We reasoned that multiplexed, calibrated imaging of FRET biosensors can be achieved by introducing calibration standards into a subset of barcoded cells. The fluorescence signal of calibration standards can be used to normalize the FRET ratio of biosensors to compensate for variability in imaging conditions. We carried out theoretical modeling to evaluate this prediction. Our analysis indicated that the normalized FRET ratio slightly decreases with increasing excitation intensity; however, this can be corrected by calibrating against standards with both low and high FRET efficiencies. To test this approach, we generated calibration standards using the CFP-YFP FRET pairs locked in “FRET-ON” and “FRET-OFF” conformations. Experimental results aligned with the theoretical predictions, demonstrating that the calibrated FRET ratio is unaffected by imaging settings. Moreover, simultaneous imaging of CFP and YFP allows for the calculation of actual FRET efficiency. Notably, our calibration method recovered the reciprocal variations in donor and acceptor FRET signals, which are frequently compromised by imaging parameter fluctuations and photobleaching over time. This provided essential validation for the observed biosensor responses. Together, our study introduces a simple and efficient approach for calibrated, multiplexed imaging of FRET biosensors, enabling consistent comparisons across experiments and ensuring reliability for long-term imaging applications.

Results

Theoretical modeling of FRET calibration

Here, we focus on unimolecular FRET biosensors consisting of a single donor and acceptor pair, which is the most commonly used design. We first outline the kinetic theory of FRET^44,45 for the simple case of a single donor and a single acceptor molecule. We assume a two-level system to describe both the donor (D) and acceptor (A) molecules. We use P_X and P_X∗ to denote the probability that molecule species X (here X = D or A) is in the ground or excited state, respectively. Probability conservation of the two-level system implies that P_X + P_X∗ = 1. The time evolution of these probabilities is described by a set of coupled rate equations:

$$\frac{d{P}_{D\ast }}{dt}={\gamma }_D^{ex}{P}_D-\left({\gamma }_D^R+{\gamma }_D^{NR}\right)P_{D\ast }-{\gamma }^{tr}{P}_{D\ast }{P}_A$$ (Equation 1a)

$$\frac{d{P}_{A\ast }}{dt}={\gamma }_A^{ex}{P}_A-\left({\gamma }_A^R+{\gamma }_A^{NR}\right)P_{A\ast }+{\gamma }^{tr}{P}_{D\ast }{P}_A$$ (Equation 1b)

Here, ${\gamma }_X^{ex}$ is the rate of excitation of fluorescent species X initially in their ground state, and ${\gamma }_X^{NR}$ and ${\gamma }_X^R$ denote the rates of excited-state depopulation through nonradiative (e.g., internal conversion) and radiative (i.e., emission of a photon) processes, respectively. The FRET rate from donor to acceptor is represented by γ^tr, which is dependent on the relative orientation and distance between the two fluorescent species. The net energy transfer is also proportional to the probability of the donor being in the excited states as well as the probability that the acceptor is in the ground state. The various electron transitions are summarized in Figure 1A. Numerical simulations of the probabilities of the donor and acceptor in the excited state over time is shown in Figure 1B. The time unit in our simulations is nano-second (ns), which means all the transition rates are in the unit of 10⁹(1/s). The simulated transition rates are of the order of 10⁸(1/s), consistent with those reported earlier.⁴⁵

Figure 1.

Theoretical modeling

(A) Schematic diagram of the various transitions in an FRET system with one donor and one acceptor molecule. D and A denote donor and acceptor molecules, respectively. Asterisk denotes excited species.

(B) Probabilities of the donor and acceptor in the excited state over time. The parameters used in the simulations are ${\gamma }_D^{ex}$ = 0.07, ${\gamma }_D^R$ = 0.5, ${\gamma }_D^{NR}$ = 0.1, ${\gamma }_A^{ex}$ = 0.02, ${\gamma }_A^R$ = 0.45, ${\gamma }_A^{NR}$ = 0.12, and γtr = 0.3.

(C) Ratio of the probability of finding donor in the excited state to that of acceptor in the steady state versus laser intensity I at steady state for varying FRET rate γtr. The parameters used in the simulations are ${\gamma }_D^R$ = 0.5, ${\gamma }_D^{NR}$ = 0.1, ${\gamma }_A^R$ = 0.45, ${\gamma }_A^{NR}$ = 0.12, and extinction coefficients ε_D = 0.1 and ε_A = 0.05.

(D) Calibrated ratio of the same simulation data based on ${\gamma }_1^{tr}$ = 0.001 and ${\gamma }_2^{tr}$ = 1.

For single-photon excitation, the excitation rates of each fluorescent species is given by

$${\gamma }_X^{ex}={\epsilon }_X\left({\lambda }_{ex}\right)I$$ (Equation 2)

where I is the laser irradiance, and ε_X(λ_ex) is the wavelength-dependent extinction coefficient, with λ_ex being the excitation wavelength.

For biosensor applications, continuous wave (CW) light sources are often employed, indicating a constant laser intensity I. We are interested in steady-state solutions, which can be obtained by setting dP_D∗/dt = dP_A∗/dt = 0 in (Equation 1a), (Equation 1b). However, in the presence of FRET, the resultant coupled nonlinear equations cannot be solved analytically. Numerical integration of the differential equations is used to obtain the steady state solutions.

The fluorescent signal S_X is proportional to the number of photons ${\gamma }_X^R{P}_{X\ast }$ emitted per unit time through the radiative transition, as well as the number N of fluorescent molecules:

$$S_X={\kappa }_{X}N{\gamma }_X^R{P}_{X\ast }$$ (Equation 3)

where the species-dependent coefficient κX is determined by the settings of image acquisition. For FRET-based biosensors, we are interested in the ratio of the acceptor to donor fluorescent signals:

$$r=\frac{S_A}{S_D}=\frac{{\kappa }_{A}N{\gamma }_A^R{P}_{A\ast }}{{\kappa }_{D}N{\gamma }_D^R{P}_{D\ast }}=C\frac{P_{A\ast }}{P_{D\ast }}$$ (Equation 4)

Here, the proportional coefficient C depends on the excitation and emission wavelengths and settings of image acquisition but is independent of the laser intensity I or the FRET transfer rate γtr. On the other hand, the ratio of the excited state probability depends nontrivially on both I and γtr.

Figure 1C shows the probability ratio P_A∗/P_D∗ at the steady state as a function of laser intensity I for varying FRET transfer rate γtr. With a small γtr, P_A∗/P_D∗ stayed relatively unchanged with increasing laser intensity. However, a more pronounced dependence on laser intensity is obtained for a larger FRET rate.

Here, we present a calibration method to minimize the dependence of the ratio of acceptor to donor signal on the laser intensity I, while also removing other extrinsic factors related to image acquisition. We select two reference FRET rates, γ₁^tr and γ₂^tr, and define a normalized ratio as

$$R=\frac{\left(S_A/S_D\right)-{\left(S_A/S_D\right)}_{{\gamma }_1^{tr}}}{{\left(S_A/S_D\right)}_{{\gamma }_2^{tr}}-{\left(S_A/S_D\right)}_{{\gamma }_1^{tr}}}=\frac{\left(P_{A\ast }/P_{D\ast }\right)-{\left(P_{A\ast }/P_{D\ast }\right)}_{{\gamma }_1^{tr}}}{{\left(P_{A\ast }/P_{D\ast }\right)}_{{\gamma }_2^{tr}}-{\left(P_{A\ast }/P_{D\ast }\right)}_{{\gamma }_1^{tr}}}$$ (Equation 5)

where the subscript ${\gamma }_2^{tr}$ and ${\gamma }_1^{tr}$ indicate quantities obtained from the two reference FRET coefficients. The calibrated ratios based on the same simulation data are shown in Figure 1D. The results show a negligible dependence on the laser intensity for all FRET rates.

Linear approximation

To gain further insight into the basis of the calibration formula, we considered linear approximation to the steady state solutions of the FRET equation. As demonstrated in Strohhöfer et al.,⁴⁵ the rate-equation modeling based on linear approximation agrees well with the experimental results for a CFP-YFP system. Assuming a small excitation probability for the acceptor P_A∗ ≪ 1, the nonlinear term in (Equation 1a), (Equation 1b) can be approximated as P_D∗P_A = P_D∗(1 − P_A∗) ≈ P_D∗. We first obtain the excitation probability for the donor

$$P_{D\ast }=\frac{{\gamma }_D^{ex}}{{\gamma }_D^{ex}+{\gamma }_D^R+{\gamma }_D^{NR}+{\gamma }^{tr}}$$ (Equation 6)

We consider a simplified situation of weak excitation, i.e., γ_D^ex ≪ γ_D^R, γ_D^NR, and neglect the excitation rate in the denominator. In the absence of FRET, the number of photons emitted per unit time is given by

$${\gamma }_D^R{P}_{D\ast }=\frac{{\gamma }_D^R}{{\gamma }_D^R+{\gamma }_D^{NR}}{\gamma }_D^{ex}=Q_D{\epsilon }_{D}I$$ (Equation 7)

where Q_D = γ_D^R/(γ_D^R + γ_D^NR) is the quantum yield of the donor molecule. The fluorescent signal is thus proportional to the product of the quantum yield and the extinction ratio. Similar expression also applies to the acceptor molecule. The steady-state solution for P_A∗ in the linear approximation is

$$P_{A\ast }=\frac{{\gamma }_A^{ex}+{\gamma }^{tr}{P}_{D\ast }}{{\gamma }_A^{ex}+{\gamma }_A^R+{\gamma }_A^{NR}}=\frac{{\gamma }_A^{ex}}{{\gamma }_A^{ex}+{\gamma }_A^R+{\gamma }_A^{NR}}+\frac{{\gamma }^{tr}{\gamma }_D^{ex}}{\left({\gamma }_A^{ex}+{\gamma }_A^R+{\gamma }_A^{NR}\right)\left({\gamma }_D^{ex}+{\gamma }_D^R+{\gamma }_D^{NR}+{\gamma }^{tr}\right)}$$ (Equation 8)

The ratio of the two excitation probabilities is

$$\frac{P_{A\ast }}{P_{D\ast }}=\frac{{\gamma }^{tr}}{{\gamma }_A^{ex}+{\gamma }_A^R+{\gamma }_A^{NR}}+\frac{{\gamma }_A^{ex}}{{\gamma }_D^{ex}}\frac{{\gamma }_D^{ex}+{\gamma }_D^R+{\gamma }_D^{NR}+{\gamma }^{tr}}{{\gamma }_A^{ex}+{\gamma }_A^R+{\gamma }_A^{NR}}$$ (Equation 9)

Note that ${\gamma }_A^{ex}$ and ${\gamma }_D^{ex}$ are both proportional to the laser intensity I, while ${\gamma }^{tr},{\gamma }_X^R$ and ${\gamma }_X^{NR}$ are independent of I. The first term explains the inverse relationship between P_A∗/P_D∗ and I at high levels of γtr, since ${\gamma }_A^{ex}$ in the denominator is the only variable that is proportional to I, while the second term is less dependent on I due to the presence of ${\gamma }_A^{ex}$ and ${\gamma }_D^{ex}$ in both the numerator and denominator (Figure 1C).

Substitute Equation 9 into the calibrated ratio, we obtain

$$R=\frac{{\gamma }^{tr}-{\gamma }_1^{tr}}{{\gamma }_2^{tr}-{\gamma }_1^{tr}}$$ (Equation 10)

Importantly, this calibrated ratio in the limit of linear approximation is now independent of laser intensity or even the parameters that are related to image acquisition.

Generation of FRET-ON and FRET-OFF calibration standards

Our theoretical analysis indicates that normalizing the FRET ratio using low and high FRET calibration standards yields a value that depends solely on FRET efficiency but not imaging settings. To test this prediction, we generated calibration standards with both low and high FRET efficiencies. Given that the CFP-YFP pair is the most commonly utilized in FRET biosensors, we created two calibration constructs: one locked the CFP-YFP pair in the FRET-active (FRET-ON) configuration and the other in the FRET-inactive (FRET-OFF) configuration.

To generate the FRET-ON construct, we started with a biosensor that undergoes loss of FRET upon target activation. One such biosensor is the CytoFAK biosensor, which contains a focal adhesion kinase (FAK) substrate sequence (Figure 2A).⁴⁶ Activation of FAK through EGF stimulation led to a loss of FRET (Figure 2B), as phosphorylation of the substrate at the tyrosine residue induced a conformational change in the biosensor. Conversely, treatment with the FAK inhibitor VS6063 caused an increase of FRET (Figure 2C). CytoFAK displayed a slight concentration in the nucleus (Figure S1A, top). Interestingly, cells expressing CytoFAK exhibited a lower YFP/CFP ratio in the cytosol compared to the nucleus, suggesting higher FAK activity in the cytosol (Figure S1B, top). Furthermore, FRET changes in response to FAK activation and inhibition were more pronounced in the cytosol (Figures S1C and S1D). To create the FRET-ON construct, we mutated the tyrosine in the FAK substrate to phenylalanine (Y349F) to prevent its phosphorylation by FAK (Figure 2A). FRET-ON exhibited similar cellular distribution patterns as CytoFAK (Figure S1A, bottom). As expected, the YFP/CFP ratio of FRET-ON remained unchanged upon EGF stimulation or FAK inhibition (Figures 2B and 2C). Unlike CytoFAK, the FRET-ON construct displayed a uniformly high YFP/CFP ratio across the cell (Figure S1B, bottom).

Figure 2.

Construction of FRET calibration standards

(A) Construction of FRET-ON calibration standard by Y349F mutation of Cyto-FAK.

(B) Plot of YFP/CFP ratio of CytoFAK and FRET-ON (mean ± SEM, n = 20 cells) in response to 100 ng/mL EGF added at 6 min.

(C) Plot of YFP/CFP ratio of CytoFAK and FRET-ON (mean ± SEM, n = 20 cells) in response to FAK inhibitor VS-6063 over 24 h.

(D) Construction of FRET-OFF calibration standard by removal of the H3 domain of the H3K9me9 (W45A) biosensor.

(E) Plot of YFP/CFP ratio of FRET-OFF and H3K9me3 biosensor (mean ± SEM, n = 20 cells) in response to 5 μM TCP over 48 h.

The FRET-OFF construct is derived from the H3K9me3 biosensor, which exhibits an increase in FRET upon histone H3 lysine-9 trimethylation.⁴⁷ Similar to the CytoFAK-based FRET-ON construct, H3K9me3 utilizes the same donor-acceptor pair (ECFP and YPet). A non-responsive control version of H3K9me3 was previously created by mutating a tryptophan to alanine (W45A) in the HP1 domain.⁴⁷ To generate the FRET-OFF construct, we further modified H3K9me3 (W45A) by removing the histone H3 domain to eliminate its nuclear localization (Figure 2D). While the H3K9me3 biosensor responded to the demethylase inhibitor tranylcypromine (TCP) with an increase in the YFP/CFP ratio over 48 h, FRET-OFF showed a lower basal YFP/CFP ratio and did not respond to TCP (Figure 2E). Together, these results demonstrate that FRET-ON and FRET-OFF maintained stable FRET ratios at high and low levels, respectively, under perturbations.

Relative FRET ratio of FRET-ON and FRET-OFF under different excitation intensity

To validate our theoretical prediction of the FRET ratio’s dependence on excitation intensity (Figure 1C), we examined the relative YFP/CFP ratio of FRET-ON and FRET-OFF under varying excitation intensities by adjusting the laser power. High excitation intensities can deplete ground-state fluorophores, leading to a non-linear relationship between fluorescence and excitation. Therefore, to ensure the excitation remained within the linear range, we first measured the fluorescence signal in cells expressing only CFP while varying the laser power. In our imaging setup, as the laser power for CFP excitation increased from 1% to 100%, the CFP fluorescence signal showed a linear increase (R² = 0.999) (Figures S2A–S2C), indicating that the laser power was within the linear range for CFP excitation.

We used the “biosensor barcoding” technique to simultaneously image FRET-ON and FRET-OFF states under various conditions.^41,42,43 This approach involves creating a set of barcodes using blue or red FPs that are targeted to distinct subcellular locations. The barcodes are spectrally distinguishable from CFP and YFP and are designated by letter-number combinations that indicate their color and location, respectively (Figure 3A). Cells are transfected separately with an FRET construct (e.g., FRET-ON, FRET-OFF, or FRET sensors) along with a unique barcode and then mixed for simultaneous imaging (Figures 3B and 3C). The FRET construct within each cell can be identified based on the associated barcode (Figure 3C). Representative images of cells expressing different barcodes are presented in Figure 3D.

Figure 3.

Simultaneous imaging of FRET-ON and FRET-OFF under different excitation intensity

(A) Spectra of the YFP-CFP FRET pair and FPs used for barcoding cells. Barcodes are made from combinations of BFPs or RFPs targeted to different subcellular locations. Each barcode is designated using a letter-number combination to denote its FP and subcellular location. The spectra are generated using the spectra viewer from FPbase.⁴⁸ The wavelengths of lasers are indicated by the red vertical lines, and the range of emission acquisition for barcodes are indicated by black boxes.

(B and C) FRET-OFF and FRET-ON plasmids were transfected into barcoded cells, and the cells were mixed and imaged for (1) the barcodes and (2) CFP and YFP signals under 458 nm excitation.

(D) Representative images of cells expressing different barcodes. Scale bars, 20 μm.

(E) The YFP/CFP ratio of FRET-OFF and FRET-ON (mean ± SEM, n = 50 cells) measured at varying laser intensities. Detector gains were adjusted for high laser intensities (50%–100%) due to YFP signal saturation.

(F) The YFP/CFP ratio of FRET-ON normalized to the average ratio of FRET-OFF shown in (E).

We simultaneously imaged cells expressing FRET-ON or FRET-OFF under varying intensities of CFP excitation (Figure 3E). The laser power was incrementally increased from 10% to 100%, and the YFP/CFP ratios for FRET-ON and FRET-OFF cells were calculated (Figure 3E). Due to YFP signal saturation at higher laser intensities, detector gain adjustments were necessary for 50% and 100% laser power. The YFP/CFP ratio of FRET-ON was then normalized to that of FRET-OFF (Figure 3F). Consistent with our theoretical modeling (Figure 1C), the normalized YFP/CFP ratio of FRET-ON decreased monotonically with increasing excitation intensity (Figure 3F).

Calibration of FRET biosensor signal under various imaging settings

We next tested whether FRET-ON and FRET-OFF can be used to calibrate FRET biosensors under different imaging settings. To do this, we selected two FRET biosensors, extracellular signal-regulated kinase (ERK) activity reporter (EKAR) and Cyto-FAK, which detect the activities of ERK and FAK, respectively.^22,46 Barcoded cells expressing four FRET constructs (FRET-ON, FRET-OFF, Cyto-FAK, and EKAR; Figure 4A) were mixed and imaged under six settings with varying laser power and detector gains (Figure 4B). Under the same imaging setting, YFP/CFP ratios were largely unaffected by the expression level (indicated by YFP fluorescence) for individual FRET constructs, except in cells with near-saturating YFP signals, which showed reduced YFP/CFP ratios (Figure S3A). We therefore excluded cells with near-saturating YFP signals in any of the six imaging settings from our analysis. As expected, CFP and YFP fluorescence intensities varied widely across different imaging conditions (Figure S3B), resulting in significant variations in YFP/CFP ratios between imaging conditions (Figures 4C and 4D).

Figure 4.

Calibration of FRET biosensors

(A) Schematic of FRET calibration standards (FRET-OFF and FRET-ON) and biosensors (CytoFAK and EKAR) as well as the CFP-YFP pair used in each construct.

(B) The level of laser power and gain settings for CFP and YFP.

(C) YFP/CFP ratio of individual cells expressing the four constructs plotted against the YFP signal.

(D) YFP/CFP (mean ± SEM) across cells expressing the FRET constructs under different imaging settings.

(E) YFP/CFP normalized to the calibration standards (mean ± SEM) of the FRET constructs, with FRET-OFF and FRET-ON set to 0 and 1, respectively, under different imaging settings. The cell numbers for FRET-OFF, FRET-ON, CytoFAK, and EKAR are 10, 8, 11, and 11, respectively. ns denotes no statistical significance (for CytoFAK and EKAR) of S2 to S6 compared with values from S1; ∗ indicates p < 0.05 by two-tailed Student’s t test assuming unequal variances.

Following Equation 5, we calibrated the YPF/CFP ratios by applying a linear scaling that sets FRET-OFF and FRET-ON to 0 and 1, respectively, using the following formula:

$$CalibratedFRETratio=\frac{{\left(\frac{YFP}{CFP}\right)}_{biosensor}-{\left(\frac{YFP}{CFP}\right)}_{FRET-OFF}}{{\left(\frac{YFP}{CFP}\right)}_{FRET-ON}-{\left(\frac{YFP}{CFP}\right)}_{FRET-OFF}}$$ (Equation 11)

Notably, after calibration, the YFP/CFP ratios for CytoFAK were consistent across all imaging settings (Figure 4E). For EKAR, which utilizes a distinct CFP-YFP pair (Figure 4A), the calibrated FRET efficiency exhibited minor variations across settings; however, these differences were not statistically significant (Figure 4E), likely due to the closely similar spectral profiles of the various CFP and YFP variants. These findings indicate that FRET-ON and FRET-OFF can be used for robust and effective calibration of FRET biosensor signals under varying imaging conditions.

Determination of FRET efficiency

We reasoned that simultaneous imaging of cells expressing various FRET constructs, along with cells expressing CFP or YFP alone, could provide a convenient way to estimate the actual FRET efficiency, i.e., the fraction of energy transferred from CFP to YFP. To illustrate this strategy, we simultaneously imaged barcoded cells expressing CFP (ECFP), YFP (YPet), FRET-OFF, FRET-ON, and CytoFAK under 458 and 514 nm excitation, acquiring emission signals in the cyan (458–500 nm) and yellow (507–544 nm) ranges (Figure 5A). The four channels of acquisition are designated as Ch1–Ch4 as shown in Figure 5B.

Figure 5.

Estimation of FRET efficiency

(A) Excitation (Ex) and emission (Em) spectra of CFP and YFP. The wavelengths of lasers are indicated by the red vertical lines, and the range of emission acquisition is indicated by red boxes. The spectra are generated using the spectra viewer from FPbase.⁴⁸

(B) Signals of the four channels normalized to that with the highest signal for cells expressing the four constructs (mean ± SD, n = 20 cells).

(C) Schematic of Ch1-4 signals (normalized to Ch4) for a construct containing one copy each of CFP and YFP, in the absence or presence of FRET.

(D) Plots of predicted (cyan dashed line) and observed FRET ratio vs. FRET efficiency for the mean ± SD of 20 cells expressing each construct (left), as well as individual cells expressing different constructs (right).

(E) Plots of predicted (cyan dashed line) and observed FRET ratio vs. normalized CFP signal (left) and FRET ratio vs. YFP signal (right) for individual cells expressing different constructs.

As expected, CFP and YFP showed maximal signals in Ch1 (458 nm excitation, cyan emission) and Ch4 (514 nm excitation, yellow emission), respectively (Figure 5B). However, CFP exhibited bleed-through into Ch2 (458 nm excitation, yellow emission), and YFP also showed signal in Ch2 due to partial excitation by the 458 nm laser. Ch3 (514 nm excitation, cyan emission) exhibited negligible signal for all constructs. We designate the signal ratio of Ch2 to Ch4 for YFP as Y2/4, and the signal ratio of channel 2 to channel 1 for CFP as C2/1 (Figure 5B).

For a construct containing CFP and YFP in a 1:1 ratio in the absence of FRET, the signal intensity of different channels, normalized to Ch4, can be expressed in terms of the signal ratio of Ch1 to Ch4, which we designate as c (Figure 5C). In the presence of FRET, a decrease in CFP fluorescence is accompanied by a proportional increase in YFP fluorescence. This leads to changes in Ch1 and Ch2 signals that can be expressed with two additional quantities: (1) the FRET efficiency X and (2) a constant of proportionality α, which relates CFP loss to YFP gain (Figure 5C). Note that Ch3 stays at 0.

The values of c, x, and α cannot be directly measured for any individual FRET construct. However, using Ch1 and Ch2 signals for FRET-ON and FRET-OFF, four equations were obtained and could be used to determine the four unknowns c, xOFF, xON, and α. The values of c and α can then be used to calculate x of individual cells using Ch1 signal (see STAR Methods for details of the calculation). Under our imaging settings, we obtained the following values:

$$Y_{2/4}=0.349$$

$$C_{2/1}=0.217$$

$$c=0.581$$

$$\alpha =3.481$$

These values allowed us to determine the FRET efficiency for FRET-OFF, FRET-ON, and CytoFAK (mean ± SD of 20 cells) as:

$$x_{OFF}=7.4\pm 2.1\%$$

$$x_{ON}=47.7\pm 3.2\%$$

$$x_{CytoFAK}=25.8\pm 3.2\%$$

Using the formulas in Figure 5C, we plotted the theoretical relationship between FRET ratio (Ch2/Ch1) and the FRET efficiency (x) (Figure 5D). Our analysis revealed that the FRET ratio becomes increasingly sensitive to changes in FRET efficiency at higher levels, asymptotically approaching infinity as FRET efficiency reaches 100%. Notably, the experimental FRET ratio and FRET efficiency values for the three constructs agreed well with the predicted curve (Figure 5D, left). For each construct, this relationship was also observed at the single-cell level (Figure 5D, right).

The loss of CFP fluorescence due to FRET is evident by plotting the FRET ratio (Ch2/Ch1) against normalized CFP emission (Ch1/Ch4). A good fit was observed between the prediction and measured values of individual cells expressing different FRET constructs (Figure 5E, left), despite the wide range of expression levels of these constructs (Figure 5E, right). Together, these results demonstrate how FRET efficiency can be determined by simultaneous imaging of FRET calibration standards and CFP and YFP constructs using biosensor barcoding.

Calibration of FRET biosensor signal over time

During time-lapse imaging, the fluorescence signals of biosensors can fluctuate due to factors such as variable excitation intensity or photobleaching over time. We reasoned that calibration standards could be utilized to correct for these variations. To test this idea, we monitored the activity of the ERK biosensor EKAR in unstimulated cells. The original ERK biosensor, EKAR, utilized mCerulean and mVenus as its FRET pair,²² which differs from the ECFP-YPet pair used in FRET-ON and FRET-OFF (Figure 4A). Given this difference, it may exhibit a different photobleaching rate. To improve consistency, we replaced the FRET pair in EKAR with ECFP and YPet to create a modified biosensor, EKAR1. We validated the responsiveness of EKAR1 to changes in ERK activity by demonstrating its activation upon treatment with EGF and its inhibition with GDC-0994 (Figure S4).

We simultaneously imaged FRET-ON, FRET-OFF, EKAR, and EKAR1 using barcoded cells under an unstimulated condition over 30 min. While the FRET ratio (YFP/CFP) for FRET-OFF was stable, it showed a gradual decrease for FRET-ON (Figure 6A). The FRET ratio of EKAR1 showed undulating activity, even after calibration with Equation 11 (Figures 6A and 6B). This fluctuating EKAR1 FRET ratio may be explained by the pulsatile ERK activation previously observed using both FRET and translocation-based ERK biosensors.^49,50,51 The ERK pulses are triggered by localized Ras activation on cellular protrusions.⁵² If so, an increase in the FRET ratio should be caused by a rise in YFP and a reciprocal decrease in CFP, and vice versa. However, the YFP and CFP signals for EKAR1 moved in the same direction throughout most of the imaging duration, contradicting this prediction (Figure 6C). This raises the concern that the fluctuating EKAR1 FRET ratio is due to imaging artifacts rather than actual changes in FRET.

Figure 6.

Calibration of FRET biosensor signals to account for photobleaching and imaging parameters over time

(A) YFP/CFP ratio from barcoded cells expressing FRET-OFF, FRET-ON, EKAR1, and EKAR over 30 min. The results for EKAR are shown in Figure S5.

(B) YFP/CFP ratio of EKAR1 calibrated using FRET-ON and FRET-OFF (Equation 11).

(C) Raw CFP and YFP intensity (normalized to initial values) of FRET-OFF, FRET-ON, and EKAR1.

(D) CFP and YFP intensity of FRET-ON normalized to those of FRET-OFF and initial values.

(E) CFP and YFP intensity from EKAR1 signal after applying a two-step calibration to correct for photobleaching and imaging parameters (see text).

(F) Raw YFP and CFP intensity from individual EKAR1-expressing cells.

(G) YFP and CFP intensity from cells in (F) after applying a two-step calibration to correct for photobleaching and imaging parameters. Values in (A–E) represent mean ± SEM of 10 cells each.

Interestingly, while the YFP and CFP signals for FRET-OFF also exhibited fluctuations, they were highly coordinated, resulting in a stable ratio throughout the imaging period (Figure 6C). This fluctuation is likely due to variations in image parameters (e.g., laser power), which would consistently affect the entire imaging field. However, the CFP and YFP signals of FRET-ON, when normalized to those of FRET-OFF, did not remain steady and instead demonstrated increasing and decreasing trends over time, respectively (Figure 6D).

The discordant changes in CFP and YFP signals between FRET-ON and FRET-OFF can be attributed to photobleaching effects. In FRET-OFF, photobleaching results in a decrease in both CFP and YFP signals, regardless of their presence within the same molecule. In contrast, for an FRET-ON molecule, photobleaching YFP alone causes a decrease in the YFP signal accompanied by an increase in CFP, while photobleaching CFP causes a decrease in both YFP and CFP signals. Given the small fraction of CFP and YFP that undergo photobleaching, the likelihood of having both in the same molecule is negligible. Assuming that the same fraction of CFP and YFP is photobleached in both FRET-OFF and FRET-ON, this results in a relative increase of CFP in FRET-ON relative to that of FRET-OFF as well as a relative decrease of YFP. The magnitude of these changes is dependent on the specific photobleaching rates of CFP and YFP.

Based on this analysis, we carried out a two-step calibration: (1) normalizing FP signals to those of FRET-OFF, which corrects for fluctuating image parameters and baseline photobleaching and (2) adjusting for the reciprocal trends in CFP increase and YFP decrease due to photobleaching in the presence of FRET (see STAR Methods for details). Remarkably, after applying these calibrations to YFP and CFP signals for EKAR1, their reciprocal trends became evident (Figure 6E).

We also examined the signals from individual cells. While the raw YFP and CFP signals frequently moved in the same direction (Figure 6F), the calibrated signals consistently exhibited opposing trends, even when the activity was not synchronized across different cells (Figure 6G). The duration of ERK activity pulses in individual cells aligned with those reported previously.^49,50,51 This analysis provided crucial support for the conclusion that changes in the YFP/CFP ratio of EKAR1 are attributable to changes in FRET, rather than imaging artifacts. Collectively, these results demonstrate how including calibration standards can provide critical validation for FRET biosensor imaging experiments.

We applied the same analysis to EKAR in the same imaging experiment (Figure S5). The YFP/CFP of EKAR, as well as the calibrated YFP/CFP showed similar temporal profiles to those of EKAR1, indicating synchronized ERK activity across the cell population (Figure S5A). This synchronization is likely driven by cell-to-cell propagation of ERK activity.^41,50 As in EKAR1, the YFP and CFP signals for EKAR moved in the same direction throughout most of the imaging duration (Figure S5B). However, calibration with FRET-ON and FRET-OFF failed to fully restore the reciprocal changes of CFP and YFP signals in EKAR, both at the population average level (Figure S5C) and in individual cells (Figures S5D and S5E). This finding suggests that FRET-ON and FRET-OFF calibration standards are insufficient to fully correct the reciprocal relationship between donor-acceptor pairs that exhibit different photobleaching rates.

Discussion

While FRET biosensors are versatile and powerful tools for studying biological processes in real time, their ability to provide a comprehensive view of molecular networks over extended time periods is limited by challenges in multiplexing and complex calibration processes. As a result, they are typically used for monitoring individual processes over relatively short durations. In this study, we addressed these limitations by introducing calibration standards in a subset of cells using a recently developed biosensor barcoding method for multiplexed imaging of biosensors.

Our theoretical analysis indicated that the ratio of FRET acceptor to donor fluorescence depends on excitation intensity, particularly at high levels of FRET efficiency. Therefore, calibrating the FRET ratio across different excitation intensities requires standards in both low and high FRET states. We validated this prediction by constructing FRET-OFF and FRET-ON calibration standards and showed that calibrated FRET ratios of biosensors remain consistent regardless of imaging conditions.

Based on our results, we propose a strategy for calibrated, highly multiplexed imaging of FRET biosensors using barcoded cells (Figure 7). In this approach, FRET-ON and FRET-OFF calibration standards, along with various biosensors, are introduced into cells expressing distinct barcoding proteins—spectrally separable fluorescent proteins (e.g., blue fluorescent proteins [BFPs] and red fluorescent proteins [RFPs]) targeted to specific subcellular locations (Figure 7A). These cell populations are then mixed and imaged simultaneously to capture the barcodes and biosensor signals. Cells expressing biosensors and calibration standards are identified based on their barcodes (Figure 7B). The FRET ratios for the calibration standards are then used to normalize those of the biosensors (Figure 7C). This calibration enables direct comparison of FRET ratios for each biosensor across different time points or experimental conditions.

Figure 7.

Strategy for calibrated and highly multiplexed imaging of FRET biosensors

(A) Calibration standards (FRET-OFF and FRET-ON) and biosensor plasmids are introduced into cells expressing different barcodes.

(B) Cells are mixed and imaged for 1) the barcodes; and 2) CFP and YFP under CFP excitation.

(C) The YFP/CFP ratio of calibration standards and biosensors are calculated for images taken at different time points (T1-T3). The YFP/CFP of FRET-ON and FRET-OFF calibration standards are used to normalize those of the biosensors using Equation 12.

We also demonstrated a simple strategy for determining actual FRET efficiency, which typically requires sequentially imaging the donor alone, the acceptor alone, and at least two constructs with varying FRET levels under conditions of donor or acceptor excitation.^53,54,55 This multi-step imaging process is labor-intensive and poses challenges for routine implementation. Additionally, as illustrated in Figure 6C, imaging parameters can fluctuate over time, which may introduce errors in FRET efficiency calculations. We demonstrated that by including cells expressing YFP and CFP alone, alongside the calibration standards, the actual FRET efficiency for each biosensor can be easily determined (Figure 5). Although the mathematical basis of our method is similar to previous techniques, the use of barcoded cells for simultaneous imaging of multiple constructs significantly streamlines the process. This approach enables the concurrent determination of FRET efficiencies for numerous FRET biosensors under consistent imaging conditions, reducing variability and improving experimental efficiency.

The inclusion of calibration standards also provides crucial validation that observed changes in the FRET ratio are indeed due to changes in FRET, rather than artifacts arising from fluctuations in imaging parameters. This is demonstrated by the observation that uncorrected YFP and CFP signals often shift in the same direction, masking the expected reciprocal patterns. Only after correcting for imaging parameters and photobleaching using the calibration standards were these expected reciprocal signals restored.

Our approach may also be applicable to ratiometric biosensors that do not rely on FRET. Examples include the ATP sensor iATPsnFR and the NADH/NAD sensor Peredox, which link a circularly permuted GFP (cpGFP) sensing module with an RFP for normalizing the expression level.^56,57 For such biosensors, calibration can be achieved using a single GFP-RFP fusion protein as a standard to normalize the relative fluorescence of the two fluorophores under varying imaging conditions.

In conclusion, our method offers a simple and effective solution for the calibrated and multiplexed measurement of biosensors, facilitating consistent comparisons across experiments and enabling robust long-term imaging studies for a wide range of applications.

Limitations of the study

In this study, calibration was tested only for CFP-YFP FRET biosensors. In addition to CFP-YFP, other FRET pairs have also been employed in biosensor construction.³⁹ For example, GFP-RFP or far-red FRET pairs have been developed to extend the spectral range and address various experimental requirements.^32,58,59 However, the barcoding strategy used in this study is only designed to be compatible with biosensors in the CFP-to-YFP emission range.⁴¹ A recently developed barcoding system can overcome this limitation to accommodate spectrally diverse biosensors.⁶⁰ Additionally, since FPs differ in photobleaching rates, distinct calibration standards may be needed for biosensors using different FRET pairs, especially during frequent imaging over extended period. While our calibration standards are robust to perturbations known to affect the biosensors they are derived from, other changes in the cellular environment could potentially impact the FRET efficiency of FRET-ON, particularly if they induce minor shifts in donor-acceptor orientation. This possibility should be considered if significant changes are seen in FRET-ON but not in FRET-OFF.

Resource availability

Lead contact

Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Chuan-Hsiang Huang (chuang29@jhmi.edu).

Materials availability

All unique reagents developed in this study are available upon request to the lead contact.

Data and code availability

• •All pertinent data are included in the main text and supplemental information. Further details are available from the lead contact upon request.
• •The code used to simulate the FRET rate equations is available in the supplemental information (fret-kinetics.cpp).
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work was supported by funding from R01GM136711 (to C.-H.H.), Cervical Cancer SPORE P50CA098252 Career Development Award (to J.-M.Y.), Pilot Project Award (to C.-H.H.), the Sol Goldman Pancreatic Cancer Research Center (to C.-H.H.), and the Johns Hopkins Discovery Award (to C.-H.H.). Additionally, the purchase of the Zeiss LSM 780 and 880 confocal microscopes by Johns Hopkins University School of Medicine Microscope Facility was made possible through NIH grants S10OD016374 and S10OD023548, respectively.

Author contributions

Conceptualization, C.-H.H. and J.-M.Y.; methodology, J.-W.W., J.-M.Y., C.-C.C., G.A., S.W., Y.X., G.-W.C., and C.-H.H.; investigation, J.-W.W., J.-M.Y., C.-C.C., G.A., S.W., Y.X., G.-W.C., and C.-H.H.; formal analysis, J.-W.W., J.-M.Y., C.-C.C., S.W., Y.X., and C.-H.H.; data curation, J.-W.W. and J.-M.Y.; software, J.-W.W., J.-M.Y., C.-C.C., G.-W.C., and C.-H.H.; writing – original draft, J.-W.W., J.-M.Y., and C.-H.H.; writing – review and editing, J.-W.W., J.-M.Y., C.-C.C., G.A., S.W., Y.X., G.-W.C., and C.-H.H.; funding acquisition, C.-H.H. and J.-M.Y.; supervision, C.-H.H. and J.-M.Y.; resources, C.-H.H. and J.-M.Y.

Declaration of interests

The authors declare that they have a patent application based on this work (PCT Application No. PCT/US24/38295).

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
EGF	Sigma-Aldrich	Cat#E9644
Tranylcypromine (TCP)	MilliporeSigma	Cat#616431
Critical commercial assays
Gateway BP Clonase II	Invitrogen	Cat#11789020
Gateway LR Clonase II	Invitrogen	Cat#11791020
GenJet In Vitro DNA Transfection Reagent	SignaGen Laboratories	Cat#SL100489
Lenti-X Concentrator	Takara Bio	Cat#631231
Experimental models: Cell lines
Human: HEK293T	ATCC	CRL-3216; RRID:CVCL_0063
Human: HeLa	ATCC	CCL-2; RRID:CVCL_0030
Human: SiHa	ATCC	HTB-35; RRID:CVCL_0032
Oligonucleotides
Forward primer to add attB1 for EKAR1, CytoFAK, FRET-ON: GGGGACAAGTTTGTACAAAAAAGCAGGCTTCACCATGG TGAGCAAGGGCGAGGAGC	This paper	N/A
Reverse primer to add attB2 and stop codon for EKAR1, CytoFAK, FRET-ON: GGGGACCACTTTGTACAAGAAA GCTGGGTCTCATTTGTACAATTCATTCATACCCTCGG	This paper	N/A
Forward primer to add attB1 for H3K9me3, H3K9me3 (W45A), FRET-OFF: GGGGACAAGTTTGTACAAAAAAGCAG GCTGCCACCATGTCTAAAGGTGAAGAATTATTC ACTGGTGTTGTCCC	This paper	N/A
Reverse primer to add attB2 and stop codon for H3K9me3, H3K9me3 (W45A), FRET-OFF:GGGGACCACTTTGTACAAGAAA GCTGGGTCTAGGCGGCGGTCACGAACTCC	This paper	N/A
Forward primer for CytoFAK Y349F mutation: GGGTTCTGAGACCGACGACTTCGCTGAGA TCATCGACGAGG	This paper	N/A
Reverse primer for CytoFAK Y349F mutation: CCTCGTCGATGATCTCAGCGAAGTCGTCG GTCTCAGAACCC	This paper	N/A
Recombinant DNA
EKAR	Addgene	Cat#18679
CytoFAK	Addgene	Cat#78300
H3K9me3 biosensor	Addgene	Cat#120802
H3K9me3 (W45A) biosensor	Addgene	Cat#120808
FRET-ON	This paper	N/A
FRET-OFF	This paper	N/A
EKAR1	This paper	N/A
pDONR221 vector	Invitrogen	Cat#12536017
pLenti-CMV-Neo destination vector	Addgene	Cat#17392
pMD2.G	Addgene	Cat#12259
psPAX2	Addgene	Cat#12260
Software and algorithms
Zen Microscopy Software	Zeiss	https://www.zeiss.com/microscopy/us/home.html
ImageJ/Fiji	Schindelin et al.61; Schneider et al.62	https://imagej.net/software/fiji/

Experimental model and study participant details

HeLa, HEK293T, and SiHa cells were purchased from ATCC and grown at 37°C, 5% CO₂ in DMEM high glucose medium (Gibco, #11965092) supplemented with 10% FBS (Corning Cellgro, 35-010-CV), 1 mM sodium pyruvate (Gibco, #11360070), and 1X nonessential amino acids (Gibco, #11140076). All imaging experiments were performed in HeLa cells (except SiHa for Figure S1), while HEK293T cells were used as the host to generate lentiviruses.

Method details

Theoretical simulation

Details of the theoretical simulation are provided in the main text. The coupled rate equations were solved using the standard fourth-order Runge-Kutta method⁶³ with a constant step size dt = 0.01 nanosecond. The time unit in the simulations is 1 nanosecond, which means the various transition rates in the rate equations are measured in units of 10⁹/sec. The parameters used in this study are not intended to represent specific FRET donor and acceptor fluorophores. Rather, they were selected to reflect experimentally observed ranges of relevant quantities. For example, according to Equation 2, the ratio ${\gamma }_D^{ex}/{\gamma }_A^{ex}$ (0.07/0.02=3.5) is equivalent to εD/εA , which closely matches the ratio calculated using known extinction coefficients for ECFP and YPet at 458 nm excitation ((32,500∗0.91)/(104,000∗0.08)=3.55, where the values were obtained from FPbase⁴⁸). Also, the transition rates for CFP and YFP are on the order of 10⁸/sec, consistent with previously reported values.⁴⁵

Chemical reagents

The EGF stock solution was prepared by dissolving EGF (Sigma-Aldrich, E9644) in 10 mM acetic acid to a final concentration of 1 mg/ml. TCP (tranylcypromine) stock solution was prepared by dissolving TCP (Millipore Sigma, 616431) in DMSO to a final concentration of 5 mM. All stocks were stored at -20°C.

Plasmid construction

Plasmids for transient expression were obtained from Addgene: EKAR (#18679),²² CytoFAK (#78300),⁴⁶ H3K9me3 biosensor (#120802),⁴⁷ and H3K9me3 (W45A) biosensor (#120808).⁴⁷ Using gene synthesis (GenScript), EKAR1 was generated by replacing mCerulean and mVenus in EKAR with ECFP and YPet, respectively. For lentiviral vector versions of these biosensors, the Gateway cloning system was used to first generate the entry vectors and then the destination vectors. To create entry vectors, we designed a pair of PCR primers with an attB1 site at the 5’ end and an attB2 site at the 3’ end to amplify the biosensor genes using EKAR1, CytoFAK, H3K9me3 biosensor, and H3K9me3 (W45A) biosensor DNA as templates. The PCR products were then cloned into the pDONR221 vector (Invitrogen, #12536017) using Gateway BP Clonase II (Invitrogen, #11789020). The resulting entry vectors were recombined with the pLenti-CMV-Neo destination vector (Addgene, #17392)⁶⁴ to generate lentiviral expression vectors using Gateway LR Clonase II (Invitrogen, #11791020). The resulting plasmids were used for both transient transfection and as transfer plasmids for lentivirus production.

Cell transfection

For short-term imaging experiments, HeLa cells were transiently transfected using GenJet™ In Vitro DNA Transfection Reagent (Ver. II) following the manufacturer's protocol (SignaGen Laboratories, #SL100489). Briefly, 4x10⁵ cells were seeded in one 3.5 cm culture dish the day before transfection. On the day of transfection, the culture medium was replaced with 1 mL fresh complete medium without antibiotics. The GenJet-DNA complex was prepared by diluting (1) 1 μg of DNA in 50 μL serum-free DMEM, and (2) 3 μL GenJet reagent in 50 μL serum-free DMEM in two separate tubes, and then mix the two together by adding GenJet into DNA. The complex was allowed to form at room temperature for 15 min, and then added dropwise onto the cells. Target protein expression in cells can be observed 24 to 48 h post transfection.

Lentivirus production and transduction

4×10⁶ HEK293T were seeded in a 10-cm dish with DMEM supplemented with 10% FBS and allowed to attach overnight. Cells were transfected with 1.25 μg of pMD2.G (Addgene, #12259), 3.75 μg of psPAX2 (Addgene, #12260), and 5 μg of lentiviral plasmids using GenJet™ In Vitro DNA Transfection Reagent (Ver. II), and the cell culture medium was replaced with fresh medium 5 h after transfection. The supernatants containing lentiviral particles were harvested 24 h after transfection and filtered through a 0.45 μm PVDF filter (Millipore, #SLHVM33RS). Lentiviral particles were concentrated 100 times with Lenti-X Concentrator (Takara Bio, #631231) following the manufacturer's protocol. Virus stocks were aliquoted and flash frozen in liquid nitrogen for long term storage at -80’C.

For longer imaging experiments that lasted over 24 hr, HeLa cells were co-transduced with lentiviruses encoding barcodes and biosensors. Cells were seeded in 6-well plates with DMEM supplemented with 10% FBS and allowed to attach overnight. Cells were transduced with lentivirus in the presence of 8 μg/ml polybrene (Millipore, #TR-1003-G). Cells were incubated at 37°C and 5% CO₂ for 48 h before mixing with other transduced cells for imaging experiments.

Microscopy

Cells were transfected with plasmids encoding the barcodes and CFP-YFP constructs (i.e. calibration standards or biosensors) in separate wells. Prior to imaging experiments, cells transfected with different barcode/CFP-YFP constructs combinations were harvested, mixed, and reseeded. Live cell imaging was performed on either a Zeiss 780 confocal microscope equipped with a spectral detector (for imaging barcoding proteins) and a motorized stage (for capturing multiple viewing fields). The detailed imaging protocol has been described in our earlier publication.⁴² Briefly, for the barcodes, BFP images were acquired under 405 nm excitation and spectral images (550-700 nm) of the red FPs under 561 or 633 nm excitation. For the CFP-YFP constructs, CFP and YFP images were acquired under 457 nm excitation.

For FRET efficiency determination, signals from four channels were acquired: 1) Ch1: 458 nm excitation/458–499 nm emission; 2) Ch2: 458 nm excitation/507–544 nm emission; 3) Ch3: 514 nm excitation/458–499 nm emission; and 4) Ch4: 514 nm excitation/507–544 nm emission. The excitation/emission separation was achieved on the Zeiss LSM780 using the main beam splitter (MBS) 458/514, which reflects the 458 nm and 514 nm laser lines, corresponding to reflection bands around 450–470 nm and 505–525 nm, while transmitting emission in the adjacent regions (470–500 nm and >530 nm) based on the data from FPbase.⁴⁸ For emission detection, dedicated bandpass filters were applied. This configuration ensures that excitation light was effectively blocked at the dichroic level, while the bandpass filters confined the detection windows to the defined emission ranges.

Image analysis

Images of barcodes and CFP-YFP constructs were processed and analyzed with NIH ImageJ and Fiji.^61,62 Cells were manually segmented to identify individual cells. The barcode for each cell was determined based on the color and subcellular localization of the barcoding proteins, allowing for the identification of the specific CFP-YFP constructs expressed in each cell. Pixel values for YFP and CFP were measured, and the YFP/CFP ratios were calculated using Microsoft Excel. Microsoft Excel was also used for statistical analysis and to generate graphs.

Determination of FRET efficiency

The analysis is for constructs containing CFP and YFP in a 1:1 ratio. In the absence of FRET, all the Ch1 signal originates from CFP, while all the Ch4 signal originates from YFP. Suppose that the signal ratio of Ch1 to Ch4 is c. Without loss of generality, we can normalize all signals to that of Ch4. Thus, Ch1 and Ch4 signals are c and 1, respectively. We can calculate Ch2 signal as the sum of those from CFP and YFP, which equals C2/1·c+Y2/4·1 (Figure 5C).

In the presence of an FRET efficiency x, CFP decreased by to c(1-x)by definition. The YFP fluorescence increases by an amount proportional to x, leading to a change in Ch2 signal, which can be expressed as C2/1·c(1-x)+Y2/4·(1+αx), where α is a constant of proportionality and depends on the imaging setting and the fluorophores (Figure 5C).

Using the measurements in Figure 5B,

$$Y_{2/4}=0.349$$

$$C_{2/1}=0.217$$

For FRET-OFF,

$$\text{Ch}1=c\left(1-x_{OFF}\right)=0.538$$

$$\text{Ch}2=C_{2/1}·c\left(1-x_{OFF}\right)+Y_{2/4}·\left(1+\alpha x_{OFF}\right)=0.555$$

For FRET-ON,

$$\text{Ch}1=c\left(1-x_{ON}\right)=0.303$$

$$\text{Ch}2=C_{2/1}·c\left(1-x_{ON}\right)+Y_{2/4}·\left(1+\alpha x_{ON}\right)=0.995$$

Solving the system of equations gave:

$$c=0.581$$

$$x_{OFF}=0.0737$$

$$x_{ON}=0.477$$

$$\alpha =3.481$$

Using the values c and α, we calculated the values of x for individual cells using the following equation of Ch1 signal:

$$\text{Ch}1=c\left(1-x\right)$$

We obtained the following x values (mean ± SD, n = 20 each) for cells expressing FRET-OFF, FRET-ON, and CytoFAK:

$$x_{OFF}=7.4\pm 2.1\%$$

$$x_{ON}=47.7\pm 3.2\%$$

$$x_{CytoFAK}=25.8\pm 3.2\%$$

The FRET ratio (Ch2/Ch1) can be calculated as the following:

$$\begin{array}{cc}FRETratio& =\left(C_{2/1}·c\left(1-x\right)+Y_{24}·\left(1+\alpha x\right))/c(1-x\right)\\ & {=C}_{2/1}+Y_{2/4}\left(1+\alpha x\right)/c\left(1-x\right)\end{array}$$ (Equation 12)

Calibration for CFP and YFP in time-lapse images

Calibration for CFP and YFP in time-lapse images (Figure 6) involves the following steps. (Note that all FP signals are normalized to the first time point in this analysis.) First, the CFP and YFP signals are normalized to those of FRET-OFF to correct for fluctuating imaging parameters (e.g., excitation intensity) and baseline photobleaching. Second, in the presence of FRET, normalized CFP and YFP signals in the first step show characteristic increase and decrease over time due to photobleaching (see Figure 6C and the related text). This effect arises from the increase in donor (CFP) emission upon acceptor (YFP) photobleaching, and is proportional to the level of energy transfer (FRET efficiency).³² Note that the FRET ratio only deviates slightly from a linear relationship with FRET efficiency within the range covering our FRET constructs (Figure 5D). Therefore, we can reasonably assume that the effect of photobleaching scales linearly with the calibrated FRET ratio defined in Equation 11. This allows us to calibrate CFP and YFP signals using the following equations:

$$Calibrated{YFP}_{biosensor}=\frac{\frac{{YFP}_{biosensor}}{{YFP}_{FRET-OFF}}}{1+(\frac{{YFP}_{FRET-ON}}{{YFP}_{FRET-OFF}}-1)\left(calibratedFRETratio\right)}$$ (Equation 13)

$$Calibrated{CFP}_{biosensor}=\frac{\frac{{CFP}_{biosensor}}{{CFP}_{FRET-OFF}}}{1+(\frac{{CFP}_{FRET-ON}}{{CFP}_{FRET-OFF}}-1)\left(calibratedFRETratio\right)}$$ (Equation 14)

Note that for FRET-ON and FRET-OFF, the calibrated YFP and CFP values equal 1, since their FRET level is expected to remain unchanged over time.

Quantification and statistical analysis

Image analysis was performed with ImageJ/Fiji^61,62 and Microsoft Excel. Cells were manually segmented, and the mean signal over the cell area was used to calculate YFP/CFP ratios. FRET ratios were calibrated using FRET-ON and FRET-OFF standards (Equation 11), and FRET efficiency was calculated from simultaneous imaging of CFP-only, YFP-only, and calibration constructs (Equation 12). For time-lapse experiments, signals were normalized to FRET-OFF and further corrected for donor–acceptor photobleaching effects (Equations 13 and 14). Statistical comparisons used two-tailed Student’s t-tests, with p < 0.05 considered significant. Statistical details of experiments can be found in figure legends.

Published: October 11, 2025