2.1.

Optical Setup

Our setup is depicted in Fig. 2. Two non-polarized LEDs excite the sample : a blue LED at 440 nm to excite the donor fluorophore mTFP1 and a green LED at 505 nm to excite the acceptor mVenus (see Fig. 3(a) for reference). Both are collimated using aspheric lenses with 32-mm focal lengths (Thorlabs ref ACL-50832U-A). The current in each LED can be adjusted independently to reach the desired illumination intensity in the sample plane. A neutral density filter is inserted on the common path to reduce both illumination intensities. Intensities between $13.5\mathrm{mW}/{\mathrm{mm}}^2$ and $41\mathrm{mW}/{\mathrm{mm}}^2$ for the blue LED (440 nm) and between $19\mathrm{mW}/{\mathrm{mm}}^2$ and $55.5\mathrm{mW}/{\mathrm{mm}}^2$ for the green LED (505 nm) were used in the different measurements. Fluorescence is collected through a microscope objective (Nikon Plan Fluor $100\times$, $\mathrm{NA}=1.3$, oil immersion) and the donor and acceptor emissions are detected simultaneously on two CMOS cameras (Basler acA2040-$55\mu \mathrm{m}$, $2048\times 1536\text{pixels}$, $3.45\mu \mathrm{m}$ pixels, quantum efficiency 65% at 550 nm), one for the donor and one for the acceptor. The two lenses used to relay the image from the exit port of the microscope to the two cameras are achromats with focal lengths of 150 mm (Thorlabs AC254-150A).

Fig. 2

Optical setup: both LEDs are conjugated with the back focal plane of the microscope objective (dashed line). The sample is imaged on three cameras: cameras 1 and 2 for fluorescence of the donor and acceptor, respectively, and camera 3 for phase contrast (dotted line). Illumination for the phase contrast image is done with a halogen lamp with a red filter (high pass $>590\mathrm{nm}$), an annulus ring in the condenser, and a phase ring deported on the imaging path. All dichroic and filters references are listed in Fig. 3(a).

Fig. 3

(a) List of the filters. (b) Spectra of mTFP1 and mVenus excitation and emission and transmission of the filters.

For each FRET measurement, we acquire three images in two steps.

The spectral transmission bands of the excitation and detection filters are shown in Fig. 3(b), overlayed on the absorption and emission spectra of the two fluorophores. References of all filters and beamsplitters used in the setup are shown in Fig. 3(a). Phase contrast imaging is performed with red light illumination (a halogen lamp with a OG590 Schott filter) to avoid photobleaching of the fluorophores during initial sample focusing. An annulus ring is placed in the back focal plane of the condenser and is conjugated with an external phase ring on the imaging path to avoid modifying the microscope objective. Adjustment of the relative position of the two cameras is accomplished using fluorescent microbeads (Fluoresbrite^®YG Microspheres $0.50\mu \mathrm{m}$, Polysciences) to overlap the two images with the same magnification. Due to chromatic aberration, this position does not correspond to good focusing on both cameras. A displacement of 200 nm of the sample could recover a sharp focus, which gives an estimate of the overall chromatic aberration of the fluorescence imaging path between the 482 and 562 nm detection bands. To avoid moving the sample, we adjusted the distance between the dichroic beamsplitter DBS4 and camera 1 by 2 mm (200 nm multiplied by the square of the $100\times$ lateral magnification) to reach precise focusing on the two cameras. Additional fine tuning of this distance is done using images of the focal adhesions themselves. The magnification is not exactly the same on both cameras and is corrected during image processing (DD is 2% smaller than DA and AA). Compared with a standard setup with a single camera, we do not need to move the sample between DD and AA acquisitions to compensate for the chromatic aberration.

We characterized the bleedthrough between channels using cells transfected with acceptor only or donor only. The cells were isolated from the background by thresholding, and pixel-by-pixel heatmaps of DA versus AA or DA versus DD were plotted. A linear fit of these heatmaps gave a factor $a=0.054$ for the direct excitation of the acceptor when illuminating with the blue LED at 440 nm (measured with the acceptor only construct in the cell) and a factor $d=0.384$ for the long tail of the donor fluorescence above 550 nm that passes through the acceptor detection channel (measured with the donor only construct). These lead to a corrected FRET intensity of

To simplify notations, we call DA, AA, and DD the intensities of the corresponding images.

The FRET efficiency measurements made on this simplified setup (Fig. 2) located at Paris-Saclay are validated against the same measurements performed on a more standard FRET microscopy setup located at Rutgers utilizing filter wheels and a high-end scientific CMOS camera¹⁷ (PCO Edge 4.2bi, $2048\times 2048\text{pixels}$, $6.5\mu \mathrm{m}$ pixels, quantum efficiency 95% at 550 nm, cooled at $-25°\mathrm{C}$).

2.2.

Sample Preparation

Chinese Hamster ovary cells (CHO-K1) (ATCC CCL-61) were grown and maintained at 37°C, 5% ${\mathrm{CO}}_2$ in high glucose Dulbecco’s Modified Eagle’s Medium supplemented with 10% fetal bovine serum, 1% penicillin/streptomycin, and 1% L-glutamine (all from Gibco) for Paris-Saclay experiments or F12K (ATCC) supplemented with 10% fetal bovine serum (Gemini) at Rutgers. Cells were plated at $\mathrm{10,000}/{\mathrm{cm}}^2$ if transfected the next day or at $\mathrm{5,000}/{\mathrm{cm}}^2$ if transfected 3 days after plating. Cells were seeded on 25 mm diameter glass coverslips (1.5H Marienfeld Superior), previously cleaned with nitric acid (18-mm coverslips cleaned with chromic sulfuric at Rutgers), rinsed 10 times with sterile water, then incubated 1 min in alcohol, and air dried. Coverslips were then incubated with $10\mu \mathrm{g}$ of poly-L-lysine from sciences sell [poly-D-lysine (PL) at Rutgers, P0899, Sigma] for 30 min at 37°C and then washed three times with sterile water or incubated with $10\mu \mathrm{g}$ of FN (F0895, Sigma) for 30 min at 37°C. Cell transfections were performed following the Lipofectamine LTX protocol (Invitrogen) with a DNA ratio of $3.75\mu \mathrm{g}/5\mu \mathrm{l}$ of plasmid/lipofectamine per coverslip (plasmid descriptions are given in Table S1 in the Supplementary Material). Cells are imaged between 24 and 48 h after transfection in live cell imaging solution (Invitrogen) or growth medium in Paris-Saclay or Leibovitz’s + 10% serum at Rutgers.

2.3.

Image Processing

The three raw images (DD, DA, and AA) are first background corrected by subtracting the mean value of a background region selected manually far from any cell. Each background-substracted image is flat-field corrected to account for the illumination profile of each LED that is not perfectly uniform over the full field of view.^18,19 A reference image is obtained for each LED by averaging and smoothing 15 to 20 images of a microscope slide with uniform fluorescence (fluorescent marker).

To compensate for translation and rotation between the two cameras and correct for the change in magnification due to chromatic aberrations, DD has to be registered to perfectly match pixel-by-pixel with DA and AA.

The transformation is calculated for each coverslip tested and applied to all of the DD images. DA and AA are already perfectly overlaid because they are acquired on the same camera without any mechanical movement between the acquisitions. The non-background pixels selected from cells expressing mTFP1, mVenus, or TSMod are segmented using a thresholding. For VinTS and VinTL, simple thresholding cannot extract focal adhesions as they appear over a fluorescence background from the cytoplasm. A binary mask to isolate focal adhesions is calculated based on the subtraction of a spatially varying background calculated with a rolling ball algorithm (ball diameter = 20 pixels),²⁰ followed by a Gaussian blur ($\sigma =3\text{pixels}$) and thresholding. Only the FAs consisting of at least 30 pixels ($\approx 0.04\mu {\mathrm{m}}^2$ in the object plane) are kept for the analysis. FRET efficiencies are averaged over each segmented FA. Details of the image processing are given in the Supplementary Material.

3.1.

Calibration

To obtain an absolute FRET efficiency measurement that can be compared to other experiments, calibration of both the Paris-Saclay and Rutgers sensitized emission setups was done using a single FRET construct TSMod, following the method described in Menaesse 2020.¹⁶ This TSMod construct is the one that will be inserted in the vinculin protein to create the tension sensor VinTS. Its FRET efficiency has been previously measured to be 28.6%,²¹ but we measured its value in our CHO-K1 cells as explained further in this section. We determined the calibration factors^22–25 $G$ and $k$ specific for each of our imaging setups. The given constructs and setups are listed as follows.

$F_c$ is the FRET corrected from bleedthroughs [see Eq. (1)], DD is the donor emission when donor is excited, and AA is the acceptor emission when the acceptor is excited. From the slope of the $F_c/\mathrm{AA}$ versus $\mathrm{DD}/\mathrm{AA}$ plot, and the FRET efficiency $E$ measured by FLIM, we obtain the $G$ calibration factor [red line in Fig. 4(a)].

Fig. 4

(a) Pixel-by-pixel heatmap of Fc/AA versus DD/AA of CHO-K1 cells expressing TSMod. 269 cells from 6 samples with exposure times of 100 or 200 ms and ND filter of OD 1 (around 48 millions pixels). The pixel values are fitted with a 2D Gaussian: $f(x,y)=A\exp (-{((x-x_0)/w_x)}^2-{((y-y_0)/w_y)}^2)$. $R^2=\mathrm{0,967}$. The black cross represents the center of the fitted Gaussian, and the dashed ellipse is the 95% confidence bound of the Gaussian. The center of the Gaussian is given by $x_0=0.3366\pm 1.2e^{-4}$ and $y_0=0.2005\pm 8.0e^{-5}$. The red and white lines represent Eqs. (2) and (3), respectively. (b) Projections of the 2D histogram along $x$ and $y$ axes.

Tracing the line of slope $-G$ passing through the data points, we get $k$ as the intersection with the axis where $\frac{F_c}{\mathrm{AA}}=0$ [white line in Fig. 4(a)].

Two phenomena can bias the calibration: photobleaching and a high concentration of fluorophores in the cells. First, to avoid photobleaching during focusing, cell images are focused using phase contrast with red light illumination. We quantified photobleaching with each LED on cells expressing mTFP1 or mVenus only. With the OD = 1 that was used for the calibration, the illumination intensities are $19.0\mathrm{mW}/{\mathrm{mm}}^2$ for the blue LED (440 nm) and $13.5\mathrm{mW}/{\mathrm{mm}}^2$ for the green LED (505 nm). The photobleaching decay half-time was 450 s for mTFP1 and 21 s for mVenus (Fig. S1 in the Supplementary Material). Because we used exposure times of 100 or 200 ms, photobleaching is negligible. Second, cells expressing very high concentration of fluorophores have a higher FRET index $F_c/\mathrm{AA}$, probably due to intermolecular FRET as a consequence of molecular crowding. To address this issue, we chose a range of intensity over which the cells have constant FRET index over $\mathrm{AA}/t_A$ (AA intensity normalized to exposure time) while keeping a good signal to noise ratio (Fig. S2 in the Supplementary Material).

The corrected FRET and donor intensities both normalized by acceptor intensity are plotted pixel-by-pixel for the non-background pixels of 269 TSMod cells [Fig. 4(a)]. Data are fitted with a 2D Gaussian model.

As a reference for FRET efficiencies in our CHO-K1 transfected cells, we used a FLIM setup²⁶ to measure the donor lifetime for TSMod. Measured lifetimes ${\tau }_D$ for donor (mTFP1) only and $⟨{\tau }_{\mathrm{DA}}⟩$ (weighted average of two contributions) for donor in the presence of acceptor for TSMod are shown in Fig. 5(a). The FRET efficiency of TSMod deduced from these measurements ($E=1-\frac{⟨{\tau }_{\mathrm{DA}}⟩}{{\tau }_D}$) is $27.9\pm 1.0\%$, in good agreement with published values for this construct.²¹ From the coordinates of the center of the Gaussian and this TSMod FRET efficiency of $27.9\pm 1.0\%$ from our FLIM measurement, we obtain : $G=1.54\pm 0.08$ and $k=0.47\pm 0.03$. The same method was used to calibrate the Rutgers setup, and it gave $G=2.97\pm 0.10$ and $k=2.14\pm 0.08$ (see corresponding heatmap in Fig. S3 in the Supplementary Material). The uncertainty on $G$ is mostly due to the uncertainty on the FRET efficiency of TSMod measured by FLIM.

Fig. 5

(a) Lifetime measurements for mTFP1 (37 cells), TRAF (20 cells), TSMod (15 cells), GGS2 (26 cells), and GGS1 (8 cells). Average lifetime values ± standard deviation are respectively : $2.78\pm 0.07\mathrm{ns}$, $2.62\pm 0.06\mathrm{ns}$, $2.02\pm 0.07\mathrm{ns}$, $1.42\pm 0.08\mathrm{ns}$, and $1.42\pm 0.03\mathrm{ns}$. (b) Sensitized FRET efficiencies measured on the Paris-Saclay setup for TRAF (26 cells), GGS2 (8 cells), and GGS1 (28 cells), based on the calibration using TSMod only. The FLIM efficiencies are calculated from the lifetimes shown in (a). Average efficiencies values ± standard deviation. The dashed line has a slope of 1, showing that the sensitized and FLIM efficiencies are the same.

As a confirmation of the calibration of our Paris-Saclay setup, we measured the FRET efficiencies of constructs with low (TRAF) and high (GGS1 and GGS2) FRET values (see Table S1 in the Supplementary Material for all plasmids). Figure 5(b) shows that the efficiencies calculated with our calibration factor $G=1.54\pm 0.08$ coincide with the efficiencies deduced from lifetime measurements for these constructs performed with our FLIM setup [Fig. 5(a)]. They also agree with published values.²⁵ This agreement confirms the validity of our calibration with TSMod only.

3.2.

Vinculin Tension Sensor

FRET efficiency measurements of VinTS and VinTL on the focal adhesion sites of CHO-K1 cells plated on poly-L-lysine, using the simplified Paris-Saclay setup (Fig. 2), are shown in Figs. 6 and 7 (blue bars). Figure 6 shows examples of raw AA images (left) and corresponding segmented images in which FRET efficiency is calculated based on the calibration described above and averaged over each focal adhesion site for VinTS (top panels) and VinTL (bottom panels). FRET efficiency is higher for VinTL, whereas for VinTS the presence of force reduces FRET efficiency. Average FRET efficiencies over 10,000 adhesion sites from 60 cells (VinTL) and 19,000 adhesion sites from 87 cells (VinTS) are shown in the blue bars of Fig. 7. The VinTL FRET efficiency is $30.4\%\pm 5\%$, slightly above the TSMod efficiency of 27.9%. For VinTS on poly-L-lysine, the average efficiency is significantly lower, at $22.0\%\pm 4\%$ (p-value $<{10}^{-3}$ by one-way ANOVA). Uncertainties are standard deviations on all adhesion sites.

Fig. 6

(a) VinTS consists of TSMod inserted between the vinculin head and tail. (b) Vinculin tail-less sensor (VinTL) is a control construct only linked to the vinculin head. (c), (d), (g), and (h) Fluorescence images in the acceptor channel of cells expressing VinTS and VinTL. (e), (f), (i), and (j) The FRET efficiencies averaged over the segmented FAs of cells expressing VinTS and VinTL. Images (c-f) are taken on the Paris-Saclay setup with an exposure time of 1900 ms. Images (g-j) are cropped images of $1100\times 1100\text{pixels}$ from the total field of view of $2048\times 2048\text{pixels}$, taken on the Rutgers setup with an exposure time of 1000 ms.

Fig. 7

FRET efficiencies measured over FAs of cells expressing VinTS or VinTL. We repeated the same experiment on Paris-Saclay and Rutgers setups calibrated as explained in 3.1. FRET efficiency is averaged over each segmented FA. Paris-Saclay: CHO-K1 cells seeded on poly-L-lysine. Rutgers: CHO-K1 cells seeded on PL or FN. Our results can be compared to other measurements: Li:²⁸ lifetime measurements on TMD cells. Gates:²⁵ ratiometric measurements on vinculin -/- MEFs cells. Rothenberg:²¹ ratiometric measurements on Vin -/- MEFs cells. Dumas:²⁹ lifetime measurements on iBMK cells. The coverslip treatment of each experiment is given by PL: polylysine, FN: fibronectine, or NT: no treatment. Mean ± standard deviation. *** denotes significance of $\mathrm{p}<{10}^{-3}$ by ANOVA followed by multiple comparison test.

Compared with the TSMod, mVenus, and mTFP1 experiments, higher illumination intensities and longer exposure times were required for the VinTS and VinTL experiments due to the lower fluorophore concentration on the focal adhesion sites. An OD = 0.6 was used and yielded $55.5\mathrm{mW}/{\mathrm{mm}}^2$ for the blue LED (440 nm) and $41.0\mathrm{mW}/{\mathrm{mm}}^2$ for the green LED (505 nm), and exposure times were increased to 1900 ms. In our analysis, we discarded adhesion sites that had low fluorescence signals (AA lower than 5 gray levels with our exposure time of 1.9 s). This situation corresponds to a signal over background equal to 1 on the DA image, the one with the highest background compared with DD and AA, due most likely to the autofluorescence of the sample. We checked that the choice of this threshold changes the value of the average efficiency by $<2\%$ for VinTL (see Fig. S10 in the Supplementary Material).

Assuming a supralinear dependence of the photobleaching rate with the illumination intensity as reported in Ref. 27 (exponent $\alpha =1.27$ for mVenus with widefield LED illumination at 505 nm), we estimated the photobleaching half-time to be $t_{A,1/2}=5.1\mathrm{s}$ for the green LED (505 nm) at $41\mathrm{mW}/{\mathrm{mm}}^2$. Following Ref. 23, we corrected our FRET efficiencies by $E_{\mathrm{corr}}=E\times 2^{\frac{t_{\text{exp}}/2}{t_{A,1/2}}}$ with $t_{A,1/2}$ being the acceptor photobleaching half-time. We divided the exposure time by 1/2 to take into account that the signal integrated over an exposure time while photobleaching is approximately the same as a signal with no photobleaching at half the exposure time. This approximation is valid to within 0.2% in our conditions. For our exposure time of 1.9 s, this correction increases the FRET efficiencies by a factor of 1.12 and has been taken into account in the values for VinTS and VinTL given above.

Using the same protocol, the same cells and same image processing method, we repeated the same measurements on the Rutgers setup (described in Ref. 17). The excitation intensity for the acceptor was much lower ($0.8\mathrm{mW}/{\mathrm{mm}}^2$) with an exposure time of 1 s, so no correction for photobleaching was necessary. In addition, we tested the cells attachment on either PL or fibronectin (FN) to allow for comparisons with previously published values found in the literature. The FRET efficiency of the tension-insensitive VinTL control (Fig. 6) was in close agreement on both setups: $31.7\%\pm 3\%$ on PL (Rutgers) and $30.4\%\pm 5\%$ on PL (Paris-Saclay) (p-value= 0.99 by two-way ANOVA). VinTL FRET efficiency was also not significantly affected by the different substrate treatment: $31.7\%\pm 3\%$ on PL and $31.9\%\pm 3\%$ on FN (p-value = 0.96). However, the different substrate coatings lead to a significant difference in FRET efficiency for VinTS ($25.1\%\pm 3\%$ on PL and $23.7\%\pm 3\%$ on FN, p-value $<{10}^{-3}$). The orange bars in Fig. 7 show those results for VinTL and VinTS. The FRET efficiency of VinTS was higher on the Rutgers setup ($25.1\%\pm 3\%$) compared with the Paris-Saclay ($22.0\%\pm 4\%$) setup (p-value $<{10}^{-3}$). Thus the absolute FRET efficiency difference between VinTL and VinTS was 8.4% on the Paris-Saclay setup versus 6.6% on the Rutgers setup.