2.1.

Multiphoton Fluorescence Lifetime Imaging in the Time Domain

Fluorescence lifetime imaging microscopy (FLIM) methods currently available can be broadly classified into two groups, namely, (1) time-domain and (2) frequency-domain FLIM. For a detailed discussion of these methods and a comparative survey of relative merits∕demerits of these methods, see an earlier review article.⁹ Briefly, in the case of time-domain methods, the specimen is illuminated with pulsed light and the subsequent fluorescence emission decay is recorded by fast detectors. By fitting the observed emission decay to an appropriate mathematical model, the fluorescence lifetimes of the different fluorescent species within the specimen are extracted. Regardless of the method adopted, either single-photon or multiphoton excitation can be used to measure lifetimes. In this article, we limit our discussion only to time-domain FLIM systems employing multiphoton excitation.

The most explored method in measuring lifetime in a scanning microscope is by employing time-gated detection pioneered by Sytsma ¹⁰ and Gerritsen ¹¹ The principle of detection is based on sampling fluorescence intensity decay curve with two or more time gates (typical $\text{width}=2\mathrm{ns}$ ), which can be opened sequentially after every laser pulse. Figure 2(a) shows a schematic of the detection with four time gates. This system has proved capable of high photon efficiency and high maximum count rate capability in measuring rapid lifetime determination in living cells. It has also been found that the overall performance of multigate detection FLIM system is dominated by the transition-time spread (TTS) characteristics of the photomultiplier tube detector.

Fig. 2

Schematics of time-domain fluorescence lifetime imaging microscopy. (a) Multigate detection method showing four time windows with a typical gate width of $\sim 2\mathrm{ns}$ each. The specimen is excited with femtosecond laser pulses and the fluorescence intensity decay curve is sampled by sequentially opening the time windows and by collecting the integrated intensities $(I_1,I_2,\dots )$ . Lifetime $\tau$ is computed by ratiometry. (b) In the time-correlated single-photon counting method, following multiphoton excitation of the specimen, individual photons are detected. By sorting the collected photons based on the time elapsed between the excitation pulse and their detection time into individual time bins (width $\sim 20\mathrm{ps}$ ) by fast electronics, a photon histogram is built based on the individually counted photons, as shown in (c). Lifetime is then calculated by fitting the profile of this histogram. (d) In the StreakFLIM method, the streak camera detects the entire fluorescence decay (time resolution $\sim 15\mathrm{ps}$ ) without any time gating or prolonged exposure to build the photon histogram. The lifetime is calculated by fitting the decay curve at each pixel. The sequence of events in data acquisition is schematically depicted in (e).

Another approach that is particularly suitable in low-light applications is time-correlated single-photon counting¹² (TCSPC). The principle of this method involves detection of single photons of a periodic light signal, the measurement of their detection times and constructing a histogram of the detection times of the single emitted photons from these individual time measurements. Figure 2(b) demonstrates the principle of TCSPC approach. Photon-counting detectors are well suited for low-light-level detection as well as for providing quantized pulses for every photon event.^{13, 14} This makes the measurement of lifetimes more accurate, although the detectors suffer from poor dynamic range as compared to their area detector counterparts. The maximum count rates detectable by the microchannel plate photon detectors in TCSPC are limited to ${10}^6\text{photons}∕\mathrm{s}$ , which limits the dynamic range in this approach. Improved fast detectors, compatibility with laser scanning microscope modules, and advanced data analysis algorithms have made TCSPC FLIM systems a reliable option. Furthermore, in low-light situations, the histogram approach used in TCSPC systems can lead to significantly longer data acquisition times, thereby making live-cell FLIM imaging difficult.

2.2.

Multiphoton FLIM Using a Streak Camera

Our group recently developed a novel multiphoton FLIM system using a streak camera^{15, 16} (StreakFLIM). An important feature of this system is that the entire decay in every pixel is collected without any time gating or prolonged excitation to build photon count histogram. The lifetime parameter can be extracted in each pixel by fitting to the intensity decay in every pixel. Figure 2(c) demonstrates the principle of operation of StreakFLIM system, and Fig. 3 provides the optical schematic of the system. A more detailed technical description of the system can be found in earlier publications.^{15, 16} Briefly, the heart of the StreakFLIM system is the streak camera (or streakscope), a device that can measure ultrafast light phenomena with a temporal resolution¹⁷ as low as $2\mathrm{ps}$ . The streak camera consists of a photocathode surface, a pair of sweep electrodes, a microchannel plate (MCP) to amplify photoelectrons coming off the photocathode, and a phosphor screen to detect the amplified output of MCP. Individual optical pulses (fluorescence emission) are collected from every point along a single line as the beam scanner scans across the $x$ axis. When these optical pulses (originating at different points in space and time) strike the photocathode surface, photoelectrons are emitted. As the photoelectrons pass between the pair of sweep electrodes, a high voltage, synchronized with the excitation pulse, is applied to these sweep electrodes. This sweep voltage steers the electron paths away from the horizontal direction at different angles, depending on their arrival time at the electrodes. The photoelectrons then get amplified in MCP and reach the phosphor screen, forming an image of optical pulses arranged in the vertical direction according to the time of their arrival at the sweep electrodes. The earliest pulse is arranged in the uppermost position and the latest pulse is in the bottom-most portion of the phosphor image. The resulting streak image has space as the $x$ axis and time as the $y$ axis. Besides intensity and time information, the streakscope can be used to obtain either spatial or wavelength information. The higher time resolution of the streakscope (Model C4334: $15\mathrm{ps}$ , Hamamatsu Photonics, Japan) as compared with the fast photomultiplier tube (PMT) $(\sim 250\mathrm{ps})$ ∕microchannel plate $(\sim 50\mathrm{ps})$ makes this system ideal for measuring very short lifetimes reliably well. Figure 2(c) demonstrates the principle of operation of Streak-FLIM.

Fig. 3

Optical layout of StreakFLIM system: G1, horizontal galvanometer mirror; G2, vertical galvanometer mirror; L1, focusing relay lens; C, coupling lens for sideport optics; I, imaging lens; O, microscope objective lens; D1, dichroic mirror; and PC, photocathode.

2.3.

Cells, Tissues, and Microscopy

HEK 293T and HeLa cells were obtained from the American Type Culture Collection (ATCC). HEK 293T cells were cultured in Dulbecco’s modified eagle medium (DMEM) containing 10% fetal bovine serum on poly-d-lysine (Sigma, St. Louis, Missouri) coated coverglasses (Nalge Nunc International, Napervile, Illinois). HeLa cells were cultured in minimum essential medium (MEM), supplemented with $2\mathrm{mM}∕\mathrm{L}$ of L-glutamine and 10% of fetal calf serum. Primary hepatocytes were isolated from C57BL∕6 mice by perfusing the liver, as described in Ref. 18. Mice were obtained from the National Institute of Aging. Mice were anesthetized with Ketamine∕S.A. Rompun cocktail and liver was perfused through the hepatic portal vein with $0.5\text{‐}\mathrm{mM}$ EGTA in ${\mathrm{Ca}}^{2+}$ -free Earle’s balanced salt solution (EBSS) for $5\min$ , followed sequentially by wash buffer (0.5% bovine serum albumin and $1.8\text{‐}\mathrm{mM}$ $\mathrm{Ca}{\mathrm{Cl}}_2$ in ${\mathrm{Ca}}^{2+}$ -free EBSS) and by collagenase buffer (0.06% collagenase, 0.5% bovine serum albumin, and $5\text{‐}\mathrm{mM}$ $\mathrm{Ca}{\mathrm{Cl}}_2$ in ${\mathrm{Ca}}^{2+}$ -free EBSS) for $10\text{to}15\min$ . The liver was then further perfused with wash buffer for $5\min$ and hepatocytes were collected. After washing three times with cold wash buffer, the hepatocytes were cultured in William’s E media supplemented with 5% fetal bovine serum (FBS), $4\mu \mathrm{g}∕\mathrm{ml}$ insulin, $1000\text{unit}∕\mathrm{ml}$ penicillin G sodium, $100\mu \mathrm{g}∕\mathrm{ml}$ streptomycin sulfate, and $250\mathrm{ng}∕\mathrm{ml}$ amphotericin B in collagen coated chambered cover glasses. All cell lines were maintained at $37°\mathrm{C}$ in an atmosphere of 5% of $\mathrm{C}{\mathrm{O}}_2$ and 95% air. Tissue samples were collected from anesthetized C57BL∕6 mice. Freshly collected tissues (skin, heart, kidney) were cut to pieces ( $\sim 6\text{to}8\mathrm{mm}$ thickness) and placed between two coverslips separated by parafilm filling for obtaining optically flat specimens. Tissue specimens were used immediately for microscopy. Nematode worms (Caenorhabditis elegans) expressing green fluorescent protein in neurons were maintained in an imaging chamber supplemented with nutrient buffer and $5\text{‐}\mathrm{mM}$ sodium azide. Streak-FLIM or TCSPC FLIM systems were employed to obtain autofluorescence images from cells and tissue specimens. Both these systems are equipped with a Ti:sapphire laser (Coherent Inc., California) for multiphoton excitation and dedicated scanning systems. Data acquisition∕analysis was performed by AquaCosmos software (StreakFLIM; Hamamatsu Photonics, Japan) and SPCM∕SPC Image software (TCSPC FLIM; Becker & Hickl, Berlin, Germany). Multicomponent analyses were performed by profile fitting in TCSPC FLIM system and by a global analysis module in StreakFLIM system.

3.1.

Autofluorescence Discrimination in C. elegans

Green fluorescent protein (GFP) and its various mutants have been extensively used in recent times for monitoring a variety of cellular processes in fixed and living cells.^{19, 20} In most circumstances, these proteins are overexpressed in cells so that the weak autofluorescence contributions can be ignored as “noise,” or suitably subtracted prior to quantification while imaging cells. However, in the case of tissues the comparable intensities of autofluorescence and the signal from fluorescent proteins make the quantification less straightforward. Figure 4 shows a section of a live nematode worm (C. elegans) expressing GFP in neurons. The multiphoton intensity image shows neuronal GFP with a bright ganglion. One can also see autofluorescence granules with similar intensities in the gut of the worm. This intestinal autofluorescence can be artifactually interpreted as the leakage of GFP inside the lumen if one compares only the intensity. On the other hand, the lifetime image unequivocally discriminates the neuronal GFP from the surrounding autofluorescence based on the lifetime decay curves which are shown in the accompanying graphs [Figs. 4(d) and 4(e)]. The whole-field histogram can be deconvolved to give two peaks corresponding to autofluorescence and one peak corresponding to GFP. Individual pixels selected each at GFP and autofluorescence region (lumen) show distinct decay characteristics with a high contrast. Beyond cuvette measurements, many studies attempted to measure autofluorescence in cells and tissues using imaging modalities where the primary limitation was found to be the excitation sources∕filters in UV region (most of autofluorescence gets excited between 270 and $450\mathrm{nm}$ ) and poor microscope optics causing aberrations in UV region. In addition to imaging in cells, fluorescent probes are now designed for staining tissues, and more recently, tumor cells expressing bright fluorescent proteins are implanted inside mouse models to monitor, for example, the progression of tumor and metastasis within the animal.²¹ All these methods have been primarily limited to wide-field, filter-based, single-photon excitation approaches. Furthermore, these imaging modalities were being employed more as a visualization tool rather than for precise quantification. The major hurdle in quantitative imaging of fluorescence in tissues is the huge autofluorescence background that reduces the signal-to-background enormously. As can be seen from Fig. 4, fluorescence lifetime contrast can provide a better reliability in quantification of fluorescent regions despite a large autofluorescence background.

Fig. 4

Autofluorescence discrimination in C. elegans. (a) to (c) Transmission, multiphoton intensity, and lifetime images obtained from a live nematode worm (C. elegans). The specimen was excited at $850\mathrm{nm}$ and fluorescence emission was collected with a short-pass filter $\left(\mathrm{SP}640\mathrm{nm}\right)$ . Lifetime images were obtained with the commercial TCSPC module coupled to a Zeiss 510 microscope. A lifetime histogram of the whole field of view (d) shows two distinct autofluorescence peaks and one peak corresponding to green fluorescent protein. (e) Intensity decay at individual pixels marked in (c), showing distinct decays for GFP and autofluorescence.

3.2.

Autofluorescence in Living Cells∕ex vivo Tissues

To investigate the molecular origin of cellular autofluorescence in living cells∕ex vivo tissues, multiphoton fluorescence lifetime imaging measurements were carried out on living cells maintained in culture media at room temperature. Three mammalian cell lines were used: (1) HEK 293T epithelial cell line, (2) primary hepatocyte cultures, and (3) HeLa cancer cells. Figure 5(a) shows the representative lifetime images obtained by exciting the specimens at $790\mathrm{nm}$ . Similar measurements were performed at multiple excitation wavelengths and the accompanying lifetime histogram [Fig. 5(b)] shows the lifetime distributions obtained by exciting the specimens at three representative wavelengths. It was observed that each cell line has its own characteristic lifetime signatures of autofluorescence. Figure 6 shows the lifetime images and histograms obtained from three freshly excised ex vivo tissues from a $20\text{‐}\text{month}$ -old female mouse. All these images were obtained by exciting the specimens at $790\mathrm{nm}$ and by collecting fluorescence emission with a short pass filter (SP $640\mathrm{nm}$ ). One can clearly see that there are obvious structural differences in different tissues that are accompanied by distinct lifetime signatures. Besides these structural differences, the autofluorescence intensities for all these tissues were similar in magnitude.

Fig. 6

Representative lifetime images obtained from mouse tissues for (a) transmission, multiphoton intensity, and multiphoton lifetime images obtained from heart, kidney, and skin of a mouse. Specimens were excited at three excitation wavelengths; 750, 790, and $840\mathrm{nm}$ . (b) Representative lifetime histogram (at $790\text{‐}\mathrm{nm}$ excitation) showing single component lifetime analysis in these tissues, demonstrating clear lifetime contrast in tissues with varying metabolic activity. Baselines for histograms of skin and heart are shifted in the $y$ axis (arbitrary scale) for the sake of clarity.

3.3.

Molecular Origins of Autofluorescence in Cells∕Tissues

The major molecular players in governing the cellular autofluorescence are the reduced nicotinamide adenine dinucleotide (NADH) and the flavoproteins.^{22, 23} Minor contributions from elastin and collagen are also reported.²⁴ Table 1 shows the common autofluorescence molecules and their typical absorption∕emission maxima with single photon excitation. Historically, the endogenous autofluorescence of cells was used to monitor the metabolic state of living tissues. It is known that the ratio of oxidized and reduced electron carriers gives a measure of steady state cellular metabolism. A quantitative estimate of this redox ratio can be calculated by separating out the contributions from reduced pyridine nucleotides (NADH and NADPH) and oxidized flavoproteins (FP) in cells. There have been numerous reports in the past employing either spectroscopic methods or intensity-based imaging methods for estimating this ratio.^{1, 2} As one can envisage, the reliability in estimating the redox ratio based on spectroscopic∕intensity variations is greatly affected by the artifacts, which include absorption and scattering of the excitation light as well as due to spatial variation of concentration of metabolic cofactors∕enzymes in the cell. Lifetime measurements are relatively insensitive to these artifacts and therefore can provide a reliable estimate of metabolic processes in living cells. Lifetime histograms in Fig. 5(b) were obtained by assuming a single exponential decay model in the analysis of lifetime images. Although multiexponential decays were observed in all the cells studied, this monoexponential approximation illustrates a few intriguing features: first, normal cell lines (primary hepatocyte cultures and HEK293T epithelial cell line) have relatively higher lifetime than the cancer cell lines (HeLa), irrespective of the excitation wavelength used. Second, the mean lifetime values calculated with different excitation wavelengths vary with different excitation wavelengths, suggesting that multiple lifetime contributions with varying amplitudes are present in different cell lines. Finally, the lifetime images are independent of wide variation in concentration that was observed across the cell. A preliminary analysis of multicomponent lifetime was carried out to estimate the different contributions and is summarized in Table 2 . Two major lifetime contributions were found in all the cells: (1) ${\tau }_1\sim 0.48\pm 0.1\mathrm{ns}$ and (2) ${\tau }_2\sim 2.4\pm 0.3\mathrm{ns}$ . The relative amplitudes of these contributions vary among the different cell lines investigated. The short lifetime component ${\tau }_1\sim 0.48\mathrm{ns}$ corresponds to free NADH and closely resembles that observed in other studies as well as to that observed for $\beta$ -NADH solution in our own experiments (data not shown). Earlier studies in NADH solutions with binding proteins have reported^{25, 26} a longer lifetime value for protein-bound NADH with a mean lifetime of $2.0\text{to}2.4\mathrm{ns}$ , and this corresponds to the second component of lifetime observed in our measurements. A recent multiphoton absorption study in NAD(P)H and flavoproteins by Huang ²³ reported that at $750\text{‐}\mathrm{nm}$ excitation, only NADH gets effectively excited, whereas flavoprotein fluorescence contributions become appreciable only for excitation wavelengths $>790\mathrm{nm}$ . In accordance with these findings, we can anticipate that only NAD(P)H gets excited at $750\mathrm{nm}$ , whereas flavoproteins are the main contributors for excitation wavelengths $>790\mathrm{nm}$ . Contributions from both NAD(P)H and flavoproteins are detected for wavelengths $>790\mathrm{nm}$ . At 790- and $830\text{‐}\mathrm{nm}$ excitations, a third lifetime component with a wider distribution and smaller amplitude $(\sim 15\%)$ was observed with ${\tau }_3\sim 3.2\pm 0.5\mathrm{ns}$ in all the cell lines. It may be attributed to oxidized flavoproteins, as observed in some of the earlier spectroscopic studies and reported to have mean lifetime values of $3\text{to}5\mathrm{ns}$ in their free state and $\sim 1\mathrm{ns}$ in the protein-bound state.^{23, 27} Although this third component was qualitatively found in all the cells for excitation wavelengths larger than $790\mathrm{nm}$ , we found that detected photon counts in our measurements were not sufficient to enable quantitative interpretation of this third component reliably. We would therefore limit our further discussion only to a two-component analysis without overstating the importance of the minor third lifetime component in cells∕tissues. As can be seen from both Fig. 5 and Table 2, normal cells have larger contributions from free flavoproteins and protein-bound NADH, whereas the cancer cells have predominantly contributions free NAD(P)H and possibly protein-bound flavoproteins. It has been recently shown that metabolic rate of tumor tissue can be correlated with a higher ratio of reduced pyridine nucleotides to oxidized flavoproteins.²⁸ Unlike these earlier studies, which relied on spectroscopic redox ratio, our results demonstrate a method to obtain a direct estimate of redox ratio spatially resolved in cells based on intrinsic lifetime parameter and with a single excitation wavelength $\left(790\mathrm{nm}\right)$ . We calculated the redox ratio from the amplitudes of two lifetime components given in Table 2 for the cell lines used in this study. Figure 5(c) shows qualitatively the differences in metabolic activity in these cell lines. This possibility is further substantiated in Fig. 6, where mouse tissues with different metabolic activities are distinguished based on their lifetime signatures. It is possible that minor lifetime contributions from collagen and elastin are also detected in this observation window. While this paper was under review, an interesting study by Blinova ²⁹ reported steady state kinetics of mitochondrial matrix NADH and the authors observed three distinct lifetime pools of NADH [0.4 (63%), 1.8 (30%), and $5.7\mathrm{ns}$ (7%)] in isolated mitochondria based on time-resovled single photon, spectroscopic measurements. Although these results are similar to our data, note that flavoprotein contributions cannot be neglected in interpreting cellular autofluorescence, and our analysis is more realistic in that regard.

Fig. 5

Representative intensity and lifetime images obtained in live cells in culture for (a) transmission, multiphoton fluorescence intensity, and fluorescence lifetime images obtained from HEK293T epithelial cell line, HeLa cancer cell line, and primary cultures of mouse hepatocytes excited at $790\mathrm{nm}$ and (b) lifetime histograms obtained from these cells assuming a single-component lifetime analysis and with multiple excitation wavelengths (750, 790, and $830\mathrm{nm}$ ). (c) Redox ratio calculated from the amplitudes of two lifetime components given in Table 2.

Table 1

Common sources of autoflourescence in cells∕tissues.

Table 2

Two-component lifetime analysis of cellular autofluorescence.

3.4.

Photon Detection Efficiency, Photon Statistics, and Multicomponent Lifetime Analysis

Cells in culture and turbid tissues can display complex, wavelength-dependent autofluorescence signatures that can have multiple molecular sources. Some of these signals have significantly overlapping emission spectra, which makes it a challenging task to isolate different autofluorescence signals reliably. Filter-based intensity methods can detect only a convolved, time-averaged signal from the different contributions. As demonstrated from the lifetime imaging histograms [Figs. 5(b) and 6(b)], although the emission spectral window was kept the same in all the measurements, characteristic lifetime differences were observed in different cell lines and in tissues. Any improvisation in the accuracy of the multicomponent lifetime analysis presented in Table 2 depends mainly on detector time resolution and photon statistics.

The photon detectors that are currently being used in lifetime imaging are the fast photomultiplier tubes $(\sim 150\text{to}300\mathrm{ps})$ and microchannel plate $(\sim 50\text{to}100\mathrm{ps})$ . Streak camera detection improves the time resolution to $\sim 2\text{to}15\mathrm{ps}$ . In the multigate detection approach, the lifetime is extracted by measuring the fluorescence signal in at least two different time-gated windows. It is known that the photon utilization and time resolution of the multigate detection approach are still limited by the detector performance.¹¹ One can also not avoid background noise arising from scattered fluorescence emission when an imaging (area) detector is used in multigate detection. Regardless of these limitations, the multigate detection FLIM methodology has been found to be very successful with single-photon excitation in the past. For multiphoton applications, it is imperative to use point-scanning detectors to eliminate background noise. Streak imaging essentially gives 4-D (intensity, time, $x$ and $y$ axes) information with high photon efficiency. In streak imaging, there is no “gating” of fluorescence emission so that the complete fluorescence decay curves can be collected and then used to extract lifetime information at individual pixels. The streak camera has good radiant sensitivity over a wide bandwidth of wavelengths, making it suitable for both single-photon and multiphoton imaging applications without compromising on detection sensitivity.³⁰ A more detailed exposition of this aspect will be published elsewhere.

In this section, we briefly describe two approaches for extracting multicomponent lifetimes in scanning microscopes. Traditionally, multicomponent lifetime analysis is performed by solving a time-dependent intensity function $F(x,y,t)$ at each pixel $(x,y)$ :

Fig. 7

Principle of global analysis for multicomponent lifetime analysis. A pollen grain specimen, stained with two different dyes, was excited at $840\mathrm{nm}$ . Lifetime images were obtained with the StreakFLIM system coupled to an Olympus IX-70 microscope. A single-component lifetime analysis (a) and (b) according to Eq. 1 shows two distinct peaks in the histogram (c). On the other hand, a second-order global analysis according to Eq. 2 shows two lifetime components, 0.3 and $0.8\mathrm{ns}$ , with varying amplitudes (d) and (e) across the field of view. (f) Lifetime histogram showing the relative amplitudes of the observed two components.