2.1.

Imaging System

In order to investigate multispectral FLIM of autofluorescent molecules, a tunable Ti:Saphir laser (Mai Tai AX HPDS, Spectra Physics, Santa Clara, California) was coupled to a laser scanning microscope (LSM710, Carl Zeiss, Jena, Germany). The Mai Tai laser is a mode-locked 80 MHz “high performance” DeepSee laser with a tuning range of 690 to 1040 nm, a maximum optical output power of about 2.8 W at 800 nm and a pulse width below 100 fs. For autofluorescence imaging, cells were excited with two photons at 720 nm. The average power at the input of the microscope was reduced and measured at the output of the objective lens to be 25 mW. A beam dwell time per pixel of 1.27 μs, which corresponds to a scanning time of 782 ms, has been chosen. Typically, the acquisition time was in the range of several seconds. Taking into account the very short dwell time, which could be achieved due to the use of the newest generation of hybrid detectors (see Fig. 2), the overall irradiation was less than or in the same order of magnitude as in comparable investigations. Moreover, mitochondrial staining with rhodamine 123 after scanning proved that our scanning conditions were not toxic to the cells (data not shown). Spectral detection of FLIM was done using narrow band pass filters at 436 nm (BP $436\pm 10\mathrm{nm}$), 470 nm (BP $470\pm 10\mathrm{nm}$), and 490 nm (BP $490\pm 10\mathrm{nm}$) from AHF Analysentechnik (Tübingen, Germany) and a two channel time-correlated single photon counting (TCSPC) system (Becker & Hickl GmbH, Berlin, Germany) consisting of a Simple-Tau-152-DX, SPC-Image, and the hybrid detector HPM-100-40. This system was coupled to the nondescanned detector (NDD) port of the LSM 710 (see Fig. 2).

Fig. 2

Multiphoton and single photon fluorescence lifetime imaging/ spectrally resolved fluorescence lifetime imaging (FLIM/SLIM) at the laser scanning microscope (LSM) 710.

For each photon, the TCSPC module determines the location within the scanning area and the time of the photon within the laser pulse period. These parameters were used to build up a two-dimensional photon distribution over the scan area and the time in the fluorescence decay. For the results presented below, we used a high resolution image size of $256\times 256\text{pixels}$ and 256 time channels. The microscope objective lens was a $20\times$ magnification NA 0.8 lens (“Plan-Apochromat” $20\times /0.8$, Carl Zeiss, Germany). With a zoom factor of 2, an area of $212\times 212\mu {\mathrm{m}}^2$ was imaged.

2.2.

Cell Culture Studies

The TERT-immortalized human oral keratinocyte line OKF6/ TERT-2^29,30 and two different human oral squamous carcinoma cells (SCC-25, ATCC-Nr. CRL-1628 and SCC-4, ATCC-Nr. CRL-1624) were grown in Dulbecco's Modified Eagle's Medium (DMEM) [Invitrogen (Gibco), Karlruhe, Germany] in the case of SCC cells and in keratinocyte serum-free medium (SFM) [Invitrogen (Gibco), Karlruhe, Germany] in the case of OKF6, both supplemented with 10% fetal bovine serum at 37°C and 5% ${\mathrm{CO}}_2$. SCC-25 and SCC-4 differed in their expression of fatty acid synthase (FAS),³¹ which is overexpressed in several human epithelial cancers. FAS expression in SCC-4 is lower than that in SCC-25, and the epidermal growth factor expression is stronger in SCC-25 than in SCC-4. As discussed in Ref. 31, chemotherapy using FAS as potential target could be more effective for SCC-25. OKF6/TERT-2 cells were used as a nonmalignant, however immortalized control cell line. OKF6/TERT-2 cells yet retained normal growth and differentiation and showed normal phenotype.

For microscopy, all cells were seeded on glass bottom microwell dishes with a coverglass of 0.16 to 0.19 mm (MatTek, Ashland, Massachusetts) at a density of $150\text{cells}/{\mathrm{mm}}^2$ and were allowed to grow for 48 h. Microscopic observation was immediately performed after removing the incubation medium and rinsing twice with indicator free and fetal bovine serum (FCS) free medium at 37°C. All three cell types were imaged under identical experimental conditions (same laser power, objective lens, zoom factor, temperature, time schedule, etc.), which were essential for statistical comparison of the recorded data. To confirm cell viability after scanning, the cells were stained with $1\text{-}\mu \mathrm{M}$ rhodamine-123 (Sigma-Aldrich, St. Louis, Missouri). Intact mitochondria could be demonstrated in cells, which were irradiated with the Ti:Saphir laser under the described experimental conditions (data not shown).

2.3.

Data Analysis

The analysis of FLIM data was done with SPCImage software (Becker & Hickl GmbH). We performed a $3\times 3$ spatial pixel binning (binning factor 1) to obtain decay profiles with $\sim 5000$ photons in total. This is assumed to be the approximate minimum number of photons required to perform multiple-component lifetime fits.¹⁸ A two-exponential fitting procedure with $I(t)=a_1{\mathrm{e}}^{-t/{\tau }_1}+a_2{\mathrm{e}}^{-t/{\tau }_2}$, where $I(t)$ is the fluorescence intensity at time $t$, ${\tau }_1$ and ${\tau }_2$ are the lifetimes of the first and second fluorescence component, and $a_1$ and $a_2$ are their relative contributions was used. The goodness of fit, the ${\chi }^2$ value, was $\le 1.2\pm 0.1$ (single-sigma standard deviation). In a first approximation, the first (short) component can be attributed to free NADH, and the second (long) component to protein-bound NADH.^19,20 The widths of the intervals $\mathrm{\Delta }{\tau }_m$, $\mathrm{\Delta }{\tau }_1$, and $\mathrm{\Delta }{\tau }_2$ containing 66% of the corresponding values were also calculated with SPCImage software. The mean lifetime ${\tau }_{\mathrm{mean}}$ was calculated to be $(a_1\times {\tau }_1+a_2\times {\tau }_2)/(a_1+a_2)$. We also calculated the intensity ratio $I({\tau }_1)/I({\tau }_2)$, which is equal to $a_1{\tau }_1/a_2{\tau }_2$. The total fluorescence intensity per pixel was calculated by summing individual photon counts with MATLAB software. The calculations were done with a binning factor of 1 for a region of interest and were divided by the pixel number. Alternatively to the two-exponential fitting procedure, different decaying states were confirmed using global analysis program SimFCS 3.0.³² This approach leads to faster solutions and might be important for various applications.

2.4.

Statistics

For statistical evaluation, four different samples were used for every spectral channel (436, 470, and 490 nm, respectively) for every cell type. In total, 150 single images were measured in the case of OKF6/TERT-2, 144 single images in the case of SCC25 and 163 images of SCC4 cells. For statistical calculations SigmaPlot 11.0 (Systat Software Inc., San Jose, California) was used. Statistical differences between different groups were analyzed with Kruskal-Wallis one-way analysis of variance on ranks followed by a pairwise multiple comparison procedure. The differences in the mean values among the treatment groups were taken as significant when they were greater than would be expected by chance; the difference is statistically significant when $P\le 0.001$. When the normality test (Shapiro-Wilk) showed a normal distribution of the data, the Holm-Sidak method was used as a pairwise multiple comparison procedure. When the normality test or the equal variance test failed, the Dunn’s method was used.

3.1.

FLIM within Spectrally Different Emission Channels

As discussed in Sec. 1, the fluorescence spectra of free and protein bound NADH possess maxima at 470 and 440 nm, respectively, when excited between 340 and 360 nm.

Therefore, FLIM was performed around $436\pm 10\mathrm{nm}$, $470\pm 10\mathrm{nm}$, and $490\pm 10\mathrm{nm}$ using narrow band-pass filters (BP) of the same optical density. The spectral range from 470 to 500 nm is related to the emission maximum of free NADH. The reason for the additional use of $490\pm 10\mathrm{nm}$ was to determine whether or not the separation of free and protein bound NADH is improved when the spectral gap between them is increased. Two-photon excitation was done in all cases at 720 nm. Since the emission spectra of free and bound NADH are highly overlapping, the lifetimes within every channel represent both components with different relative contributions. A two-exponential fitting procedure was used to calculate the fluorescence lifetimes of bound and unbound NADH.

The first (short) fit component was correlated to free NADH, whereas the second (long) component contributed more to protein bound NADH. FLIM was studied in OKF6/TERT-2, SCC-4, and SCC-25 cells. The OKF6/TERT-2 cell served as control (see Sec. 2.2). FLIM of the mean lifetime ${\tau }_{\mathrm{mean}}$ and the distribution histogram for the two spectral regions 436 and 470 is shown in Fig. 3 for all cells investigated. The lifetime is represented in false colors. Unbound NADH should be found in a higher amount around 470 nm. In fact, the mean lifetime in the spectral region of 470 nm shifts to shorter values compared to that of 436 nm for all cells. A higher contribution of the short lifetime component within 470 nm was also confirmed with the phasor approach [see Figs. 4(a) and 4(b) for SCC-4 cells]. A further shortening, although statistically not significant, could be observed for 490 nm (see Fig. 5). This means that the contribution of bound NADH decreases from 470 to 490 nm, which could improve the spectral separation between free and bound NADH. A minor contribution of FAD fluorescence to the spectral region around 490 nm cannot be excluded.

Fig. 3

FLIM and distribution histogram of the mean lifetime of NADH of OKF6/TERT-2, SCC-25, and SCC-4 cells within two emission channels BP 436 ($436\pm 10\mathrm{nm}$) and BP 470 ($470\pm 10\mathrm{nm}$). The histograms show the distribution of the ${\tau }_m$-value within the range 400 to 1100 ps.

Fig. 4

(a) NADH maps of FLIM images by the phasor approach of SCC-4 cells within two emission channels BP 436 and BP 470. (b) Phasor plot shows location of free/bound cellular NADH as well as location of free NADH in solution. The NADH maps (c) and phasor plot (d) of FLIM of OKF6/TERT-2, SCC-25, and SCC-4 cells, measured within emission channel BP 436, demonstrate a decrease of free/bound NADH within the cells from nonmalignant OKF6/TERT-2 to malignant SCC-25 and SCC-4. The color scale from red to white indicates the change of the relative concentration of free and bound NADH, from high (red) to low (white) ratios of free/bound NADH.

Fig. 5

Statistical evaluation (box plot) of the lifetime parameters for OKF6/TERT-2, SCC25, and SCC4 measured for the three emission channels BP 436, BP 470, and BP 490, in detail, the mean lifetime ${\tau }_m$ (a), the first (short) fluorescence compound ${\tau }_1$ (b), the second (long) compound ${\tau }_2$ (c), and $a_2$, which is the relative contribution of the protein bound NADH (d).

As demonstrated in the box plot in Fig. 5, the difference in the mean lifetime proven to be statistically significant in most cases using the Kruskal-Wallis one way analysis of variance on ranks, followed by a pairwise multiple comparison procedure (Holm-Sidak method). Not significant was the difference between 470 and 490 nm, especially for the SCC-4 cells. The shift to shorter lifetimes correlates well with the decrease of the amplitude $a_2$ of the long component. The contribution of bound NADH was, in general, less than that of unbound NADH and decreased by approximately 2% to 2.5% at 470 nm compared with 436 nm then further decreased by 0.5% to 1% at 490 nm due to a decreased influence of bound NADH. Although the fluorescence quantum yield of bound NADH is four times higher than of unbound NADH, the fluorescence intensity was highest for all cell types at 470 nm, as demonstrated in Fig. 6. The differences for the different channels were statistically significant ($P\le 0.001$) with the Kruskal-Wallis one way analysis of variance on the ranks. Since optical density of all three filters was proven to be the same (data not shown), the intensity difference confirms the higher contribution of unbound NADH to the total fluorescence intensity.

Fig. 6

Total fluorescence intensity for OKF6/TERT-2, SCC-25, and SCC-4 at 436, 470, and 490 nm, calculated as total photon counts/pixel with binning 1.

3.2.

FLIM of Different Cell Lines

The fluorescence intensities of OKF6/TERT-2, SCC-25, and SCC-4 cells, which were calculated as total photon counts during imaging within an area of $3\times 3$ pixels (binning 1), are demonstrated in Fig. 6. The intensity was always lower for both SCC cell lines compared with OKF6/TERT-2. The reduced fluorescence intensity could indicate a decrease in the total amount of NADH. This is in contrast to a variety of cancers, including breast cancer,^33,34 whereas it was similar for squamous carcinoma as well as oral and cervical dysplasia.^35–37 Whether this is due to a change in the ratio of free/bound NADH was proved further by inspecting the fluorescence lifetimes. As demonstrated in Fig. 3, the value for ${\tau }_m$ was increased from the nonmalignant OKF6/TERT-2 to malignant SCC-25 and SCC-4 cells, respectively, which was statistically significant for all spectral channels ($P\le 0.001$), as proven by Kruskal-Wallis one way analysis followed by a pairwise multiple comparison procedure (Holm-Sidak method, see Fig. 5). The phasor approach confirmed an increase of longer decaying components for malignant versus nonmalignant cells [see Figs. 4(c) and 4(d)]. An elongation was observed for ${\tau }_1$ which was statistically significant ($P\le 0.001$, Dunn’s method), whereas a slight decrease for ${\tau }_2$ was observed for SCC-4 compared with SCC-25. The decrease, however, was not statistically significant. As shown in Fig. 5, SCC-4 cells exhibited the longest ${\tau }_1$, whereas OKF6/TERT-2 decayed in the shortest way. ${\tau }_1$ is correlated with the lifetime of free NADH. A lifetime of $420\pm 10\mathrm{ps}$ was measured by us for NADH in water (data not shown). The slight difference for the cells could be due to different microenvironments such as pH and viscosity and may reflect changes in dynamic quenching.³⁸ For ${\tau }_2$, the difference was significant only for the pairs OKF6/TERT-2 versus SCC-4 and SCC-25 (Dunn’s method). ${\tau }_2$ is correlated with the lifetime of bound NADH. The longer values for SCC-4 and SCC-25 could be due to a different protein binding compared with OKF6/TERT-2. Whether the slight difference between SCC-4 and SCC-25 is due to different metabolic activities has to be proven further. A different ability to express metabolic oncogenes has been reported.³¹ In addition, SCC-4 showed a strong tendency to form large progressively growing colonies, which could be related to their high tumorigenic behavior.³⁰

Although the lifetime is a critical parameter for the definition of the microenvironment, for characterization of different cell types the relative contribution of free/bound components probably plays a more essential role. A change in the metabolic activity could not only provide a change in the fluorescence lifetime of both free and bound NADH, but also a change in their relative contributions.^20,39 As shown in Fig. 5, SCC-4 cells exhibited the highest contribution of bound NADH. The differences in $a_2$ were proven to be statistically significant ($P\le 0.001$, Dunn’s method for BP 436 and BP 470 and Holm-Sidak method for BP 490) for all cells. The larger amplitude $a_2$ is also responsible for the longer lifetime ${\tau }_m$ in the case of SCC-4 cells. Reduced fluorescence intensity as demonstrated in Fig. 6 for SCC cells is, therefore, not due to reduced bound NADH, but possibly to a reduced amount of total NADH.

Figure 7 represents (in false colors) the relative intensity of the short compound versus the long compound $I({\tau }_1)/I({\tau }_2)$, which is equal to $a_1{\tau }_1/a_2{\tau }_2$. For SCC-4, the long lifetime compound is prominent and the concentration is high in the cytoplasm and seems to be correlated with mitochondria (see orange color at BP 436 in Fig. 7). In the case of nonmalignant OKF6/TERT-2, a short compound which is significantly localized in the cell nucleus, is observed in a high concentration (blue color in Fig. 7 at BP 470). This fact is interesting because free NADH in the cell nucleus is known to play a role in the regulation of gene expression and is enhanced in cells where growth is stimulated.¹⁴ The box plot in Fig. 8 demonstrates that the mean ratio $I_{\text{short}}/I_{\text{long}}$ shows a significant difference ($P\le 0.001$, Dunn’s method) for all cells at 436 and 490 nm. At 470 nm, the difference is significant only for OKF6/TERT-2 versus SCC-25 and SCC-4 cells. The contribution of the different decaying components for the cells investigated was also confirmed with the phasor approach [see Figs. 4(c) and 4(d)].

Fig. 7

Distribution of $a_1{\tau }_1/a_2{\tau }_2$ in false colors for the three cell lines in the two spectral regions BP 436 and 470.

Fig. 8

Box plot of the ratio of the intensity $I_{\text{short}}/I_{\text{long}}=I_1/I_2$.

In Fig. 9, the box plot of the width of the interval including 66% of the lifetime values is demonstrated. $\mathrm{\Delta }{\tau }_m$, $\mathrm{\Delta }{\tau }_1$, and $\mathrm{\Delta }{\tau }_2$ were, in most cases, enhanced for SCC-25 and SCC-4 compared with OKF6/TERT-2 for all spectral channels. The differences for $\mathrm{\Delta }{\tau }_m$ for all three cell lines was statistically proven significant ($P\le 0.001$) with Kruskal-Wallis one way analysis followed by a pairwise multiple comparison procedure (Dunn’s method for BP 436, BP 470, and Holm-Sidak method for BP 490). The smaller intervals in the case of OKF6/TERT-2 could be explained by attributing the better statistics of these cells to higher photon counts (see Fig. 6). Additionally, this also supports the assumption that both cancer cells showed higher diversity and, therefore, various binding states of NADH.

Fig. 9

Box plot of widths of interval including 66% of all measured lifetime values for OKF6/TERT-2, SCC-25, and SCC-4 measured for the three emission channels BP 436, BP 470, and BP 490: $\mathrm{\Delta }{\tau }_m$ (a), $\mathrm{\Delta }{\tau }_1$ (b), $\mathrm{\Delta }{\tau }_2$ (c).