Demonstration of FLIM flow cytometry at over 10,000 eps

To experimentally demonstrate the high-throughput capability of the FLIM flow cytometer, we measured 6-μm beads (Polysciences 17156-2) and live cells (Jurkat) stained with Calcein-AM flowing on the microfluidic chip at high flow speeds (over 3m/s). Specifically, we repeated the acquisition of 10,000 consecutive events (each data acquisition time: 0.82–0.89seconds) triggered by detecting fluorescence signals emitted from the flowing objects (Fig. 2a and Supplementary Fig. ³) and calculated the event rate (detailed in the Methods section). Consequently, we found that the actual event rates for the beads and cells were 11,297 eps and 10,371 eps, respectively³¹ (Fig. 2b). These event rates were at least one order of magnitude higher than the highest value reported to date²¹ (although the claimed event rates were physically unattainable based on the reported flow speed and conventional event rate definition). Additionally, we confirmed that the triggered events followed the Poisson distribution for both beads and cells (Fig. 2c)^25,32. The absence of time intervals from 0 to 20 μs was due to the data length of each event (20.48μs). Moreover, the abundance of time intervals around 20.5 μs suggests that a larger number of events were triggered once the FLIM flow cytometer was ready for the next data acquisition after completing the previous one. This was primarily attributed to the high sample concentrations (~10⁷ objects/mL), suggesting that the demonstrated event rates approached the upper limits attainable with the FLIM flow cytometer.

Open in a separate window

Fig. 2

Demonstration of FLIM flow cytometry at over 10,000 eps.

The panels (a, line graphs in b, c–e) were generated from one of the image acquisition trials of 10,000 consecutively triggered events, from which 6-μm fluorescent beads (n=9659) and Jurkat cells (n=9036) were extracted (Supplementary Fig. ³). a, Representative images of 6-μm beads obtained at 10,862 eps and of Jurkat cells stained with Calcein-AM obtained at 10,877 eps. Scale bars: 10μm. b Demonstrated event rates. The line graphs show the relationship between the instantaneous (per 100 events), mean, and actual event rates calculated from 10,000 event acquisitions. n represents the number of image acquisition trials conducted. c Distributions of time intervals (n=9999) between adjacent trigger events for beads (top, blue) and cells (bottom, red). Each dot in the scatter plots represents the total number of time intervals within every 5μs as the vertical value and their average as the horizontal value. Histograms show the details of the dots representing 20–25μs time intervals. Solid lines are exponential fits to the dots based on the least absolute residuals. d Distributions of beads and cells in fluorescence intensity and fluorescence lifetime. The double-headed arrows indicate the coefficient of variations (CVs) for fluorescence intensity and fluorescence lifetime. e Distributions of beads (n=9659) and cells (n=9036) in the CVs of fluorescence intensity and fluorescence lifetime pixels within each object. Difference values between two samples represent effect sizes (Cohen’s d), with their standard errors less than 0.04. The violin plots display the median values with white dots, the first and third quartiles with box edges, and 1.5 times the interquartile range (IQR) with whiskers. Source data are provided as a Source Data file.

Next, we statistically evaluated the obtained 10,000 images, from which the beads (n=9659) and the cells (n=9036) were extracted for analysis. We found that fluorescence lifetimes for beads and cells were 1.66±0.06ns [mean±standard deviation (SD)] and 2.78±0.26ns, respectively (Fig. 2d), almost consistent with the literature value of Calcein³³. The negative correlation between fluorescence lifetime and fluorescence intensity for the cells can be attributed to the self-quenching of fluorescent molecules³⁴. Additionally, comparing the coefficient of variations (CVs) for fluorescence intensity and fluorescence lifetime in Fig. 2d, the fluorescence lifetime showed smaller variation between objects than the fluorescence intensity for both samples. Similarly, comparing the CVs of the fluorescence intensity and fluorescence lifetime pixel values within each object, the fluorescence lifetime exhibited smaller variation for both samples (Fig. 2e). These results indicate that FLIM flow cytometry is robust against factors affecting not only inter-object but also intra-object fluorescence intensity variations, compared to fluorescence intensity imaging flow cytometry.

Differentiation of inter-object fluorescence lifetime components

To characterize the capability of the FLIM flow cytometer for distinguishing inter-object fluorescence lifetime components, we measured a mixture of 6.5-μm polymer beads, each characterized by fluorescence lifetime values of 1.72ns, 2.71ns, and 5.54ns (PolyAn 11000006, 11001006, 11002006)³⁵. The images, pseudo-colored in fluorescence lifetime values, revealed the presence of the three types of beads (Fig. 3a), which was supported by the clear subpopulations in the scatter plot (Fig. 3b). Additionally, these beads exhibited three well-separated peaks in the fluorescence lifetime histogram while two out of the three peaks overlapped in fluorescence intensity. This emphasizes the limitation of distinguishing these beads solely based on fluorescence intensity measurements including traditional flow cytometry. Moreover, differentiation of these polymer beads based on their fluorescence wavelengths was not feasible due to the spectral overlap in the 550–800nm range³⁵. Thus, fluorescence lifetime was the sole effective parameter for this classification. Notably, the fluorescence-lifetime-based classification was not impacted by doublet beads or inhomogeneous fluorescent staining, as indicated by the arrows in Fig. 3b (see Supplementary Fig. ⁴ for the scatter plot obtained by measuring each type of beads separately). After segregating the three subpopulations of the beads using gates at 1.6ns and 3.0ns, we calculated the mean and SD of the fluorescence lifetimes for each section. The resulting values were 1.25±0.05ns, 2.09±0.13ns, and 4.70±0.54ns (mean±SD), which were closely aligned with, albeit slightly smaller than, the nominal values (see Discussion section for the discrepancy).

Open in a separate window

Fig. 3

Differentiation of inter-object fluorescence lifetime components by FLIM flow cytometry.

Scale bars: 10 μm. a Representative fluorescence lifetime images (50 rows × 25 columns) obtained from a mixture of 6.5-μm polymer beads with fluorescence lifetime values of 1.7ns, 2.7ns, and 5.5ns. The flow speed was 2m/s. Two independent experiments were performed, resulting in similar results. b Distributions of the fluorescent beads (n=9945). Dashed lines represent 1.6ns and 3.0ns. The right images show the beads randomly picked from the corresponding red boxes in the scatter plot. FL fluorescence intensity, LT fluorescence lifetime. Source data are provided as a Source Data file.

Differentiation of intra-object fluorescence lifetime components

To characterize the ability of the FLIM flow cytometer to distinguish the intracellular environmental differences, we analyzed the fluorescence lifetime of Euglena gracilis cells (a single-celled microalgal species) stained with SYTO16 for labeling their nuclei. While the images revealed the presence of multiple conditions of E. gracilis cells (Fig. 4a), we specifically selected the elongated cells with an eccentricity exceeding 0.95 for further analysis. Their fluorescence lifetime images (Fig. 4b) and the corresponding scatter plots (Fig. 4c) clearly illustrate that SYTO16 in the nuclei shows longer fluorescence lifetimes than autofluorescence from intracellular chlorophyll, irrespective of fluorescence intensity values. Additionally, the fluorescence lifetime of chlorophyll was estimated from the histogram of fluorescence lifetime image pixels (Fig. 4d), resulting in a value of 0.34±0.17ns (mean±SD), aligning with a literature value³⁶. Moreover, to statistically distinguish intracellular local environments based on fluorescence lifetime measurements, we partitioned the acquired fluorescence lifetime images (n=1290) with a cutoff value of 0.9ns and extracted, via CellProfiler³⁷, morphological features of areas with fluorescence lifetime >0.9ns (considered equivalent to the nucleus region) and areas with fluorescence lifetime <0.9ns (considered equivalent to the chloroplast region) in the E. gracilis cells. Consequently, both areas exhibited distinctive values in several morphological parameters that characterize the intracellular structure and composition of the cells (Fig. 4e). This underscores the power and effectiveness of FLIM flow cytometry in assessing intracellular local environments of many cells.

Open in a separate window

Fig. 4

Differentiation of intra-object fluorescence lifetime components by FLIM flow cytometry.

a Representative fluorescence lifetime images (10 rows × 25 columns) of E. gracilis cells stained with SYTO16 and flowing at 2m/s. The images were segmented by a specific threshold value of fluorescence intensity (160 arb. units). Only E. gracilis cells with elongated shapes (eccentricity>0.95) were used for subsequent analyses (b–e). b Representative images of E. gracilis cells with an eccentricity of >0.95, which includes images in Fig. 1e. FL fluorescence intensity, LT fluorescence lifetime. Scale bar: 10μm. Two independent experiments were performed, resulting in similar results for panels a and b. c Scatter plots showing pixel distributions in fluorescence intensity and fluorescence lifetime corresponding to the images in panel b. n represents the number of image pixels. d Histogram of the pixel values from the fluorescence lifetime images in panel b. n represents the number of image pixels. e Representative morphological features of objects occupied with <0.9-ns pixels (blue, n=1290) and with >0.9-ns pixels (red, n=1079). The area ratio is defined as the ratio of an object area with <0.9ns or > 0.9ns to the entire cell area. Source data are provided as a Source Data file.

Investigation of cellular heterogeneity in rat glioma

Investigating the phenotypic diversity of cells within a malignant tumor in vivo is crucial for understanding the pathological features of the tumor³⁸. In this application, we used high-throughput FLIM flow cytometry to characterize intratumor phenotypic heterogeneity in rat glioma. Specifically, we induced tumor formation by implanting rat glioma cells (GS-9L) into a male rat brain, allowed them to develop for 18 days, dissociated the removed tumor, and stained the nuclei, followed by subsequent analysis with FLIM flow cytometry (detailed in the Methods section) (Fig. 5a and Supplementary Fig. ⁵). As shown in Fig. 5b, while flow cytometry (black histogram) and fluorescence lifetime flow cytometry (blue histogram) observed only one and two subpopulations, respectively, FLIM flow cytometry revealed a third subpopulation which derived from intranuclear variations indicated by the fluorescence lifetime gradient from nuclear periphery to center (Supplementary Fig. ⁶). Given that the cultured glioma cells did not show these subpopulations (Supplementary Fig. ⁷), they could result from interactions with diverse tumor microenvironments. To delve deeper, we analyzed the morphological features of cells and their nuclei for each subpopulation (Fig. 5c). Consequently, the subpopulations exhibited variations in morphological features such as size, compactness, and eccentricity. Specifically, the first subpopulation (SP1) showed overall smaller variations. In contrast, the second (SP2) and the third subpopulations (SP3) showed increased morphological complexity, as indicated by factors including the percentage of cells with multiple nuclei (see Methods to refer to calculation methods used). Additionally, the SP3 exhibited larger cell and nucleus areas. These results indicate that FLIM flow cytometry directly probed the diversity of intranuclear physical states within a single tumor, originating from various cell types including glioma cells and immune cells (e.g., lymphocytes and macrophages)³⁹. Remarkably, this separation of the subpopulations, achieved without the use of antibodies (e.g., CD markers), remained unattainable using any combination of fluorescence intensity, fluorescence lifetime, or morphological features of cells and their nuclei (Supplementary Fig. ⁸), suggesting the capability of FLIM flow cytometry in detecting cellular characteristics unidentified with conventional flow cytometry and imaging flow cytometry.

Open in a separate window

Fig. 5

Investigation of cellular heterogeneity in rat glioma with FLIM flow cytometry.

a Experimental procedure. b Distributions of tumor-derived rat glioma cells in fluorescence intensity (black histogram), fluorescence lifetime (blue histogram), and fluorescence lifetime gradient from nuclear periphery to center (red histogram). SP: subpopulation. n=5732. c Comparison of subpopulations in morphological features of cells and their nuclei. TP: total population. Values between two samples represent effect sizes (Cohen’s d), with their standard errors <0.04. The violin plots display the median values with white dots, the first and third quartiles with box edges, and 1.5 times the IQR with whiskers. For the violin plots, the sample sizes are as follows: TP: n=5732; SP1: n=1054; SP2: n=1991; SP3: n=2687. The bar plot illustrates the percentage of cells with multiple nuclei calculated for each cell population, accompanied by the upper and lower bounds of its 95% confidence interval. Source data are provided as a Source Data file.

Detailed schematic of high-throughput FLIM flow cytometry

The detailed schematic of the high-throughput FLIM flow cytometer is presented in Supplementary Fig. ¹. A continuous-wave 488-nm laser (Coherent Genesis CX 488 STM) served as the light source. Two linearly polarized, intensity-modulated beam arrays were produced through the interference of two sets of 50 beams, each deflected by an acousto-optic deflector (AOD, Brimrose TED-300-200-488) in a Mach-Zehnder interferometric configuration. The AODs were driven by multi-tone signals containing 50 frequency components (ranging from 207.12890625 to 307.6171875MHz and from 293.26171875 to 393.75MHz, with a spacing of 2.05078125MHz), generated by an arbitrary waveform generator at 1.2 Giga-samples per second (GS/s) (Signatec PXDAC4800). In one of the optical paths in the interferometer, a laser beam underwent a frequency shift of 207.6171875MHz using an acousto-optic frequency shifter (Isomet 1250C) before reaching the AOD. This resulted in the intensity modulation frequency of each beam array spanning from 20.9960937 to 221.972656MHz, with a 4.1015625MHz spacing. After the generation of the intensity-modulated beam arrays, a pair of mirrors was employed to spatially invert one of the beam arrays. Subsequently, they were orthogonally polarized and recombined in the same optical path by utilizing a half-wave plate and a polarizing beam splitter. After that, they were separated in the direction orthogonal to the array beam by a Wollaston prism (Thorlabs WPQ10). A dichroic filter reflected most of the light for excitation toward a customized microfluidic chip (Hamamatsu Photonics) to excite fluorescent molecules, whereas the weak transmitted light was used as a baseline for calculating the phase delays of the fluorescence signals. After being expanded by a 4-f relay-lens system to align the beam diameter with the pupil of an objective lens, the excitation beams were focused by the objective lens (Olympus UPLSAPO20X; NA, 0.75) into the microfluidic channel to scan the flowing object whose speed was controlled by syringe pumps (Harvard Apparatus, H70-4500). Transmitted light, epifluorescence light from the two beam arrays, and reference light (not incident on the sample) were all separately detected by Si avalanche photodetectors (Thorlabs APD430A/M for the detection of transmitted and the reference light and APD430A2/M for the detection of the fluorescence light). The signals from the photodetectors were converted into digital data by a digitizer with a sampling rate of 1.25 GS/s (Spectrum M4i.2212-x8) and processed using a lab-made LabVIEW program (LabVIEW 2016 and LabVIEW 2021). For our system, it is possible to implement a laser with a different wavelength as an alternative excitation source, as long as we carefully redesign the optical system, including components sensitive to wavelengths, such as AODs and a dichroic filter. Implementing spectrally resolved fluorescence lifetime imaging is also possible by adding dichroic filters or gratings to separately detect fluorescence light based on their wavelengths.

Fluorescence lifetime calibrations

It was necessary to execute fluorescence lifetime calibrations for the FLIM flow cytometer before or after conducting FLIM flow cytometry. This involved measuring a fluorescence lifetime standard sample exhibiting a mono-exponential fluorescence decay. Specifically, we measured 10-µM Rhodamine B in ethanol, a well-known example demonstrating a mono-exponential fluorescence decay with a fluorescence lifetime of around 2.69ns⁵¹, and recorded the phase differences between reference signals and fluorescence signals. To determine the phase differences that originated from the FLIM flow cytometry setup, we subtracted the phase differences stemming from a fluorescence lifetime of 2.69 ns from the recorded phase differences. These calibration phases were subsequently subtracted from the phase differences between the reference signals and fluorescence signals in FLIM flow cytometry, leading to the generation of phase differences that originated from the fluorescence lifetime of the target sample.

Signal processing for image reconstruction

Signal processing to reconstruct images from temporal signals from APDs is a key procedure for FLIM flow cytometry, which is outlined in Supplementary Fig. ². First, we performed Fourier transformation on the temporal signal to obtain its frequency spectrum. Then, we extracted a sub-band spectrum centered on each modulation frequency using the frequency window equivalent to the modulation frequency spacing between neighboring intensity-modulated beam spots. The extracted sub-band spectrum was low-pass filtered by shifting its sub-band to the zero-frequency position and eliminating approximately 10% of its high-frequency components, which can effectively reduce crosstalk noises induced by undesired interference pairs of beam spots while retaining spatial resolution in a typical setting of our system²⁶. After that, we performed inverse-Fourier transformation on the spectrum to obtain a complex line profile of a flowing object. To obtain a bright-field image, we took the absolute value of the profile generated from transmitted light signals and concatenated them. After performing phase calibration of the profile generated from fluorescence signals by using the phases of reference signals and the already obtained calibration phases of Rhodamine B in ethanol, we concatenated amplitude and phase line images and generated an amplitude image pair and a phase image pair. Doubling the number of the image pixels by interpolation, we estimated the horizontal and vertical pixel differences between the image pairs obtained from both fluorescence channels by calculating the 2D cross-correlation of the amplitude image pair after one of the amplitude images was flipped. For Fig. 1d–g and Supplementary Figs. ⁴, and ¹³, a 3×3 median filter was used for phase images after the interpolation. Then, we superposed the amplitude and phase image pairs by translation. The superposed amplitude image (i.e., fluorescence intensity image) was obtained as a simple average of the amplitude image pair. On the other hand, the superposed phase image was obtained as phase values at the center modulation frequency (i.e., the average of the complementary modulation frequency pair) by fitting each phase pair to be superposed under the assumption of the existence of two different fluorescence lifetime components. Lastly, fluorescence lifetime values were calculated based on the $\tau \mathit{=}\tan \theta /{2\pi f}_{\mathrm{center}}$, where $\tau$ is fluorescence lifetime, $\theta$ is the estimated phase, and $f_{\mathrm{center}}$ is the center modulation frequency (around 120MHz). Alternatively, the simple average of two fluorescence lifetime values was calculated if the assumption of the existence of two fluorescence lifetime components was not feasible with the phase pair being superposed.

The fluorescence lifetime images in Fig. 1d, Fig. 2, Fig. 3, and Supplementary Fig. ⁴ were segmented with a threshold value of I_max/2.5, where I_max is the intensity maximum in each fluorescence intensity image. Similarly, the fluorescence lifetime images in Fig. 1e were segmented with a threshold value of I_max/4.5 while those in Figs. 1f, g, 4b–d, 5, and 6, Supplementary Figs. ⁵, ⁹, and ¹⁰ were segmented with a threshold value of I_max/3.5. The original pixel size of the fluorescence lifetime images, without interpolation, is 0.8 μm (in a beam spot alignment direction) and 0.2–0.8 μm (in a flow direction), depending on the flow speed^26,29,30. For example, the original pixel size and the number of pixels of each image in Fig. 2a are 0.8μm×0.7μm and ~48×11 pixels for beads and 0.8 μm×0.8μm and ~50×22 pixels for cells, in the horizontal and vertical (flow) directions, respectively.

Jurkat cell preparation

Jurkat cells (E6.1) used in this study were derived from ECACC and purchased through KAC (Cat. No. EC88042803-F0), with authentication by STR profiling. They were routinely cultured in RPMI-1640 (FUJIFILM Wako Pure Chemical Corporation) supplemented with 10% fetal bovine serum (FBS) (MP Biomedicals), 1% penicillin-streptomycin (FUJIFILM Wako Pure Chemical Corporation) at 37°C in a 5% CO₂ atmosphere. To obtain the data of Jurkat cells in Fig. ​Fig.22 and Supplementary Fig. ³, the cells in the culture medium were centrifuged, suspended in PBS (FUJIFILM Wako Pure Chemical Corporation) at a concentration of 3.2×10⁶ cells/mL, and stained with 5-µM Calcein-AM (Funakoshi, Cat. No. 341-07901) at 37°C for 55min in darkness. Then, they were centrifuged again and resuspended in about 600µL PBS, followed by FLIM flow cytometry. To obtain the data in Supplementary Fig. ^11a-e, the cells in the culture medium were centrifuged and suspended in PBS with 0-µM (negative control) and 3.45-µM doxorubicin (FUJIFILM Wako Pure Chemical Corporation, Cat. No. 040-21521) for 250min. Viability tests with trypan blue after the doxorubicin treatment showed > 90% viability of both samples. 6.90-µM doxorubicin was also used in Supplementary Fig. ^11f. Then, they were stained with 4-µM SYTO16 (Thermo Fisher Scientific, Cat. No. S7578) at 37°C for 65min in darkness, centrifuged, and resuspended in PBS, followed by analysis with FLIM flow cytometry. To obtain the data shown in Supplementary Fig. ¹², the cells were stained with 4-µM SYTO16 in PBS at 37°C for 1hour in darkness, followed by being centrifuged and resuspended in PBS. To obtain the data shown in Fig. 6, Supplementary Figs. ⁹, and ¹⁰, the cells in the culture medium were centrifuged and suspended in the FBS-free culture medium. The cell solution (about 0.9 × 10⁶ cells/mL) was divided into six 1mL samples. Then, the 1mL FBS-free culture medium with 10-µM doxorubicin was added to five out of the six samples so that the samples were treated with 5-µM doxorubicin for 20min, 50min, 80min, 110min, and 140min, respectively. All six samples were centrifuged and suspended in 4-µM SYTO16 FBS-free culture medium at 37°C for 50min in darkness at the same time, centrifuged, and resuspended in PBS, followed by FLIM flow cytometry. To avoid potential systematic errors, we measured all six samples in a random order (measurement order: 20min, 140min, 80min, 50min, 0min, and 110min samples). Also, the Jurkat cells used for Supplementary Fig. ¹¹ and Fig. 6 were imaged on completely different days, two months apart. The range of doxorubicin concentrations and incubation times used in this study was determined based on previous research^42,52.

Euglena gracilis cell preparation

Euglena gracilis NIES-48 cells were obtained from the Microbial Culture Collection at the National Institute for Environmental Study⁵³. They were routinely cultured in AF-6 culture medium at 25°C while being illuminated in a 14:10-hour light: dark pattern. To obtain the data in Fig. 4e. gracilis cells were stained with 2.5-µM SYTO16 for 50min, followed by FLIM flow cytometry.

Rat glioma cell preparation

Glioma-established rat glioma cell line GS-9L was purchased from ECACC and was not authenticated. The GS-9L cells were maintained in Dulbecco’s minimum essential medium (DMEM) (Wako) with 10% FBS and penicillin-streptomycin (100U/mL, Nacalai). The cells were seeded into a tissue culture dish (Invitrogen) 2 or 3 days before implantation into a rat brain. They were cultured in an incubator at 37°C in a humidified atmosphere composed of 95% air and 5% CO₂. Cells were harvested by trypsinization, washed once with DMEM, and resuspended in PBS (Nacalai) for implantation.

Ten-week-old male Fisher 344 (F344/NSlc) rats (Kumagai-Shigeyasu Co., Ltd.), each weighing between 200 and 250g, were housed in standard facilities and provided free access to water and food. Sex differences were not considered in this study, as its aim was the proof-of-principle demonstration of the developed FLIM flow cytometer. For intracranial tumor implantation, rats were anesthetized with 2% isoflurane in a mixture of 30% oxygen and 70% nitrous oxide. During surgery, the rectal temperature was maintained at 37°C ± 0.5°C using a feedback-regulated heating pad (Bio Research Center, BWT-100). They were then placed in a small-animal stereotaxic frame (Kopf). A sagittal incision was made through the skin, and a burr hole was created in the skull by using a twist drill. The cannula coordinates were 0.5mm anterior to the bregma and 3mm lateral from the midline. A cell suspension (5×10⁴ cells/2μL) was used for tumor implantation. This suspension was injected at a depth of 4.5mm from the brain surface. The needle was then removed, and the wound was closed using a nylon suture. For tumor extraction, the rats were anesthetized by aspirating excess isoflurane 18 days after tumor implantation. They were transcardially perfused with saline to remove blood. Tumors were extracted from the rat brain. The tumor tissues were cut into pieces approximately 5mm in diameter and dissected into single cells by using a gentleMACS^TM Dissociator (Milteyi Biotec). The single-cell suspensions were frozen and stocked with a cell freezing buffer (CELLBANKER 1 plus, Takara Bio) to transport from Miyagi to Tokyo in a frozen state along with frozen cultured GS-9L cells. After being defrosted, the tumor-derived rat glioma cells were incubated in RPMI-1640 supplemented with 10% FBS and 1% penicillin-streptomycin at 37°C in a 5% CO₂ atmosphere for 0min (Supplementary Fig. ¹⁴) and for 7hours and 40min (Fig. 5 and Supplementary Figs. ⁵ and ⁸). The cells used for Supplementary Fig. ¹⁴ and for Fig. 5 were from different samples. They were stained with 2.5-µM SYTO16 for 1hour, followed by analysis with FLIM flow cytometry (Fig. 5 and Supplementary Figs. ⁵ and ⁸) or confocal FLIM analysis (Supplementary Fig. ¹⁴). On the other hand, cultured GS-9L were stained with 3-µM SYTO16 for 1hour after being thawed, followed by analysis with FLIM flow cytometry (Supplementary Fig. ⁷). Also, the defrosting protocol was as follows; we warmed ~1mL of the frozen cell solution with 37°C water bath for 1–2min, diluted it with 10mL of 37°C pre-warmed RPMI-1640 supplemented 10% FBS and 1% penicillin-streptomycin, centrifuged the solution, removed the supernatant, and resuspended the cells in the RPMI-1640 solution.

Image analysis

Image analysis for FLIM flow cytometry was performed by using CellProfiler 4.2.1³⁷. The bright-field, fluorescence intensity, and fluorescence lifetime images obtained with the FLIM flow cytometer were respectively aligned (e.g., 100 × 100 image tiles) and output as combined images that were input into CellProfiler. Segmentation of the target objects in Figs. 2 and 3, and Supplementary Fig. ³ was based on fluorescence intensity images while, in Fig. 4, it was based on both bright-field and fluorescence intensity images. In Figs. 5 and 6, Supplementary Figs. ⁷, ⁸, and ¹¹, the segmentation of cells and their nuclei was based on the bright-field images and fluorescence intensity images, respectively. Here, the nuclei were segmented by using a cut-off value of I_max/3.5, where I_max is the maximum fluorescence intensity for each image tile, and subsequent morphological operations such as dilation, erosion, and closing. After the segmentation, morphological features, intensities (i.e., transmitted light intensity, fluorescence intensity, and fluorescence lifetime), and intensity distributions were extracted. When more than two nuclei were segmented within a cell, the one with the smaller eccentricity was chosen for subsequent analyses of nucleus morphology. Despite a limited number of instances, when aggregated multiple cells were segmented as a single object by CellProfiler, we treated them as single cells with multiple nuclei. Conversely, when multiple nuclei were segmented as a single object inside a cell, we treated it as a single cell with a single nucleus. To calculate fluorescence intensity and fluorescence lifetime in Figs. 2, ​,3,3, ​,5,5, and 6, Supplementary Figs. ⁴, ⁷, ⁸, and ¹¹, we simply averaged pixel values within the segmented objects. Regarding the fluorescence lifetime gradient and fluorescence intensity gradient from nuclear periphery to center in Figs. 5 and 6, and Supplementary Figs. ⁶, ⁷, ⁸, and ¹¹, we divided the single nucleus into 4 ring-shaped regions by using the MeasureObjectIntensityDistribution module of CellProfiler, calculated fractional fluorescence lifetime and fluorescence intensity in each contour, performed linear fitting with these 4 fractional values, and determined the value changes from the outermost contour (i.e., peripheral area) to the innermost contour (i.e., central area) using the fitted line (Supplementary Fig. ^6a). Morphological features such as size, compactness, and eccentricity were extracted by using CellProfiler in Figs. 4–6 and Supplementary Figs. ⁷, ⁸, and ¹¹.

Event rate calculation of high-throughput FLIM flow cytometry

In Fig. 2, we conducted the acquisition of 10,000 consecutively triggered events and calculated the event rate, defined as the number of images containing recognizable object(s) acquired per second. The criteria of recognizable objects were an eccentricity of <0.6 for beads and an eccentricity of <0.72 for Jurkat cells (Supplementary Fig. ³). Fewer than one object was recognized per image, even when multiple objects were imaged in a single image acquisition. Consequently, the actual event rates were always smaller than the average values as shown in the line graphs of Fig. 2b. In addition, along with the event rates, we calculated the digital data generation speed of our FLIM flow cytometer³¹. For beads in Fig. 2, the demonstrated event rate of 11,297 eps corresponds to a digital data generation speed of 108.3 MB/s, which was calculated as follows: 11,297 event/sec×3352 Byte/event×3 channels (i.e., bright-field, fluorescence, fluorescence lifetime images) since an effective event acquisition time was 2.6819 μs [11 pixels (the number of pixels of a bead image in a flow direction) divided by 4.1015625MHz (the line acquisition rate)], which corresponds to 3352 Byte/event using the digitizer with a sampling rate of 1.25GHz and 8-bit resolution. Similarly, the demonstrated event rate of 10,371 eps for cells in Fig. 2 corresponds to 198.9 MB/s.

Statistics and reproducibility

Python 3.10.9 was used for statistical analysis. In Figs. 5 and 6, Supplementary Figs. ⁷, ¹¹, and ¹⁴, the effect sizes (Cohen’s d) between the kth and ith samples were calculated, where k<i, as follows⁵⁴:

where $M_i\left(M_k\right)$, ${\mathrm{SD}}_i\left({\mathrm{SD}}_k\right)$, and $n_i\left(n_k\right)$ represent the mean, standard deviation, and sample size of the ith (kth) sample, respectively. In addition, standard errors of the effect sizes were calculated as

In Supplementary Fig. ^{14c, p} values were calculated with Welch’s t-test (two-sided).

The number of images used in this study per sample typically ranged from 5000 to 10,000 to ensure statistical significance comparable to the standard practice in flow cytometry analyses (see each figure for the precise sample size for each experiment). Objects that were not identified by CellProfiler in the images or that returned NaN (Not a Number) values for their morphological features were excluded from the subsequent analysis to prevent the inclusion of noise. Additionally, while a small number of outliers were not shown in the figures to ensure better visibility of the data distributions, all data points are provided in the Source Data. To avoid potential systematic errors during data collection by the FLIM flow cytometer, the measurement order of the multiple samples was randomized where applicable (e.g., Fig. 6). The investigators were not blinded to allocation during experiments and outcome assessment because a single investigator conducted both the measurements and the analysis.

FLIM flow cytometry of cultured rat glioma cells (GS-9L)

We measured cultured GS-9L cells, intended for implantation into a rat brain, with FLIM flow cytometry, allowing for comparison with tumor-derived glioma cells shown in Fig. 5. As depicted in Supplementary Fig. ^7a, distinct subpopulations were not observed in the cultured cells in any parameters used in Fig. 5b. This indicates that the discernible higher-order phenotypic differences observed in the tumor-derived cells may result from interactions with various tumor microenvironments. The differences in the percentage of cells with multiple nuclei and in the variation of the nuclear-to-cell area ratio between the tumor-derived and cultured cells (Supplementary Fig. ^7b) further support this notion. Additionally, the cultured cells exhibited longer fluorescence lifetimes compared to the tumor-derived cells. This disparity could be partly attributed to the larger cell and nucleus areas in the cultured cells (Supplementary Fig. ^7b), potentially resulting in diminished self-quenching for the fluorescent molecules, in addition to the smaller cellular heterogeneity in a cultured condition.

FLIM flow cytometry of Jurkat cells treated with doxorubicin

To further investigate the nucleus changes induced by doxorubicin, we performed FLIM flow cytometry of Jurkat cells treated with doxorubicin for 250min, which was much longer than the duration used in Fig. 6, and calculated the fluorescence lifetime, fluorescence intensity, and transmitted light intensity for each nucleus (Supplementary Fig. ^{11a, b}). Consequently, the distribution of control Jurkat cells exhibited two characteristic subpopulations, suggesting different nucleus states. In addition, higher fluorescence intensity corresponded to a shorter fluorescence lifetime due to the self-quenching of SYTO16⁴⁵. Moreover, the subpopulation with higher fluorescence intensity (i.e., shorter fluorescence lifetime) exhibited larger cell areas compared to the one with lower fluorescence intensity (i.e., longer fluorescence lifetime) (Supplementary Fig. ^11c). After we treated the cells with 3.45-μM doxorubicin for 250min, the two peaks in fluorescence lifetime became shorter and merged toward a single peak although the peaks in fluorescence intensity and transmitted intensity remained partly overlapped (Supplementary Fig. ^11b). This indicates that the surrounding environment of the fluorophores (i.e., SYTO16) expressed by their fluorescence lifetime was varied toward a single state by the doxorubicin treatment, regardless of DNA content. The potential cause of this change may be non-radiative energy transfer between SYTO16 and doxorubicin molecules⁴² supported by chromatin condensation⁵², which is known to be induced by doxorubicin. Meanwhile, it is understandable that since the fluorescence intensity and transmitted light intensity are mainly determined by the amount of the DNA content, the relative positions of the two peaks of their values showed no obvious change before and after the doxorubicin treatment. However, the decreased transmitted light intensity after the doxorubicin treatment may indicate the effect of chromatin condensation or other nucleus structure alterations that contribute to the increased scattering of intensity-modulated light by the nucleus.

Next, we statistically investigated the intranuclear fluorescence lifetime distribution of the Jurkat cells before and after the doxorubicin treatment. Specifically, we divided the area of each nucleus into four ring-shaped regions (i.e., contours) and measured fractional values in each contour with respect to the entire nucleus in the same manner as in calculating fluorescence lifetime gradient (Supplementary Fig. ⁶). As shown in Supplementary Fig. ^11d, the variations in fluorescence lifetime distributions among intranuclear areas were smaller compared to those in fluorescence intensity, indicating the robustness of fluorescence lifetime. Moreover, the fractional fluorescence lifetime values decreased from the center of the nucleus to its periphery. This trend in intranuclear fluorescence lifetime distribution could be explained by the relatively loose and sparse chromatin structure at the central nucleus area⁴³, attributed to frequent gene transcription, leading to the reduced self-quenching of SYTO16. On the other hand, as shown in Supplementary Fig. ^11e, the trend in fluorescence lifetime distribution reversed (i.e., the central areas of the nuclei showed shorter fluorescence lifetimes) after doxorubicin treatment while fluorescence intensity distribution remained largely unchanged. This reversal could be explained by chromatin condensation or other intranuclear structural alterations induced by doxorubicin, leading to changes in the intranuclear distributions of non-radiative energy transfer efficiency between SYTO16 and doxorubicin⁴².

Furthermore, we calculated the same parameters used in Fig. 6b for control cells and those treated with 3.45-μM and 6.90-μM doxorubicin for 250min (Supplementary Fig. ^11e). In comparison to the results in Fig. 6b (0–140min doxorubicin treatment), we observed more pronounced changes in cell and nucleus areas for the 250min doxorubicin treatment while similar changes in fluorescence lifetime and fluorescence lifetime gradient from nuclear periphery to center. This supports the notion, as suggested in Fig. 6b, that the change in fluorescence lifetime precedes the change in cell and nucleus areas.

Validation with TCSPC-based confocal FLIM

To validate some of our findings with conventional FLIM, we conducted experiments using a TCS SP8-FALCON confocal laser-scanning microscope (Leica Microsystems, Wetzeler, Germany) equipped with a pulsed diode laser (PDL 800-B, 470nm; PicoQuant, Berlin, Germany) operating at a repetition rate of 20MHz. The emitted fluorescence from 500 to 640nm was obtained through an HC PL APO 63×/1.40 Oil CS2 objective (Leica Microsystems). We generated a fluorescence lifetime image for Supplementary Fig. ^12a in a 256×256-pixel format with 100 repetitive measurements, while we generated images for Supplementary Fig. ^14a in a 512×512-pixel format with 15 repeat measurements. To support the findings in Fig. 6, we obtained a fluorescence lifetime image of Jurkat cells stained with SYTO16. The TCSPC-based confocal FLIM revealed not only internuclear but also intranuclear differences in fluorescence lifetime (Supplementary Fig. ^12a,b). In particular, the dim regions near the center of the nuclei showed longer fluorescence lifetimes, as evidenced by the negative correlations between fluorescence intensity and fluorescence lifetime pixel values (Supplementary Fig. ^12c), which is consistent with our findings in Fig. 6b. To support the findings in Fig. 5, we obtained fluorescence lifetime images of SYTO16-stained cells derived from distinct male rat glioma. The TCSPC-based confocal FLIM revealed clear cell-cell differences in fluorescence lifetime (Supplementary Fig. ^14a). Furthermore, by applying a gate of a fluorescence lifetime of 7ns to segregate nuclei with shorter and longer lifetimes (Supplementary Fig. ^14b) and analyzing their morphological features, we found consistency in nucleus morphology with our findings in Fig. 5c (Supplementary Fig. ^14c).

To analyze images obtained with the TCSPC-based confocal FLIM, we used Python 3.10.7 and the OpenCV 4.5.1.48 library. Initially, we converted fluorescence intensity images to grayscale and applied a median filter (cv2.medianBlur, ksize=3) to eliminate noise. Next, we calculated an adaptive threshold using the Gaussian method to binarize the images, aiming to segment the nuclei from the background. Then, we applied morphological closing twice (cv2.MORPH_CLOSE, ksize1=5, ksize2=3) to fill in abnormal nuclear pixels and used a median filter (cv2.medianBlur, ksize=7) once more to filter out background noise generated during processing. After that, we utilized the Laplacian edge detection function to compute the second-order gradient of the images, obtaining boundary information for each nucleus. The edge information was then converted into contours using the contour detection function. These extracted contours serve as a mask that can be used for contour extraction of fluorescence lifetime images (Supplementary Fig. ^12b). Finally, using morphology calculation functions such as cv2.contourArea and cv2.arcLength, we obtained morphological parameters of the nucleus.

Sensitivity

Our FLIM offers unique sensitivity compared to conventional laser-scanning imaging techniques. This is due to the multiplexing of image pixel acquisition in our FLIM, which allows for an increase in pixel dwell time, and thus signal strength per pixel by a factor of N_spots, the number of beam spots, compared to conventional techniques with the same line-scan rate²⁶. However, the superposition of shot noises from all beam spots in our FLIM leads to a decrease in the signal-to-noise ratio (SNR) by a factor of √N_flu, where N_flu represents the number of beam spots used for fluorescence excitation per line scan, assuming a uniform fluorophore distribution. As a result, the SNR improvement can be increased by a factor of $\sqrt{N_{\mathrm{spots}}}/\sqrt{N_{\mathrm{flu}}}$.

This work was supported by JSPS Core-to-Core Program (K.G.), JSPS KAKENHI Grant Numbers 19H05633 and 20H00317 (K.G.), Ogasawara Foundation (K.G.), Nakatani Foundation (K.G.), Konica Minolta Foundation (K.G.), Humboldt Foundation’s Philipp Franz von Siebold Award (K.G.), Humboldt Association of Japan (K.G.), Precise Measurement Technology Promotion Foundation (H.M.), Konica Minolta Imaging Science Award (H.M.), JST PRESTO (JPMJPR1878) (K.H.), JST FOREST (21470594) (K.H.), JSPS Grant-in-Aid for Scientific Research (B) (23K23297) (K.H.), JSPS Grant-in-Aid for Young Scientists (20K15227) (K.H.), Research Foundation for Opto-Science and Technology (K.H.), JSPS KAKENHI Grant Numbers 21J10600 and 24K18149 (H.K.), Konica Minolta Light Future Incentive Award (H.K.), and JSPS KAKENHI Grant Number 23K27432 (S.R.). We thank Mayu Sehara for her help with the cell sample preparation. The manuscript underwent editing with the assistance of a large language model (LLM).