Fluorescence is a ubiquitous readout of molecular localisation in the life sciences and has enabled the elucidation of many biological processes, particularly since the development of genetically expressed fluorophores has facilitated direct observation of signalling processes in live cells. Co-localisation of different fluorophores via fluorescence microscopy has provided evidence for protein interactions but this has been (diffraction) limited to the resolution of the microscope used. Although super-resolved imaging techniques[1
] can break this diffraction limit, there is still a need for stronger direct evidence of protein interaction and this is afforded by Förster resonance energy transfer (FRET), which only occurs between fluorophores separated by less than ≈10 nm.[2
] As well as being used to detect and monitor the binding of appropriately labelled proteins, either as an end point in fixed cells or as a dynamic process in live cells, FRET is also utilised in a range of genetically expressed intracellular biosensors of which the “Cameleon” calcium sensor[3
] is the first and best known. A wide array of intramolecular FRET biosensors have now been reported, reading out, for example, calcium,[4
] and others.[10
] In addition, there are “cleavage” sensors for which the FRET signal disappears upon activation, such as the calpain FRET sensor for monitoring calpain proteolytic activity,[11
] to complement the vast range of intermolecular FRET readouts of protein interactions.
While there are many approaches[12
] to detect FRET, the most widely used are probably spectral ratiometric imaging and fluorescence lifetime imaging (FLIM). Fluorescence lifetime measurements are particularly robust because they are usually independent of the fluorophore concentration, the excitation and detection efficiencies, and the impact of scattering and sample absorption. Spectral ratiometric measurements have the advantage that they are often easier to implement in terms of instrumentation and require fewer detected photons (and therefore shorter data acquisition times) than lifetime measurements. Their chief drawback is that they require calibration of the spectral response of the optical system, which can include the sample itself. This calibration is usually achieved through measurements of samples labelled with donor only and acceptor only as well as the sample under investigation, for example.[13
] Once this is realised, spectral ratiometric measurements can then provide information concerning the effective FRET efficiency (i.e. the product of the actual FRET efficiency and the fraction of FRETing donor/acceptor molecules) and the relative total numbers of donor and acceptor molecules.[14
] With an independent measurement of the actual FRET efficiency of the interaction under investigation, spectral ratiometric FRET can then also provide the fractions of the FRETing donor and acceptor populations. The independent measurement of FRET efficiency (positive control) can be determined by a spectral ratiometric measurement of a FRETing sample exhibiting the same interaction with known stoichiometry (e.g. donor directly linked to acceptor) or by using another method, such as acceptor photobleaching or FLIM.[14
Fluorescence lifetime measurements are relatively robust in the presence of spectral crosstalk, being insensitive to donor–acceptor stoichiometry since it is only the donor fluorescence that is measured. They therefore do not require parallel spectral calibration measurements and are independent of the optical system, which makes them attractive for more challenging biological samples such as live animal models, for which the spectral properties can change in space and time. A determination of the mean fluorescence lifetime provides the effective FRET efficiency and fitting fluorescence lifetime measurements to a double-exponential decay model can provide an estimate of both the actual FRET efficiency and the fraction of FRETing donor population, which is suitable for quantitative assays such as dose–response curves. It should be noted, however, that fitting lifetime data to complex fluorescence decay models requires significantly more detected photons than measurements of single-exponential decay times, although this can be mitigated for FRET by using global analysis as discussed below. For this reason it is desirable to utilise fluorophores that exhibit monoexponential decay profiles.
FLIM thus provides a widely applicable means to read out FRET in order to map protein interactions and so to study cell signalling[16
] and other processes. It can also be used to map other variations in the local fluorophore environment, for example utilising probes that can report on the concentration of analytes such as calcium,[17
] or physical changes such as temperature[18
] or lipid order.[19
] The robust nature of FLIM readouts, including of FRET, means that they may potentially be translated along the drug-discovery pipeline from in vitro assays to animal models. In spite of this potential, however, FLIM has not yet made a significant impact on drug discovery. This is at least partly due to concerns about data acquisition times and to a lack of available instrumentation. Herein, we present a rapid automated optically sectioning FLIM multiwell-plate reader that we are developing for high content analysis (HCA) and illustrate its potential with an exemplar assay for aggregation of the Gag protein during the human immunodeficiency virus (HIV) cycle. To illustrate the future potential, we demonstrate the capability to simultaneously image multiplexed FLIM (FRET) readouts and to extend 3D FLIM and FRET to endoscopy and tomography of disease models.
The ability to read out protein interactions using FLIM and FRET in automated HCA could provide significant new opportunities for drug discovery and basic research, for example, providing the opportunity to utilise siRNA libraries for screening genetic perturbations with respect to signalling networks. A major obstacle to the application of FLIM to HCA has been the speed of data acquisition. To date, most FLIM experiments in cell biology have followed the first demonstration of FLIM microscopy,[20
] and utilise time-correlated single photon counting (TCSPC)[21
] implemented in a laser scanning confocal or multiphoton microscope. While this approach provides high-quality data, the imaging speed for this sequential pixel acquisition is limited by the constraints of single photon counting detection and by the nonlinear increase in photobleaching and photodamage that ensues as the power of the scanning laser beam is increased linearly. TCSPC has been implemented in a laser scanning multiwell-plate reader[22
] but this instrument did not acquire fluorescence lifetime images, instead delivering a single lifetime measurement per well. Recently, an imaging multiwell-plate reader utilising multiphoton TCSPC FLIM was reported,[23
] but this was for secondary measurements following identification of “hits” by steady-state polarisation-resolved anisotropy imaging since the FLIM was considered to be too slow for rapid measurements.
Wide-field FLIM achieves faster imaging rates than laser scanning FLIM with lower photobleaching due to the parallel pixel interrogation. The first demonstration of wide-field FLIM to read microarrays was demonstrated using time-gated imaging to improve the sensitivity for DNA profiling.[24
] More recently, an automated instrument for unsupervised FLIM of multiwell-plate sample arrays was reported[25
] that exploited the frequency domain (FD) fluorescence lifetime determination of FRET in a wide-field (non-sectioned) microscope. This provided an elegant demonstration of the potential of automated FLIM FRET and the opportunities afforded by statistical analysis of such FLIM array data. More recently, a wide-field FD FLIM plate reader has been applied to image post-translational modifications (tyrosine phosphorylation) in situ—specifically uncovering components that transduce signals from epidermal growth factor receptor.[26
] While these wide-field FD FLIM instruments have been successful, they can be limited in signal-to-noise ratio by the lack of optical sectioning, which improves quantitative readouts by rejecting contributions from out-of-focus fluorescence (as is inherent in laser scanning confocal/multiphoton systems). FD FLIM systems can also be limited in speed by the need for sufficient temporal sampling of the fluorescence signal to avoid aliasing artefacts that can arise from pulsed excitation or nonlinearities in the excitation/detection modulation. This aliasing issue can, however, be addressed using special demodulation functions, as in the technique of 2
We recently demonstrated that wide-field time-gated imaging using a gated optical intensifier (GOI) can be combined with a Nipkow disc confocal microscope to provide high-speed optically sectioned FLIM that is significantly faster, for a given signal-to-noise ratio (S
), than current TCSPC instrumentation. With this instrument we demonstrated the acquisition of time-gated FLIM FRET images of live cells labelled with FRET constructs at up to 10 frames per second.[27
] Although wide-field time-gated FLIM generally yields a lower S
ratio per photon emitted by the sample than TCSPC, the parallelism provides a greater S
per unit acquisition time, which is important for studying dynamics and for higher-throughput applications. Wide-field time-gated imaging can facilitate FLIM with as few as three time-gates for monoexponential and five for double-exponential decay models in the presence of a background contribution. We previously published the first report of an automated high-speed optically sectioned FLIM multiwell-plate reader for HCA,[28
] by utilising time-gated imaging via a home-built Nipkow disc microscope for automated (unsupervised) FLIM FRET with acquisition times of less than 10 s per well including sample motion, autofocus, cell finding and system calibration (or less than 16 min/96-well plate), and demonstrated its application to fixed and live cell imaging. Herein, we demonstrate that this approach can be implemented on a commercial wide-field multiwell-plate reader (GE Healthcare IN Cell Analyzer 1000) typically acquiring optically sectioned FLIM images in a few seconds and automatically reading a 96-well plate in ≈10 min, within which time we typically acquire several hundred photons per pixel.
Our in-house software controls the automated image acquisition and data analysis, including image segmentation and global analysis. Approximately 300 photons are sufficient to fit a fluorescence decay profile to a monoexponential decay model but not to more complex models. For many applications it is still useful to fit complex fluorescence decays to a monoexponential model and use the change in effective lifetime as an indicator, for example, of FRET. Alternatively, one can obtain more detailed information by applying global analysis to the same data. The simplest approach is global binning, which involves binning all the detected photons from a region of interest (ROI) and fitting the composite signal to, for example, a double-exponential decay model, to determine the fluorescence lifetime components under the assumption that the lifetimes are invariant across the ROI. This can be combined with automatic image segmentation to define the regions of interest. A more sophisticated approach is to apply global fitting,[29
] which again assumes that the component lifetimes are invariant across the image but here the entire image data set is fitted in parallel to the model by minimising a global χ2
. Global fitting is computationally intensive and so is normally applied in post-processing, while more immediate readouts are typically provided by fitting to a monoexponential decay model or by using analytical approaches such as rapid lifetime determination[30
] or phasor analysis.[31
] Phasor analysis can also be applied to FLIM FRET data and provides useful graphical representations with no need for iterative fitting. It can be extended to include a “cluster-style” analysis of the phasor plot distribution that allows the signal at each pixel to be decomposed into FRETing and non-FRETing components, thus allowing the stoichiometry at each pixel to be determined.[32
] Phasor analysis and global fitting techniques make the same basic assumption that the signal from each pixel can be described as a linear sum of two spatially invariant decay profiles. Phasor analysis imposes lower computational and memory requirements than global analysis, but a quantitative comparison of the information content, signal-to-noise performance and degree of bias for these two different approaches under practical experimental conditions has, to the best of our knowledge, not yet been undertaken.
The optically sectioning multiwell-plate reader described is being developed for FLIM and FRET assays and here we discuss its performance, including with respect to an exemplar application concerning the aggregation of HIV-1 Gag proteins in the late life cycle of the HIV-1 virion.[33
] HIV-1 Gag is the major structural protein within HIV-1 virions and is thought to interact with other viral proteins, the viral genome and with a large number of host cell factors, to orchestrate the formation of new virions at the plasma membrane of the target cell.[35
] The expression of HIV-1 Gag alone within living cells leads to the formation of virus-like particles (VLPs), which provides a convenient means to study this late stage of the HIV cycle. By tagging HIV-1 Gag with either cyan fluorescent protein (CFP) or yellow fluorescent protein (YFP) at its C terminus, we are able to read out oligomerisation using FLIM FRET since co-transfection of Gag–CFP and Gag–YFP results in the production of VLPs that undergo FRET from Gag–CFP to Gag–YFP due to the close packing of Gag proteins within newly formed VLPs. While FRET has previously been used to monitor clustering of Gag proteins in this context,[36
] we believe that this is a useful test of the utility of our new Nipkow FLIM multiwell-plate reader, noting that previous work was confined to confocal microscopy. The application of an automated FLIM plate reader could permit the screening of compounds designed to interfere with this important stage in the HIV life cycle.
For drug discovery, an automated FLIM multiwell-plate reader can be used to apply FRET to read out protein–protein interactions that can report on the impact of drug candidates on a cell-signalling network. While this is already a significant advance on the state of the art, such readouts would be more specific and informative if multiple components of cell-signalling networks could be probed simultaneously. Spectral ratiometric imaging has previously been used to simultaneously map two separate FRET readouts,[39
] but the spectral crosstalk remains a challenge when using this approach. To address this issue, we previously[41
] multiplexed a FLIM FRET readout of Ras-Raf protein binding, using TagRed/mPlum as the donor–acceptor pair, with a spectral ratiometric readout of a “Cameleon” FRET probe[3
] utilising CFP/YFP, to demonstrate simultaneous monitoring of two components of the epidermal growth factor (EGF) signal pathway in live COS-7 cells. We believe that this combination of fluorophores is widely applicable and offers the lowest degree of crosstalk for multiplexed FRET yet reported. Unlike ratiometric approaches to FRET, FLIM is able to make use of relatively dim acceptors such as mPlum—or even dark acceptors[42
]—since only measurement of the donor is required. Indeed, the use of dark acceptors is desirable for FLIM FRET and can enable a higher density of multiplexed readouts. This earlier work was limited to wide-field microscopy by the equipment then available to us. We have now demonstrated an optically sectioned multiplexed FLIM microscope based on the approach of our earlier work[41
] but using a Nipkow disc unit incorporating an electronically controlled dichroic changer and electronically controlled excitation and emission filter wheels. This enables interleaved FLIM acquisitions in two spectral channels, thus permitting the spatiotemporal mapping of two fluorescence lifetime readouts. For a first demonstration of this capability, we have simultaneously imaged the calcium transients induced in HEK293T cells following the addition of ionomycin using both a genetically expressed calcium sensor, Troponin TN-L15[43
] labelled with CFP/Citrine, and a calcium-sensitive dye, GFP-Certified FluoForte (Enzo Life Sciences Ltd, UK). In the future we hope to implement multiplexed FLIM readouts on our multiwell-plate reader.
There is increasing concern that biological processes observed in monolayers of cultured cells are not necessarily reproduced in live model organisms, or human patients. This is of crucial importance to drug discovery and testing, but practical constraints—particularly the desire for convenient readouts—have nevertheless led most assays to be undertaken in conditions that are far removed from their native or intended physiological contexts. This has been particularly true of optical readouts, which have been widely used in high-throughput screening, but now there is an increasing trend to translate assays from cell monolayers to “more realistic” 3D cell and tissue cultures, to transparent live organisms that are amenable to genetic manipulation such as drosophila and zebrafish, and to small mammals. Imaging plays an important role in assays for such disease models although, to date, most of the readouts have been concerned with morphology and ultimate phenotypes, which can be difficult to directly correlate with cellular responses. We aim to translate FLIM and FRET along the drug-discovery pipeline and image functional changes in signalling networks that can be directly correlated with cell-based assays using the same readouts.
To this end we are developing tomographic FLIM instrumentation based on the technique of optical projection tomography (OPT),[44
] which is analogous to X-ray computed tomography but uses visible radiation rather than X-rays and so requires the samples to be more or less transparent. This is usually achieved using a chemical clearing process that eliminates the scattering of visible radiation by refractive index matching throughout the sample using BABB (a 1:2 mixture of benzyl alcohol and benzyl benzoate) or other index-matching solutions. Our “tomoFLIM” approach utilises wide-field time-gated fluorescence imaging combined with a rotating sample stage. Reconstruction of 3D transmission and fluorescence images of transparent samples is realised using standard back-projection techniques to produce a series of time-gated images generated as a function of time delay after excitation. This can be processed to give 3D fluorescence intensity and lifetime distributions, which, in turn, can be used to map FRET in 3D throughout a sample. To date, we have demonstrated the ability to map FLIM in 3D in chemically cleared mouse embryos[45
] and have demonstrated 3D FLIM of FRET in transparent phantoms. The prospect of reading out FRET-based assays using tomoFLIM in disease models such as zebrafish offers many opportunities in drug discovery, including lead validation and toxicology studies, as well as for molecular cell biology in general, but there remain some significant hurdles. The need for chemical clearing precludes the use of live samples and, unfortunately, the chemical clearing process itself appears to degrade the fluorescence of genetically expressed fluorescent proteins.[46
] Very recently, however, we have succeeded in applying tomoFLIM to a fixed but not cleared zebrafish embryo labelled with enhanced green fluorescent protein (EGFP). This is an important step towards our ultimate goal of imaging live zebrafish.
Having identified drug candidates using cell-based assays, it is necessary to evaluate their performance, for example with respect to efficacy and toxicology, in animal models such as mice. Unfortunately such animals are far from transparent but the prospect of translating cell-based assays to animal models is stimulating increasing interest in addressing this challenge. Our tomoFLIM set-up can also be applied to imaging scattering samples but then the image reconstruction cannot be realised using back projection. Instead, it is necessary to take into account the multiple scattering of the detected photons and to use statistical diffuse light propagation algorithms to reconstruct the 3D fluorescence quantum efficiency and lifetime distributions based on the probable trajectories of the detected photons. Unlike OPT, which can deliver diffraction-limited images, diffuse fluorescence tomography (DFT) provides significantly decreased spatial resolution. Nevertheless, we have shown that DFT is able to reconstruct fluorescence lifetime distributions in scattering phantoms.[47
] If the technique can be extended to live samples, such as genetically manipulated mice, then it would provide a further opportunity to correlate FLIM-based readouts, for example of cell-signalling networks, across the drug-discovery pipeline. We note that tomographic FLIM in diffuse media has received considerable attention recently, particularly with respect to the development of appropriate reconstruction algorithms (see, e.g., ref. [48
]), but tomographic FLIM of FRET in a live subject is still yet to be reported.
For many investigations, it would be valuable to also obtain morphological information and to be able to image cell biological processes in vivo in live disease models, such as mice, with subcellular resolution. This can be realised using intravital microscopy, as is, for example, undertaken in neurological studies in live rodents,[49
] or by using endoscopy. Although we believe that FLIM endoscopy provides a promising route to study signalling processes and disease mechanisms in vivo, as well as providing a new molecular imaging modality for clinical diagnosis, we note that there have been few reports of FLIM endoscopes since the first demonstration of FD FLIM applied to ex vivo and in vivo tissue imaging.[50
] In the time domain, we have demonstrated wide-field time-gated FLIM in rigid[51
] and flexible[52
] endoscopes and note that, in general, the larger achievable fields of view and imaging speeds may make wide-field FLIM endoscopes more suitable for clinical diagnostic (screening) applications than laser scanning instruments, although the latter can provide optical sectioning and subcellular resolution to study signalling processes. For the latter application, we developed the first confocal laser scanning FLIM endomicroscope, which provides optically sectioned subcellular resolution that we applied to tissue autofluorescence and FRET in fixed cells.[53
] We are not aware of any reports of multiphoton FLIM endoscopy but we note that multiphoton FLIM has been extensively used for intravital imaging of skin (see, e.g., ref. [54
]) and has been applied to FRET studies of Alzheimer’s disease.[55
Our confocal FLIM endoscope is based on a commercially available endomicroscope[56
] (Cellvizio GI, Mauna Kea Technologies) to which we have added an ultrafast excitation laser and TCSPC detection to facilitate FLIM, thus realising acquisition times of a few seconds. This approach enables endoscopic confocal FLIM via a range of optical-fibre-bundle probes that either work in a “contact” imaging mode (i.e. working depth 0 μm) or use distal micro-optics to achieve optically sectioned imaging at a fixed working depth (e.g. from 20 to 80 μm). Subsequently a similar instrument, albeit limited to “contact imaging”, was demonstrated for FLIM of FRET in live cells immobilised in a gel-based matrix, and reporting FLIM acquisition times of 300 s.[57
] Our earlier system[53
] was assembled on a large optical table in a physics laboratory and we now describe a refined prototype intended for clinical imaging and animal experiments that is developed for integration into a trolley of area 70×100 cm2
. To illustrate its potential for in vivo FLIM FRET studies, we applied it to live cells expressing EGFP and mCherry, either separately or joined by a short peptide linker to provide a FRET sample. Unambiguous FLIM FRET readouts in these live cells were obtained with acquisition times of a few seconds. To provide a real-time FLIM preview mode to guide clinicians and experimenters, we are now working to implement the first in, first out (FIFO) mode of TCSPC acquisition, which will also permit acquisition of larger image data sets. We note that the “time-tagged” photon detection associated with FIFO TCSPC will potentially enable us to use image registration techniques if FLIM of samples moving within the total acquisition time is required. We are also working on FLIM montaging techniques to combine multiple fields of view in a single data set.