Phasor S-FLIM: a new paradigm for fast and robust spectral fluorescence lifetime imaging

Abstract

Fluorescence lifetime imaging microscopy (FLIM) and spectral imaging are two broadly applied methods for increasing dimensionality in microscopy. However, their combination is typically inefficient and slow in terms of acquisition and processing. By integrating technological and computational advances, we developed a robust and unbiased spectral FLIM (S-FLIM) system. Our method, Phasor S-FLIM, combines true parallel multichannel digital frequency domain electronics with a multidimensional phasor approach to extract detailed and precise information about the photophysics of fluorescent specimens at optical resolution. To show the flexibility of the Phasor S-FLIM technology and its applications to the biological and biomedical field, we address four common, yet challenging, problems: the blind unmixing of spectral and lifetime signatures from multiple unknown species, the unbiased bleedthrough- and background-free Förster resonance energy transfer analysis of biosensors, the photophysical characterization of environment-sensitive probes in living cells and parallel four-color FLIM imaging in tumor spheroids.

Fig. 1 |. Phasor S-FLIM workflow and robustness to noise of the phasor approach.

a, Schematic representation of the Phasor S-FLIM workflow. From left to right, fluorescence emission is chromatically separated by a diffraction grating and collected by a 32-channel PMT detector array, ranging from 404 to 670 nm. Single photon pulses are timed by two 16-channel constant fraction discriminators (CFDs) and their digital output is processed by an FPGA-based DFD architecture. Data are streamed via USB 3.0 to a computer for postprocessing and a 5D matrix (2D space, wavelength, time in nanoseconds, time in frames) is transformed to the phasor space and blindly unmixed to identify the pure spectral and lifetime components. b,c, Bias (b) and standard deviation (c) as a function of the number of photons used in the simulation of a 0.5 ns decay with varying number of photons. d, Execution time (exe) as a function of the number of points analyzed, numbers of iterations higher than 10³ were not considered for the fitting due to the long execution time. Gray shaded areas correspond to an approximate range for a typical FLIM image, considering 2 to 100 photons per pixel (b and c) and a number of pixels ranging from 256 × 256 to 1,024 × 1,024 (d). Curves refer to phasor transformation (blue) and LMS fit (red). e–g, Bias (e), standard deviation (f) and execution time (g) as a function of the FWHM of the simulated Gaussian IRF and the SBR for phasor transformation (right) and LMS fit (left). Surfaces are obtained from 1,000 realizations of a 2.5 ns decay obtained from 1,000 photons. Average profiles for bias (h), standard deviation (i) and execution time (j) as a function of the FWHM of the simulated Gaussian IRF (left) and the SBR (right) for phasor transformation (blue) and LMS fit (red). SBR is defined as the number of photons used to simulate the decay $N_{\text{Ph}}^{\text{delay}}$ over the number of photons used to simulate the background $N_{\text{Ph}}^{\text{bkg}}$.

Fig. 2 |. Phasor S-FLIM spectral and lifetime blind unmixing of cellular samples.

a, RGB image, corresponding to approximately 30 photons per dye per pixel, and unmixed FLIM images of live cells labeled with ViaFluor 488 (tubulin, top right), MitoTracker Orange (mitochondria, bottom left) and LysoTracker Red (lysosomes, bottom right). A gamma correction is applied to the RGB image to make the tubulin and lysosome staining visible. b, Average lifetime (left) and spectra (right) for tubulin (green), mitochondria (yellow) and lysosomes (red) of the unmixed images shown in a (open circles) compared with pure components (solid lines). c, RGB image and unmixed FLIM images of a fixed cell labeled with NucSpot 488 (DNA, top right), anti-TOMM20/Alexa555 (mitochondria, bottom left) and Phalloidin/Alexa564 (actin, bottom right), corresponding to 11.3 photons per dye per pixel. d, RGB image and unmixed FLIM images of the same fixed cell depicted in c with reduced photon amount, stated on the right. e,f, Average spectra (e) and lifetime (f) obtained for DNA (top), mitochondria (center) and actin (bottom) with photon amount varying from 0.6 (purple) to 11.3 (orange) photons per dye per pixel, compared with pure components (black dash-dotted lines). All cells depicted are representative of three biological replicates yielding similar results. Scale bars, 10 μm.

Fig. 3|. Phasor S-FLIM approach to FRET standards and biosensors.

a, Schematic representation of the C32V FRET standard. b, FRET efficiency as a function of the expression level ($<N_{\mathrm{P}\mathrm{h}}>$, number of photons per pixel for the unmixed mCerulean) for Phasor S-FLIM (left, yellow and right, gray) and FLIM (left, gray and right, red). c, Difference in FRET efficiency between FLIM and Phasor S-FLIM as a function of the expression level (blue dots) and fitting (red dotted) with a function $f=\frac{A}{<N_{\mathrm{p}\mathrm{h}}>}$, where $A$ is a constant scale factor. d, Schematic representation of the H3K9me3 chromatin compaction FRET biosensor. e, FRET images obtained by FLIM (left) and Phasor S-FLIM (right) for cells in hypo-osmotic medium (top), control (center) and cells on hyper-osmotic medium (bottom). f, Distribution of FRET efficiency obtained by Phasor S-FLIM (left) and FLIM (right) for the cells depicted in e. g, Box plots of FRET efficiency obtained by Phasor S-FLIM (yellow) and FLIM (pink) for the cells depicted in e. Median FRET efficiencies are 40.0% (hypo-osmotic), 34.1% (control) and 39.7% (hyper-osmotic) for FLIM and 33.8% (hypo-osmotic), 38.1% (control) and 46.8% (hyper-osmotic) for Phasor S-FLIM. In g, red lines represent the median, the edges of the box the 25th and 75th percentiles and the whiskers extend to the most extreme datapoints. The numbers of pixels analyzed for f and g are $n=\mathrm{15,029}$ (control), $n=\mathrm{16,402}$ (hyper-osmotic) and $n=\mathrm{41,847}$ (hypo-osmotic). The number of cells analyzed for a–c is 28, obtained from three biological replicates, while cells in e are representative of four biological replicates yielding similar results. Scale bars, 10 μm.

Fig. 4 |. Phasor S-FLIM characterization of solvatochromic probes in living cells.

a, RGB (left) and GP (right) images of cells labeled with Laurdan, symbols represent lipid droplets (circle), high-GP internal membranes (diamond) and low-GP internal membranes (circle). b, RGB (left) and GP (right) images of cells labeled with Nile Red, symbols represent lipid droplets (circle) and internal membranes (diamond). c,d,f,g,i,j, TRANES (c,d), $\tau$ phasor (f,g) and $\lambda$ phasor (i,j) for Laurdan (c,f,i) and Nile Red (d,g,j) for the regions depicted in a and b, respectively. e,h, $\tau$ phasor plot (e) and $\lambda$ phasor plot (h) showing the position of the Laurdan (blue) and Nile Red (red) regions. The parula (yellow to blue) colormap represents temporal evolution whereas the RGB colormap (purple to dark red) is an RGB representation of the central wavelength for the associated spectral channel. Cells in a,b are representative of three biological replicates yielding similar results. Scale bars, 10 μm.

Fig. 5 |. Phasor S-FLIM single-cell physiological profiling of living tumor spheroids.

a, Schematic representation of the analysis pipeline. b, MDA-MB231 tumor spheroids are labeled with Nile Red and JC-1 as shown in the RGB image. Using the calibrated spectral signature of the dyes, four FLIM images are unmixed. c, The unmixed images are associated with lipid droplets (top left), internal membranes (top right), mitochondria (bottom left) and active mitochondria (bottom right). d,e, Average intensity images are combined to obtain ${\text{GP}}_{\lambda }$ maps corresponding to lipid droplet concentration (d) and mitochondria activity (e). f,g, FLIM images of the Nile Red spectral emission are used to map lipid order of the internal membranes (f) and hydrophobicity of the lipid droplets (g). The spheroid depicted is representative of three biological replicates yielding similar results. Scale bars, 50 μm.