Fluorescence Lifetime Correlation Spectroscopy (FLCS): concepts, applications and outlook

Abstract

Fluorescence Lifetime Correlation Spectroscopy (FLCS) is a variant of fluorescence correlation spectroscopy (FCS), which uses differences in fluorescence intensity decays to separate contributions of different fluorophore populations to FCS signal. Besides which, FLCS is a powerful tool to improve quality of FCS data by removing noise and distortion caused by scattered excitation light, detector thermal noise and detector afterpulsing. We are providing an overview of, to our knowledge, all published applications of FLCS. Although these are not numerous so far, they illustrate possibilities for the technique and the research topics in which FLCS has the potential to become widespread. Furthermore, we are addressing some questions which may be asked by a beginner user of FLCS. The last part of the text reviews other techniques closely related to FLCS. The generalization of the idea of FLCS paves the way for further promising application of the principle of statistical filtering of signals. Specifically, the idea of fluorescence spectral correlation spectroscopy is here outlined.

Figure 1

Illustration of removal of background-related artefacts in autocorrelation functions G (τ) by Fluorescence Lifetime Correlation Spectroscopy (FLCS). (a) The time correlated single photon counting (TCSPC) histogram of photon detection times (each channel corresponds to 16 ps); uniformly distributed background caused by detector after pulsing, thermal noise or photons of stray light adds over 100 counts to each channel. (b) FLCS filters calculated for fluorescence signal (red) and for the uniform background (black). (c) Comparison of autocorrelation function calculated without (black) and with FLCS filtering (blue); the decay on μs timescale is caused by detector after pulsing, while the lowering of autocorrelation amplitude is a result of uncorrelated background (thermal noise, stray light). The data were measured in a supported lipid bilayer on glass containing fluorescently labelled lipid as a tracer for lipid diffusion. More details on the experiment can be found in reference [32].

Figure 2

Simulated, realistic decay curve, its components and their calculated FLCS filter functions. (a) Two bi-exponential decay functions with the indicated amplitudes and lifetimes were convoluted with a real instrument response function (IRF not shown) resulting in Components 1 and 2, respectively. Both components have an average lifetime of 3.00 ns, in spite of different shapes. Uniform background (200 counts in each TCSPC channel) is added. The total decay curve is an arithmetic sum of these three components, plus the Poisson noise. (b) Calculated filter functions for decay components of the Total decay curve. Although the goal is to resolve Component 1 from Component 2, finding and subtracting the decay background (the third “decay” pattern) is so trivial that it always should be included in the filter calculation. The advantage is that the resulting ACFs for Components 1 and 2 will be purged of detector after pulsing and uncorrelated background contributions.

Figure 3

Simulated, realistic decay curve, its components and their calculated FLCS filter functions. (a) Two bi-exponential decay functions with the indicated amplitudes and lifetimes were convoluted with a real instrument response function (IRF not shown) resulting in Components 1 and 2, respectively. They have different average lifetimes, 3.00 ns and 3.60 ns, respectively, in spite of the same lifetime constituents. Uniform background (200 counts in each TCSPC channel) is added. The total decay curve is an arithmetic sum of these three components, plus the Poisson noise. (b) Calculated filter functions for decay components of the Total decay curve. Although the goal is to resolve Component 1 from Component 2, finding and subtracting the decay background (the third “decay” pattern) is so trivial that it always should be included in the filter calculation. The advantage is that the resulting autocorrelation functions (ACFs) for Components 1 and 2 will be purged of detector after pulsing and uncorrelated background contributions.

Figure 4

Illustration of possible sources of anticorrelation in FLCS. (a) to (c): Two-dimensional diffusion was simulated in a 6 μm × 6 μm square simulation box with periodic boundaries with 1 μs step; the Gaussian detection area had a radius of 240 nm; the simulated sample contains a mixture of two components A and B characterized by diffusion coefficients 20 μm²s⁻¹ and 80 μm²s⁻¹ and fluorescence lifetimes 4 ns and 2 ns, respectively. (a) Influence of the excluded volume; the ideal situation with point-like particles (dotted lines) is compared with the situation in which each particle occupies 4.4% of the detection area (solid lines). (b) Influence of electronic dead-time; brightness was 400 kHz per particle, average count rate 394 kHz and electronic dead-time 1 μs. Three cases are compared: correlation functions calculated using all photons (dead-time free, ideal situation, dotted lines), calculated considering the dead-time and using the knowledge which of the simulated molecules emitted each photon (dashed lines) and finally considering dead-time and using FLCS filtering to separate the components (solid lines) (c) Influence of detector timing jitter due to intensity-dependent IRF shift on ACFs and CCFs (solid lines). The signal rate dependent jitter was simulated as follows: IRF shift in ps units was 2.4 times the number of photons counts in the last 1 ms. With 41.284 kHz average count rate it means approximately 100 ps average (but fluctuating) shift. Ideal correlation functions calculated using the knowledge which of the simulated molecules emitted each photon instead of FLCS filtering is shown for comparison (dotted lines). Note that the parameters of the shift were chosen arbitrarily and do not simulate any particular type of single photon avalanche diode (SPAD). (d) Real experimental data, examples of FLCS CCFs obtained from a mixture of Cy5 and Atto655 [39]; negative CCF amplitude appears even with well matched TCSPC patterns (green solid line) while a deliberate shift of one of the patterns by 60 ps gives rise to a positive cross-correlation (black solid line). FLCS filtered, separated component ACFs of the two compounds are plotted for comparison, in order to demonstrate the relative amplitude of the effects (blue and red lines).The input data were taken from a sample workspace of a commercial software (SymPhoTime, version 5.3.2, PicoQuant, Berlin, Germany, 2010).

Figure 4

Illustration of possible sources of anticorrelation in FLCS. (a) to (c): Two-dimensional diffusion was simulated in a 6 μm × 6 μm square simulation box with periodic boundaries with 1 μs step; the Gaussian detection area had a radius of 240 nm; the simulated sample contains a mixture of two components A and B characterized by diffusion coefficients 20 μm²s⁻¹ and 80 μm²s⁻¹ and fluorescence lifetimes 4 ns and 2 ns, respectively. (a) Influence of the excluded volume; the ideal situation with point-like particles (dotted lines) is compared with the situation in which each particle occupies 4.4% of the detection area (solid lines). (b) Influence of electronic dead-time; brightness was 400 kHz per particle, average count rate 394 kHz and electronic dead-time 1 μs. Three cases are compared: correlation functions calculated using all photons (dead-time free, ideal situation, dotted lines), calculated considering the dead-time and using the knowledge which of the simulated molecules emitted each photon (dashed lines) and finally considering dead-time and using FLCS filtering to separate the components (solid lines) (c) Influence of detector timing jitter due to intensity-dependent IRF shift on ACFs and CCFs (solid lines). The signal rate dependent jitter was simulated as follows: IRF shift in ps units was 2.4 times the number of photons counts in the last 1 ms. With 41.284 kHz average count rate it means approximately 100 ps average (but fluctuating) shift. Ideal correlation functions calculated using the knowledge which of the simulated molecules emitted each photon instead of FLCS filtering is shown for comparison (dotted lines). Note that the parameters of the shift were chosen arbitrarily and do not simulate any particular type of single photon avalanche diode (SPAD). (d) Real experimental data, examples of FLCS CCFs obtained from a mixture of Cy5 and Atto655 [39]; negative CCF amplitude appears even with well matched TCSPC patterns (green solid line) while a deliberate shift of one of the patterns by 60 ps gives rise to a positive cross-correlation (black solid line). FLCS filtered, separated component ACFs of the two compounds are plotted for comparison, in order to demonstrate the relative amplitude of the effects (blue and red lines).The input data were taken from a sample workspace of a commercial software (SymPhoTime, version 5.3.2, PicoQuant, Berlin, Germany, 2010).

Figure 5

Spectral patterns, associated filter functions and spectrum-specific autocorrelation functions of dye labelled liposomes (unilamellar vesicles). (a) Reference spectra (spectral patterns) of 3,3′-dioctadecyloxacarbocyanine perchlorate (DiO) and 4,4-difluoro-1,3,5,7-tetramethyl-4-bora-3a,4a-diaza-s-indacene-8-propionic acid (BODIPY) dyes used for labelling various sized vesicles. Included here is the pattern of the spectral “baseline” caused by fast electron multiplying charge coupled device (EM-CCD) readout as well as the accumulated steady state emission spectrum of the vesicle mixture. Note the complete overlap of the spectra. (b) FSCS filter functions calculated for the 3 spectral components. (c) Comparison of spectrum-specific ACFs obtained by FSCS of a vesicle mixture and reference ACFs obtained by classical FCS of simple control samples. Although the match (especially for DiO) is not perfect, these results demonstrate a remarkable separation capability of FSCS. Spectral separation of DiO from BODIPY is an extremely difficult case, given the complete overlap of emission spectra.