FLIM-FRET for Cancer Applications.

There are various techniques by which one can detect if FRET has occurred in the context of microscopy. Based on the measurement of the fluorescence intensities of donor and acceptor molecules, intensity-based FRET is one of the most commonly employed FRET microscopy techniques and it relies on the phenomenon that when the donor is excited, the fluorescence intensity of donor will be reduced (“quenched”) and simultaneously the fluorescence intensity of acceptor will be increased (“sensitized”). Intensity-based FRET detection method requires a simpler setup, such as standard confocal or wide-field fluorescence microscopes, but there are some drawbacks to this method such as donor and acceptor bleedthrough, which requires careful correction measurements [ 2 – 4 ]. Also intensity-based E depends on the excitation intensity and the fluorophore concentration, and can determine whether a specific treatment or condition affects the proximity and the FRET signal between donor and acceptor molecules [ 5 – 9 ]. Another method to detect FRET is lifetime based, where in case of FRET occurrence the fluorescence lifetime of donor will be shortened. Although, both fluorescence lifetime imaging (FLIM) and intensity based FRET measurements are dependent on the acceptor: donor ratio, as shown previously [ 9 – 11 ], FLIM-FRET behaves independently of the donor concentration since fluorescence lifetime is inherent to each fluorophore and its surrounding environment in a concentration independent manner. This review is focused on FLIM for FRET applications, in particular in in vitro cell-based and in vivo cancer research. The imaging techniques and data analysis for FLIM are described in the following sections.

FITTING METHOD

Fitting methods extract the lifetimes of the different fluorescence components and the corresponding amplitudes by fitting the measured fluorescence decay curves to an appropriate model. To fit the measured data correctly, the instrument response function (IRF) of the system is used to convolve with the fitting model. IRF is treated as the decay curve recorded by the FLIM system for an infinitely short fluorescence lifetime. For in vitro cell-based applications, the IRF can be obtained from a mono-compound with known (very short lifetime), scattering medium, second harmonic generation, or a piece of paper. For in vivo applications, depending on the optical properties in the tissue, the IRF can be obtained in two ways. If the tissue is slightly turbid, such as a nude mouse model, the IRF can be obtained using the same method as for the in vitro cell-based application. If the tissue is strongly scattering, such as a rat model, the measured decay curve is modified by the tissue transfer function. Thus, using a fitting model convolved with the IRF-alone is not an accurate analysis approach. To overcome this problem, when fluorescence lifetime is 1.5 times larger than the full width at half maximum (FWHM) of IRF, the fluorescence lifetimes can be obtained by fitting the decay tail without deconvolving IRF [ 35 – 37 ]. However, this approach has not been used for FRET applications. In order to alleviate the effect of optical properties in strongly scattering tissues, IRF can be replaced by decay curve measured at the excitation wavelength which can accurately combine the effect of optical properties with the model. This approach has been used for FRET applications [ 38 ]. Based on the different fitting models, fitting methods include pixel-wise nonlinear analysis, global analysis, maximum likelihood analysis, and Bayesian analysis.

Pixel-wise Nonlinear Analysis

Pixel-wise nonlinear analysis [ 39 , 40 ] is the most common method used in FLIM analysis. It is a gold-standard for other methods to determine their accuracy. Pixel-wise nonlinear analysis uses the bi-exponential fluorescence decay model to fit the measured data and generates four main parameters, which are the lifetime of donor molecules not involved in FRET (non-FRETing donors), and the lifetime of donor molecules involved in FRET (FRETing donors) and the corresponding fractional amplitudes of the lifetimes, respectively. The E can be extracted from these parameters from the equation introduced above. The nonlinear least-squares method is used for this analysis to interactively minimize the reduced goodness-of-fit parameter. This analysis usually is performed pixel-by-pixel, leading to a slow speed in the execution of the algorithm. The E is normally obtained with a large standard deviation, due to the SNR generated from each pixel. Thus, to obtain more accurate results, only photon counts larger than 1000 should be used for bi-exponential fitting. To overcome this limitation, active illumination has been developed to increase SNR and enhance the dynamic range [ 22 , 41 ].

Global Analysis

Global analysis [ 42 – 44 ] uses all the pixels under the assumption that the lifetimes of FRETing and non-FRETing donors are invariant, and therefore the FRET efficiencies are constant over all pixels, but also allows fractional amplitudes of the lifetimes to vary on each pixel. Due to the assumption, the algorithm is accelerated significantly. A global goodness-of-fit parameter is minimized by the whole dataset simultaneously. Generally, the global analysis is faster than the pixel-wise nonlinear analysis. However, the common global analysis is difficult to estimate more complex decay models under FRET conditions, such as the donor with a multi-exponential decay (more than one lifetime components), i.e., enhanced cyan fluorescent protein (ECFP) is a donor with two lifetime components. To overcome this challenge, a rapid global fitting [ 45 ] has been developed. It is able to fit the data to more complex decay models and it is possible to get quantitative information from large datasets with high speed.

NON-FITTING METHOD

The non-fitting method allows the recovering of the lifetime values and amplitude fractions directly, including the rapid lifetime determination, minimal fraction of interacting donor and phasor analysis. These methods have been used for in vitro cell-based FRET applications, and they provide great potential for in vivo FRET applications with fast acquisition speed. Rapid Lifetime Determination (RLD) RLD is a non-fitting technique to calculate the fluorescence lifetime using the integrated fluorescence intensity during different time-gate windows. It has been widely used for fast-FLIM imaging based on time-gated FLIM [ 18 , 19 , 50 ]. The primary application has been used for fluorophores exhibiting mono-exponential decays by taking only two time gates. It also extends to bi-exponential decay model for FRET analysis by taking four time gates in time-gated FLIM. The fluorescence image at the first time-gate image (D 0 ) was obtained with a time-gate window of ΔT. Similarly, the fluorescence images at other time gates (D 1 , D 2 and D 3 ) were acquired with the same ΔT. The lifetime can be obtained from τ 1 = −Δ T / ln( y 2 ); τ 2 = −Δ T / ln( x 2 ), where τ 1 and τ 2 are the short lifetime and long lifetime of the donor, in the presence of acceptor, x and y are determined by D 0 , D 1 , D 2 and D 3 . (refer to more details in [ 50 , 51 ]). However, since the fluorescence decay curve is greatly under sampled, it is impossible to estimate the lifetimes from multi-exponential decay curves. Thus, it makes difficult to evaluate the amplitude fractions of quenched and unquenched donors under FRET conditions [ 28 ].

Minimal Fraction of Interacting Donor Approach

Minimal fraction of interacting donor (mf D ) [ 52 , 53 ] is used to get the minimal percentage of donor involved in FRET. The mf D calculation is expressed as [ 52 ] mf D = 1−< τ > / < τ > D . < τ > is mean lifetime of the donor, in the presence of the acceptor. < τ > D is the mean lifetime of donor alone. For time-gated FLIM, both < τ > and < τ > D are calculated based on the following equation using several time gates (~5 time-gate images) [ 52 ], < τ > = ∑ Δ t i · I i / ∑ I i where Δt i is the time delay of the i th measured fluorescence and I i is the fluorescence intensity of each pixel in the i th image. The advantage of mf D method is that it can directly extract the quantitative information related to the relative concentration of interacting donor with fast acquisition time, without fitting procedure, because it only depends upon experimental parameters, such as time gate selection and measured fluorescence intensity. Moreover, due to the independence of donor behavior (or E), it can be used for both single- and multi-lifetime decay models, which is a difficult task for the general decay curve fitting methods. The disadvantage of mf D method is that it is not suitable for the high donor fraction under FRET conditions.

Phasor Analysis

Phasor analysis [ 54 – 56 ] is a non-fitting method, and can be used for both TD-FLIM and FD-FLIM. The measured fluorescence decay curve is transferred to a new coordinate system (phasor plot) with g and s coordinates. When the measured data is multi-exponential components, the coordinate g and s are given by [ 54 ] g ( ω ) = ∑ k f k 1 + ( ω τ k ) 2 , s ( ω ) = ∑ k f k ω τ k 1 + ( ω τ k ) 2 , where τ k is lifetime , f k is intensity-weighted fractions for τ k , ω is the angular frequency of laser repetition or light modulation [ 54 ]. In the phasor plot of FLIM images, non-FRETing donors and FRETing donors are distributed in specific regions on the semicircle, without considering autofluorescence. On this semicircle, a phasor corresponding to the FRETing donor is close to the point (1,0), while a phasor corresponding to the non-FRETing donor is close to the point (0,0). Thus, it is easy to recognize the population of the molecules undergoing FRET. The large benefit of phasor-plot is that FLIM data analysis provides an intuitive interface for beginners. The phasor approach has the potential to analyze large FLIM datasets, since it does not require exponential analysis for each pixel in the image. In time-gated FLIM, only 4 time gates are needed for calculating two lifetime components by phasor analysis [ 55 ]. However, it is not sensitive to low photon counts (less than 100 counts) [ 49 ].