Overcoming limitations of FRET measurements

Abstract

FÖrster resonance energy transfer (FRET) is a non-radiative process that has been extensively applied for probing molecular events involved in a range of cellular and biological processes 1. FRET provides a means for measuring intermolecular spacings on the order ∼1–10 nm 2, an ability that has allowed measurement of cellular events such as protein folding, substrate interactions, enzyme kinetics, and intercellular signaling molecules. The advent of a wide range of fluorescent proteins from which to construct FRET reporters has enabled a variety of live-cell FRET studies to be readily conducted. In addition, newer techniques for gene manipulation have allowed generation of transgenic animal models expressing FRET reporters. The use of FRET-based reporters is now widespread, and in some cases FRET measurements have become the “gold-standard.” The potential for developing new genetic approaches allowing localized intracellular biochemical assays that make use of FRET measurements is high. Unfortunately, FRET approaches are associated with several limitations inherent in the physical process of FRET as well as the technologies used to measure FRET. For example, it is widely recognized that the fluorescence properties of many labels, especially fluorescent proteins, are sensitive to changes in local environment, such as changes in pH, ionic concentrations, oxidation, temperature, and refractive index. FRET measurements make use of two or more fluorescent labels, or proteins, each of which may present different sensitivities for changes in environmental factors. Hence, FRET measurements may be skewed by agonist-induced or unappreciated changes in local environment. Arguably the largest limitation of FRET measurements is the low signal-to-noise ratio (SNR) associated with FRET imaging. While many technologies have been developed to measure FRET, in general, all of them suffer from poor SNR when compared to imaging of a single fluorescent label. The governing factors behind the reduced SNR include the loss of energy associated with the FRET process and the fact that, whether directly or indirectly, two fluorescent molecules contribute to the measured FRET signal. These factors carry over into many different technologies that have been developed to measure FRET: filter set-based (traditional fluorescence microscopy), acceptor photobleaching, spectral FRET, and fluorescence lifetime imaging microscopy (FLIM) FRET. While we, and others, have shown that current state-of-the-art techniques such as spectral imaging and FLIM can reduce variance in FRET measurements 3, 4, these measurements still have low SNR when compared to imaging of a single label (Fig. 1). This low SNR, and the techniques of spectral and FLIM microscopy developed to overcome it, has often resulted in long exposure times (>1–2 sec) and uncertainty of the ability to differentiate between small differences in FRET efficiencies or interpret results based on small changes in FRET efficiency 7. Current state-of-the-art FRET measurement technologies suffer from reduced signal-to-noise ratios (SNR) when compared to measurement of a single fluorescent label or protein. A: Image of single-label acceptor emission using acceptor excitation (Venus, excited at 488 nm, expressed in pulmonary microvascular endothelial cells); (B) image of calculated FRET efficiency using hyperspectral confocal imaging shows reduced SNR when compared to the single-label image in Panel (A)—image shows a single expressing cell (Turquoise-Epac-Venus cAMP reporter5, excited at 405 nm, expressed in pulmonary microvascular endothelial cells); (C) image of fluorescence lifetime where greyscale intensity represents lifetime, masked to illustrate a single cell (Turqouoise-Epac-Venus cAMP reporter 5, expressed in HEK-293 cells); (D) the image from Panel (B), masked to remove nuclei, false-colored, and scaled to highlight changes in FRET efficiency; (E) the image from Panel (C), false-colored and scaled to highlight changes in fluorescence lifetime. Images (A, B, and D) were taken using identical objectives and acquisition parameters using a Nikon A1R spectral confocal microscope, with the exception of using a 488 nm laser for direct excitation of Venus in Panel (A). Images were then linearly unmixed using a non-negatively constrained unmixing algorithm implemented in MATLAB software, and converted to 8-bit and scaled linearly for display. Images (C, E) were provided by courtesy of Dr. Kees Jalink, van Leeuwenhoek Centre of Advanced Microscopy, Amsterdam, The Netherlands. SNR was calculated as described by Bernas and colleagues 6—in brief, pixels in regions of similar (but not oversaturated) intensity were identified using an 8-way high-pass filter and the mean intensity (signal) was calculated from the most homogeneous regions, while noise was calculated as the standard deviation of these regions and the SNR calculated as the ratio of mean intensity to standard deviation. [Color figure can be