1. Introduction

The study of cell signaling pathways is of crucial importance in understanding the means by which cells communicate with one another and respond to different stimuli. Such insights are of particular concern for medicine, where the identification of specific interactions between proteins and other cellular components opens up new possibilities for targeting disease at the molecular level. In recent years the advent of new fluorescent probes has made it possible to image a host of cellular processes including gene transcription, the trafficking of proteins and the colocalisation of different species in subcellular compartments [1]. In conjunction with this, advanced spectroscopic techniques including spectrally resolved imaging, polarisation resolved and fluorescence lifetime imaging have vastly expanded the information available from fluorescence microscopy, e.g. [2, 3, 4]. Fluorescence lifetime imaging (FLIM) in particular is fast emerging as an important tool in the biomedical sciences. The excited state lifetime of a fluorophore can be a sensitive function of its local environment and so FLIM, which is the measurement of fluorescence lifetime at each spatially resolvable element in an image, can provide a map of the molecular environment of a fluorophore. In this way, one can obtain information on many cellular parameters, including calcium concentration, pH, temperature and viscosity [5]. A topic of current interest is the use of fluorescent membrane probes that report on variations in lipid order, as discussed in section 4.3.

Perhaps the major current application of FLIM for imaging variations in the local fluorophore environment is the detection of Förster resonant energy transfer (FRET) between molecules. Here, the fluorescence of a particular molecular species (the donor) can be significantly affected by its proximity to a second chromophore (the acceptor). Where the donor emission spectrum overlaps with the absorption spectrum of the acceptor, the donor may transfer energy non-radiatively from its excited state to the acceptor, with an accompanying decrease in the mean donor excited state lifetime [5]. The rate of this energy transfer is inversely proportional to the 6^th power of the donor-acceptor separation and so in practice FRET is only observed for separations of <10nm. This is of the order of the distance between proteins bound together in a complex. Measurements of FRET between appropriately labeled species can therefore report on binding events, as well as on conformational changes in dual labeled molecules, and the spatial distribution of these sub-resolution events can be mapped out using fluorescence imaging techniques including FLIM [6].

While FRET can be studied by analysing many aspects of the fluorescence signal (e.g. intensity, excitation and emission spectra and polarisation anistropy [5, 7]), in practice, fluorescence lifetime often provides the most robust measurement [2]. The donor fluorescence lifetime is largely independent of factors such as fluorophore concentration, excitation and detection efficiency, inner filter and multiple scattering effects, which can complicate or degrade absolute intensity measurements. In particular, the insensitivity of fluorescence lifetime measurements to fluorophore concentration make FLIM the preferred technique when imaging samples for which the precise stoichiometry of the donor and acceptor species is unknown — as is the case when monitoring interactions between separately labeled species (binding partners, enzyme/substrate pairs etc), discussed in section 4.4. Conventionally, however, FLIM has been a relatively slow imaging modality, particularly when implemented in laser scanning confocal or multiphoton microscopes to provide high quality optically sectioned images, typically requiring image acquisition times of minutes. FLIM-FRET has therefore largely been restricted to studies of fixed cells. Higher imaging rates have been achieved with wide-field detection using gated or modulated optical intensifiers for both time domain and frequency domain FLIM but this was at the expense of optical sectioning. Recently, optically sectioned FLIM with wide-field detection has been reported using Nipkow disc microscope systems [8, 9]. We have refined this approach, combining a high power supercontinuum excitation source with Nipkow disc microscopy and an optimized wide-field FLIM acquisition strategy, such that we can acquire depth-resolved fluorescence lifetime images of live cells at unprecedented rates. This permits us to capture fast time-lapse FLIM FRET sequences of live cell dynamics, as well as 3D FLIM stacks at frame rates up to 10 fps.

2. Considerations for live cell fluorescence lifetime microscopy

For live cell FLIM studies it is desirable to acquire optically sectioned fluorescence lifetime images at frame rates fast enough to avoid motion artifacts and to resolve temporal dynamics of interest, with sufficient photons being detected to permit accurate lifetime determination. As well as permitting 3-D image stacks to be recorded, optical sectioning discriminates against background (out of focus) light, which could otherwise corrupt fluorescence lifetime measurements [9, 10]. The enhanced image contrast allows one to delineate different subcellular compartments more easily, and so localize signaling events with greater accuracy. The achievable imaging speed and signal to noise ratio depend on the on the sample, the efficiency of the photon detection and on the particular imaging approach employed, such as single or multiphoton excitation, laser scanning confocal or wide-field microscopy. For FLIM, the required light budget is significantly greater than for intensity imaging. Assuming Poisson noise, for example, achieving an accuracy of +/-10% when fitting a simple monoexponential fluorescence decay profile has been calculated to require at least 225 photons [11]. To fit complex decay profiles to models with multiple decay components, the number of photons required may exceed tens of thousands per pixel, depending on the ratios of the component lifetimes across the image [12]. A priori knowledge can alleviate this signal requirement, as can techniques such as global analysis, where the component lifetimes are assumed to be invariant across the image, or restricted information requirements such as only fitting the fractional populations in each pixel [13].

On balance, we favour single photon rather than multiphoton excitation for in vitro live cell imaging experiments, unless we are working with u.v. excited fluorophores. While multiphoton imaging is preeminent for in vivo studies, providing increased imaging depth in biological tissue compared to single photon excitation, the higher nonlinear photobleaching and photodamage make multiphoton microscopy less appropriate when rapid imaging of a single image plane is required. We note, however, that since multiphoton excitation causes negligible out-of-plane photobleaching compared to conventional or confocal microscopy, it can be advantageous when acquiring dense z-stacks for 3-D imaging, e.g. where a high resolution reconstruction of a tangential slice through a cell is needed.

Optically sectioned FLIM is routinely implemented using laser scanning confocal or multiphoton microscopy, often in conjunction with time-correlated single photon counting (TCSPC) detection [14]. This is a time domain method that utilizes fast electronics to measure the arrival times of individual photons with respect to the excitation pulses, building up a histogram that represents the fluorescence decay profile. While this approach provides high quality FLIM data, it is difficult to implement at high frame rates for biological samples, with typical image acquisition times for live cell FLIM being of the order of several minutes. This is partly because of the limitations of single photon counting detection electronics but it is also because of the increased photobleaching, photodamage and phototoxicity considerations that follow from the use of higher excitation power for high frame rate imaging. At ‘low’ excitation powers, these undesirable effects scale linearly with excitation power but at high excitation intensities the onset of non-linear in-plane photobleaching, phototoxic and photodamage effects becomes evident. Depending on the sample brightness, the need to avoid pulse pile up [14, 15] may also be an issue, although this problem can be addressed to large extent by novel TCSPC hardware in which multiple photomultiplier signals can be combined to facilitate a higher maximum count rate [15].

Parallel pixel illumination and acquisition can permit higher imaging speeds without the penalties associated with photobleaching and photodamage, since the parallel light collection permits a greater number of photons to be acquired for a given peak excitation power [16]. Optically sectioned FLIM with multiple parallel excitation beams has been demonstrated in multiphoton microscopes [17, 18] as well as in single photon excited Nipkow disc based microscopes [8, 9]. These systems utilized wide-field FLIM detection but did not achieve high speed imaging. By exploiting the high-speed wide-field FLIM techniques that we previously demonstrated with endoscopic [19] and microfluidic applications [20], we have developed a new Nipkow disc based FLIM microscope that incorporates a high power supercontinuum excitation source and provides optically sectioned FLIM images of live cells at up to 10 fps. We believe this instrument to be well-suited to rapid FLIM of live cell protein-protein interactions and are applying it to FLIM and FRET studies, both for time lapse imaging of cell dynamics, and with a view to developing high throughput imaging of live cells in a multiwell plate array.

We note that wide-field optically sectioned imaging may also be implemented using structured illumination [21]. This technique works well for conventional intensity imaging and has been combined with FLIM [10]. However, since the optically sectioned images are calculated from multiple image acquisitions that contain both in-focus and out-of-focus light, they exhibit reduced signal to noise ratios compared to confocally sectioned images and this compromises the fitting of fluorescence lifetimes. The calculation of the fluorescence lifetime images is also rather sensitive to any artifacts arising during the calculation of the optically sectioned image.

3. Experimental configuration of the high speed optically sectioned FLIM microscope

Our system is built around an inverted wide-field fluorescence microscope (Olympus IX81) fitted with a Nipkow disc microscope head (Yokogawa CSU10) to provide optically sectioned images. As well as a spinning disc arrayed with pinholes to provide parallel confocal illumination and detection, this version of the Nipkow disc microscope incorporates a second disc of microlenses that focus the excitation light more efficiently through each pinhole, thereby improving the light transmission without compromising on cross-talk between pixels. This optical arrangement requires a spatially coherent excitation source, for which gas lasers or frequency doubled solid-state lasers are usually employed. For time domain FLIM we require an ultrafast excitation source and this can be conveniently provided using supercontinuum generation in microstructured fibres [22]. We previously reported the first application of a supercontinuum-based excitation source for Nipkow disc FLIM microscopy in [8] where a home-built super continuum source pumped by a femtosecond Ti:Sapphire laser was employed. While this system could acquire FLIM images of bright samples such as stained pollen grains in a few seconds, it was much too slow and inefficient to be applied to FLIM of live cells labeled with e.g. fluorescent proteins. The instrument reported here provides some two orders or magnitude improvement in imaging speed/efficiency. This is partly because of the more powerful excitation source, which is now a commercially available supercontinuum source (Fianium Ltd, model SC450). This provides picosecond pulses at 50 MHz repetition rate with spectral power density ~2mW/nm in the wavelength range 470–490nm. We have also optimized the spectral characteristics of the filters and dichroic beamsplitter for the fluorophores used and undertaken an extensive analysis of the noise characteristics of the gated optical image intensifier, allowing us to determine the optimum settings of the gain and other acquisition parameters of the system.

 

Fig. 1. Experimental set up

Download Full Size | PDF

The experimental set-up is shown in Fig. 1. A home-built spectral selection unit based on a prism spectrometer [23] is used to direct radiation in the required spectral range into a single mode optical fibre (cut-off wavelength: 450nm) using a 0.17 NA aspheric objective for fibre coupling. The output of this fibre is coupled into the CSU10 Nipkow disc scan head, which is mounted at the left hand camera port of the Olympus 1X81 microscope. For FLIM the sectioned image is relayed onto the photocathode of a gated optical image intensifier (Kentech Instruments, model number HRI) incorporating a variable delay generator to facilitate capture of time gated intensity images at different intervals after pulsed excitation. The output image from the phosphor screen of the intensifier is relayed onto a cooled scientific grade CCD camera (Hamamatsu ORCA-ER) for data acquisition and subsequent processing.

In order to maximize the photon economy, and therefore the imaging rate of time-gated FLIM, we aim to use time gates with the maximum practical width. The HRI provides time gates with rising and falling edges of less than 100 ps and conveniently adjustable gate widths up to 1ns. The relative steep rising edge of this gate profile permits us to start collecting light almost immediately after excitation. Depending on the particular FLIM experiment, we can use as few as two time gates to estimate the lifetimes of single exponential decay profiles using an analytic rapid lifetime determination (RLD) algorithm (providing the background signal is known) to achieve real-time (up to ~10 fps) FLIM. To study complex decay profiles or samples exhibiting significant variation in fluorescence lifetime across the field of view, we tend to record up to ~10 time gates for each FLIM image, thereby decreasing the FLIM rate accordingly. Lifetime decays are fitted to the experimental data using a nonlinear least squares iterative Levenberg-Marquardt algorithm. We typically record the temporal instrument response prior to measurements on the sample, and can convolve this with the decay in the fitting software. The full frame rate (336×256 pixels) of our system is currently limited to 30 fps by our read-out CCD camera and this sets an upper limit for our time-gated imaging system. We note, however, that very few biological samples would be sufficiently bright to permit faster full-field imaging.

4. Results and discussion

4.1 Comparison of the system with confocal TCSPC

To illustrate the improved speed of FLIM acquisition, we compared this Nipkow disc FLIM microscope with a laser scanning confocal TCSPC FLIM microscope system using the same excitation wavelength. While the latter system has many attractive features, including the spatial resolution and contrast afforded by confocal microscopy, the signal/noise benefits of single photon counting detection - which allows the (Poissonian) noise to be accurately modelled when fitting the decay profiles - and the ability to perform highly temporally resolved measurements with no loss in signal (as occurs with gated detection), it is less well suited for rapid imaging, as discussed in section 2. For these experiments, we used a TCSPC module (Becker and Hickl GmBH, SPC-830) in conjunction with a cooled photomultiplier tube (Hamamatsu PMC-100) to obtain fluorescence lifetime images. The PMT was connected to the external port of a Leica SP2 confocal microscope. Single photon excitation was provided by a frequency doubled femtosecond Ti:Sapphire laser (Spectra Physics, Tsunami), the output pulses of which were stretched to ~10 ps before coupling into the microscope. A 490 nm dichroic mirror was used in both the Nipkow disc microscope and SP2 confocal microscope, together with a 500–550nm band pass filter in the emission channel. A x60 objective was used in each microscope with NA of 1.40 and 1.30 for the confocal and Nipkow disc microscopes respectively. To account for the slight difference in axial resolution, the pinhole in the SP2 microscope was opened to obtain an equivalent axial PSF to that of the Nipkow disc — this ensured an equal volume of the sample was probed in both microscopes. In the Nipkow disc microscope, the excitation source was the supercontinuum source described in section 3.

Although both systems will ultimately be limited by photobleaching considerations (particularly with repeated acquisitions for time lapse sequences), the goal here was to compare the signal to noise achievable with the two instruments operating near their maximum practical imaging speed. Therefore the laser power was adjusted to provide the approximate maximum signal for each experimental configuration. For the confocal TCSPC system, this limit was taken to be the pulse pile up limit of 10⁶ counts per second, which corresponded to 8 µW at the sample. For the Nipkow disc system, the laser power at the sample was 0.9 mW which was limited by the maximum available from the continuum source. The system was set to acquire 6 time-gated images for each FLIM acquisition and the gain on the HRI was set to 550 volts.

Images of cells were acquired at integration times of 1–30 seconds with the number of image pixels set to 256 x 256 in the confocal microscope, and 256×336 on the Nipkow disc system. Images were smoothed with a 3x3 kernel and lifetimes fitted in a custom written LabView program (National Instruments). The intensity in each image was thresholded so as only to take into consideration those pixels which had 25% or more of the peak intensity and fluorescence lifetimes were fitted using a least squares iterative Levenberg Marquadt algorithm. Figure 2(a) presents representative FLIM images acquired on each system and Fig. 2(b) shows how the mean and standard deviation of fluorescence lifetime averaged over the cells varied as a function of acquisition time. It is clear that the Nipkow FLIM system provides superior FLIM images for acquisition times of 10 seconds and shorter: the mean lifetime recorded on the Nipkow FLIM microscope is approximately constant over the range of acquisition times down to 1 second, while the confocal TCSPC microscope presents an artifact for acquisition times below 10 seconds, owing to insufficient photons being detected. This shortage of fluorescence photons is reflected in the plot of standard deviation, where the increased width of the lifetime distribution is evident across the full range of acquisition times. For a one second acquisition, only the Nipkow FLIM microscope provides useful lifetime images.

 

Fig. 2. (a). Representative images of EGFP expressing cells acquired on the Nipkow disc microscope (left column) and confocal system (right column) with different acquisition times. The white pixels are those where an (erroneous) lifetime has been calculated that lies beyond the bounds of the color scale. (Scale bars =10 µm)

Download Full Size | PDF

 

Fig. 2. (b). Plots of the mean fluorescence lifetime and standard deviation measured across cells expressing EGFP using confocal TCSPC and time gated Nipkow disc microscopy.

Download Full Size | PDF

4.2 Theoretical comparison of wide field and TCSPC performance

The results shown in Fig. 2 are empirical though nonetheless informative measures of relative performance. To support this comparison, simulations were undertaken to estimate the percentage error in measured fluorescence lifetime of a “standard” fluorophore as a function of acquisition time for the two approaches. These simulations included the expected noise for a given flux at the detector. For the case of TCSPC FLIM, where one measures photon events per second and the signal is shot noise-limited, both the noise and the flux on the detector can be straightforwardly calculated from Poissonian statistics.

For the wide field FLIM system the situation is more complex, as the noise will vary with the gain on the multichannel plate (MCP). Modeling the wide field detector therefore requires knowledge of the variation of intensifier noise as a function of MCP gain. This was characterized in a separate experiment by observing the intensity distribution of images recorded at different MCP gains when the photocathode was uniformly illuminated by a continuous wave diode laser. At each gain setting, measurements were made for 10 different CCD integration times, the maximum of which was chosen so that the CCD just reached saturation. The noise at each integration time was then defined to be the standard deviation of the intensity at each pixel across a series of 100 images acquired with the same settings. For the single MCP GOI used in these experiments, it was found that the variance in intensity across the 100 images scaled linearly with camera integration time at all gain values, although with differing constants of proportionality. To include noise in these simulations, we used the values obtained for a gain setting of 550 V, this being the setting most frequently used for cell and tissue imaging with our FLIM system.

To determine the flux incident on the photocathode, single photon counting measurements were carried out using the CCD to record the signal from the phosphor of the GOI. By using neutral density (ND) filters to provide calibrated attenuation of the detected flux and reducing the camera integration time to 5 ms with the MCP gain set at its maximum value of 850 V, it was possible to resolve individual photon events on the CCD (a comparison of the variance in their number with the mean detected across a series of 100 images confirmed that the distribution in photon number was Poissonian). The incident flux could then be calculated from knowledge of the ND filter transmission. Thus a look-up table could be constructed to provide the noise associated with any gain setting, together with the corresponding flux.

For these simulations the flux per pixel per second was taken to be that obtained when using the maximum practical laser power. For the case of TCSPC, this was again considered to correspond to a count rate of ~10⁶ counts per pixel per second. For the Nipkow disc microscope, one could envisage using a much higher total excitation power distributed across the many excitation foci. In practice, however, we are limited by the available laser power. For this comparison we took the flux to correspond to the maximum observed from the cells expressing EGFP, discussed in the previous section. This flux was calculated from knowledge of the integration time required to reach camera saturation at a gain of 550V and was of the order 10⁵ counts per pixel per second.

Using these values of incident flux and taking into account the noise characteristics of each detector, we compiled a series of detected fluorescence decay profiles for the two systems for a “standard” fluorophore with a monoexponential decay profile of 2.0 ns lifetime. For the Nipkow FLIM system, 6 time-gates were used to sample the fluorescence decay profiles. The number of photons in each time bin was determined from the maximum flux per pixel and noise was added accordingly. This was repeated for different acquisition times and this series of simulated data was fitted using our in-house software to provide a lifetime histogram for each acquisition time. The results of this simulation are summarized in Fig. 3 below. It will be seen that the error in fitted lifetime is lower over the range of acquisition times for the wide-field Nipkow FLIM with a detection rate of 10⁵/s than for the TCSPC system operating at a count rate of 10⁶/s. The results for a TCSPC count rate of 10⁵/s are also illustrated in Fig. 3. We note that the Nipkow disc microscope system could further improved performance for many samples if higher excitation powers were used.

 

Fig. 3. Accuracy in lifetime as a function of acquisition time for three cases: i) confocal time correlated single photon counting with a count rate of 10⁶s^-1; ii) confocal time correlated single photon counting with a count rate of 10⁵s^-1; iii) the Nipkow disc system, assuming a flux per pixel equal to that calculated from cells expressing EGFP. (Lines drawn here correspond to raw, unsmoothed image data).

Download Full Size | PDF

Figures 2 and 3 illustrate how the advantages of parallel pixel acquisition outweigh the reduction in photon efficiency imposed by the sampling inherent in time-gated imaging, which is itself significantly mitigated by using wide (~ns) time gates to sample the decay. We note that this analysis has assumed that the fluorophore exhibits a monoexponential decay of known lifetime, which is the optimal scenario for time-gated FLIM. These assumptions are reasonable approximations when working with the short acquisition times required for live cell FLIM, for which it is difficult to obtain sufficient photon counts for multiexponential decay analysis. Where more complex decay analysis is required, e.g. for some FRET experiments, global analysis methods can be used that allow the relative populations of different lifetime components to be determined using a small (<6) number of gates.

4.3 Demonstration of the system on live cells (cholesterol depletion)

As an example of the utility of this Nipkow FLIM system for rapid imaging of live cell behaviour, we present a time-resolved study of membrane dynamics following cholesterol depletion. FLIM has recently been applied to the study of membrane microdomains (sometimes termed ‘lipid rafts’), thought to be cholesterol- and sphingolipid-enriched ordered-phase domains in the plasma membrane [24]. Such microdomains have been implicated in a range of cellular processes including acting as platforms for cell signaling and membrane trafficking and are thought to be involved in a number of disease processes [25–27]. These membrane microdomains may be studied using fluorescence probes, such as the dye Di-4-ANEPPDHQ, which has been shown to undergo a blue-shift in fluorescence emission in the ordered phase of lipid membranes [28]. Recently we showed that, in addition to this spectral shift, Di-4-ANEPPDHQ exhibits a change in fluorescence lifetime between the ordered and disorded phases [29]. Measurements of the fluorescence lifetime of this probe can provide higher contrast images than those obtained from changes in emission spectrum and when used with the Nipkow FLIM microscope this provides an exciting possibility for imaging changes in lipid membrane order on fast timescales.

 

Fig. 4. Fluorescence lifetime images of HEK 293 cells stained with Di-4-ANEPPDHQ (a) prior to and (b) 9 minutes after addition of 7mM methyl-β-cyclodextrin; (c) change in the mean fluorescence lifetime in the plasma membrane at intervals following addition of methyl-β-cyclodextrin. Lifetime values and error bars are the mean and standard deviation calculated from a region of interest around the plasma membrane, averaged across several cells (Scale bar =10 µm)

Download Full Size | PDF

A classic way to investigate the importance of microdomains to a particular biological system is observe changes in that system following disruption of these domains. This can be achieved by depleting the cholesterol content of the plasma membrane by incubating the cells with methyl-β-cyclodextrin. Figure 4 above shows representative fluorescence lifetime images (inset) of cells labeled with Di-4-ANEPPDHQ imaged (a) prior to and (b) after 9 minutes incubation with methyl-β-cyclodextrin. The change in probe lifetime indicates the expected decrease in lipid order of the cell membranes. Below, the graph (c) shows how the mean lifetime (averaged across the cells) varied over a time-course experiment for which FLIM images were acquired every 17 s (the image acquisition time was 8 seconds).

This experiment allowed us to follow the dynamics of cholesterol depletion in the membrane, and illustrates the Nipkow FLIM microscope’s potential for rapid imaging of membrane lipid order. Interestingly, the reduction in lipid order was observed to occur significantly faster than has previously been assumed when using methyl-β-cyclodextrin for cholesterol depletion experiments [30–32], falling to a plateau after 5 minutes, in contrast to other protocols that used exposure times of the order of 10–20 minutes or longer. This could be an important consideration for cell viability following prolonged exposure to such reagents. This preliminary experiment was undertaken at room temperature (20°C) and we plan to repeat it at physiological temperatures, where the rate of cholesterol depletion is likely to be significantly faster, and to investigate the effects of cyclodextrin concentration on the dynamics.

4.4 High speed FLIM-FRET of protein interaction in live cells

As discussed above, while FLIM is widely recognized as one of the more robust techniques to image FRET [2], optically sectioned FLIM has not been widely applied to live cells owing to the limited imaging speed of current instrumentation. Probably the most widespread technique for rapid FRET is spectral ratiometric imaging, which involves comparing the fluorescence intensities in two emission channels centred on the donor and acceptor peak emission wavelengths. This approach is most successful when the donor and acceptor concentrations are equal and provides the basis for a range of FRET biosensors, in which the donor and acceptor fluorophores are attached to the same molecule. Such biosensors have been used successfully to image a number of cellular parameters, including calcium fluxes, phosphorylation events, receptor binding and production of downstream signal molecules [33, 34]. Unfortunately the spectral ratiometric method is less effective when the donor and acceptor are labeled on separate molecules. This is because of problems with spectral bleed-through (detection of donor fluorescence emission in the acceptor channel), direct excitation of the acceptor, and the difficulties associated with distinguishing between changes in emission ratio arising from FRET and those arising from differences in local concentration of donor and acceptor. Although correction algorithms have been developed to address these issues, they require the comparison of multiple images acquired at different excitation wavelengths and can be highly sensitive to noise.

For situations where the precise stoichiometry of the donor and acceptor species is unknown, it is preferable to make measurements of the donor fluorescence lifetime, which is independent of fluorophore concentration. The rapid FLIM capability of our Nipkow disc based microscope makes it an effective tool for studying protein interactions by FLIM-FRET, permitting the recording of live time-lapse, sectioned FLIM-FRET images with integration times significantly less than 10 seconds.

To demonstrate the potential of Nipkow FLIM for such studies, we imaged the activation state of the small GTPase H-Ras in MDCK and COS 7 cells. Ras proteins are an important subject of study in cancer research, due to the high proportion of tumor cells that carry mutations in their Ras genes, and identifying the spatial and temporal aspects of interactions between Ras proteins and their effectors is an active field of research. The binding of a fluorescently labeled Ras protein to a coexpressed effector binding domain of Raf (Raf-RBD) is a favoured method for observing changes in activation profile and has formed the basis of several similar studies [35–38]. Ras activation can thus be monitored by detecting FRET between the bound Ras protein and the fluorescently labeled Raf-RBD, as represented schematically in Fig. 5.

 

Fig. 5. Monitoring Ras activation by FRET: coexpression of mRFP-labeled H-Ras, together with an EGFP-labeled Ras binding domain from C-Raf Kinase (Raf-RBD) permits Ras activation to be studied by observing FRET between the two fluorophores. Prior to activation, Ras is in a GDP-bound state but in response to upstream signaling, dissociates from GDP and binds to GTP, resulting in recruitment of Raf-RBD to the membrane and ensuing FRET signal.

Download Full Size | PDF

For the experiments described below, live MDCK cells coexpressing H-Ras-mRFP and an EGFP-labeled Raf-RBD were imaged at different time points following EGF stimulation. The EGFP donor was excited with a wavelength excitation band of 470–490nm and its fluorescence was selected using a 500–550nm band pass filter. To confirm the presence of H-Ras in the cell membrane, images of H-Ras-mRFP were obtained by mercury lamp excitation, using a 550–590nm excitation filter with 600LP emission filter. Figure 6 shows donor EGFP fluorescence lifetime images of the live MDCK cells pre stimulation and post stimulation with EGF, together with the image obtained of H-Ras-mRFP by mercury lamp excitation (top). Within 30 seconds of adding epidermal growth factor (EGF), a shortening of the EGFP donor lifetime was observed at the cell membranes, indicating activation of H-Ras-mRFP. The maximum decrease in donor lifetime was seen at 10 minutes, after which the lifetimes began to rise, indicating that this was a transient activation profile.

In addition to time lapse imaging studies, the rapid optically sectioned FLIM capability offers opportunities for high throughput imaging of cells for screening applications. The high signal to noise achievable in short FLIM acquisition times is useful for pharmaceutical screening applications including, for example, evaluating the effect of different inhibitors on Ras activity. This rapid optically sectioned FLIM also permits the acquisition of z-stacks to provide 3-D fluorescence lifetime images as shown in Fig. 7.

 

Fig. 6. Time lapse fluorescence lifetime imaging of Raf-RBD-EGFP interacting with H-Ras-mRFP at the cell membrane in MDCK cells. Left column: Donor fluorescence lifetime (continuous scale); middle column: donor fluorescence lifetime (binary scale, thresholded at 2400ps); right column: merged fluorescence lifetime with intensity; bottom: H-Ras-mRFP localization. White arrows indicate regions of lifetime shortening in the plasma membrane (Scale bar =10µm). Each image was acquired in 6s.

Download Full Size | PDF

 

Fig. 7. Sectioned fluorescence lifetime image stack through a COS 7 cell expressing H-RasmRFP and Raf-RBD-EGFP, displaying FRET at the plasma membrane following stimulation by EGF. Each image was recorded in 6s, with a 100s total acquisition time (Scale bar =10 µm).

Download Full Size | PDF

4.5 High speed FLIM of model EGFP-mRFP FRET constructs in live cells

The previous section illustrated the potential to record biologically significant optically sectioned images of intermolecular FRET with acquisition times below 10 seconds. The outcome of this experiment depends not only on the FLIM microscope but also on the sample itself. To further investigate the imaging speed of our Nipkow FLIM system, we used a model FRET system - a tandem construct of EGFP linked to mRFP by a 6 amino acid linker — in live MDCK cells. This construct should exhibit a strong FRET signal, which we confirmed by comparing the EGFP donor fluorescence lifetime with that observed in live MDCK cells expressing EGFP alone. For rapid FLIM we set the binning of the read out camera (Hamamatsu ORCA-ER) to 4×4, which yields a minimum frame read out time of 30 ms. The maximum FLIM speed can be achieved using only two time gates to sample the fluorescence decay profiles. Lifetimes may then be calculated analytically from the data using a rapid lifetime determination (RLD) algorithm [39–41]. This therefore implies a maximum FLIM acquisition rate of ~15 fps. In practice, however, longer integration times at each gate are required to obtain sufficient signal to noise for accurate lifetime determination.

Figure 8 shows fluorescence lifetime images of two cells, one expressing EGFP alone and the other expressing the tandem FRET construct, acquired at frame rates up to 10 fps, together with the lifetime histograms. These images were acquired from two successive time-gates and the lifetimes were calculated analytically using an RLD fitting procedure [42]. It will be seen that the difference in fluorescence lifetime between the cell expressing the FRET construct and that expressing EGFP alone is clear in all the FLIM images and that for frame rates up to 5 fps, the lifetime histograms exhibit two clearly resolved peaks. These results are encouraging for high throughput assay applications. We note that by averaging multiple cells, one would be able to observe lifetime contrast reporting FRET at even higher frame rates.

 

Fig. 8. (a). Images of live MDCK cells (with Mercury lamp excitation) expressing either EGFP or a tandem construct of EGFP-mRFP; (b) fluorescence lifetime images of the same field of view, captured at frame rates of 1 fps (top row), 5 fps (middle row) and 10 fps (bottom row). Also shown are the lifetime histograms corresponding to each image. (Scale bar =10µm).

Download Full Size | PDF

5. Conclusions

We have demonstrated the application of an optically sectioning Nipkow FLIM microscope to studies of live cell signalling events. The acquisition times are significantly shorter than that achievable by a laser scanning confocal TCSPC FLIM microscope on equivalent samples and are sufficiently short to permit the study of dynamic changes in cell activation and function. It is the parallel pixel acquisition of the Nipkow disc microscope that provides this imaging speed advantage — even though the signal to noise ratio per photon detected is lower than that obtained from TSCPC FLIM.

A key step in achieving this performance has been the incorporation of a high power fibre-laser-pumped supercontinuum source to provide spatially coherent ultrafast excitation light. We note that the spectral versatility of such sources will enable multiplexed FLIM and FRET experiments with simultaneous or interleaved excitation of multiple fluorophores. This versatility and imaging speed could have a significant impact in high content screening. The rapid, optically sectioned, time-gated FLIM could be combined with global analysis techniques to assay protein interactions and signal pathways in an automated manner analogous to the work reported in [43].

6. Materials and methods

6.1 Construction of plasmids for expression of fluorescent proteins

Enhanced Green Fluorescent Protein (EGFP) was expressed in live cells using the pEGFP-C1 vector (Clontech). Constructs expressing human Ras were prepared in the pTriEx4 vector (Novagen) and incorporated the full-length open reading frame (ORF) fused at the N-terminus to the monomeric Red Fluorescent Protein (mRFP) ORF. The fluorescent tag was separated from the Ras protein by a linker incorporating the sequence GGSGGS. The Raf-RBD-EGFP construct was constructed in the pEGFP-C1 vector and comprised amino acids 51–200 of human Raf-RBD from C-Raf Kinase, separated from EGFP by the same GGSGGS sequence. The tandem construct of EGFP-mRFP was prepared in the pTriEx4 vector, and contained the GGSGGS sequence as a linker between the two fluorophores.

Standard splicing PCR was used to generate fused expression constructs. In brief, the ORFs of fluorescent proteins and Ras were amplified with an overlapping region consisting of a glycine-glycine-serine-glycine-glycine-serine linker (in the final expressed construct). The two gel purified PCR products were mixed and a second PCR initiated with oligos allowing amplification of a fused construct and allowing cloning into pTriEx4 by Ligation Independent Cloning (LIC). The fused PCR product was gel purified and cloned into pTriEx4 following the manufacturers protocol. Cloned constructs were sequence verified and expressed

6.2 Cell culture and transfections

COS 7 and MDCK cells were cultured in Dulbecco’s modified Eagle’s medium DMEM (Gibco, Invitrogen) with 10% added foetal bovine serum. HEK 293 cells were cultured in DMEM medium with 4500mg/ml glucose, l-glutamine and supplemented with 10% foetal calf serum (FCS) and 100µg/ml penicillin/streptomycin.

For the comparison between FLIM on the Nipkow disc microscope and confocal TCSPC, COS 7 cells were seeded onto 30mm MatTek dishes with glass coverslip bases. Cells were transfected with 1.0µg of pEGFP-C1 using a standard Lipofectamine PLUS protocol from Invitrogen and left overnight in DMEM containing 10% FBS. Cells were imaged the following day in phenol red-free MEM (Gibco, Invitrogen), containing 30mM HEPES.

For the FRET experiments between H-Ras-mRFP and Raf-RBD-EGFP, MDCK cells were serum starved overnight in 0.1% FBS and microinjected the following morning with 10µgml^-1 DNA pEGFP-C1/Raf-RBD-EGFP and pTriEx4/H-Ras-mRFP. Microinjection was performed using a FemtoJet system from Eppendorf. Cells were imaged 4–6 hrs post microinjection in phenol red-free MEM with 30mM HEPES. Cells were stimulated with 100ngml^-1 human recombinant EGF (Calbiochem).

For the cholesterol depletion experiments, HEK 293 cells were seeded onto 50mm glass bottom dishes (Intracel), and left overnight in DMEM containing 10% FBS. 1 hr prior to imaging, the medium was replaced with serum free medium supplemented with 5µM di-4-ANEPPDHQ. Cholesterol depletion was achieved by addition of varying concentrations of methyl-β-cyclodextrin (Sigma-Aldrich) in phosphate buffered saline (PBS).

For the RLD experiments, MDCK cells were injected with 10µgml^-1 DNA pEGFP-C1 or pTriEx4/EGFP-mRFP and imaged 4–6 hours later on the microscope in phenol red-free MEM with 30mM HEPES.

6.3 Live cell time-lapse fluorescence lifetime imaging

For the live cell time-lapse imaging experiments, a x100 1.3NA oil immersion Olympus objective and 1X81 Olympus microscope was used, in conjunction with the CSU10 Yokogawa confocal scan head (Perkin Elmer Life Sciences UK). Cells were imaged on a 37°C heated stage unless otherwise stated. FLIM images were recorded as described previously [8].