2.1.

Swiss Ranger-2 Camera and Lock-In Imaging

The lock-in imager is the SwissRanger SR-2 camera whose sensor has $124\times 160$ pixels with an optical fill factor of $\sim 17\%$ and is manufactured in $0.8\text{‐}\mathrm{\mu }\mathrm{m}$ combined CMOS/buried-channel charge-coupled device (BCCD) semiconductor technology.¹⁰ Each lock-in pixel has two independent charge-storage sites (Fig. 1 ). The front-illuminated CCD has a quantum efficiency of $\sim 30$ , 50, and 70% at 500, 600, and $700\mathrm{nm}$ , respectively.

Fig. 1

Pixel architecture, mode of operation, and camera design of the SR-2 imager. (a) Each lock-in pixel has a photosensitive region that converts photons into photoelectrons. (c) Electrons are integrated in two different storage areas, depending on the phase of the signal applied to the integration gates. Here, the integration is performed at opposite phases. This allows the cross-correlation and read-out (at the sensing-nodes) of two images, acquired simultaneously, and at opposite phases. (b) All the electronics for FLIM are embedded in a compact design.

The CCD gate potentials are modulated in opposite phase configuration at a frequency of $20\mathrm{MHz}$ . This allows the simultaneous phase-dependent accumulation of photogenerated electron hole pairs in the two separate storage sites (Fig. 1). Thus, the camera records two images in parallel at opposite phases. The electronics of the SR-2 generate both the modulation signals for the gate control and for the external light source. The SR-2 is controlled by a computer via a USB connection. In its original implementation, an infrared LED array integrated within the camera housing is used as the modulated light source. For the application to FLIM, the LED array and its electronic drivers were removed to reduce the generation of heat inside the camera housing. The original camera computes the distance map on-chip. For our purposes, the SR-2 firmware was modified to allow the read-out of the two phase-dependent images. However, the camera firmware can be modified to directly provide lifetime images. We previously operated the SR-2 with an external delay unit for the collection of four or eight phase images.¹¹ In the present work, the SR-2 was operated without the external delay unit, and only two and four phase images were recorded.

2.2.

Microscopy

The SR-2 was connected to the binocular port of a fully automated Axiovert200M (Carl Zeiss Jena GmbH, Jena, Germany) microscope by a $0.4\times$ C-mount adapter. This microscope is part of an in-house developed frequency-domain MCP-based FLIM system, described elsewhere.¹³ A 405-nm solid-state Compass laser (Coherent Incorporated, Santa Clara, California) was used for excitation of the turbo-sapphire (TS) green fluorescent protein (GFP). LED excitation was achieved using a NSPB500S LED (Nichia Corporation, Japan) with emission peaking around $470\mathrm{nm}$ in combination with a $470\pm 20\text{‐}\mathrm{nm}$ bandpass filter (AHF Analysentechnik AG, Tübingen, Germany). The driving signal for the modulation of the light sources was obtained directly from the lock-in imager. Further details of the system can be found elsewhere.¹¹ The fluorescence was observed through a $515\pm 30\text{‐}\mathrm{nm}$ optical bandpass filter and a 495-nm dichroic mirror (AHF Analysentechnik AG). Cy3 images were collected using a filter cube fitted with a $560\pm 20\text{‐}\mathrm{nm}$ excitation filter, a $595\text{‐}\mathrm{nm}$ long-pass dichroic mirror, and a $630\pm 30\text{‐}\mathrm{nm}$ emission filter. The MCP-FD-FLIM was controlled by in-house developed software programmed in the DaVis imaging suite (Lavision GmbH, Göttingen, DL). The acquisition and analysis of the data by the SR-2 was controlled by Matlab (Mathworks, Natick, Massachusetts). Both the SR-2- and MCP-based FLIM systems were calibrated on a pixel-by-pixel basis by imaging a fluorescent slide (Chroma Corporation) with a lifetime of $4.81\mathrm{ns}$ . The emission of the slide overlaps with the emission of the fluorophores used in this study.

2.3.

Data Analysis

The SR-2 enables the simultaneous acquisition of two images at relative phases of 0 and $180\mathrm{deg}$ $(I_{0°},I_{180°})$ , respectively. Subsequently, two additional images at relative phases of 90 and $270\mathrm{deg}$ $(I_{90°},I_{270°})$ , respectively, can be acquired by using the SR-2 internal delay line. The same acquisition protocol was used for the MCP detection, but in this case all the images $(I_{0°},I_{90°},I_{180°},I_{270°})$ were acquired sequentially. The four-phase data stack was processed by the analysis of the zero and first Fourier coefficient¹⁴ to estimate the phase delay $\left(\varphi \right)$ and the demodulation $\left(m\right)$ of the fluorescence emission:

The two-phase acquisition data were analyzed by the rapid lifetime determination algorithm (RLD).¹² This algorithm allows calculation of the demodulation factor:

All data analysis was carried out with the Matlab toolbox “ImFluo,” whose lifetime functions are available from http://www.quantitative-microscopy.org/pub/sr2.html. Dark images were acquired prior to lifetime detection. This background was subtracted from each following acquisition.

2.4.

Sample Preparation

The fluorescent beads were prepared by covalent conjugation of recombinantly produced TS-GFP to CnBr-activated Sepharose beads (Amersham Biosciences Europe GmbH, Freiburg, Germany) as previously described.¹¹ A FRET experiment was performed on TS-GFP beads conjugated with Cy3 monofunctional reactive dye (Amersham Bioscience Europe GmbH) according to a previously published protocol.¹³ These beads exhibit FRET efficiencies of about 30% (as detected by the MCP-based FD-FLIM). CHO cells were transfected with a pcDNA3.1 expression vector encoding the TS-GFP using the Effectene transfection reagent according to the protocol supplied by the manufacturer (Qiagen, Hilden, Germany). EGFP and TS-GFP were purified from bacteria using the N-terminal hexahisitidine tag encoded by the vector, by immobilized metal affinity chromatography on Talon resin (Promega) according to the protocol supplied by the manufacturer. Rhodamine 6G (R6G) was from Sigma-Aldrich (Deisenhofen, Germany).

3.1.

Time-of-Flight Detection and Fluorescence Lifetime Sensing

The SwissRanger SR-2 camera used in this study was originally developed and optimized for the recording of distance maps. This sensor estimates distances by illuminating a scene with intensity-modulated light and measuring the (phase) delay of the reflected photons (time-of-flight detection). With increasing object distance, longer delays will be detected in its reflected light [Fig. 2a ]. At every pixel, the lock-in imager computes the average intensity of the scene [Fig. 2b], the relative phase delays [Fig. 2c] of the reflected light field, and a distance map [Fig. 2d]. The camera is able to measure picosecond time lags. Therefore, the timing resolution is sufficient to measure the nanosecond fluorescence lifetime of biologically relevant fluorochromes such as fluorescent proteins.

Fig. 2

Time-of-flight imaging: (a) and (d) The original application of the lock-in imager, time-of-flight detection for real-time 3-D vision is shown. (b) A lock-in imager produces an intensity map of the imaged scene and (c) the phase delays of the reflected illumination light. From these values, (d) the distances of the objects in the 3-D scene are calculated. (a) Increasing distance gives rise to a longer delay of the reflected light.

Fundamentally, FLIM differs from time-of-flight imaging only in the measurement of fluorescence emission instead of reflected light. The emitted fluorescence exhibits a lifetime-dependent phase delay and demodulation. With increasing fluorochrome lifetime, the relative emission (phase) delay and the attenuation of its modulation depth increase^{1, 6, 14} [Fig. 2a].

3.2.

Comparison with State of the Art Technology

The 20-MHz modulation frequency of the SR-2 is sufficiently high to obtain reliable lifetime images of fluorochromes with nanosecond lifetimes.^{11, 15} To investigate the advantages of the lock-in imager further, we compared images of the same sample, imaged by the SR-2 and by the conventional multichannel-plate-based detection system (Fig. 3 ). The test sample, a turbo-sapphire green fluorescent protein (TS-GFP) conjugated bead, was illuminated by a 405-nm laser diode and images were acquired with 20-ms integration time. The lifetimes determined from the phase delay and demodulation of the fluorescence signal were similar for both the MCP and the lock-in imager. The slight systematic mismatch between the MCP and SR-2 estimation can be explained by the initial phase bias dependence that can occur as a result of aliasing when less than five phase steps are imaged.¹⁶

Fig. 3

Comparison between lifetime detection using an MCP-based imager (left column) and the novel CCD/CMOS lock-in imager (right column). A TS-GFP-conjugated bead was imaged by both systems (first row). The phase (second row) and modulation (third row) lifetimes (see also distributions) are quantitatively detected by both systems. Phase lifetimes: $2.67\pm 0.09\mathrm{ns}$ for the SR-2 and $2.6\pm 0.2\mathrm{ns}$ for the MCP. Modulation lifetimes: $3.7\pm 0.2\mathrm{ns}$ for the SR-2 and $3.2\pm 0.4\mathrm{ns}$ for the MCP; the presence of an edge artifact in the SR-2 image is caused by the dark current of the SR-2. The fluorescence intensities show the typical “chicken-wire” artifact that can be observed with the multichannel plate (first row, left panel).

A major advantage of this lock-in imager-based FLIM is the possibility to capture two modulated images at opposite phases simultaneously (see Secs. 2.1, 2.3). In contrast, the MCP method requires the sequential acquisition of images at different relative phases. Figure 3 shows images derived from four sequential intensity images by the MCP. The SR-2 acquires these images with only two exposures. Furthermore, the MCP exhibits artifacts in the detected intensities due to the fiber coupling between the MCP and the CCD that are not present in the image obtained by the SR-2.

3.3.

Application to Cellular and Förster Resonance Energy Transfer Imaging, and High-Throughput High-Content Screening

The feasibility of lifetime imaging on a biologically relevant sample using the SR-2 technology was investigated by imaging CHO cells transiently expressing TS-GFP [Fig. 4a ]. The sample was illuminated by a modulated 405-nm laser diode, the exposure time was $40\mathrm{ms}$ , and four phases were acquired during two sequential exposures of the SR-2. The TS-GFP phase-lifetime estimation was found to be $2.0\pm 0.3\mathrm{ns}$ . The lifetime map is somewhat noisy, but it clearly demonstrates the feasibility of all solid-state FLIM on single cells. Similarly, a TS-GFP fluorescent bead labeled with the Cy3 was imaged [Fig 4b]. The Cy3 fluorophore was photobleached by prolonged exposure with the HbO lamp and a partially closed aperture field iris of the microscope (see circle and arrow). The decreased Cy3 fluorescence emission correlates to an increase of the TS-GFP emission by the inhibition of FRET through the local depletion of the acceptor chromophore. Accordingly, the (phase) lifetime image and corresponding histogram revealed a lifetime reduction of TS-GFP of $\sim 28\%$ due to FRET with the Cy3 fluorophore.

Fig. 4

Cellular and FRET imaging: (a) Already the present prototype can retrieve correct lifetimes from biologically relevant samples. CHO cells were transfected with TS-GFP, whose phase lifetime estimation was measured to be $2.0\pm 0.3\mathrm{ns}$ . The scale bar indicates $10\mathrm{\mu }\mathrm{m}$ . (b) Donor (TS-GFP) and acceptor (Cy3) fluorescence images of a TS-GFP/Cy3 bead acquired by the SR-2. The Cy3 fluorophore was photobleached on the area marked by the circle and the arrows. In this region, the SR-2 measured a (phase) lifetime of $2.5\pm 0.4\mathrm{ns}$ higher (see histogram and arrow) than the lifetime of the TS-GFP in the regions where the acceptor was not photobleached $(1.8\pm 0.5)$ . This corresponds to a FRET efficiency of $\sim 28\%$ . The scale bar indicates $20\mathrm{\mu }\mathrm{m}$ .

Subsequently, we integrated the SR-2 in a screening platform. The microscope employed here is capable of operating in an unsupervised fashion as required for high-throughput screening (HTS). This automated MCP-based FLIM system (described elsewhere) is capable of acquiring and analyzing $\sim 100,000$ FLIM images per day. Figure 5 shows a (phase) lifetime image of an array of a $3\times 3$ group of wells in a 1536 multiwell plate. The wells were filled with enhanced green fluorescent protein (EGFP, bottom row), rhodamine-6G (R6G, top row), and a mixture of EGFP and rhodamine-6G (middle row). Here, the specimen was excited using a modulated 470-nm LED light source, and four phase measurements were acquired during two sequential acquisitions with an exposure time of $20\mathrm{ms}$ each. The lifetime images demonstrate that the SR-2 produces accurate and reproducible fluorescence lifetime contrast. The coefficients of variation ranged from 2 to 5%, and the results are in good agreement with those using MCP-based lifetime imaging.

Fig. 5

Application of cost-effective FLIM instrumentation to high-throughput screening. A $3\times 3$ section of a 1536-well plate, loaded with different concentrations of purified EGFP protein, rhodamine6G, and mixtures of both, is shown. The phase lifetime estimation histograms of the samples are given at the respective position along the rows and columns. The histograms at the top show the contrast between different lifetimes. The histograms on the side demonstrate the reproducibility of the measurements for identical wells.

The integration of the SR-2 technology into an automated microscopy platform results in a valuable tool for high-content high-throughput screening.

3.4.

Video-Rate Single-Shot Fluorescence Lifetime Imaging Microscopy

The simultaneous acquisition of two images at opposite phases by the SR-2 is a perfect match for the rapid lifetime determination algorithm.¹² The RLD algorithm is a robust and fast method for calculating fluorescence lifetimes that requires only two images recorded at opposite phases. Importantly, since both phase images are acquired simultaneously, the lifetimes recorded with the SR-2 are not affected by motion artifacts or photobleaching.

In contrast, MCP-based FLIM requires the sequential acquisition of images at different phases. Therefore, samples exhibiting movements that are faster than the acquisition time will suffer from motion-induced lifetime artifacts. In addition, the sequential recording of the phase images makes the MCP-based FLIM approach sensitive to photobleaching, in particular when only two phase images are recorded as required for the RLD. At present, it is impossible to correct for bleaching artifacts when only two images are required.

The SR-2 technique does not suffer from these drawbacks because of its parallel acquisition. Therefore, the SR-2 allows optimal use of the RLD algorithm. The combination allows a direct visualization of lifetime images even at video-rate imaging. Figure 6 shows images of TS fluorescent beads acquired at a frame rate of $\sim 24\mathrm{Hz}$ . The exposure time was $20\mathrm{ms}$ and the 405-nm laser diode was used for the excitation. Every second frame is shown. The computer-assisted microscope stage translated the sample along one axis at about $250\mathrm{\mu }\mathrm{m}∕\mathrm{s}$ (Fig. 6, middle column) or $\sim 500\mathrm{\mu }\mathrm{m}∕\mathrm{s}$ (right column). The imaging of a stationary bead (left column) is shown as a control. This example clearly shows that the lifetime images are unaffected by sample motion, and that this technology is well suited for FLIM of fast processes. Another interesting application is the detection of rare events. Here, a sample is continuously scanned, searching for a single cell that is undergoing a specific event, e.g., a protein-protein interaction that occurs at a specific point in the cell cycle. The latter application would require the on-line calculation (and displaying) of the fluorescence lifetime images. This could be easily realized using the RLD algorithm in combination with the on-chip computational capabilities of the camera.

Fig. 6

FLIM at video rate $(\sim 24\mathrm{Hz})$ : turbo-sapphire beads are imaged with the rapid lifetime determination algorithm, while the microscope stage translates the sample at different indicated speeds (0, 250, and $500\mathrm{\mu }\mathrm{m}∕\mathrm{s}$ from left to right columns) in the downward direction. Three frames out of six are shown (upper panel) and overlaid (bottom panel) to show the sample trajectory.

3.5.

Photon Economy

The photon economy of different FLIM techniques can be expressed by a figure of merit (F). Here, we employ a figure of merit defined as the ratio of the relative error in the lifetime determination and the square root of the number of detected photons.^{15, 17} When pulsed excitation is used, time-correlated single photon counting and lock-in detection can achieve an F value of 1. This is the lower limit; higher values imply a decreased performance. In frequency domain FLIM, the figure of merit strongly depends on the ratio of the fluorescence lifetime and the modulation frequency of the excitation light. The optimum modulation frequencies for fluorescence lifetime detection are $\sim 0.11$ and $\sim 0.20{\tau }^{-1}$ for the phase and modulation lifetime estimation,¹⁵ respectively, which corresponds to $\sim 40$ and $\sim 80\mathrm{MHz}$ for a fluorescence lifetime of $2.5\mathrm{ns}$ , typical for fluorescent proteins. The operation of the SR-2 at $20\mathrm{MHz}$ is therefore sufficient for lifetime sensing of nanosecond fluorophores, although the quality of the modulation lifetime estimation is not optimal. However, the technology employed in the SR-2 has already been tested at modulation frequencies of about $50\mathrm{MHz}$ , and different electronics could provide a modulation frequency of up to $100\mathrm{MHz}$ .¹⁸ Operating the detector at such frequencies would further improve the sensitivity, in particular for short lifetimes. Interestingly, with the combination of the rapid lifetime determination algorithm and the parallel acquisition of the SR-2, it is possible to realize a very high photon economy. No photons are lost, as is the case for the sequential acquisition of the different phase-dependent images. The RLD photon economy was not previously characterized. We estimated by Monte-Carlo simulations F values of $\sim 2.4$ and $\sim 1$ when using a sinusoidal wave or pulsed excitation, respectively. The use of the RLD algorithm in combination with the parallel acquisition of the camera therefore requires the collection of $7\text{‐}$ to 8-fold less photons compared with the standard multiple-phase and sequential acquisition.