Improving Quality, Reproducibility, and Usability of FRET-Based Tension Sensors

Abstract

Mechanobiology, the study of how mechanical forces affect cellular behavior, is an emerging field of study that has garnered broad and significant interest. Researchers are currently seeking to better understand how mechanical signals are transmitted, detected, and integrated at a subcellular level. One tool for addressing these questions is a Förster resonance energy transfer (FRET)-based tension sensor, which enables the measurement of molecular-scale forces across proteins based on changes in emitted light. However, the reliability and reproducibility of measurements made with these sensors has not been thoroughly examined. To address these concerns, we developed numerical methods that improve the accuracy of measurements made using sensitized emission-based imaging. To establish that FRET-based tension sensors are versatile tools that provide consistent measurements, we used these methods, and demonstrated that a vinculin tension sensor is unperturbed by cell fixation, permeabilization, and immunolabeling. This suggests FRET-based tension sensors could be coupled with a variety of immuno-fluorescent labeling techniques. Additionally, as tension sensors are frequently employed in complex biological samples where large experimental repeats may be challenging, we examined how sample size affects the uncertainty of FRET measurements. In total, this work establishes guidelines to improve FRET-based tension sensor measurements, validate novel implementations of these sensors, and ensure that results are precise and reproducible. © 2018 International Society for Advancement of Cytometry.

Keywords: FRET efficiency; FRET-based biosensor; mechanotransduction; sensitized emission.

Figure 1.

(A) A cytosolic module consisting of mTFP1 (blue) linked to Venus (yellow) via a glycine-glycine-serine repeat (GGS2), which results in high FRET due to the short linker length. (B) A cytosolic module consisting of mTFP1 (blue) linked to Venus (yellow) via a derivative of tumor necrosis factor alpha (TNFα) receptor associated factor (TRAF), which results in low FRET due to the long linker length. Localization of the constructs in vinculin −/− MEFs is shown in the acceptor channel (I_aa). The corresponding acceptor-normalized corrected-FRET (F_c/I_aa) image and acceptor-normalized donor (I_dd/I_aa) image are shown with pseudo-coloring. (C) Heatmap of F_c/I_aa vs. I_dd/I_aa pixel data used to estimate the G factor. Data comes from all pixels segmented by cell masks for GGS2 (n = 74 cells) and TRAF (n = 39 cells) from a single experiment. Triangles, squares, and circles used to denote the mean, median, and mode, respectively, of each population, and the slope of the black dashed line corresponds to the estimation of G based on the mode. (D) Comparison of the four methods used to estimate G. Methods include a previously described cell-averaged approach (43) and our newer pixel-based approach using the mean, median, or mode. Data and error bars are the mean and standard error, respectively, from four independent experiments. Due to unequal variances as determined by the Levene’s test, Welch’s ANOVA was used and a statistically significant difference in means was not found.

Figure 2.

(A) Sample acceptor images of vinculin −/− MEFs expressing either GGS2 or TRAF and the associated cell-masked, pseudo-colored images of the calculated k factor. Scale bars are 25μm. (B) Histogram of pixel data corresponding to either GGS2 (blue, n = 74 cells) or TRAF (orange, n = 39 cells) from all cells in a single experiment. (C) Comparison of the four methods used to estimate k. Methods include a previously described cell-averaged approach (43) and our pixel-based approach using the mean, median, or mode. Data and error bars are the mean and standard error, respectively, from both constructs across four independent experiments. Following a Levene’s test, an ANOVA did not find statistically significant differences in the means.

Figure 3.

Comparison of different methods to characterize FRET efficiency in live, TSMod-expressing vinculin −/− MEFs. For each cell, the FRET efficiency was characterized from either the mean, median, or mode, with and without bootstrapping. Data and error bars are the mean and standard error, respectively, of the entire cell population (n = 454 cells, N = 11 experiments). Dashed line represents a published FRET efficiency value of 28.6% for unloaded mTFP1-(GPGGA)₈-Venus (TSMod) (43).

Figure 4.

(A) The tension sensor module (TSMod) consists of two fluorophores separated by a flagelliform linker sequence (GPGGA)₈. Localization and estimated FRET efficiency of TSMod in (B) live and (C) paraformaldehyde-fixed vinculin −/− MEF cells. (D) The actin-binding vinculin tension sensor mutant (VinTS-I997A) consists of TSMod inserted after aa 883 of vinculin and a point mutation at aa 997. Localization and estimated FRET efficiency of VinTS-I997A in (E) live, (F) paraformaldehyde-fixed, and (G) vinculin immunofluorescently-labeled vinculin −/− MEFs. (H) The vinculin tension sensor (VinTS) consists of TSMod inserted after aa 883. Localization and estimated FRET efficiency of VinTS in (I) live, (J) paraformaldehyde-fixed, and (K) vinculin immunofluorescently-labeled vinculin −/− MEFs. FRET distributions of all sample images are provided in Supp. Fig. 3. (L) Mean FRET efficiency of TSMod (live, n = 454, N = 11; fixed, n = 378, N = 11), VinTS-I997A (live, n = 161, N = 11; fixed, n = 50, N = 2; Vinculin IF, n = 272, N = 5), and VinTS (live, n = 173, N = 13; fixed, n = 164, N = 6; vinculin IF, n = 98, N = 3) under various conditions. Following a Levene’s test, differences between (a) TSMod and VinTS-I997A and (b) VinTS was determined by a Welch’s ANOVA and Games-Howell Test. Error bars represent one standard error.

Figure 5.

(A) Internal vinculin intermolecular FRET controls contain Venus or mTFP1 inserted at aa 883. (B) VinTS dark intermolecular FRET controls contain point mutations on either Venus or mTFP1 that disrupt the fluorescent properties of the fluorophore. (C) Localization of internal vinculin constructs co-expressed in a vinculin −/− MEF. (D) Localization of VinTS dark constructs co-expressed in a vinculin −/− MEF. The corresponding masked FRET efficiencies are also shown. (E) Difference in the mean FRET efficiency of vinculin internals and VinTS darks determined by t-test (p = 0.0014). Error bars are one standard error.

Figure 6.

(A) Uncertainty in the estimation of the population’s mean FRET efficiency as a function of sample size for TSMod expressed in live (n = 454) and fixed (n = 378) vinculin −/− MEFs. For each sample size, the uncertainty is the width of the 95% confidence interval of the mean’s distribution, which arises from the simulated experimental sampling explained in the Materials & Methods. (B) The same analysis was performed on data from live vinculin −/− MEFs expressing either TSMod (n = 454), VinTS-I997A (n = 161), or VinTS (n = 173).