Autofluorescence Lifetime Imaging of cellular metabolism: Sensitivity towards cell density, pH, intracellular and intercellular heterogeneity

Abstract

Autofluorescence imaging (AFI) has greatly accelerated in the last decade, way past its origins in detecting endogenous signals in biological tissues to identify differences between samples. There are many endogenous fluorescence sources of contrast but the most robust and widely utilized have been those associated with metabolism. The intrinsically fluorescent metabolic cofactors Nicotinamide adenine dinucleotide (NAD⁺/NADH) and Flavin adenine dinucleotide (FAD/FADH₂) have been utilized in a number of AFI applications including basic research, clinical and pharmaceutical studies. Fluorescence lifetime imaging microscopy (FLIM) has emerged as one of the more powerful AFI tools for NADH and FAD characterization due to its unique ability to non-invasively detect metabolite bound and free states and quantitate cellular redox ratio. However, despite this widespread biological use, many standardization methods are still needed to extend FLIM based AFI into a fully robust research and clinical diagnostic tools. FLIM is sensitive to a wide range of factors in the fluorophore microenvironment and there a number of analysis variables as well. To this end there has been an emphasis on developing imaging standards and ways to make the image acquisition and analysis more consistent. However biological conditions during FLIM based AFI imaging are rarely considered as key sources of FLIM variability. Here we present several experimental factors with supporting data of the cellular microenvironment such as confluency, pH, inter/intra cellular heterogeneity, and choice of cell line that need to be considered for accurate quantitative FLIM based AFI measurement of cellular metabolism.

1. Introduction

Autofluorescence Imaging (AFI) based diagnostics in the past half a century relied on a few key facts such as 1) oxidized forms of riboflavin, flavin mononucleotide, and flavin dinucleotide emit a strong green fluorescence around 520 nm 2) reduced Nicotinamide adenine dinucleotide (NADH) emission in a cell reflected healthy metabolically active cells. These early ideas evolved into identifying the right fluorescence emitters and realizing that these molecules can possess different states that can be distinguished by their spectrum, fluorescence lifetime and polarization states. One of the first decisive steps in advancing AFI was recognizing that the oxidized nicotinamide adenine dinucleotide (NAD+) is a non-fluorescent species and upon reduction to NADH, the sample will present an absorption peak at 350nm and emission peak at 460nm. This principle is currently used to identify the NAD+/NADH redox ratio in a sample in many biochemistry labs. NAD+ participates in multiple signaling processes, production of second messengers, and these pathways are essential for mitochondrial function and to keep the redox balance or the oxidative stress on the cell. The second important step was key research on Flavin adenine dinucleotide (FAD), which was identified and isolated by its spectral properties with a distinct green emission, compared to the NADH blue emission. However, both Flavin mono nucleotide (FMN) and FAD have absorption peaks at 355nm and 450nm and emission peaks at 530nm and they are rather indistinguishable (1). In a similar fashion, both NADH and Nicotinamide adenine dinucleotide phosphate (NADPH) are quite indistinguishable by their fluorescence properties and their combined emission is denoted by ‘NAD(P)H’. Britton Chance elucidated that the simultaneous observation of NADH and FAD provide presumptive evidence that the signals are from mitochondria rather than cytosol(2) and accredited most of the NADH fluorescence to the mitochondrial bound NADH states(3).

The implications of measuring the NAD+/NADH ratio from a biological sample revealed multiple metabolic pathways through the pioneering works of Otto Warburg(4), who estimated the difference in metabolism of cancer cells, Hans Kreb, who described the Citric acid-Krebs’s Cycle(5), Gregorio Weber who studied NADH optical emission properties upon binding to enzymes(6,7), and Britton Chance who paired NADH-FAD emissions into forming the optical redox ratio(8). This optical readout of metabolism has been followed in recent years by the development of a number of metabolic imaging modalities including those based on ratiometric imaging(9), fluorescence lifetime imaging(FLIM)(10), fluorescence anisotropy(11), optical metabolic indices(12), optical redox ratios(13), and Forster resonance energy transfer(FRET) based metabolic biosensors(14). Initially, NAD(P)H lifetime based metabolic interpretations were very ambiguous because of low time resolutions and other technical difficulties in scanning and imaging optics and photon collection which limited the signal to noise ratio of the system. This made it unable to achieve single molecule sensitivity to validate the NAD(P)H response with spatial resolution. The key breakthrough of NAD(P)H based lifetime imaging(FLIM) was the identification of two distinct fluorescence lifetime for NAD(P)H upon binding to different enzymes and substrates. This is demonstrated in Figure 1 for an enzyme, lactate hydrogenase(LDH) binding to NADH. The LDH catalyzed reaction for lactate production in glycolysis uses NADH as a cofactor (Panel A). This NADH binding can be identified by the changes in the NADH fluorescence lifetime (Panel B). This rise in LDH levels in a solution-based measurement as presented in Sharick et.al., increases the mean measured fluorescence lifetime. Multiple techniques and methods re-established these lifetime based observations, sometimes even reporting equivocal absolute values of bound lifetimes for same enzyme (NADH-LDH as 1.8ns ,2.8, 3.2ns) (15–17). Another prevailing explanation is the possibility of four NADH lifetimes in a cell (0.4 free, 1-conformation of free NADH, 1.7-bound NADH, 3.2-NADH bound to an enzyme in ETC(Electron Transport Chain)(18). The terms “free NADH” is used in this article for describing reduced NAD+ that is unbound to a substrate and “bound NADH” for the ones that are bound to a substrate. However, the different methods currently used to estimate the free NADH: bound NADH ratio based on lifetime are not extremely sensitive to the absolute bound lifetime values. Owing to this reason, most of the lifetime based metabolic assays work to quantify the free NADH present in the system with respect to its bound fraction.

Figure 1). Fluorescence lifetime of NADH.

A) NAD⁺ generation: The pyruvate-lactate cycle catalyzed by LDH. This reaction uses NADH as a cofactor and generates NAD⁺. B) NADH Lifetime curve with addition of LDH: [Figure adapted from Sharick et al. (15)] the IRF (black curve) is convolved with the fluorescence of NADH (purple) and compared as an overlay. LDH was added to obtain 1.4uM, 3.5uM and 7.3uM and respective lifetime curves are plotted as an overlay in colors blue, green and red respectively.

Lifetime based methods have been predominantly developed and applied for cellular imaging after the advent of multiphoton microscopy, which uses a multiphoton laser excitation at a higher wavelength and higher excitation photon density. This imaging modality allowed research groups to pursue bioimaging at larger depths and excite samples in ultraviolet range with lesser phototoxicity (19,20). Instead of the traditional NADH excitation at 340-390nm, multiphoton excitation at 740nm-780nm was employed, which presented no toxicity (unlike the UV radiation) at moderate laser powers(<30mW)(21). Another advantage was that most multiphoton microscopes used a mode-locked laser which had excellent temporal instrument response function that helped FLIM to separate fluorescence signal from excitation laser. FLIM analysis and principles are briefly explained in Methods section.

Among the AFI methods and techniques, FLIM holds a unique position because FLIM quantification is independent of the intensity. FLIM is also unique in its ability to differentiate enzyme bound and free forms of key metabolites such as NADH. Nevertheless, the lack of standardization protocols still poses a great challenge to the development of this field. Researchers need FLIM measurements such as metabolic mapping experiments of cancer (22) to work across FLIM variants(23–25) and model systems and yield similar, comparative results. The concern has been raised about the ability of FLIM to be used for defining new robust quantitative image-based biomarkers for normal and diseased processes. Much of the development work in trying to make FLIM more rigorous has focused on developing good imaging standards (26,27)for the FLIM instrumentation and improved software analysis(28,29). However, despite the great known influences of cell microenvironment fluxes on FLIM, there have been few research studies that have formally examined this beyond being part of the result. The cellular microenvironment can not only be characterized by AFI FLIM methods but it also could affect some of the findings if not well accounted for. For example, metabolic measurement could be strongly affected by cellular density and its implications in the redox state or metabolism. Particularly, cell density and confluency has been shown to have an impact on cholesterol metabolism in cancer cells and recent studies show that NADH lifetime is sensitive to oxidized lipids(30–32). Cell growth is correlated to metabolic markers like glucose uptake and lactate production in epithelial cells(33). Physiological parameters play an important role in AFI FLIM because of the molecular level sensitivity of FLIM. In this research study we analyze and quantitate several of the key environmental influencers of AFI-metabolic FLIM. These include the dependence on the A) cell density and proliferation rate, B) imbalance in the pH, C) intercellular heterogeneity: differences in the same cell type and D) intra-cellular heterogeneity: differences between different cell types and need for careful selection of cell lines in a comparative study. We show how these cell scale variances can affect AFI FLIM measurements for metabolism and give recommendations for accounting for this in study design.

2. Materials & Methods

2.1. Cell culture

The mammary epithelial cell lines MCF-10A (referred herein as 10A and MCF10-CA1d (referred herein as CA1D) were cultured in media made of 1:1 ratio of DMEM/F12. The media was supplemented with 5% Horse Serum (Life Technologies Corporation, Carlsbad, CA), 10ug/mL bovine insulin (Cell applications Inc, San Diego, CA), 500ng/mL hydrocortisone (BioXtra, Sigma Aldrich, St. Louis), 20ng/mL epithelial growth factor (Millipore Sigma, Temecula, CA). The CA1D is a transformed human breast epithelial cell line from pre-malignant tissue (34,35). The pancreatic cell lines HPV16 E6/E7 cells and PANC1 cells were cultured in regular cell media mixture of DMEM (Gibco, Life Technologies, Carlsbad, CA) + 10% FBS (Gibco, Life Technologies, Carlsbad, CA) + 1% penicillin-streptomycin (Gibco, Life Technologies, Carlsbad, CA). HPV16 (HPDE6-E6E7c7 cells) are immortalized human pancreatic duct epithelial cells (36) and PANC1 are epithelioid carcinoma cells derived from human pancreatic duct. Cancerous epithelial metastatic MDA-MB-231 (referred herein as MB231) cells were also cultured in regular media (Low glucose media is optional). The co-culture media for MCF10A and MB231 used the MCF10A media with supplements. Cells were maintained in 55mm dishes in incubator at 37degC. Regular passaging protocol was followed for all cell lines as follows: The dish was washed twice with PBS and incubated for 4 mins in 2.5mL 0.25% trypsin. 10A cells were found to adhere stronger than other cells used. Post incubation, the cells were mixed with media and centrifuged at 3000rpm for 2.5 mins in a 15mL centrifuge tube to separate the trypsin media. The cells were diluted at a desired media volume and plated to imaging dishes or culture dishes. The confluency data presented in this article are all measured in 12 well tissue culture dishes (Corning, NY). The confluency bar graph (shown in Figure 10) categorized into two confluency is derived from the same 12-well plate, with 6 wells seeded at higher confluency and 6 with lower seeding density. The cell density was estimated by calculating the number for cells for 100% confluency and reducing the cell numbers in the following wells as 54%, 29%, 15%, 8%, 4%, 2%, 1%, 0.66%, 0.35%, 0.18%, 0.1%. The seeding densities are in logarithmic fashion. The density below 4% seeded density was plotted as the lower confluency data and the other six wells were plotted as higher confluency for simplicity (This sub-categorization plot into 3×4 wells is shown in Figure 8). The pH -time lapse data was conducted with cells plated on uncoated 35mm x No 1.0 glass bottomed dishes (MatTek, Ashland, MA). The MCF10A, MDA-MB-231 and MCF10A-CA1D cells were provided by Suzanne M. Ponik of the UW-Madison, PANC1 cells were provided by Jillian K Johnson of the UW-Madison and the HPV-16 cells were provided by Joe T. Sharick/Melissa Skala of the Morgridge Institute for Research. The MCF10A cells, MB231 cells and PANC1 cells can be purchased from ATCC, Manassas, VA. HPV16 cells can be purchased from ABM, Richmond, BC, Canada and CA1D cells can be procured from Karmanos Cancer Institute, Detroit, MI, USA(37). All the cells were plated 24-36 hours before the experiment.

Figure 10). The differences in the cell type persists with the confluency.

The mutant cancerous cell line ca1d, proliferates like cancer cell and presents a higher tm than 10a cells. In the tumorigenic cell lines, the PANC1 cell lines are dominantly lower at all confluency conditions. The significance levels in the bar graphs are represented by asterisk (p<0.01). The Students test was performed over ensemble of samples from 3 replicates with 6 wells (<3images per well) with 42< n <54 per bar.

Figure 8). Confluency series for 10A cells.

MCF10A cells are plated in 12 wells in a logarithmically decreasing seeding cell population in the wells labelled A to L (A highest confluency to L lowest confluency). The lifetime distribution of each row of the 12-well plate is plotted on the left using gradient of colors representing the wells (A-blue, L-red) as decreasing order of confluency. The mean lifetime image of all three representative wells (Well D, Well H and Well L) are shown in panels on the right with the corresponding well names. The grayscale corresponds to the intensity of photons represented in the lifetime images. The lower value of the gray scale is not applicable because of the thresholding applied in the image based on amplitude of the intensity decay curve (by the analysis software SPC-Image). The upper value is constant between all image files as 700 photons per 3×3 pixel. The mean lifetime painted over the intensity image uses color scale from 0.70 ns to 2.7 ns. The scale bar for the images shown is 300 microns.

2.2. FLIM

FLIM for this study was done using time resolved photon counting methods and each pixel in a FLIM image contains its temporal intensity decay in the order of picoseconds-nanoseconds. This pixel data is fit into an exponential decay curve and the fitting parameters such as mean lifetime, amplitude coefficients, fractional coefficients are determined. This is illustrated in Figure 2B. The image showing the total number of photons in a pixel is represented in the intensity image. Upon pixel-by-pixel analysis, a lifetime mapped image (panel C) can be obtained by color mapping the pixel to the lifetime values present in the image. The fitting was performed by multiexponential (n=2) fitting and reducing the chi-square value for every curve with determined temporal instrument response function measured daily before the experiment. The software uses an Levenberg-Marquardt based search algorithm to find the minimum of the weighted chi-square(38). As an alternative to fitting, the exponential curve can also be transformed into first order Fourier coefficients and plotted as phasor coordinates (g, s). Phasor plots are obtained by the numerical calculation of the discrete Fourier transform of the time domain lifetime decay curve. The conversion between two different schemes have been described previously(39–41). An alternative graphical plotting scheme can be used from the fitted values to generate these g and s values(42).

Figure 2). Fluorescence lifetime Imaging:

In a FLIM image, every pixel records the intensity decay curve at picosecond resolution. The image mapping the total number of photons from the entire recoded time is called the intensity image (A). Each pixel has an intensity decay curve that can be fit to an exponential function after a deconvolution step with the instrument’s timing response function (B). The fitting result from each pixel, the mean lifetime value can be color mapped into an image and the histogram can be plotted in parallel as shown in panel (C).

The data presented in this article was analyzed using SPCImage (Becker Hickl, Germany), SimFCS (LFD, UC-Irvine, CA) and customized plotting tools developed in Python(43). There are many analysis methods that can obtain a clean distribution of lifetimes from a FLIM measurement. This has been extensively discussed in the multiple FLIM reviews(28,44) and is out of the scope of this work. TCSPC instrumentation dictates the use of intensity weighted lifetime (itm) for separating NADH based fractions(45). However, most of the published results in TCSPC fitting based analysis use mean lifetime (tm). Phasor plots are generated based on the intensity weighted lifetime which appear higher than the mean lifetime value. Mean lifetime is calculated as follows: ${\tau }_{mean}\left(tm\right)=\sum _{i=1}^N{a}_i{\tau }_i/\sum _{i=1}^N{a}_i$, for every component i with an amplitude of ai and lifetime ti. Intensity weighted lifetime or apparent lifetime is calculated by substituting the amplitude with integral contribution of each component i. ${\tau }_{apparent}\left(itm\right)=\sum _{i=1}^N{a}_i{{\tau }_i}^2/\sum _{i=1}^N{a}_i{\tau }_i$. The phasor plots from fitting data can be obtained by placing the lifetime point at the intensity weighted fraction between the intensity weighted lifetimes(42). Mean lifetime (tm) based histograms and tm-images are default outputs for SPCImage analysis and are used in this article. For other means of comparison, we always depend on itm so that the lifetime values have an absolute interpretation rather than for SPCImage analysis users.

2.3. Imaging

Imaging was carried out on a custom-built multiphoton microscope that was previously described (46–48). Refer Figure 3 for the microscope scheme. Autofluorescence lifetime imaging was excited using a mode-locked tunable (690-1040nm) ultrafast laser working at 80MHz (Mai Tai DeepSee, Spectra Physics, Santa Clara, CA). The system is built around a microscope frame (Nikon Eclipse TE300, Nikon) and employs a photon counting GaAsP photo multiplier tube/PMT (H7422-40P, Hamamatsu, Japan) at a constant high photomultiplier gain for single photon efficiency (DCC gain of 85.0). For images that require no timing signal used a lower PMT gain (45.0-70.0). The detector signal and the laser clocking signal was routed to a time correlated single photon counting electronics (SPC-150, Becker and Hickl, Germany) to estimate photon arrival time for every photon. Home-built scanning software (WiscScan, LOCI, UW-Madison, http://loci.wisc.edu/software/wiscscan) was used to scan the laser using the galvanometric mirrors and create the image. WiscScan uses the hardware specific dynamic libraries to create the image from the FIFO data stream and save the timing histogram into a time-resolved FLIM image.

Figure 3). Schematic of the multiphoton microscope used.

The microscope uses an ultrafast laser controlled for its power, beam size and polarization are scanned on the sample through an objective lens. The fluorescence collected from the excitation spot is recorded live at high time resolution by the PMT and the timing electronics. The filter laser combination 740nm-450/70 is specific for NADH imaging. Abbreviations used <PMT>photomultiplier tube, <EOM> Electro-optic-modulator; <TCSPC>time correlated single photon counting.

The microscope is equipped with multiple objective lenses. Four of them were used for the images presented in this article (Nikon 60× Plan Apo water-immersion 1.2NA.; 20× S Fluor VC 0.75NA, 10× P Fluor 0.5NA and 10× S Fluor 0.5NA). The images were collected to obtain a minimum of 1000 photons per 3×3 pixel which was accounted to 120 seconds using the 0.5NA, 90sec using 0.75NA lenses and 60sec using the 1.2NA lens. NADH was excited using 740nm and collected using a 450/70 nm band pass filter. The laser power was maintained below 70mW for the 10× objective and below 22mW for the 20× objective. Imaging for the 12 well dishes was through plastic so higher powers had to be used. The glass bottom dishes were imaged at powers below 10mW using the 20× and 60× objective lenses. The photon counting and lifetime estimation was calibrated with fluorescence from dye solutions [Urea crystals for SHG(0ns), Coumarine6 in ethanol 2.5ns(27), Rhodamine110 in water 4.0ns(25), Rhodamine-B in water 1.7ns(27), (Sigma Aldrich, St Louis)] regularly. The system timing instrument response (IRF) was periodically measured to ensure the working of laser and maximal resolution from detector-timing electronics. The IRF was measured as the time correlated single photon histogram from the second harmonic signal generated of Urea. This was measured to be between 200-400 ps with different time-amplitude conversion gain and offset used.

3. Results

AFI-FLIM is a robust method in measuring the autofluorescence lifetime decay curves from a biological sample at a molecular level. However, this molecular level sensitivity necessitates careful sample environment controls (49). Especially in cell-based metabolic imaging, these differences are very crucial and one need to be careful in selecting the right set of controls and analysis tools for metabolic imaging. The importance to these analysis methods can also play a crucial role in finding the changes using metabolic imaging but is not implored in this study. FLIM sensitivity towards different environmental features are well studied and demonstrated in literature previously (50,51). Many of the new metabolic imaging methods demonstrated in the past decade dwell in a still evolving area of FLIM based contrast rather than absolute definitions and we address few of the main causes of these potential FLIM anomalies seen in AFI metabolic interpretations.

The differences in cancer cell metabolism visualized in FLIM is illustrated in Figure 4. Panel A shows the mean lifetime image of a co-culture of two breast epithelial cell lines (cancerous and non-cancerous), which can be morphologically identified as normal cells forming colonies and cancerous cells remain separated from each other. The two cell lines can be distinguished by their lifetime values (color scheme: green-normal/higher lifetime, red-cancer/lower lifetime) in panel A. The cells shown in this Figure 4 are 1) MCF-10A, a non-tumorigenic epithelial immortalized cell line derived from human mammary glands and 2) MDA-MB231 cells that are epithelial adenocarcinoma cells derived from a metastatic site. The time resolved intensity decay curve of the two cell types are compared in panel B, where the normal cells present a higher lifetime than the cancer cells. Pixels from the respective cell group can be fit to exponential curve and a lifetime distribution can be obtained for both groups as shown in Figure 4C. The alternative to mathematical exponential fitting analysis is achieved by using phasor plots for graphical plotting of the lifetime curves as presented in Figure 4D, and its zoomed-in plot. The parallel plots shown in Figure 4C and Figure 4D to resolve two or more species in FLIM is the quintessential endpoint for most of the FLIM analyses. This data represents general FLIM based measurements shown in literature(10,52–54), when lower lifetime presented in cells are correlated to metabolism relying on glycolysis rather than oxidative phosphorylation. However, this is not justified by the LDH-NADH binding shown in Figure 1, because a higher lactate production associated with increased glycolysis should correlate with higher bound NADH levels. Regardless of this ambiguous NADH-LDH metabolic interpretation used in literature, FLIM is able to correlate cancer cells to a lower NADH lifetime; in multiple cases of cancer, stem cell differentiation and so on. Some recent work correlate this bound fraction to NADH-dehydrogenase (Complex I) binding because of abundance of bound lifetimes correlating to the mitochondrial structure(55). This suggests a rather complicated bound-NADH scenario that correlates to oxidative phosphorylation in the cells than just NADH-LDH binding step in anaerobic respiration. This could be further studied by following the NAD+ production pathways and distinguishing other substrates of NADH that affects bound lifetimes, which could be as well be a function of cell type and its genealogy. An increase in free NADH fraction because of a higher lactate production can be explained as either by increased NADH production associated to balance the NADH oxidation or the lack of spatial observational discrimination of cytoplasmic ratios from mitochondrial fraction in a cell. This increasing level of complexity kept FLIM based NADH studies at bay as a tool for distinction and not interpretation. However, the presence of increased free NADH ratio correlates to increased glycolytic response and is used AFI-FLIM for identifying the fraction of free NADH available.

Figure 4). FLIM in cancer:

MCF10A cells are compared against MD-MB231 cells in a co-culture. The cancerous MB231 cells present freer NADH ratio or anaerobic respiration which is viewed in color coded image panel A, in the Intensity decay shown in panel B, the lifetime distribution after fitting the decay curves shown in panel C or using graphical means as shown in phasor plot in panel D.

3.1. Time Course to study pH imbalance

The simplest way to verify the redox ratio of any system is by changing the pH of the system. Redox ratio [NADH]↔[NAD+] in the cell is coupled to the pH and is often reflected on the respiration [pyruvate]↔ [lactate] reaction. Higher levels of oxidation of lactate to pyruvate happens at higher pH(56) and this would give higher quantity of free NADH. Cell culture media is often supplemented with pH indicator such as phenol red and a pH balancing agent such as CO₂ is regularly used in cell culture to maintain the pH levels in live-cell experiments. CO₂ based bicarbonate buffering system works by readily dissolving CO₂ that gets ionized to H⁺ and HCO₃^- in water. Atmospheric CO₂<0.5% and maintaining CO₂ levels at 5% will regulate the pH of the media and upon imbalance, the pH will turn alkaline. The cells were plated on a glass bottom dish at >50% confluency and incubated for 24 hours before imaging. The imaging was carried out at a laser power measured to be 10mW at the back aperture of the objective lens. Both cells were plated at same seeding number of cells (~300,000 cells) and imaged at identical conditions.

In order to study the pH effect without addition of a secondary pH conditioner, we removed the pH regulator (CO₂) from the system. Upon removal of CO₂ balance from 5%, the pH of the media will rise above pH of 8 in a short duration and this can be seen in the color of the pH indicator. In order to study the effects of pH balance, we employed this strategy: the CO₂ incubation was removed and cells were maintained at 37degC and we recorded FLIM for ~7 hours. The intermittent imaging of the same cells within the same field of view did not kill these cells (cells recovered after imaging), but the effect on mean lifetime and intensity is very drastic in less than 60 mins. The cells were placed back into 5% CO₂ levels after the 7 hours imaging after brief washing and media replacement.

The same cells in the field of view were imaged for ~7 hours with intermittent intervals of 90 secs and presented in Figure 5. Panel A shows the lifetime change over time and the panel B shows the intensity change over time. The drop of lifetime in <60 mins of pH imbalance is illustrated and this lifetime stabilizes during the experiment. The effect of pH on NADH based FLIM has been performed before to understand the stress response and involvement of NADPH production in related to the ROS(reactive oxygen species) production(57,58). Although there has been report of FLIM-NADPH response with ROS(59), currently there are no accurate ways to robustly characterize this response. The intensity drop between time 2.4 hours and 6.6 hours shows possible photobleaching over time or couple of apoptotic cells which reduced the number of cells in the image over time. The pH stress could put cells into senescent states or apoptosis induction which are not studied further in this article but presents a promising way to probe oxidative stress and cell cycle arrests. The importance of bicarbonate buffer system or using an alternative HEPES buffer (10mM-25mM) used to maintain the pH at 7.2-7.6 pH is relevant for AFI-FLIM based metabolic imaging. We also performed controls with dye solutions (Rhodamine 6G in water), to estimate the effects of possible bleaching and standardize the cell analysis. The dye solutions made in water were imaged over 20 hours to find the variation in the lifetime to be below 0.1ns. The potential cellular apoptosis or quiescence role in these measurements cannot be ruled out. However, this variability can be pointed as another reason to maintain pH levels for AFI FLIM. In order to maintain the total number of cells by senescence rates, the cells were plated at 60% confluency. The ETC (electron transport chain) uncoupler reactions like addition of CCCP(Carbonyl cyanide m-chlorophenyl hydrazone) or FCCP (Trifluoromethoxy carbonyl cyanide phenyl hydrazone) to cells can be related to this pH sensitivity because mitochondria stay at a higher pH for ETC and will present a drop in the mean fluorescence lifetime. The photo-toxicity or excitation induced stress on the cells and the consequent NADPH response is outside the scope of this study and has been previously discussed in several studies such as mitoflash(60) and irradiation damage(61) studies. However we did not see a strong correlation of the damage using intermittent excitation scheme over time. The representative FLIM images at 0 and 4 hours for 10A cells are shown in the inset of panel A.

Figure 5). Time course of pH imbalance.

The cells were imaged for 6+ hours intermittently. Panel A shows the lifetime change over time and the panel B shows the intensity change over time. The sudden drop of lifetime in <60 mins of pH imbalance is illustrated and this value stabilize over 7 hours. Two representative lifetime images from 10A cells are shown 4 hours apart to demonstrate no significant apoptosis to the adherent cell population. The intensity shows potential photobleaching over time or apoptosis in the cells and hence decreased NADH signal. The connection differences at 5.3hour time point is an instrumentation limitation in measuring numerous files in sequence.

3.2. Intercellular heterogeneity

Cells grown in the culture dish over time show proliferation and differences in NADH lifetime images. The lifetime distribution of a proliferating cell colony will present two or more different lifetimes. The distinct higher lifetime shows increased bound NADH levels present in the proliferating cells. This variability presents a need for careful controls for reports suggesting differentiation in cells based on AFI-FLIM. The small variability in the lifetime in these cells growing from a colony can be identified using both fitting analysis and phasor approach. For reversible mapping to the lifetime species observed, we suggest use of phasor plots as shown in Figure 6. The mean lifetime image from MCF-10A cells (panel A) is plotted parallel to the phasor representation (panel B) of the results. Notice that these phasor plots are just graphical representation of the lifetime fitting and not frequency domain transformation of the individual exponential curves. The phasor plot can be used to make phasor cursor-based masks of the image to identify different lifetime distributions. We can identify the sub species in both phasor plot and the mean lifetime image. The phasor plots use a universal circle (semi-circle) which shows pure single exponential species and the line connecting 0.4ns and 3.2 ns represents the average free to bound NADH trajectory reported in literature. The cell colony and the differentiated cells fall on this trajectory and we assume these smaller population is the dividing cells in the colony, with the senescent glycolytic cells at the bottom layer. In Figure 7, three different cell shapes were chosen by morphology and compared against the full image to find the differences in lifetime. The four different images of cells were analyzed :1) long cells shown in blue colored circle, 2) regular cells in red colored circle and 3) spread larger cells in green colored circle and 4) an ensemble of all cells in gray colored square. The changes between the population were not significant, but the rod shaped MB231 showed lower bound NADH levels (mean lifetime) than the other two populations (panel E). These cells are also presented in the phasor plot. The distinction seen in both analysis is different because phasor plots are based on intensity weighted mean lifetime(itm) rather mean lifetime(tm, which is widely used for TCSPC fitting plots)(62). The color codes red-green-blue-gray for the images are followed through both mean lifetime distribution plots, median lifetime values and the phasor plot presented in the panel. The mean lifetime images for a series of cell densities can be seen in Figure 8 and the differentiated lifetimes in each cell colony can be observed. Another key intercellular difference is the different morphology of cells from the same cell culture. This was best presented in the MB231 cells. The intensity levels in the images in Figure 8 for different cell density are comparable but shows distinctive changes in the lifetime (the look up table for intensity and lifetime used are the same for all the FLIM images).

Figure 6). Intercellular heterogeneity – differences in MCF10A colony of cells:

The cells in the colony shows differences in the mean lifetimes. A) Mean lifetime Image. B) Phasor plot for the image with three cursors selected red, green blue C) The image masked by the position of three cursors-red, green and blue.

Figure 7). Intercellular heterogeneity – comparison of MB231 cells by morphology.

A) The image of different cell morphology in a single image. Three insets are zoomed in as B, C and D panels. The cells are visibly different. The colors are used to denote the differences between the three morphologies. The cells are masked for distinct type. E) The color-coded panels (A-gray, B-blue, C-red, D-green) are plotted by them mean lifetime distribution and phasor plot (panel F).

3.3. Cell Density-Confluency

Confluency is the frequently estimated as the number of adherent cells on a 2-dimensional cell culture plate. This is generally measured by the area or the surface of the cell culture dish which is covered by cells. Cell density is the number of cells per area. Both of these parameters have similar implication in metabolic imaging and are controlled by seeding cell population. The number of cells in a culture dish can control the migration (63) and proliferation(64). Proliferative capacity of any cell is a function of the parent tissue from where the cell was derived and all the cells compared here are continuously dividing epithelial cells which would get regenerated by functional cells upon apoptosis or necrosis, if inside a tissue. We studied the cell density-based effects on AFI FLIM using different seed population.

In order to study the effect of cell density, cells were plated in 12 wells in a logarithmically ascending seeding cell number. This was done by calculating the number for cells for 100% confluency and the following wells as 54%, 29%, 15%, 8%, 4%, 2%, 1%, 0.66%, 0.35%, 0.18%, 0.1%. The cells were counted using automated cell counting methods from transmission images at a low magnification(65). These dishes were progressively imaged using FLIM with an interval of 3 days to establish a difference in their lifetime signature with both media depletion and confluency. The 12 wells are named by letters: A to L and an illustration of the wells for the MCF10A cells are shown in Figure 8. The images (A-L) are representative of the mean lifetime presented in the well and not the total number of cells in that well. The 12 wells are separated into 3 groups for visualization purposes and their respective lifetimes are plotted on the same row on the left. An optimal representation of this sample is a mean lifetime distribution plot of all 12 wells together (This scheme of plotting is used later in the article: Figure 11). Note that Figure 8 is a representative image of the dataset and Figure 11 shows the full data set (from 3 trials). The lifetime image color scheme represents red colors for lower mean lifetime values and blue for higher mean lifetime values. The lifetime-based color scheme of the images show decrease in the mean lifetime with an increase in cell density. However, if the lifetime distribution is to be examined, one would notice the bimodal behavior in certain wells. These variations in different colonies is further examined later in the intracellular heterogeneity. The wells E-H shows many proliferating cells that stay distinct from its surrounding lower lifetime cells. Notice that all the distributions are normalized and is a representative of the fraction of population.

Figure 11). Cell Lines in Distinction:

The three cell lines are compared against each other at increasing order of 12 cell density levels represented by the colors: Red shows lower seeding cell density and Blue shows higher cell density. MB231 cells vary in a smaller range and present higher free NADH ratio than others at all 12 density levels. The CA1D cells present two different modalities and 10A cells present a gradual change from bound to free NADH levels with confluency

In order to study the change in metabolism, we plotted 10A cells and MB231 cells for three different days as shown in Figure 9. The graph is plotted between the fraction of lower lifetime species to the total NADH (which is analogous to the free NADH levels) on the x axis and the mean lifetime (without intensity weighting) on the y axis. Wells (represented by green dots) in the right bottom of the graph would have lower bound NADH while the cells on top left part of the graph would show a higher bound NADH levels. Analyzing Figure 9, the cancerous MB231 cells always presented a lower bound NADH levels at all the time points. The cells presented a higher fraction of free NADH at higher cell densities (darker color represents higher cell seeding density). The lower confluent cells always presented a higher bound NADH level of that compared to higher confluent wells.

Figure 9). Change in lifetime with confluency and time:

The changes in mean lifetime and the percent of short lifetime species (free NADH) with increasing days and growth. The top panels show MCF10A cells and bottom panels show MD-MB-231 cells. The cancerous MB231 cells dominantly present free NADH ratio or anaerobic respiration. Each point corresponds to a different well and the darker color of the well shows larger confluency (in a logarithmic scheme of increasing cell population).

This data demonstrates that confluency is a key factor when comparing samples using lifetime-based metabolism. The cell population density and proliferation rate have a strong impact on the final metabolism of the cell rather than the total NADH available. The intensity levels in the images shown earlier in Figure 8 shows the total NADH levels (free + bound) and the intensity remains comparable for all density levels. The increased presence of quiescent cell population at higher confluency could also play important role in determining the average mean lifetime observed in a sample. A complete study on this account comparing cell cycle cyclin protein levels and growth factor levels against tau mean needs to be investigated for understanding these effects. This study falls out of the scope of lifetime imaging and is not pursued in this article.

3.4. Intracellular heterogeneity among cell types and selection of cancer cell line

The role of cell density was evaluated for two different conditions.

1. Between two well characterized epithelial breast cancer cell lines [10A, CA1D] which are isogenic in their genetic profile and proliferate similarly. However, CA1D leads to cancer and presents cancerous morphology and hallmarks. These cell lines are compared against MB231 cells, a triple negative (estrogen receptor (ER) negative, progesterone receptor (PR) negative and human epidermal growth factor receptor 2 (HER2) negative) metastatic epithelial cancer cell line (also possess mutant p53).

2. Between two pancreatic cell lines: PANC1, derived from pancreatic tumor epithelial cells and HPV16 from normal pancreatic duct epithelial cells immortalized using a HPV16-E6E7 gene transduction protocol.

The difference between these two sets is the organ or tissue from which the cells were derived and level of similarity between the cells. For simplicity of the analysis we are not portraying the differences between the breast cells and pancreatic cells, but the respective differences within these groups. For understanding the differences between the two cell types, comparison between HPV16 cells (HPDE cells) and MCF10A cells under metabolic reagents can be found here:(15).

MCF 10 cell line series constitute of different mutant cell lines which have multiple representative cell lines for breast cancer progression: benign hyperplasia-atypical hyperplasia-carcinoma-malignant tumor. CA1D cells are derives from xenografts after trocar transplantation of the tissue(34). CA1D was shown to form tumor 100% of the time, but showed heterogeneity (variability in the ratio of differentiated-undifferentiated-squamous elements(34)). MCF-10A cells are non-malignant cells and MCF10A-CA1d are transformed malignant cells with isogenic properties. FLIM based mitochondrial activity in these cells was characterized using FAD fluorescence(66). CA1D cells are currently used as a good tumor metabolism mimicking cells in-vitro because of its slower differentiation process, when compared against MCF10A-CA1h or CA1a cells which are invasive and faster growing cells(67).

Figure 10 shows the differences between the cell lines at the two cell density levels. Notice that this chart is tabulated using intensity weighted mean lifetimes (which is a common denomination between phasor analysis and TCSPC fitting analysis, see Analysis). All cell types at higher confluency present lower mean lifetime. This confluency dependent lower bound NADH levels could possibly result from their lesser proliferation rate with increasing cell density. The decreased proliferation rate decreases the chance for NADH binding which controls the lactate production. The CA1D cells exhibit higher lifetimes when compared against the 10A cells with same genetic background, while the MB231 cells show a lower lifetime corresponding to decreased NADH binding levels. This critical difference in the different cell types are analyzed in Figure 10. This higher bound NADH levels in CA1D cells is presented because the CA1D cells may not have differentiated and the tumor differentiation of this cell lines may require xenografting them. Regardless, if one was to compare these two cell lines for AFI-FLIM methods, the results would be baffling. With the exception of the isogenic cell line group, the sets of cell lines (10A-231, and HPV16-PANC) behaved as expected presenting a lower mean lifetime for cancerous group. The difference in proliferation regime between the three breast cancer cell lines is further illustrated in Figure 11.

The three cell lines are compared against each other in increasing order of cell density (color scale). All the cells present difference with cell density. CA1D cells shows bimodal distribution with far centers. 10A cells shows a wider distribution and MB231 cells shows bimodal distribution with closer centers. Associating metabolism directly to the mean lifetime, the MB231 cells vary in a smaller range and present higher free NADH ratio than others at all confluency. The CA1D cells present two different subspecies and changes drastically to high free NADH levels at higher confluency. 10A cells present a gradual change with confluency and does not present a bimodal distribution

4. Discussion

The current landscape for AF-metabolic imaging is a cumulative product of last 60 years of research on different pyridine nucleotide and flavin nucleotide interactions in metabolism and this field has been evolving rapidly in the last decade. NAD⁺/NADH is a good redox level indicator for cells because catabolic pathways use NAD as an electron acceptor and this redox potential balances with other cellular redox potentials. The direct relation of bound NADH levels present in the cells to the redox ratio is still abstruse because NAD+ binds to numerous substrates in different pathways, however a consensus in the lifetime measurements have been established to validate lower lifetimes presents anaerobic respiration or hypoxia. Multiple research groups are working to shed light in AFI FLIM by the targeting the metabolic pathways associated with NADH in the last decade and a union of the results is starting to be at the horizon. We feel that a key piece towards further advancing AFI metabolic imaging is better understanding the role of cell-based changes in measurements both for accuracy but also for improving the mechanistic basis for this technique.

In this study, we presented that an isogenic cancer cell line that presents higher bound lifetime than the non-cancerous partner. The higher NADH levels may correspond to its true metabolic response because the cell lines are genetically same which correlates directly to the excepted anaerobic respiration discussed before. This can be answered by comparing the other CA1a and CA1h alongside, however it was not followed in this study. The cell line of choice of study to evaluate any change in metabolism between cell groups is a very important parameter. Isogenic cell lines like CA1d may not represent the contrast in metabolism one would expect with highly cancerous cell like MB231. Careful evaluation of confluency vs metabolism is essential to realize the proliferative scheme of any cell and this is best illustrated here using lifetime imaging (Figure 11).

Cell density is a major parameter when defining any in-vitro experiment, and specifically for metabolism studies at single cell level. Multiple studies have proven that the redox ratio falls with confluency. A study on 3T3 fibroblast cell line have shown that the NADH/total NAD ratio is up to 90% in single cells and drops to 50% at higher confluency(68). Similarly in rat intestinal IEC-6 cells, confluency brought about an increase in glutathione concentrations and a significant reduction in the redox potential (69). The idea of population density vs concentration of NADH is analogous to the notion of cellular metabolic currency or the inflation of currency vs the growing population. The increasing population can either help in increasing the value of currency (increase NADH levels) by producing and supplying materials or can decrease the currency simply because materials get consumed. A Nuclear Magnetic Resonance Imaging (NMR) based study on cell growth vs metabolism was profiled in near-confluent cultures: the glucose utilization correlated with alanine production and did not correlate with lactate (supports anaerobic conditions at higher confluency). This study arbitrates the variations with pools of pyruvate from mitochondria or cytosol and growth conditions(70). One needs to keep in mind the close association of pH and redox ratio in the cell. The higher confluency and lower bound NADH levels are all indicators of possibly same cellular state. Under this assumption the two main factors that change cellular metabolic index using lifetime for cell colonies are either 1) fraction of proliferating cells with higher lifetime in the field of view of the image 2) the overall redox state of the cell that reduces the lifetime.

Regardless of the confluency and cell line choice, cancerous cell lines do present a lower lifetime than the respective non-cancerous cell line (seen for both breast and pancreatic cells). Mean oxygen consumption and lactate production are much lower in cellular cultures that is reaching confluency(68). Seahorse Assay based comparison of MB231 and MCF10A cells with oxygen tension had shown that the MB231 cells are sensitive to oxygen tension and present increased LDH levels with oxygen tension(71). An attempt to study the confluency and AFI was made previously(10,72), which attributed the depletion of metabolite model, lowering respiration and TCA cycle as a cause for lower bound NADH levels. NADH concentrations varied between logarithmic and stationary phase and the free NAD⁺:NADH levels reach a plateau at higher cell density(73). One main argument listed in these studies is that cell contact is the key factor that reduce the redox potential of the culture, rather than cell proliferation, which is an environment driven pathway. Increasing evidence of pH and hormonal regulation on proliferation rate brings this conclusion into question. This can be only investigated by coupling cell cycle arresting experiments or coupling cell cycle states into NADH lifetime imaging. Intracellular and mitochondrial matrix pH and NAD(P)H response has been studied before using mitochondrial flux inhibitors and uncouplers in range 6.5-8.0 and the results match with the results presented here(58).

pH levels are linked to the cell density as a determinantal fraction (74). The optimum growth rate is set to 6.9 to 7.4 pH for most mammalian cells and the confluent dishes will present a higher pH and the cellular multiplication halts the rate. A faster metabolism may lower the pH if using phenol red DMEM aided with Ham’s F12. There are many FLIM based pH probes to measure pH and calcium levels which can be used to calibrate the system pH before establishing metabolic control(75).

Another important aspect in cancer cell metabolic interpretation is the effect of cellular migration and microenvironment modulations associated with it. We attempted to understand if cellular differentiation from the same colony could have an associated metabolic signature. Using simple morphological differentiation seen in a colony of cells, we could identify a lower bound NADH levels in metastatic cells. The MB231 cells are metastatic cells and this migratory behavior may have a role in the lower bound NADH levels. Another study that investigated migration in a narrower micro domain showed that migratory HeLa cells present stress and exhibited lower bound NADH levels (76).

5. Conclusion

Cell scale metabolic measurements are important for a range of development and disease studies. AFI FLIM has already been shown to be a powerful tool for these measurements. This work has not only advanced rapidly in the research arena, there are now emerging biomedical and clinical applications. For example, there are tumor growth inhibitor research studies based on oncological drugs used in clinics carried out with the help of AFI FLIM. One of the major challenges for AF-FLIM moving forward is the lack of general standards and techniques to achieve metabolic imaging interpretations without mathematical modelling. With addition of newer FLIM imaging and analysis technologies, we have introduced new levels of power and complexity to the field which needs careful examination and validation of biological experiments. There is excellent research being conducted to advance the FLIM hardware and software. We submit that additional work needs to be conducted on the influences both positive and negative that cell-based fluxes and variations can have for AFI FLIM based metabolic measurements. In addition, the environmental factors like laser irradiation damage and response should be explored. Cell influences are a major factor driving the power of FLIM but can also impact its accuracy and complicate its interpretation. A further understanding of these changes could lead not only to improved robustness but potentially new AFI-FLIM based biomarkers.