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

Microscopy. All data were acquired with a custom-built microscope described in the Supplementary Information . The solution measurements and cells acquired for the FRET and Laurdan experiments were acquired with 405 nm excitation light and a 410-nm long pass dichroic mirror (Semrock) whereas measurements for the Nile Red and tumor spheroids experiments were performed with a 488-nm excitation light and a 500-nm long pass dichroic mirror (Semrock). A Nikon ×100/1.45 numerical aperture (NA) Oil objective was used for the Laurdan and Nile Red experiments and the measurements in solutions, an Olympus PlanApo N ×60/1.45 NA Oil objective was used for all other cells experiments and an Olympus UPlanSApo ×20/0.75 NA air objective was used for imaging the tumor spheroids. Cell culture and transient transfection. CHO-cells were cultured in low glucose medium (Ham’s F-12K Medium Kaighn’s modification, Gibco, ThermoFisher Scientific), supplemented with 10% v/v fetal bovine serum (heat inactivated FBS, GenClone) and 1% v/v penicillin/streptomycin solution 100× (10,000 units of penicillin and 10 mg ml −1 streptomycin in 0.85% saline solution, GenClone), in a 37 °C and 5% CO 2 incubator. MDA-MB231 cells were cultured in DMEM medium (DMEM, high glucose, with l-glutamine, GenClone), supplemented with 10% v/v FBS (heat inactivated FBS, GenClone) and 1% v/v penicillin/streptomycin solution 100x (10,000 units of penicillin and 10 mg ml −1 streptomycin in 0.85% saline solution, GenClone), in a 37 °C and 5% CO 2 incubator. One day before transfection, MDA-MB231 cells were seeded in an eight-well chambered coverglass (Thermo Scientific, Nunc Lab-Tek), previously coated with 2 μg ml −1 fibronectin in Dulbecco’s phosphate buffer solution (fibronectin human plasma 0.1% solution, Sigma-Aldrich; DPBS 1x without Ca, Mg, Phenol Red, GenClone). Transfections were performed with Lipofectamine 3000 (Invitrogen, ThermoFisher Scientific) diluted in Opti-MEM I reduced-serum medium (Gibco, ThermoFisher Scientific), according to the manufacturer’s instructions. Approximately 1 d after transfection, Opti-MEM medium was substituted with culture medium (DMEM medium supplemented with 10% v/v FBS and 1% v/v penicillin/streptomycin solution 100x). The following FRET standard plasmids, used for transfections, were a gift from S. Vogel: C5V (Addgene plasmid no. 26394), C32V (Addgene plasmid no. 26396), mVenus N1 (Addgene plasmid no. 27793), mCerulean N1 (Addgene plasmid no. 27795), VCV (Addgene plasmid no. 27788) and VCVV (Addgene plasmid no. 27789). pCAG-CFP was a gift from C. Cepko (Addgene plasmid no. 11179). YPet-N1 was a gift from P. Daugherty and M. Davidson (Addgene plasmid no. 54637). THe H3K9me3 biosensor was a gift from Y. Wang (Addgene plasmid no. 120802). For each plasmid, bacterial growth was spread on Luria-Bertani medium (LB) agar plates with the appropriate antibiotic; single colonies were inoculated in a liquid culture of LB and antibiotic and placed in a shaking incubator (37 °C, 270 r.p.m.) overnight. Plasmid DNA was then purified from bacterial growth using a mini-prep kit (QIAprep Spin Miniprep Kit, QIAGEN) and concentration was verified with Nanodrop2000 (ThermoFisher Scientific). MDA-MB231 spheroids were formed starting from a culture of 20,000 cells per spheroid mixed with their medium and a 2% Matrigel (Corning Matrigel Matrix) solution in 96-well round-bottom plates (Corning Costar Ultra-Low Attachment Multiple Well Plate). After a brief centrifugation (5 min at 1,000 r.p.m.) and 3 d of incubation (37 °C, 5% CO 2 ), the spheroids were transferred in eight-well chambers (Nunc Lab-Tek Chambered Coverglass) and embedded in a 3.0 mg ml −1 collagen Type-1 gel matrix (Corning Collagen I, Rat Tail, 100 mg). Media and dyes were added on top after collagen polymerization and incubated at 37 °C and 5% CO 2 before imaging. MDA-MB231 tumor spheroids. MDA-MB231 cells were cultured in DMEM medium (DMEM, high glucose, with l-glutamine, GenClone), supplemented with 10% v/v heat inactivated FBS (GenClone) and 1% v/v penicillin/streptomycin solution 100x (10,000 units of penicillin and 10 mg ml −1 streptomycin in 0.85% saline solution, GenClone), in a 37 °C and 5% CO 2 incubator. MDA-MB231 spheroids were formed starting from a culture of 20,000 cells per spheroid mixed with their medium and a 2% Matrigel (Corning Matrigel Matrix) solution in 96-well round-bottom plates (Corning Costar Ultra-Low Attachment Multiple Well Plate). After a brief centrifugation (5 min at 1,000 r.p.m.) and 3 d of incubation (37 °C, 5% CO 2 ), the spheroids were transferred in eight-well chambers (Nunc Lab-Tek Chambered Coverglass) and embedded in a 3.0 mg ml −1 collagen Type-1 gel matrix (Corning Collagen I, Rat Tail, 100 mg). Media and dyes were added on top after collagen polymerization and incubated at 37 °C and 5% CO 2 before imaging. Solutions, fluorescent labeling and treatments. The set of solutions used for calibration was prepared at a concentration of 7 μM in PBS pH 7.4 and consists in a solution of Alexa 405 (Thermofisher Scientific), fluorescein (Thermofisher Scientific) and ATTO490LS (ATTO-tec). The mixture of three solutions is composed by a mixture of Alexa 405 at 400 nM, SeTau 405 (Seta Biomedicals) at 2 μM and ATTO 490 LS at 8 μM in PBS pH 7.4. Alexa 405 and SeTau 405 display an absorption maximum at 405 nm, and ATTO 490 LS absorbs at 38% of its absorption maximum. The concentrations were adjusted to have a photon amount of the same order of magnitude from the three emission spectra (67, 16 and 17%, respectively). Live cells in Fig. 2a were stained with 1:5,000 ViaFluor 488 (Biotium), 200 nM MitoTracker Orange CMTMRos (Thermofisher) and 1 μM LysoTracker DND-99 Red (Thermofisher), incubated 30 min and washed twice before imaging. Dyes were diluted to have less than 0.1% DMSO in the medium. Immunostaining samples were prepared as follows. One day before immunostaining, MDA-MB231 cells were seeded in eight-well chambered coverglass (Thermo Scientific, Nunc Lab-Tek), previously coated with 2 μg ml −1 fibronectin in DPBS (fibronectin human plasma 0.1% solution, Sigma-Aldrich; DPBS 1x without Ca, Mg, Phenol Red, GenClone). Cells were washed twice with warm DPBS 1x and fixed with paraformaldehyde 4% in PBS (Ready-to-Use Fixative, Biotium) for 15 min at room temperature. Cells were washed with DPBS 1x, permeabilized with 0.1% Triton X-100 in DPBS 1x for 15 min at room temperature and washed with DPBS 1x. The samples were incubated for 1 h at room temperature with a blocking solution of 0.1% Triton X-100 + 1% BSA (albumin from bovine serum, Sigma) + 10% Normal Goat Serum (Abcam, ab7481) in DPBS 1x, then incubated in the dark for 1 h at room temperature with Phalloidin-Alexa594 (Thermofisher) diluted 1:400 in DPBS 1x + 1% BSA. After washing with DPBS 1x, the samples were incubated overnight at 4 °C with anti-TOMM20-Alexa555 (Abcam, ab221292) antibody diluted in the blocking buffer 1:250. After washing with DPBS1X, samples were incubated with NucSpot 488 (Biotium) diluted 1:30,000, mounted with Fluoromount (Sigma) mounting medium and preserved in the dark at 4 °C. For the solvatochromic dyes staining, stock solutions of Laurdan (Toronto Research Chemicals) and Nile Red (Thermofisher Scientific) at 30 mM in DMSO were prepared and diluted 1:10,000 at a final concentration of 3 μM in cellular media, left incubating in a 37 °C and 5% CO 2 incubator for 30 min. Tumor spheroids were labeled with 3 μM Nile Red and 2 μM JC-1 diluted in cellular media. JC-1 was diluted 1:1,000 from a 2 mM stock solution in DMSO. Hypo-osmolar medium was prepared by diluting DMEM 1:1 in milli-Q water, reducing the osmolarity by half with respect to DMEM. Hyper-osmolar medium was prepared by diluting PBS 10× 1:10 in DMEM, increasing by osmolarity by approximately a factor of two with respect to DMEM. Cells were plated as described above, warm hypo- or hyper-osmolar medium was substituted to culture medium and cells were imaged within 1 h. Simulations and computer code. All simulations were performed in MATLAB 2019b. Data for Fig. 1b – d were simulated as follows: a variable number of photons were drawn from a single exponential probability distribution corresponding to a lifetime of 0.5 ns, of which a histogram was computed to model a lifetime decay. To mimic a realistic measurement, the histogram was calculated over five times the actual period and the photon arrived in each of the five periods were cumulated together. This is particularly important when simulating long lifetimes or IRF convolution, in which photons may arrive in periods adjacent to the one in which the fluorescence was excited. For fitting, the initial parameters (height and exponential constant) were initialized to the true values. A set of 10,000 realizations was simulated for every number of photons and the error was computed as the standard deviation of the resulting values of lifetime, whereas the bias was computed as the average of the resulting values of lifetime minus the simulated value. Data for Fig. 1e – j were simulated as follows: a fixed number of photons (1,000) were drawn from a single exponential probability distribution corresponding to a lifetime of 2.5 ns and spread following a Gaussian distribution centered in 0 with variable FWHM to mimic the instrument IRF. A constant background affected by Poisson noise with variable average number of photons was also added. For the speed comparison, pure exponentials with a lifetime of 1 ns were simulated and fitted with the MATLAB function ‘fminsearch’. The same dataset was also transformed into phasors and the lifetime was obtained as τ = s ω g (see Supplementary Information for further details). The average computational time and standard deviation was computed from a set of 100 realizations with a varying number of iterations. For the data in Fig. 1e – j the median of 1,000 realizations was considered, the curves were fitted with a single exponential convolved with a known Gaussian IRF and affected by a constant background. The S-FLIM datasets simulated in Supplementary Fig. 9 were simulated as follows: three Gaussian spectra with center at 480, 540 and 620 nm were simulated and standard deviations of 30, 35 and 45 nm were simulated. Each spectrum was simulated with a different lifetime: spectrum one was a combination of 8 and 0.2 ns, spectrum two was a combination of 4 and 1 ns whereas spectrum three was a pure exponential with lifetime 2.5 ns. The total number of photons for the dataset was set at 500,000 and noise was added with the PoissonNoise function in MATLAB. We simulated 100 realizations, each of which was analyzed and unmixed by the Phasor S-FLIM blind unmixing algorithm described in the main text. Colormaps in Fig. 5 were obtained by the superjet function: A. Vallmitjana Lees (2020). Superjet ( https://www.mathworks.com/matlabcentral/fileexchange/74715-superjet ) RGB images were reconstructed by transforming the image relative to each spectral channel to an RGB image and successively combining them in a single RGB image. Wavelength to RGB conversion was performed using the function SpectrumRGB (J. Mather (2020). Spectral and XYZ Color Functions MATLAB Central File Exchange). Phasor S-FLIM. Phasor S-FLIM analysis was performed on the entire image (solution measurements) or on a thresholded image (cells and spheroids) with a user-defined threshold. No spectral correction was applied to any dataset in the paper. The number of photons in Fig. 2d – f was reduced by considering only a fraction of the photons detected in each pixel from 10% (0.6 photons per pixel per dye) to 100% (11.3 photons per pixel per dye, Fig. 2c ). FRET efficiency in Fig. 3 for Phasor S-FLIM was calculated as an average between the FRET efficiency image obtained by the lifetime of mCerulean after unmixing and from the spectral FRET computed from the intensity of the mCerulean and mVenus unmixed channels. We report the lifetimes of mCerulean and ECFP to be multi-exponential, as also reported elsewhere 57 , 58 ; therefore, their lifetime for the FRET calculation was fixed at 3.3 and 3.0 ns, respectively, to be consistent with the literature 29 , 59 . Regions in Fig. 4a , b were segmented by k -means clustering using GP and intensity information with seven clusters. A user-defined number of clusters was merged together for each of the considered regions. TRES from pixels assigned to the same cluster were combined to perform the analysis shown in Fig. 4c – j , a representative region is shown by the markers. Pure components for Nile Red and JC-1 in Fig. 5 were obtained by Phasor S-FLIM unmixing of spheroids labeled with the single dyes, the spectra were then stored and their first and second harmonic phasors were used to unmix the sample labeled with both dyes.