An automated multiwell plate reading ° im microscope for live cell auto ° uorescence lifetime assays

Douglas J. Kelly*†||, Sean C. Warren*†, Sunil Kumar*, João L. Lagarto*, Benjamin T. Dyer*‡, Anca Margineanu*, Eric W.-F. Lam, Chris Dunsby*¶ and Paul M. W. French* *Photonics Group, Department of Physics, Imperial College London South Kensington Campus, London, SW7 2AZ, UK Institute of Chemical Biology, Department of Chemistry Imperial College London, South Kensington Campus London, SW7 2AZ, UK National Heart & Lung Institute Imperial Centre for Experimental & Translational Medicine Du Cane Road, London, W12 0NN, UK Department of Surgery and Cancer, Imperial College London Du Cane Road, London, W12 0NN, UK ¶Centre for Histopathology, Imperial College London Du Cane Road, London, W12 0NN, UK ||douglas.kelly09@imperial.ac.uk


Introduction
The bene¯ts of automated high content assays are well established for drug discovery and are increasingly recognized for systems biology and basic research in the life sciences.The ability to acquire and analyze ¯xed point and time-lapse image data across cell populations with minimal supervision and medium to high throughput creates opportunities to screen complex biological phenotypes against compound and gene libraries, to test drug candidates and to explore disease mechanisms.In particular, the capacity to increase the number of cells that can practically be imaged and the number of \repeats" per condition in a single \experiment" mitigates for biological noise, with the ability to average readouts over large datasets enabling quantitative measurements of challenging assays such as protein-protein interactions to become statistically robust.To this end, high content analysis (HCA) has evolved from simple automated intensity-based imaging to utilize spectroscopic image-based readouts of sample arrays, particularly harnessing the molecular speci¯city of °uorescence labeling.2][3] FLIM can also be used to gain insight into cellular processes via the auto°uorescence lifetimes of endogenous cellular °uorophores and we report here the translation of such readouts to an automated FLIM plate reader.
One key endogenous cellular °uorophore is reduced nicotinamide adenine dinucleotide (NADH).NADH is an important coenzyme in cellular metabolism, being a product of glycolysis and the citric acid cycle and a substrate for the generation of ATP by oxidative phosphorylation.NADH is °uorescent under UV excitation and this property has been studied for more than 50 years with the goal of linking changes in metabolic state to concomitant changes in °uorescence parameters measured in vivo. 4In addition, in vitro studies of the °uorescence enhancement of NADH upon interaction with binding partners have been used to quantify the activity of dehydrogenase enzymes. 5This readout results from the binding of NADH in an elongated conformation that reduces quenching of the reduced nicotinamide group by the adenine moiety. 6In 1992, Lakowicz et al. imaged the °uorescence lifetime of NADH upon protein binding, 7 indicating the potential for FLIM of NADH to be used in biological contexts.Crucially, measurements of excited-state lifetimes do not depend on total °uorophore concentration, detection e±ciency or sample absorption; therefore time-resolved °uorescence data from di®erent instruments can be directly compared and may be applied to live cell imaging experiments and translated to both in vivo pre-clinical and clinical measurements.][10][11] Since aberrant metabolism is a hallmark of cancer, there has been considerable interest in exploiting measurements of cellular auto°uorescence lifetime to read out changes in metabolic state for the study, diagnosis and monitoring of cancer; such changes have been associated with a shift from oxidative phosphorylation to glycolysis. 12][15][16][17] FLIM is most commonly implemented in laser scanning confocal or multiphoton microscopes using time-correlated single photon counting (TCSPC).This can be directly applied to HCA. 2 but the need for FLIM to detect signi¯cantly more photons/pixel than are required for intensity-based imaging makes this approach, where image data are acquired sequentially (with the maximum imaging rate limited by considerations of photobleaching, phototoxicity and sometimes by °uorescence protein expression level), too slow for many HCA applications.FLIMbased FRET readouts have been utilized in HCA using wide-¯eld frequency modulated imaging 1,18 or time-gated detection with Nipkow disk-based optical sectioning. 3,19to realize automated acquisition of °uorescence lifetime images across an array (multiwell plate) requiring only $ 10 s/¯eld of view.Automated FLIM assays have included interrogation of post-translational modi¯cations downstream of epidermal growth factor receptor (EGFR), 1 a small-scale screen of inhibitors for internalization of the transmembrane receptor CXCR4 2 and an investigation of the e®ects of N-myristoyltransferase inhibitors on Gag protein aggregation in HIV models. 3To date, however, such rapid automated FLIM technology has not been applied to assays utilizing endogenous °uorophores.To the best of our knowledge, we present here the ¯rst demonstration of automated wide-¯eld time-gated FLIM being applied to readouts of changes in cellular metabolism as reported by the °uorescence of reduced NADH.We outline practical challenges associated with UV-excited FLIM, including background °uorescence and cell viability, which were addressed in the design of this instrument.We then present its application to quantitative measurements of NADH lifetime parameters to read out protein binding and changes in cellular metabolism induced by a metabolic inhibitor (rotenone) and an anti-tumor drug (cisplatin).

Imaging instrumentation and data acquisition software
Pulsed UV excitation light at 370 nm was obtained from a frequency doubled femtosecond mode-locked Ti:Sapphire laser (MaiTai HP with #3980 Frequency Doubler/Pulse Selection Unit, Spectra-Physics).Excitation intensity was controlled by means of a motorized ¯lter wheel (FW102B, Thorlabs) ¯tted with neutral density ¯lters.Excitation radiation was coupled to the back port of an inverted epi°uorescence microscope frame (IX-81, Olympus) via a UV-transmitting multimode ¯ber (M67L02, Thorlabs) followed by a lens system and a spinning disk di®user of UVT acrylic (Luminit) to provide spatially incoherent wide-¯eld illumination, as shown in Fig. 1.This con¯guration ensures maximum °exibility for coupling di®erent laser sources into the plate reading microscope.The microscope frame was equipped as described previously 3 with optical autofocus, motorized stage, automated objective nosepiece and camera port selection.A 40Â, 0.6 NA objective lens (LUCPLFLN 40Â, Olympus) was used that was speci¯ed for UV excitation and o®ered a long working distance that was compatible with a range of di®erent multiwell plates.On the left hand port, a gated optical intensi¯er (GOI) (HRI, Kentech, with an S20 photocathode for high sensitivity in the blue spectral region) provided time-resolved imaging capabilities, with the phosphor output being read out using a cooled CCD camera (ORCA ER-II, Hamamatsu).The spatial resolution of this FLIM set-up was limited by the GOI rather than the optical components or camera pixel size.The GOI can resolve 18 line pairs/mm and so features as small as 1.4 m may be resolved in the sample plane.A non-FLIM imaging capability was implemented by attaching a second CCD camera (Flea2 08S2M, Point Grey Research) to the microscope trinocular with a lens relay designed to compensate for the di®erence in chip size and so match FLIM and non-FLIM ¯elds of view.A white light LED provided illumination for bright ¯eld or phase contrast imaging and was controlled using a digital acquisition card (USB6008, National Instruments) and an LED driver (LEDD1B, Thorlabs).A syringe pump (KD Scienti¯c) could be attached to the system for automated stimulation of samples during time course experiments.The microscope is ¯tted with an incubator (Solent Scienti¯c) to facilitate live cell imaging at 37 C.The instrument's acquisition software was written predominantly in LabVIEW (National Instruments) and was designed to allow the user to control all aspects of FLIM data acquisition on the plate reader including camera parameters, gate delays and ¯lter sets.The software allows the user to design automated experiments including movement to di®erent ¯elds of view, time-lapse imaging, imaging in multiple z-planes and phase contrast imaging, all of which can be combined in arbitrary sequences.To minimize the light dose, exposure to the excitation laser was controlled by automatically shuttering the laser during the autofocus routine, during stage translation and when the acquisition is paused between time-lapse images.The acquisition software permits automatic upload to OMERO 20 servers for image data storage, sharing and subsequent analysis.
To minimize the overall data acquisition time and light exposure, a cell \pre-¯nd" function was implemented in the experimental protocol to automatically identify the most suitable ¯elds of view for subsequent FLIM, as has been described previously for our FLIM plate reader. 19For this work, the software was modi¯ed to enable cell searching to be performed at lower (e.g., 10Â) magni¯cation than the FLIM readouts, thereby increasing the speed of identi¯cation of suitable ¯elds of view.For auto-°uorescence-based studies, the pre-¯nd function has been adapted to work with phase contrast images, in which regions of cells are identi¯ed using a variance thresholding technique 21 implemented in LabVIEW.Since this method fails most often when searching near the edge of wells, where the vertical sides of the wells and the meniscus of the imaging media degrade phase contrast image quality, the search routine is typically started from the center of each well in turn.This is successful for all but the sparsest samples.If necessary, more complex cell identi¯cation functionality could be employed using CellPro¯ler, 22 an open source cell image analysis software package, which can be called from the pre-¯nd tool, although this capability was not used for the work presented here.

Imaging parameters
Data presented here were collected under excitation at 370 nm, with excitation power after the objective measured to be in the range 95-115 W. Excitation light was directed to the sample via a ¯lter cube (Semrock 377/50 excitation, 409 dichroic, 409LP emission) with an additional barrier ¯lter (Semrock, 440/60) being used to select for NADH °uorescence over background.The total acquisition per ¯eld of view is typically < 25 s, depending on the acquisition parameters: hardware autofocusing generally requires $ 3 s while 2-3 s are required to switch between FLIM and phase contrast imaging when performed as part of the automated acquisition.Fluorescence decays are sampled at eight time delays after pulsed excitation and the CCD integration times and frame accumulations are optimized for each speci¯c sample to maximize dynamic range and avoid saturation.For data presented here, the typical total exposure per ¯eld of view to excitation laser light is 10-20 s.
For preliminary experiments to investigate cell viability, cells stained with Erythrosin were imaged at 10Â magni¯cation using a standard °uorescence microscopy con¯guration (mercury lamp, 545/30 excitation; 610/75 emission) as well as bright ¯eld illumination.

Protein binding assay to demonstrate repeatability
Stock solutions of NADH (Sigma-Aldrich) and mitochondrial malate dehydrogenase (mMDH, Merck-Millipore) were prepared in Dulbecco's phosphate bu®ered saline (DPBS) without calcium or magnesium (Invitrogen) at room temperature.Concentrations were calculated based on absorption measurements performed using a spectrophotometer (UV3101-PC, Shimadzu).Final solutions for lifetime measurements were made up further in Mg/Ca negative DPBS, with NADH concentration held constant at 50 M, whilst the concentration of mMDH was varied between 0 and 12.1 M. 50 L of solution was pipetted into wells of a glass bottomed 96-well plate (Imaging Plate CG, zell-kontakt GmbH) in a step pattern.

Vital dye exclusion assay
MCF7 cells were split into 1:3 one day prior to seeding on a glass bottomed 96-well plate in standard growth media at a density of 10 4 cells per well.24 h post seeding, growth media was aspirated and replaced with imaging media following washing with DPBS.Erythrosin viability staining was performed following the protocol for adherent cells outlined in Krause et al. 23 Brie°y, erythrosin B was prepared at 5 g/mL in growth media, in which cells were then incubated at room temperature for 2 min before washing three times with imaging media.To con¯rm the e®ectiveness of the staining method, the cells were subjected to \partial heatshock" as a positive control, following Krause et al. 23 : Cells were submersed in 60 C DPBS for 10 s, then 0 C DPBS for 5 s before staining.DPBS was heated to 60 C using a water bath.Low temperature DPBS was thawed from frozen immediately prior to the experiment.Temperatures were checked (to within 2 C) using a thermocouple.DPBS was introduced to wells by pipetting down the side of wells to minimise mechanical disruption of cell adhesion.All measurements were made at 37 C, in imaging media.

FLIM data
To ¯t our experimental data to a double exponential decay model, we make use of the global ¯tting capabilities of FLIM¯t, an open source analysis package described elsewhere. 24Brie°y, lifetime data can be e±ciently and robustly ¯tted to multiexponential models by global analysis based on variable projection.Complex decay models can thus be globally ¯tted across an image, a time series or a number of repeat acquisitions for a speci¯c condition (e.g., \dose").Furthermore, FLIM¯t enables models to be adjusted to account for instrument response functions, incomplete decays resulting from repetitive pulsed excitation and spatially varying and time varying backgrounds.For the datasets presented here, such simultaneous global ¯tting of millions of pixels in parallel typically requires less than a minute of computation time using FLIM¯t running on a standard desktop computer.
Before ¯tting, an intensity threshold criterion was applied to FLIM images in order to analyze only the pixels corresponding to cells.The lifetime data was smoothed with a 3 Â 3 square kernel before ¯tting, and reference reconvolution methods were used to account for the instrument response function 25 using Stilbene 3 (a monoexponential dye, ¼ 1200 ps) as a lifetime reference.This approach to ¯tting time gated FLIM data has been validated by simulation.Time-varying °uorescent background originating from the multiwell plate or from media components (Sec.3.1) was accounted for using a °uorescence decay pro¯le measured from a well without seeded cells.For consistency with previously published literature, mean lifetimes are presented as \amplitude-weighted" means When applying global ¯tting across the experimental data presented here, errors are typically shown as standard errors calculated across repeat wells within a condition.Where relevant, signi¯cance of di®erence in measured parameters is tested using Tukey's honestly signi¯cant di®erence criterion for multiple comparisons following oneway analysis of variance (MATLAB, MathWorks).

Vital dye exclusion data
Analysis was performed in ImageJ: °uorescence images were thresholded and overlaid on phase contrast images in order to determine the fraction of cells that had been stained.Counting of cells was performed manually.Results from imaged and heat shocked cells were compared to control results using Dunnet's test, implemented in MATLAB. 26

Background °uorescence characterization
In contrast to multiphoton microscopy of NADH, which is inherently optically sectioned, wide-¯eld single photon excitation measurements of NADH auto°uorescence lifetime are prone to background °uorescence, which might arise from UV excitation of °uorescence in the microscope optics, sample holders (e.g., multiwell plates, cover slips, etc.) and culture media.In Fig. 2(a), we present an overview of the measured intensity of selected sample holders described in Table 1 under the experimental conditions outlined in Sec.2.2.Well plates with plastic bases rather than glass are preferred in visiblewavelength studies, as cells grown on plastic tend to exhibit improved adhesion and morphology and in our hands show superior expression of transfected protein constructs.However, under UV illumination, plastic surfaces present unacceptably high background °uorescence, contributing 40% of the number of counts of a typical cell sample.In contrast, background from glass substrates contributes less than 5% of cell °uorescence.

Protein binding assay to demonstrate repeatability
Experiments using mixtures of free and mMDH bound NADH in aqueous solution were used to verify the repeatability of lifetime measurements performed under UV excitation across a 96-well

Vital dye exclusion assay
An investigation was conducted to establish that the UV excited °uorescence imaging process did not lead to cell death due to phototoxicity.Cells were prepared as outlined in Sec.3.3 above and imaged over a 135-min period to approximate the duration of a typical multiwell plate experiment.Typical imaging parameters (0.1 mW excitation power at sample for 10 s) were used with each ¯eld of view being imaged twice.After imaging, cell death was quanti¯ed in four wells that had been imaged with UV excitation, four wells that had not been imaged and four wells that had not been imaged but had been \partially heat-shocked".The cell treatment, staining and image analysis is described in Secs.2.3.3 and 2.4.2.Cells that had been imaged with UV excitation showed no signi¯cant increase in staining fraction over those that had not been imaged.Cells in the wells exposed to the positive control \partial heat shock" condition exhibited a signi¯cant increase in staining fraction over both imaged and nonimaged conditions [see Fig. 4(a)].
In a separate experiment, BT474 cells were imaged every 2 h over a 16-h time course with phase contrast imaging and auto°uorescence imaging being interleaved in order to simulate a time-lapse experiment and look for changes in cell morphology that might indicate cell damage.Figure 4(b) shows time lapse phase contrast images acquired across this period for which no adverse e®ects are apparent from changes in cell morphology.

Assaying metabolic modulation: Action of rotenone on live MCF7 cells
Rotenone is a known inhibitor of Complex I in the respiratory chain.As such, rotenone treatment is expected to result in an increase in free intracellular NADH, and hence a decrease in mean auto-°uorescence lifetime.100 L of either control media or rotenone su±cient to reach a ¯nal concentration of 1 M was automatically dispensed immediately prior to imaging the fourth time point in a time course for each well in turn.Auto°uorescence lifetime data were collected across six wells per condition, for each of which automated time courses were taken in sequence.repeat wells with the standard errors being calculated across the repeat wells.No change in lifetime is observed when control media is added to the wells, showing that the act of dispensing liquid into wells alone has no e®ect on the lifetime measurement.In contrast, there is a signi¯cant decrease in measured lifetime upon stimulation with 1 M rotenone.The observed reduction in mean lifetime is consistent with previous studies in which oxidative phosphorylation has been inhibited by Complex I poisoning 8 and may re°ect an increase in free NADH in the cell.

Assaying metabolic modulation: Action of cisplatin on live MCF7 cells
A plate for investigating NADH lifetime readouts of time-and dose-dependent responses to cisplatin treatment was designed as shown in Fig. 6(a).MCF7 cells were imaged at 37 C in imaging media described in Sec.2.3.Cells were su±ciently densely seeded such that no pre-¯nd operation was required and the ¯rst 10 ¯elds of view spiraling outward from the center of each well were imaged.Figures 6(d

Discussion
The work reported here aimed to investigate the potential to apply an automated wide-¯eld FLIM plate reader to readouts of cellular auto°uorescence with a view to developing high-content FLIM assays for toxicology and to explore potential pitfalls of this approach.Currently, cellular auto°uorescence FLIM experiments are typically performed using laser scanning multiphoton microscopes with lifetimes determined using TCSPC.Such measurements bene¯t from being optically sectioned, which enables intracellular auto°uorescence to be spatially separated from other components that can present an unwanted background.However, the need to sequentially scan the excitation spot limits their applicability to high throughput studies since the nonlinear scaling of multiphoton photobleaching and photodamage with excitation intensity precludes the use of the high excitation powers required for rapid imaging.The parallel pixel acquisition of wide-¯eld time-gated FLIM does enable high imaging speed/ throughput but, as indicated by the results presented in Sec.3.1, the composition of media in which cells are imaged and the choice of growth substrate are important considerations for wide-¯eld FLIM due to their potential contributions to background °uorescence when there is no optical sectioning.Here we show that appropriate selection of media components and substrates can permit quantitative wide-¯eld FLIM experiments with minimal \background" contribution to °uorescence decays.The imaging media composition outlined in Sec.2.3.1 allows FLIM experiments to be performed over extended time periods while maintaining a stable metabolic state in control samples (data not shown).
Using glass bottomed well plates and media optimized for minimal UV-excited °uorescence, the results in Sec.3.2 assaying binding of NADH to mMDH in solution show that our FLIM plate reader achieves a high level of repeatability across a 96-well plate, validating the use of such an instrument for quantitative lifetime-based studies of UV excited °uorescence.The vital dye exclusion assay presented in Sec.3.3 shows that the number of dead cells does not increase following imaging with pulsed UV excitation in our wide-¯eld instrument and the extended time course FLIM measurements do not result in changes in cell morphology indicative of cell deterioration, as determined by the phase contrast imaging interleaved with the FLIM acquisition.
While further experiments would be required to determine the precise photochemical impact of the UV illumination on live cells, e.g., the potential generation of reactive oxygen species or DNA damage, the results presented here give us con¯dence that the UV excitation is not signi¯cantly compromising the health of the cells being imaged.
To realize quantitative FLIM-based assays of changes in metabolism using cellular auto°uorescence, it is necessary to deal with complex °uorescence decay pro¯les, including contributions from free and bound metabolites.Throughout this work we have analyzed NADH °uorescence by global ¯tting to a biexponential decay model and have accounted for background °uorescence, instrument response and incomplete decays in the ¯tting procedure, as outlined in Sec.2.4.1.However, we note that NADH exhibits a more complex °uorescence decay pro¯le for which the ¯tting of FLIM data is subject to some discussion in literature.It is generally accepted that free NADH in solution has two lifetime components associated with di®erent conformations of the molecule: folded (quenched) and extended (unquenched). 27The lifetimes of these components of NADH can be a®ected when binding to proteins and can vary depending on binding partner. 28,29There is potentially an additional source of cellular auto°uorescence detected in our experiments arising from NADPH, which is spectrally indistinguishable from NADH, although it is normally considered that the relative abundance and quantum yield of NADPH is su±ciently low that it may be ignored. 30Thus, a complete description of NADH °uorescence would require a multiexponential model with four or more components.With the limited number of auto-°uorescence photons typically detected, however, it is generally not possible to ¯t FLIM data to such complex models.Qualitative di®erences between di®erent NADH signals can be obtained by ¯tting to a single exponential model but double exponential models result in better ¯ts to experimental data (i.e., lower 2 ) and so the mean lifetime calculated from a double exponential ¯t is often used as a more reliable readout of changes in metabolism.To relate lifetime measurements to underlying metabolic changes, the short and long lifetime components of a double exponential ¯t are often considered to represent the free and bound NADH components, respectively. 7,31lternatively, it is possible to avoid the challenges associated with nonlinear iterative ¯tting photonlimited data to complex models by using phasor analysis, 32 although for quantitative readouts the resulting phasor plots need to be interpreted as linear combinations of lifetime components of a chosen model, typically representing bound and free NADH.
Figure 5 illustrates the application of a double exponential ¯t to NADH auto°uorescence decay pro¯les where the variation in the mean lifetime provides a readout of the action of rotenone on live MCF7 cells.Rotenone acts to inhibit oxidative phosphorylation by poisoning Complex I in the electron transport chain and previous studies have shown that rotenone causes a reduction in the mean lifetime. 8A similar reduction has been shown in studies using cyanide compounds 33,34 that act by inhibiting Complex IV.The observed decrease in measured auto°uorescence lifetime upon rotenone stimulation evident in Fig. 5 indicates that our automated FLIM plate reader can report changes in NADH auto°uorescence lifetime caused by inhibition of oxidative phosphorylation.
To illustrate the potential of automated FLIM assays for studies of toxicology and drug e±cacy, we investigated the action of cisplatin on live MCF7 cells.Cisplatin is a drug commonly used in anticancer therapies, being potent against a great number of tumors.6][37] In previous work, glycolysis in MCF7 cells has been shown to increase following treatment with cisplatin 38 and we therefore investigated whether treatment with cisplatin could lead to a change in the NADH °uorescence lifetime parameters.Our preliminary results presented here in Sec.3.5 are consistent with changes in NADH °uorescence lifetime parameters following a shift in cell metabolism toward glycolysis reported by Skala et al. 39 We note, however, that apoptosis induced by staurosporine has previously been linked to an increase in mean NADH °uorescence lifetime. 30,40The di®erence between this observation and our results may be due to the speci¯c modes of action of cisplatin and staurosporine and/or the rate at which apoptosis occurs with these two compounds but further work would be required to elucidate this.

Conclusions
We have reported the development and demonstration of what we believe to be the ¯rst automated multiwell plate reading FLIM microscope for application to auto°uorescence lifetime imaging under wide-¯eld UV illumination.We have shown that wide-¯eld time gated imaging can be applied to quantitative automated FLIM-based readouts of NADH °uorescence, noting that wide-¯eld FLIM provides greater signal-to-noise per unit acquisition time than laser scanning TCPSC systems, 19 making it more practical for \high throughput" unsupervised lifetime imaging-based multiwell plate assays.To illustrate its potential for assays of cellular metabolism, we have validated its reproducibility using solution-based experiments and presented exemplar measurements of the action of metabolic modulators in live cells.
We believe the automated FLIM plate reader reported here is already useful for reading out changes in cell metabolism without the need for exogenous staining.Lifetime-based readouts of cellular auto°uorescence can provide information on metabolite binding states that is not available from intensity measurements and the ratiometric character of FLIM enables such readouts to be translated across imaging platforms 41 and directly correlated with in vivo measurements.We envisage that its implementation on a high content imaging platform can extend FLIM of cellular auto-°uorescence from small scale microscope-based studies to screening contexts.This could be useful for drug discovery, including studies of toxicology and e±cacy, and for determining e±cacy of particular therapeutic treatments, 31 e.g., for strati¯ed medicine.][15][16][17] The instrument reported here could be further developed by combining the FLIM and phase contrast imaging with other imaging modalities.Indeed, spectrally resolved imaging of endogenous °uorescence from cell metabolites has previously been utilized for toxicology in an automated plate reader 42 and polarization resolved FLIM has been used to study the components of the complex NADH °uorescence signal in a laser scanning multiphoton microscope. 9However, increasing the number of °uorescence parameters to be determined must also increase the required number of detected photons per pixel, which is challenging for live cell HCA, and we believe that FLIM combined with global ¯tting to a biexponential decay model is a good compromise between the experimental limitations and the molecular information obtainable.
This instrument could also be improved by increasing the spatial resolution.Axial resolution could, in principle, be provided by implementing optical sectioning with wide-¯eld time-gated FLIM using a Nipkow spinning disk, analogous to our previous work 3,19,41 but our current Nipkow spinning disk units are not designed for excitation wavelengths below 400 nm.Alternatively it may be possible to combine parallelized multiphoton excitation with wide-¯eld FLIM 43,44 or TCSPC 45 to realize optical sectioning while avoiding the challenges associated with UV excitation, although our experience 44,45 suggests this would still result in longer image acquisition times.As discussed in Sec.2.1, the lateral resolution of our instrument is currently limited by the resolution of the GOI in the current con¯guration.This could be improved using intensi¯ers that can resolve up to 30 lp/mm (resolution determined on precommercial prototype) or by changing the magni¯cation, although this would decrease the ¯eld of view and therefore the numbers of cells to be imaged during HCA.With higher resolution, the mitochondria and cytoplasmic contributions could be separated using image segmentation and global ¯tting could be used to obtain lifetime components averaged over the same compartments across di®erent cells.This would probably require the use of higher NA objectives with immersion °uid, which would add complexity to the instrument but has been realized in HCA plate readers. 46

Fig. 1 .
Fig. 1.Schematic of experimental con¯guration of plate reader equipped for auto°uorescence lifetime imaging.The microscope chassis is equipped with motorized stage, ¯lter cube cassette and objective nosepiece as well as an autofocus unit to facilitate automated acquisition.Time resolved imaging is implemented using a GOI.(SHG: second harmonic generation, ND ¯lter: neutral density ¯lter, GOI: gated optical intensi¯er).

Figure 2 (
b) illustrates the °uorescence intensity originating from di®erent components of media, for which growth media and imaging media formulations are outlined in Sec.2.3.3, and all drug treatments are diluted in imaging media.Based on these results, we choose to use glass-based well plates in all UV-wavelength plate reader experiments, and avoid imaging in media with either phenol red or FBS.

Fig. 2 .
Fig. 2. Comparison of relative contributions from di®erent potential sources of background °uorescence.(a) Integrated °uorescence intensity in units of camera digital numbers (DN) originating from a range of sample holders.(b) Comparison of background signal from media components and treatment compounds.Blue bars show integrated intensity originating from media components, including signal from Matriplate 384 well; red bar shows integrated intensity from MCF7 cells for comparison.Error bars show standard deviation across repeat wells.

Fig. 3 .
Fig. 3. Results showing repeatability of lifetime results across a 96-well plate.(a) Map showing plating layout for NADH-mMDH measurements of lifetime repeatability; numbers associated with wells represent mMDH concentrations in M. (b) Plate map of mean ¯tted lifetime across each well.(c) Plots showing ¯tted mean lifetime averaged over repeat wells.(d) Standard deviations across repeat wells to illustrate how repeatable are the lifetime measurements between repeat wells.

Figure 5 Fig. 4 .Fig. 5 .
Fig. 4. Results of vital dye exclusion and morphology assays following auto°uorescence imaging experiments.(a) Results of erythrosin viability staining for adherent cells, illustrating that no signi¯cant di®erence is evident between control and imaged cell viability using this method.(b) Exemplar phase contrast images over a 16-h time course that were interleaved with NADH auto°uorescence FLIM acquisitions; the time elapsed since start of experiment is indicated in hours.There are no changes in cell morphology to indicate adverse e®ects caused by this extended imaging.

Fig. 6 .
Fig. 6. Results showing e®ect of cisplatin treatment dose and time on MCF7 cells.(a) Plate map indicating cisplatin treatment dose and time.(b) False-color map showing mean ¯tted lifetimes obtained for each well across the plate.(c) Exemplar intensity merged FLIM images from the dataset with false color scale showing ¯tted mean lifetime (scale bar ¼ 50 m).(d)-(g) Fitted parameters presented by condition with error bars representing standard errors over repeat wells.

)- 6 (
g) present data for cells untreated or treated with cisplatin 6 or 12 h before being imaged with mean lifetime values obtained from global ¯tting to a double exponential decay model across the images in each well.Figures6(b) and 6(d) show the changes in mean auto°uorescence lifetime obtained from global ¯tting to a double exponential decay model across all the images for each condition (i.e., across repeat wells) and indicates a signi¯cant decrease in mean auto°uorescence lifetime following cisplatin treatment.Figures6(e) and 6(f) show the variation of the short (\free NADH") and long (\bound NADH") lifetimes with cisplatin dose and exposure time and Fig.6(g) shows the corresponding changes in the fractional contribution from the long (\bound NADH") component to the auto°uorescence signal, which decreases with increased cisplatin dose and treatment time.According to Tukey's test, the short lifetime components measured in this experiment and the long lifetime components of the cisplatin treated cells do not vary signi¯cantly for di®erent cisplatin treatments, although all the cisplatin treated cells present a signi¯cantly shorter long component (\bound NADH") lifetime than the untreated cells.