1. Introduction

1.1 Fluorescence lifetime imaging microscopy

The fluorescence (or luminescence) lifetime is the average time that a luminophore spends in the excited state. The sensitivity of the fluorescence (or luminescence) lifetime to environmental factors can be exploited to investigate the physico-chemical environment of the luminophore [1].

Fluorescence Lifetime Imaging Microscopy (FLIM) has been demonstrated to be a useful tool for biology, biophysics and diagnostics, but it requires know-how and costly equipment that are limiting its use as an analytical tool in the life sciences and its adoption for large-scale screening. FLIM can be used for the quantitative imaging of Förster Resonance Energy Transfer (FRET) between two dye molecules. Importantly, FRET imaging allows imaging of the biochemistry inside living cells [2,3].

However, even given its intrinsic value, the spreading of FLIM instrumentation in laboratories is limited by its cost and by the required know-how necessary for its maintenance and operation [4]. The fluorescence lifetime can be measured in the frequency-domain (FD) by exciting the fluorescent molecules with intensity-modulated light and detecting the phase delay and the demodulation between the fluorescence and the excitation.

Because the fluorescence lifetimes of the most used organic and genetically encoded luminophores range in the nanosecond region, pico-second or femto-second pulsed sources and repetition rates in the megahertz range are required for FLIM. Major advances in solid-state technologies [5] today enable the use of cost-effective laser- (LD) and light emitting diodes (LED), as directly electronically modulated light sources. The implementation of an all-solid-state system for wide-field detection of fluorescence lifetimes is an innovation that will drive the further adoption of FLIM in the life sciences.

Wide-field FLIM setups require the use of multi-channel plate (MCP) based image intensifiers [4]. Unfortunately, MCPs are comparatively expensive, prone to photo-damage due to overexposure and require elaborate electronics. Moreover, although the timing properties of MCPs are optimal for FLIM, their spatial resolution is relatively low. Furthermore, MCPs can inject a comparatively high noise level in the measurement [6]. For these reasons, a robust solid-state detector presents a desirable alternative to MCPs.

In the recent past, pioneering work [6–8] demonstrated the possibility of direct fluorescence lifetime sensing with a modified commercial CCD. Nanosecond lifetime sensing was realized by directly modulating the gain of the sensor [6–7]. Modulation frequencies of 100-500 KHz were used and a series of images was recorded at increasing delay. In refs [6–7] a maximum modulation frequency of about 10MHz was measured. However, in the FLIM experiments modulation frequencies of only ≤500KHz could be realized. This is sub-optimal for use with typical fluorophores used in life sciences. More recently, real-time lock-in imaging using a modified commercial CCD was demonstrated [8]. Two-phase images were acquired simultaneously at a modulation frequency of 16KHz, far too low for FLIM applications.

In the present article we report on the use of a CCD/CMOS hybrid lock-in imager for nanosecond lifetime imaging at a modulation frequency of 20MHz. The imager combines the possibility of lock-in imaging [8] with modulation frequencies suitable for lifetime sensing [6,7]. The imager was originally developed for full-field 3D vision in real time [9,10]. The sensor was developed for high speed lock-in imaging, therefore further modifications of the CCD were not required.

We demonstrate that all-solid-state lock-in imagers are a viable alternative to MCP detection in the near future.

1.2 Time of flight and the lock-in imager

The sensor of the lock-in imager presented here (SwissRanger SR-2 camera) is manufactured in 0.8μm combined CMOS/BCCD semiconductor technology [9]. This allows optimal CCD performance while utilizing the flexibility of the integration of CMOS active-pixel sensor (APS) readout architectures, which allows the individual addressing of pixels and its fast readout. The imager chip (see Fig. 1(a)) is composed of an array of 124 × 160 pixels with an area of approx 40μm × 55μm per pixel and with an optical fill factor of ~17%. Each lock-in pixel consists of several CCD gates on a silicon substrate; two independent charge-storage sites (see Fig. 1(b)) and the APS readout circuitry with addressing and source-follower amplifier. The driving potentials of the CCD gates are controlled at opposite phases with a modulation frequency of 20MHz. The square gate modulation signals are generated directly by the camera electronics, which also provides the reference signal for the light source. Due to the limited bandwidth of the sensor electronics, this results in a sine-like modulation of the gain of each storage area.

 

Fig. 1. The lock-in imager sensor. Panel A shows microphotography of the sensor. Each single pixel, arranged in a 124×160 array has a dimension of about 40μm × 55μm. Each pixel has two gates that are controlled with voltages in opposite phase (Panel B). Thus, the photoelectrons generated in the photosensitive area, will accumulate in the two storage areas according to the relative phase of the photon flux and the gate potentials.

Download Full Size | PDF

Controlling the gates in opposite phase allows the photo-generated electrons to be accumulated in the two different storage sites, depending on the phase instant at which each photon impinges on the sensor. Thus, the readout process of the sensor returns two images that are simultaneously acquired at opposite phases. The imager of the SR-2 camera is a front-illuminated CCD with a typical quantum efficiency of about 30%, 50% and 70% at 500nm, 600nm and 700nm, respectively.

This lock-in imager and its use for time-of-flight (TOF) detection has been described elsewhere in more detail [9]. The frequency-domain TOF technique relies on the measurement of the phase-delay of a reflected light signal by which the distance of the reflector can be determined. In its original implementation, an infrared LED array integrated within the camera housing is used as the actively modulated light source. For the application of FLIM, the LED array and its electronic driver were removed. This was mainly done in order to reduce the generation of heat inside the camera housing. Furthermore, a larger fan as used in PC cooling was added for additional cooling of the electronics.

2. Experimental setup

2.1 Microscopy

The SwissRanger SR-2 (SR-2, Centre Suisse d’Electronique et de Microtechnique SA, Neuchâtel, Switzerland) lock-in imager CCD was connected to the binocular port of a fully automated Axiovert200M (Carl Zeiss Jena GmbH, Jena, Germany) by a 0.4x C-mount adapter. A solid-state diode laser operating at 405 nm (Compass, Coherent Inc., Santa Clara CA, USA) was used for excitation of the turbo-sapphire (TS) green fluorescent protein [11]. The Compass laser can be modulated directly. The modulation signals were obtained from the TTL output of the camera. In order to obtain a 50% duty-cycle of the laser output, the offset voltage levels of the modulation signals were adjusted by the use of a bias-T (AVCOM DCP-20, Fotronic Corp. Melrose MA, USA) and a precision DC voltage generator (Voltcraft PPS-12008 by Conrad Electronic GmbH, Germany). An external delay line (Kentech Instruments Ltd., South Moreton, UK) was used for imaging with 8 phase-steps as described below (see section 2.2). The fluorescence was collected through an optical band-pass filter centered at 515±30nm and a long-pass 495nm dichroic mirror (AHF Analysentechnik AG, Tübingen, Germany). A 530–585nm band-pass emission filter, a 530–585nm long-pass filter and a 515nm dichroic mirror were used for the detection of Rhodamine 6G fluorescence. A microscope filter cube with a beam splitter instead of a dichroic mirror and emission/excitation filters was used for the collection of reflected 405nm light. Excitation at 470nm was achieved using a NSPB500S (Nichia Corp., Japan) LED in combination with a 470±20nm band-pass filter (AHF Analysentechnik AG). Excitation at 500nm was achieved using a NSPE500S (Nichia Corp., Japan) LED in combination with a 500±20nm band-pass filter (AHF Analysentechnik AG). The LED was driven using an adapted additional RF power amplifier PA-4 (IntraAction Corp., Bellwood IL, USA). The LED was positioned at the field aperture position of the microscope in order to achieve homogeneous illumination. The LD and the LED provided square-wave and sine-wave excitation light, respectively. The acquisition time for the SR-2 images presented here was 40ms. The acquisition and analysis of the data by the SR-2 was performed using the Matlab suite (Mathwork, Natick MA, USA). FLIM data analysis software can be downloaded from www.quantitative-microscopy.org/pub/sr2.html. No image enhancement was applied. Reference fluorescence lifetime measurements for verifying the results obtained with the SR-2 were carried out using a multi-channel-plate based FD-FLIM platform, described elsewhere [4].

2.2 Data acquisition and analysis

FD data analysis [1,12] requires the collection of images at different relative phases (Φ₀) in order to estimate the phase delay (Φ) and the luminescence signal demodulation (m). Upon each exposure, the SR-2 provides two images collected at opposite phases (S₀ and S_π). A subsequent π/2 shift of the internal phase delay allows the collection of two more images (S_π/2 and S_3π/2) at a relative phase equal to π/2 and 3π/2, respectively. These four images can be used to estimate the phase delay and the demodulation of the luminescence [9]:

where Φ’ and m’ are the instrumental phase delay and demodulation, which can be estimated by measuring reflected excitation light as zero lifetime reference or by using a well-characterized luminescent sample Fig. 2).

From the quantities calculated by Eq. (1), the sample lifetime can be determined [1]:

where ω indicates the circular modulation frequency of the excitation light.

Typically FD lifetime imaging measurements are carried out using 8 or even more phase angles. This approach results in higher lifetime accuracy and, by using appropriate acquisition sequences of the measurements at different phase angles, the effects of photo bleaching can be minimized [13]. Here, 8 phase angles were recorded to demonstrate the excellent phase-sensitive detection of the SR-2 device.

The internal delay line of the SR-2 currently only provides phase delays in π/2 steps. Therefore, after having acquired the four images as described above, we made use of an external delay line to inject an additional π/4 delay into the TTL SR-2 signal. Then, we acquired four more images (S_π/4, S_3π/4, S_5π/4, S_7π/4). The data was transformed to perform Fourier analysis [14]:

Computing the phase delay and the demodulation factor from these parameters is straightforward:

Again, Eq. (2) can be used together with Eq. (3) and Eq. (4) to retrieve lifetime estimations.

Dark images of the two storage areas were collected by closing the binocular-output of the optical path of the microscope and subtracted from the images of the sample. Dark and sample images were acquired with the same imaging parameters. No photo-bleaching correction was applied as the simultaneous acquisition of two images in opposite phases eliminates photo-bleaching artifacts.

2.3 Sample preparation

Recombinantly produced Turbo-Sapphire GFP was covalently conjugated to CnBr-activated Sepharose beads (Amersham Biosciences, Uppsala, Sweden). Beads (25 μl dry mass) were hydrated and activated with 1 mM HCl for 10 min. Beads were washed with PBS supplemented with 100 mM Bicine-HCl pH 8.0 for 5 min and subsequently incubated for 30 min with 100 μl 0.1 mg/ml recombinant TS-GFP in PBS supplemented with 100 mM Bicine-HCl pH 8.0 at room temperature and under continuous mixing by inversion to ensure homogeneous binding. After the coupling, the supernatant was removed and the beads were washed three times with 1 ml PBS supplemented with Tris-HCl pH 7.5 to quench remaining reactive groups. Spectrophotometric evaluation of the reactant supernatant confirmed quantitative coupling. A small aliquot of the resulting covalently coupled beads was mounted under a cover slip on a glass slide using Mowiol sealant. The mounted beads were allowed to solidify overnight at 4 degrees centigrade before being subject to imaging.

An excess of purified vector DNA was incubated with GelStar dye (Cambrex corp., East Rutherford NJ, USA). Solutions of recombinantly produced EGFP and GelStar/DNA complex were diluted in water to obtain equally fluorescent samples. Rhodamine 6G (R6G, Sigma-Aldrich, Deisenhofen, Germany) was diluted in distilled water at 10μM from a 0.1mM methanol stock solution.

3. Results and discussion

3.1 Lock-in imager response

The lock-in imager was tested by recording excitation light reflected from a foil positioned at the sample plane. Eight different phase images were recorded (see Fig. 2(a)). Four images were acquired using the internal camera phase delay settings of 0 and π/2 (black circles). Each internal delay setting resulted in two images with a phase difference of π. Four more phase images were recorded in a similar way after introducing an additional phase shift of π/4 using an external delay line (gray circles). The resulting curve shows the expected modulation and demonstrates that the SR-2 measures the actual phase of the impinging luminescence without interfering artifacts caused by the internal electronics.

The demodulation map (see Fig. 2(c)) is somewhat noisy but fairly homogeneous across the field of view. The maximum modulation depth that can be achieved using square wave excitation and lock-in detection is ~64%. This is due to the convolution of the fluorescence signal with the sinusoidally modulated gain of the camera and the subsequent integration over the exposure time. Therefore, even a fully modulated light will be detected with modulation depth lower than 100%. The measured value was equal to 50±3% (mean ± standard deviation over the same field of view). Therefore the demodulation contrast of the sensor in the UV-blue region is ~80% of the maximum achievable. This slightly reduces the sensitivity for lifetime detection. The phase response of the sensor is less homogeneous in particular at the right corners (white arrows) in Fig. 2(e). This effect can be corrected by applying a pixel-by-pixel calibration of the imager.

The same measurement was performed on a fluorescent plastic slide, now with the appropriate filter cube (see Fig. 2(b,d,f)). This allowed the characterization of the lock-in imager response in the spectral region of the Turbo-Sapphire GFP emission. The homogeneity of the response was found to be equal to that in the UV/Blue region. The demodulation map is less noisy due to the higher quantum yield of the sensor in this spectral window. Interestingly, a higher demodulation contrast can be achieved at this longer wavelength. The measured demodulation of the fluorescence signal amounted to 55±1%.

 

Fig. 2. Response of the lock-in imager. Panel A and B show the average intensity at each detected phase. The grey curve represents the average intensity (I₀) and the circles are the experimental points connected by a spline curve (dashed). Gray circles correspond to images acquired by the injection of the additional delay by the external delay unit. The left side panels (A, C and E) represent measurements of a reflective foil, while B, D and F refer to a fluorescent slide acquisition. C and D depict the demodulation of the signal measured over the entire illuminated field of view; E and F show the correspondent phases. The latter are inhomogeneous over the field of view (arrows). Considering the lifetime of the samples, i.e. 0 ns and 4.8 ns for the reflective foil and fluorescent slide, respectively, the initial phase of the detection is shown to be constant, while the demodulations suffer from a color-effect of the lock-in imager.

Download Full Size | PDF

As the lifetime of the plastic slide is 4.8ns and homogeneous, the demodulation at zero lifetime was calculated to be 65±1%. This value is equal to the theoretical maximum that can be achieved. The different demodulation factors at different wavelengths are due to the front illumination of the SR-2; light of different wavelengths generates photoelectrons at different depths in the substrate. Consequently, the photoelectrons sense different well potentials that correspond to different demodulation contrasts. This effect requires calibration of the lock-in imager for the spectral region of the detected luminescence.

The relative sensitivities of the two storage areas were almost identical; the average value was found to be 99.02% and the standard deviation amounted to 0.02%. This high homogeneity affords calculation of the lifetime without a correction for pixel-to-pixel sensitivity fluctuations.

3.2 Fluorescence lifetime sensing

To evaluate the fluorescence lifetime imaging performance of the SR-2, we calibrated the lock-in imager with the fluorescent slide as described above. This reference sample was previously characterized with the MCP-based FLIM. Its measured lifetime of 4.8 ns was used to calibrate the SR-2 camera. Subsequently, the same slide was imaged to confirm the successful pixel-by-pixel calibration. Figure 3(a) shows the phase-lifetime map and distribution (4.8±0.4 ns) acquired at 470nm excitation (LED). Further measurements of the lifetimes of EGFP and DNA-bound GelStar solutions demonstrate the lifetime contrast that is achieved by the camera. The average phase lifetimes were 2.6±0.4 ns and 6.6±0.7 ns, respectively. Furthermore, a Turbo-Sapphire GFP bead was imaged by exciting the sample with the 405nm laser. Figure 3(b) shows the images and distributions of the phase and demodulation lifetimes, corresponding to 2.67±0.09 ns and 3.7±0.2 ns, respectively.

 

Fig. 3. Fluorescence lifetime sensing. The lock-in imager distinguishes compounds with different fluorescent lifetimes. Panel A shows the phase-lifetime maps and distributions of: EGFP in solution (gray line), a fluorescent slide (dashed curve) and DNA-bound GelStar (black solid line) in solution. The lifetimes were 2.6±0.4 ns, 4.8±0.4 ns and 6.6±0.7 ns, respectively. Both the phase- (panel B, gray curve) and demodulation- (black line) lifetimes can be measured at a modulation frequency of 20MHz. A Turbo-Sapphire GFP bead showed values of 2.67±0.09 ns and 3.7±0.2 ns, respectively. Panel B inset (R6G) shows the phase (4.3±0.2) and modulation (4.3±0.4) lifetime of the mono-exponential decaying fluorophore standard Rhodamine 6G.

Download Full Size | PDF

Both the modulation and the phase lifetime are comparable with measurements carried out with the MCP-based FD-FLIM setup operating at the same modulation frequency. Using this setup, lifetimes of 2.6±0.2 ns and 3.2±0.4 ns were found respectively. The higher noise content of the latter is explained by the poorer performance of the modulation lifetime estimator at the modulation frequency used. The inset in Fig. 3(b) shows the lifetime distribution of a solution of R6G. Here, a 500nm LED was used for excitation and fluorescence was detected at 530–585nm. R6G is known to exhibit a monoexponential decay, therefore the modulation and phase lifetimes are similar (4.3ns). The above test measurements clearly demonstrate the feasibility of nanosecond fluorescence lifetime imaging by the SR-2.

The SR-2 camera currently operates at a modulation frequency of 20MHz. The modulation frequency has a profound effect on the efficiency of lifetime determination. If, for instance, sinusoidally modulated illumination is considered, the optimal frequency for the phase based detection of a 2.5 ns lifetime is ~40MHz. The optimal frequency for the corresponding demodulation based lifetime determination is ~80MHz. The next generation SR2 chips will operate at (much) higher modulation frequencies that are more appropriate for the imaging of ns demodulation lifetimes. A thorough description of the photoeconomy of the FLIM techniques can be found elsewhere [15,16].

It was shown that a directly-modulated CCD camera can be operated at a modulation frequency of 500kHz. At this frequency, it is difficult to obtain information from the luminescence demodulation. More than 2000 times more photons are required to obtain the same signal-to-noise ratio with respect to measurements carried out at 40MHz. Directly-modulated CCDs could possibly work at 10MHz. However, at this frequency, phase detection still requires ~6 fold more photons than at 40MHz. This frequency is still too low for the accurate determination of the demodulation lifetime. Compared to the optimal frequencies (40MHz, and 80MHz), the SR-2 currently operates at a frequency which requires only ~2, and ~15 fold higher counts, respectively. The technology employed in the SR-2 can be operated at higher frequencies. In fact, the same technology has been operated at a modulation frequency of 50MHz and different electronics could provide modulation frequency up to 100MHz [10]. Nevertheless, the presented results already demonstrate a three order of magnitude gain in photoeconomy compared to the reported sensitivity of directly-modulated CCDs.

The major limitations of the camera for FLIM applications are the lack of active cooling of the sensor and its currently suboptimal optical sensitivity for fluorescence microscopy. Moreover, the lack of cooling causes a relatively high dark current, limiting the signal-to-noise ratio of the FLIM detection. Active cooling of the sensor will reduce the dark current and allow longer integration times. The use of a micro-lens array can to a large extent compensate for the poor fill factor of the array (~300% gain). In addition, improved coupling optics can increase the sensitivity by another 50%. Improved versions of the camera and imager will become available in the near future.

4. Summary

We describe the application of a CCD/CMOS lock-in imager (CSEM Swissranger SR-2) for fluorescence lifetime sensing. The imager can operate at modulation frequencies as high as 20 MHz, 40 times higher than current directly-modulated CCDs. This makes the SR-2 about 3 orders of magnitude more sensitive for nanosecond lifetime imaging than current directly-modulated CCDs. The SR-2 is the first lock-in imager to operate in a frequency range that is optimal for sensing lifetimes in the nanosecond region. The SR-2 camera is well capable of nanosecond fluorescence lifetime imaging, even though it was not designed for this purpose. The results of lifetime measurements on reference specimens are in excellent agreement with the results obtained using a standard frequency-domain lifetime acquisition system.

At present the SR-2 camera presents limited sensitivity that should be optimized for Fluorescence lifetime imaging of living cells. However, the sensitivity of the current SR-2 camera can be easily improved by an order of magnitude. Based on our results we expect that a modified imager will be capable of imaging typical biological specimens with integration times of 100ms or longer.

The technology currently employed in the SR-2 can be used to produce imagers that operate at modulation frequencies as high as 50-100 MHz and speeds of up to 30-50 frames per second. The performance of the imager can be further improved by the production of a back-illuminated chip.

The use of lock-in imagers like the SR-2 may result in compact, cost-effective and user-friendly FLIM systems. This development could lead to a larger penetration of FLIM as a routine analytical tool in the fields of drug screening, diagnostics and proteomics, where the necessity to compromise between cost, speed, robustness and quantification is more challenging.