2.1.

Spectral Unmixing of FRET Samples

For an FRET sample containing free donors at concentration $C_d$ and free acceptors at concentration $C_a$ as well as the paired donor at concentration $C_D$ and acceptor at $C_A$, the emission spectra should be a linear combination of six contributions from the free donor, paired donor, free acceptor, and paired acceptor, including direct excitation and sensitized emission, as well as BG. Similar to the Iem-spFRET method we recently developed,¹² we can sort these fluorescence components according to these emission characteristics of donor and acceptor as well as BG, respectively

2.2.

$\mathsf{\rho }$ Factor

The measured emission fluorescence intensity spectra $F_D^{i,\mathrm{ref}}$ ($\lambda$) and $F_A^{i,\mathrm{ref}}$ ($\lambda$) for the cells exclusively expressing donors or acceptors, which is essential for predeterming the spectral fingerprints of donor and acceptor, are

The emission spectra of fluorophores in fluid solvents were independent of the excitation wavelength.²⁰ We can define a $\rho$ factor to denote the relationship between two molar extinction coefficient ratios of acceptor to donor at two excitations (${\lambda }^2$ and ${\lambda }^1$)

2.3.

Formulas of IIem-spFRET

Considering ${\delta }^i$ [Eq. (3)] at two different excitations and $\rho$ factor [Eq. (6)], we can obtain the FRET formulas including the apparent FRET efficiency (${\mathrm{Ef}}_D$) and the ratio ($R^t$) of total acceptor concentration $C_A^t$ to total donor concentration $C_D^t$

2.4.

Predetermination of γ Factor

After $\rho$ factor is predetermined, ${\gamma }^i$ can be easy predetermined using a tandem construct with the same donor and acceptor fluorophores. In this case, we define the $R^t=C_A^t/C_D^t=n$. According to Eq. (10), we obtain ${\gamma }^1$

3.1.

Reagents and Plasmids

Cerulean and Venus-kras plasmid constructs were purchased from Addgene Company (Cambridge, Massachusetts). The FRET-standard constructs, including C32V (Cerulean-32-Venus, Addgene plasmid 29396), CVC (Cerulean-5-Venus-5-Cerulean, Addgene plasmid 27788), VCV (Venus-5-Cerulean-5-Venus, Addgene plasmid 27788), and VCVV (Venus-5-Cerulean-5-Venus-6-Venus, Addgene plasmid 27789), were kindly provided by the Vogel Lab (National Institutes of Health, Bethesda, Maryland). Plasmid DNA of CFP-Bax and YFP-Bax was kindly provided by Dr. Prehn.⁴ Plasmid DNA of 18AA was kindly supplied by Professor Kaminski.⁷ Staurosporine (STS) was purchased from Sigma-Aldrich Co. LLC (Santa Clara).

3.2.

Cell Culture and Transfection

Huh7 and HepG2 cells line were obtained from the Department of Medicine, Jinan University (Guangzhou, China). Cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco, Grand Island) containing 10% fetal calf serum (FCS; Sijiqing, Hangzhou, China) at 37°C in a humidified atmosphere containing 5% ${\mathrm{CO}}_2$. For transfection, Huh7 or HepG2 cells were cultured in DMEM containing 10% FCS in 35-mm glass dish with a density $4\times {10}^4\text{cells}/\mathrm{ml}$. After 24 h, all plasmids were transfected into cells for 24 to 48 h when the cells reached 70% to 90%. Turbofect^TM (Fermentas Inc., Glen Burnie, Maryland) was used as a transfection reagent.

3.3.

Microscope–Spectrometer Platform

FRET measurements were implemented on a user-built MSM platform that consists of a widefield fluorescence microscope (Axio Observer, Carl Zeiss, Oberkochen, Germany) with a $40\times /1.3\mathrm{NA}$ oil objective and a miniature fiber optic spectrometer (QE65 Pro, Ocean Optics, Florida).¹⁸ The MSM platform, equipped with a metal halide lamp (X-Cite 120, Carl Zeiss, Oberkochen, Germany) and AxioVision Rel. 4.8 (Carl Zeiss, Oberkochen, Germany) software for CCD operation as well as SpecSuite (Ocean Optics, Florida) software for spectra detection, has two detection channels: a microscopic imaging channel with a CCD camera (AxioCam MRm, Carl Zeiss, Oberkochen, Germany) for guiding cells and a spectral detection channel with a QE65 Pro spectrometer for collecting spectra from the guided cells. The spectra channel was accorded to the middle of CCD imaging channel. The detection area ($S$) of the spectrometer is $2.826\times {10}^5{\mu \mathrm{m}}^2$ that accords to a $68.99{\mu \mathrm{m}}^2$ of the sample area in this report.

The illumination intensity can be attenuated in five discrete steps ($T_1=0\%$, 12.5%, 25%, 50%, and 100%), and another neutral density filters controller with six discrete steps ($T_2=2\%$, 20%, 40%, 60%, 80%, and 100%) can be used for the same purpose. In our study, the excitation optical filters were BP405/20 (Ex405 nm) (BP $405/20\times$ Exciter, Chroma) and BP 436/20 (Ex436 nm) (BP 436/20, Carl Zeiss, Oberkochen, Germany). 455-nm dichroic mirror (455DCLP Dichroic, Chroma) and emission long pass filter of LP455 (455DCLP Dichroic, Chroma) were used as emission control. The spectral detection range is 460 to 620 nm. Each count($\lambda$) or $E(\lambda )$ at the emission wavelength $\lambda$ is related to the photos from about a 0.761-nm wavelength range. In addition, excitation bandpass filter of BP 500/25 (Ex500 nm), primary dichroic mirror FT 515, and emission bandpass filter of BP 535/30 (Carl Zeiss, Oberkochen, Germany) were used for the acceptor photobleaching.

4.1.

$\rho$ Factor and Spectral Fingerprints

The lifetime of the metal halide lamp we used for excitation is much longer than 2000 h, and thus our excitation is very stable. Furthermore, spectral responsivity of our MSM platform was carefully calibrated by using a standard lamp (LS-1-CAL, Ocean Optics, Florida), just as described previously.¹⁸ Our MSM platform has the same calibration curve [$K(\lambda )=1/\eta (\lambda )$] of spectral sensitivity at an interval of 1 month (Appendix A1), further demonstrating the stability of the MSM platform.

BG spectra were measured from 18 blank living Huh7 cells without expressing FPs with 405 and 436 nm excitations, respectively [Figs. 1(a) and 1(b)]. Figure 1(c) shows the images of cells separately expressing Cerulean and Venus with 405 nm (${\lambda }^1$) and 436 nm (${\lambda }^2$) excitations, respectively. Figures 1(d) and 1(e) show the spectra of the cell indicated by the red circles in Fig. 1(c) and the average BG spectra in Figs. 1(a) and 1(b). According to Eqs. (7) and (8), the excitation absorption coefficients $v_D=0.2738$ are calculated from 460 to 560 nm wavelength range [Fig. 1(d)] and $v_A=0.2090$ calculated from 520 to 550 nm wavelength range [Fig. 1(e)]. Statistical results from 14 cells expressing Cerulean and 13 cells expressing Venus were $v_D=0.2773\pm 0.0064$ and $v_A=0.2051\pm 0.0322$, respectively. Thus, $\rho =1.3520$ by Eq. (6). The normalized spectra of Cerulean and Venus from above cells with 436 nm excitation and the normalized BG spectra calculated from 18 blank living Huh7 cells were used as the spectral fingerprints of Cerulean ($e_D$) and Venus ($e_A$) as well as BGs ($e_B^1$ and $e_B^2$), respectively [Fig. 1(f)].

Fig. 1

Spectral fingerprints of Cerulean [$e_D(\lambda )$] and Venus [$e_A(\lambda )$] as well as BG [$e_B^i$ ($\lambda$), $i=1$ or 2] inside living Huh7 cells. (a) and (b) Spectra of 18 blank cells without FPs at 405 nm (a) ($i=1$) or 436 nm (b) ($i=2$) excitation under the same measurement conditions. (c) Images of cells separately expressing Cerulean and Venus with 405 or 436 nm excitation. Scale bar: $10\mu \mathrm{m}$. (d) and (e) Emission spectra of Cerulean (d) and Venus (e) inside the living cell indicated by the red circles in (c) and the average of spectra of 18 cells in (a) and (b), as well as the corresponding $v_D$ and $v_A$. (f) Normalized spectral fingerprints of donor ($e_D$) and acceptor ($e_A$) as well as BG ($e_B^1$ and $e_B^2$) at two excitations.

Although careful calibration of the measurement platform is necessary for IIem-spFRET and Iem-spFRET methods, this step is not required for every FRET measurement due to the stability of our platform for at least 3 months. In addition, because of the independence of ${\gamma }^i$ on the illumination intensity and emission-spectral transmission as well as FP concentration,^6,12,20 predetermining the $\rho$ factor can be a one-off event for given FPs for at least 3 months on our platform. However, most imaging systems do not have this stability, and so investigators had better not skip the calibration step.¹⁸ In addition, as the two excitations used here share the same emission channel (${\eta }^1(\lambda )={\eta }^2(\lambda )$, the calibration of our system is very simple and can be completed in a few seconds.¹⁸

We found that the normalized spectra of Cerulean ($e_D$) and Venus ($e_A$) in living cells with 436 nm excitation were consistent with those with 405 nm excitation.¹⁸ In fact, much of the literature assume that the spectra of FPs in living cells are independent of excitation wavelength.^{6,14,15,18,20–23} Therefore, Eq. (6) should be applicable to the FPs in living cells.

As the extinction coefficient of Venus at 405 nm is very low and 436 nm excitation can effectively excite both Cerulean and Venus,²⁴ so the measured normalized spectra of Cerulean and Venus with 436 nm excitation were used as their spectral fingerprints. Once $e_D$, $e_A$, and $e_B^i$ are predetermined, we need not measure them for subsequent quantitative FRET measurement. In addition, acquisition range of the spectral fingerprints must overlap well with the emission spectra of donor and acceptor to insure the accurate measurement of Iem-spFRET.¹⁴ In this study, the spectral ranges covering $\sim 96.3\%$ and $\sim 96.8\%$ of Cerulean and Venus [Fig. 1(f)] emission spectra fulfill this requirement.

Although we introduce $\rho$ factor into IIem-spFRET, there is no need to add any external references for determining the $\rho$ factor. In addition, $\rho$ factor and spectral fingerprints can be determined at the same time [Figs. 1(d) and 1(e)]. We here predetermine the $\rho$ factor by measuring the excitation absorption coefficients ($v_D$ and $v_A$), which can be obtained from the fluorescence intensities of cells exclusively expressing donor and acceptor with two excitations [Eqs. (7) and (8)]. The excitation intensity ratio ($I^{1,\mathrm{ref}}/I^{2,\mathrm{ref}}$) between two excitations must be constant for donor-only and acceptor-only reference, respectively. In fact, the $I^{1,\mathrm{ref}}/I^{2,\mathrm{ref}}$ is constant for a stable excitation light source; we thus only need to measure both $v_D$ and $v_A$ under the same instrumental conditions for determining the $\rho$ factor. In addition, $v_D$ and $v_A$ are constant for a given donor–acceptor pair and optical system and are also independent of the emission wavelength [Eqs. (7) and (8)]. In order to improve the accuracy of measurement, we chose the spectra of Cerulean in 460 to 560 nm wavelength range to calculate $v_D$ [Fig. 1(d)] and the spectra of Venus in 520 to 550 nm wavelength range to calculate $v_A$ [Fig. 1(e)] for the determination of $\rho$ factor (1.3520). Meanwhile, we also used the spectra of both Cerulean and Venus in the same wavelength range of 520 to 550 nm [Figs. 1(d) and 1(e)] to calculate the $v_D$ and $v_A$, and obtained consistent $\rho$ factor (1.3477).

4.2.

Extinction Coefficient Ratios of Venus to Cerulean in Living Huh7 Cells

To predetermine the extinction coefficient ratio (${\gamma }^i$) of Venus (V) to Cerulean (C) in living Huh7 cells on our platform, which is necessary for IIem-spFRET measurement, we transfected Huh7 cells with C32V, and the representative cell images with 405 ($i=1$) or 436 nm ($i=2$) excitation are shown in Fig. 2(a). The corresponding spectra and unmixing components by using Eq. (2) are shown in Figs. 2(b) and 2(c), and ${\delta }^1=1.4168$ and ${\delta }^2=1.1797$. Substituting these parameters and the predetermined $\rho =1.3520$ into Eq. (11), we calculate ${\gamma }^1=0.3298$ and ${\gamma }^2=0.4459$, where ${\varphi }_D=0.62$, ${\varphi }_A=0.57$,^25,26 and $n=1$. The statistical ${\gamma }^1$ and ${\gamma }^2$ values from 14 live Huh7 cells by IIem-spFRET were ${\gamma }^1=0.3055\pm 0.0108$ and ${\gamma }^2=0.4130\pm 0.0147$ [Table 1 and Fig. 2(d)]. Meanwhile, we also used the PbFRET method^7,27 to measure the FRET efficiency ($E$) of C32V and then determined the ${\gamma }^i$ values by using the Iem-spFRET method [Fig. 2(d) and Appendix A].

Fig. 2

${\gamma }^i$ values of Venus to Cerulean in living Huh7 cells. (a) Fluorescence images of representative cells expressing C32V at 405 or 436 nm excitation. Scale bar: $10\mu \mathrm{m}$. (b) and (c) Spectral unmixing of the corresponding spectra by IIem-spFRET with 405 (b) or 436 nm (c) excitation in (a), where ${\alpha }^1=0.1315$, ${\beta }^1=0.0897$, $c^1=0.0032$, ${\alpha }^2=0.5300$, ${\beta }^2=0.4493$, and $c^2=0.0008$. (d) Statistical ${\gamma }^1$ and ${\gamma }^2$ values from at 14 living cells by using Iem-spFRET and IIem-spFRET methods, respectively.

Table 1

γ values calculated by Iem-spFRET or IIem-spFRET method.

The real values of ${\varphi }_D$ and ${\varphi }_A$ (quantum yield) are the biggest unknown in our calculation due to the dependence of ${\varphi }_D$ and ${\varphi }_A$ on the cell environment. However, the fact that the $E$ values of FRET constructs in living cells obtained by a $\varphi$-dependent method are consistent with those obtained by a $\varphi$-independent method demonstrates that the cell environment does not seriously affect the $\varphi$ value (or the $\varphi$ ratio) of FPs.^{6,12,14,15,18,23} Based on our previous studies,^12,18,23 we also used the $\varphi$ values from a reference for our calculation.^25,26

$\rho$ factor not only bridges the two molar extinction coefficient ratios (${\gamma }^1$ and ${\gamma }^2$) but also can be used to determine the $\gamma$ values by using a tandem construct with unknown FRET efficiency [Eq. (11)], which is an incremental improvement for the IIem-spFRET method. Predetermination of $\gamma$ values can be considered as a one-off event for given FPs and instruments.¹² For the determination of $\gamma$ values by using a standard FRET construct, the most merit of the IIem-spFRET method is that we need not measure the FRET efficiency of this construct. Although both the Iem-spFRET and IIem-spFRET methods obtained consistent $\gamma$ values, the IIem-spFRET method showed more stable results [Fig. 2(d) and Table 1], demonstrating the robustness of the IIem-spFRET method. In contrast to the Iem-spFRET method, IIem-spFRET circumvents premeasuring the $E$ value of the tandem construct, which not only simplifies the measurement process of $\gamma$ values but also makes IIem-spFRET very applicable to the quantitative measurement of GFP-YFP pair. Accurate measurement of the $E$ values of the reference construct is crucial for the determination of $\gamma$ by using the Iem-spFRET method (Table 1). In fact, cellular motility of living cells and the degree of acceptor photobleaching may affect the PbFRET measurement,²⁸ thus resulting in the unstable $\gamma$ values measured by the Iem-spFRET method.

4.3.

Implementation of Iem-spFRET and IIem-spFRET for Cells Separately Expressing C32V, CVC, VCV, and VCVV

We next performed the Iem-spFRET and IIem-spFRET methods on our platform to measure the $E$ and $R^t$ values of C32V, CVC, VCV, and VCVV inside living Huh7 cells. Figure 3(a) shows images of representative cells separately expressing C32V, CVC, VCV, and VCVV with 405 and 436 nm excitations. The corresponding normalized spectra are shown in Figs. 3(b)–3(e), and the unmixing weights by using IIem-spFRET are listed in Table 2. Substitute these unmixing weights into Eqs. (9) and (10) to obtain the corresponding $E$ and $R^t$ values: 29.97% and 0.973 for C32V, 47.66% and 0.521 for CVC, 66.33% and 2.296 for VCV, and 74.86%, and 3.154 for VCVV, respectively (Table 2). The unmixing weights by using Iem-spFRET for the same cells are also listed in Table 2, and the corresponding $E$ and $R^t$ values were 28.36% and 0.994 for C32V, 46.70% and 0.539 for CVC, 60.87% and 2.952 for VCV, and 39.93% and 9.710 for VCVV, respectively.

Fig. 3

Implementation of both Iem-spFRET and IIem-spFRET methods on MSM platform for FRET quantification with representative cells separately expressing C32V, CVC, VCV, and VCVV in living Huh7 cells. (a) Fluorescence images of cells with 405 (upper panels) or 436 nm (lower panels) excitations. Scale bar: $10\mu \mathrm{m}$. (b) and (e) Spectra normalized to unit area of C32V (b), CVC (c), VCV (d), and VCVV (e) constructs inside the cell indicated by red circles in (a) with 405 and 436 nm excitation, respectively. (f) Statistical $E$ values of C32V, CVC, VCV, and VCVV tandem constructs in living Huh7 cells obtained by Iem-spFRET and IIem-spFRET methods, respectively. (g) Statistical $R^t$ values obtained by Iem-spFRET and IIem-spFRET methods from Huh7 cells exclusively expressing C32V, CVC, VCV, and VCVV tandem constructs.

Table 2

E and Rt values as well as unmixing parameters by Iem-spFRET or IIem-spFRET methods.

Statistical $E$ and $R^t$ values measured by IIem-spFRET were $28.34\%\pm 2.80\%$ and $0.99\pm 0.04$ for C32V ($n=16$), $48.00\%\pm 1.28\%$ and $0.54\pm 0.03$ for CVC ($n=15$), $67.89\%\pm 1.66\%$ and $2.09\pm 0.17$ for VCV ($n=15$), $69.02\%\pm 6.26\%$ and $3.68\pm 1.31$ for VCVV ($n=16$), respectively [Figs. 3(f) and 3(g)]. We also used the Iem-spFRET method to measure the $E$ and $R^t$ values of C32V, CVC, VCV, and VCVV constructs for the same cells, and the calculated $E$ and $R^t$ values were $26.85\%\pm 2.63\%$ and $1.00\pm 0.07$ for C32V, $47.48\%\pm 1.54\%$ and $0.54\pm 0.04$ for CVC, $60.47\%\pm 3.89\%$ and $3.04\pm 0.45$ for VCV, $56.33\%\pm 16.4\%$ and $5.74\pm 3.19$ for VCVV, respectively [Figs. 3(b) and 3(c)]. Here, the $E$ values measured by the two methods were basically consistent with $28.67\%\pm 0.95\%$ measured by lux-FRET for C32V,¹⁸ $41.4\%\pm 3.2\%$ for CVC measured by Vogel’s laboratory with FLIM method.⁹ Statistical $E$ values of VCV by the IIem-spFRET method are consistent with $64.5\pm 3.2\%$ measured by Vogel’s laboratory with FLIM method.⁹ The $R^t$ values calculated by the two methods are in agreement with the corresponding theoretical values of 1 for C32V, 0.5 for CVC. The statistical $R^t$ value of VCV by the IIem-spFRET method is also consistent with the theoretical value of 2. However, for VCV and VCVV constructs, the $E$ and $R^t$ values measured by the Iem-spFRET method had large deviations and fluctuations, but the IIem-spFRET method showed greatly improved measured results [Figs. 3(f) and 3(g)].

For the cells expressing low levels of VCV or VCVV with higher FRET efficiency, the measured $E$ and $R^t$ values by using the Iem-spFRET method showed large deviations and fluctuations, whereas the results measured by using IIem-spFRET exhibited significant improvements [Figs. 3(f) and 3(g)]. Although the autofluorescence intensity of cells was very low, different autofluorescence intensity of each cell might affect the FRET measurement. When the expression levels of VCV and VCVV are very low, direct subtraction of the BG signal from the measured spectra of cells expressing VCV or VCVV seriously affected the spectra in the 460 to 500 nm wavelength range, especially with 405-nm excitation [Figs. 3(d) and 3(e)]. Generally, the spectra of BG and autofluorescence should be stable; thus, they can be introduced as the BG spectral fingerprints to resolve the measured spectrum of FRET sample for removing the BG and autofluorescence. Iem-spFRET simply subtracted the averaged BG signal (including BG and autofluorescence) obtained from many cells from the measured fluorescence signal for each cell expressing FPs, whereas IIem-spFRET used the BG spectral fingerprint as a component to linearly unmix the measured spectrum for eliminating the influences of autofluorescence of each cell. Therefore, IIem-spFRET improves the anti-interference ability and robustness of FRET measurements in living cells [Figs. 3(f) and 3(g)].

4.4.

Good Robustness of IIem-spFRET Method

We next evaluated the effect of fluorescence intensity (or expression level of FPs) on Iem-spFRET and IIem-spFRET methods. Figure 4 shows the spectra of the C32V and VCV cells in Fig. 3(a) as well as the BG spectra with 405 and 436 excitations. Compared with the spectra of C32V cells [Fig. 4(a)], the 476-nm peak of VCV cells is very low, especially for the excitation of 405 nm [Fig. 4(b)], due to the high FRET efficiency of VCV. We chose 470 to 490 nm emission data as $I_0^i(CH0)$, and the relative intensities of the C32V cell in Fig. 4(a) with 405 and 436 excitations at CH0 were $I_0^{405}=42.1$, $I_0^{436}=177.9$, respectively, and the corresponding $E$ and $R^t$ values measured by using Iem-spFRET and Iem-spFRET were consistent (Table 3, cell 5). The relative intensities of the VCV cell in Fig. 4(b) with 405 and 436 excitations at CH0 were $I_0^{405}=12.6$ and $I_0^{436}=52.6$, respectively, and Iem-spFRET and IIem-spFRET obtained different $E$ and $R^t$ values (Table 3, cell 6). Just as listed in Table 3, both the Iem-spFRET and IIem-spFRET methods can obtain consistent results for C32V cells with different fluorescence intensity. However, for VCV cells with different fluorescence intensity, only IIem-spFRET obtained consistent results, indicating the good robustness of the IIem-spFRET method to fluorescence intensity.

Fig. 4

Spectra of C32V (a) and VCV (b) constructs inside the cells indicated by red circles in Fig. 3(a) as well as BG spectra with 405 and 436 nm excitation, respectively.

Table 3

E and Rt values of C32V and VCV constructs measured by Iem-spFRET or IIem-spFRET method.

The measured fluorescence intensity is proportional to the excitation intensity and the expression levels of FPs. To reduce photobleaching, the excitation intensity was controlled with $T\le 1\%$ in our system (Appendix A3). Therefore, the fluorescence intensity in this report mainly depends on the expression levels of FPs. In addition, we can choose 430 and 450 nm wavelengths as two excitations to enhance the emission intensity of Cerulean and Venus, which may further improve the robustness of the IIem-spFRET method.

4.5.

Analysis on STS-Induced Bax Oligomerization in Living Cells

Bax is a proapoptotic protein required for the process of mitochondrial outer membrane permeabilization.⁴ We here used the IIem-spFRET method to analyze the stauroporine (STS)-induced Bax oligomerization in living HepG2 cells just as described previously.¹² Figure 5 shows the fluorescence image of representative cells coexpressing CFP-Bax and YFP-Bax (left) and the corresponding fluorescence spectra with 405 nm (middle) or 436 nm (right) excitation. We found that Bax distributed evenly in cytoplasm in healthy cells [Fig. 5(a)] but translocated to mitochondria [Fig. 5(d)]. IIem-spFRET analysis showed that the corresponding $E$ value was 0.25% for control cells and 35.23% for STS-treated cells. The statistical $E$ value was $1.15\pm 0.79\%$ for healthy cells ($n=15$) and $16.02\pm 12.30\%$ for STS-treated cells ($n=15$), indicating that STS induced Bax translocation to mitochondria and subsequent oligomerization.

Fig. 5

IIem-spFRET analysis on Bax oligomerization in single live HepG2 cells. (a) Fluorescence images of a cell coexpressing CFP-Bax and YFP-Bax. Scale bar, 10 μm. (b) and (c) Spectral unmixing of the corresponding spectra with 405 (b) or 436 nm (c) excitation for the cell in (a) by IIem-spFRET. (d) Fluorescence images of a cell coexpressing CFP-Bax and YFP-Bax after STS treatment for 4 h. Scale bar, $10\mu \mathrm{m}$. (e) and (f) Spectral unmixing of the corresponding spectra with 405 (e) or 436 nm (f) excitation for the cell in (d) by IIem-spFRET.