Fluorescence correlation spectroscopy (FCS) is carried out with an electron multiplying CCD (EMCCD). This new strategy is compared to standard detection by an avalanche photo diode showing good agreement with respect to the resulting autocorrelation curves. Applying different readout modes, a time resolution of 20 µs can be achieved, which is sufficient to resolve the diffusion of free dye in solution. The advantages of implementing EMCCD cameras in wide-field ultra low light imaging, as well as in multi-spot confocal laser scanning microscopy, can consequently also be exploited for spatially resolved FCS. First proof-of-principle FCS measurements with two excitation volumes demonstrate the advantage of the flexible CCD area detection.
© 2006 Optical Society of America
Fluorescence correlation spectroscopy [1, 2] is an ultra sensitive technique to study small ensembles of fluorescent molecules in transparent fluid media. By evaluating the correlated fluctuations of their emitted photons when passing through an open laser-induced focal volume, information about the molecules and their interaction with each other and the solvent can be extracted without influencing the system. The most prominent example is pure Brownian diffusion, which leads to characteristic particle number, hence fluorescence intensity, fluctuations. Simple autocorrelation analysis of the signal yields exact concentrations and diffusion constants. Other processes inducing intensity fluctuations such as photophysical transitions  or protonation-deprotonation reactions  are also readily accessible in a standard FCS setup. An accurate measure of inter- and intramolecular dynamics is attainable by cross-correlating the concomitant fluctuations of two or more spectrally distinct fluorophores . In a standard FCS setup, all this information is collected from one specific focus position inside the sample. For the investigation of cellular mechanisms however, there is a great interest in large scale translocation and spatial dynamics which cannot be assessed directly. Therefore, parallel multi-spot excitation and detection strategies are needed. They also enable faster data acquisition which is advantageous for high throughput screening applications.
Multiplexed confocal excitation and detection strategies have been developed and commercialized in the last years for laser scanning microscopy (LSM). The excitation is, for example, realized by Nipkow spinning disk, linear multi-beam scanning or line-scanning strategies. Parallel detection is mostly performed by an EMCCD-camera. These systems are (by a factor given approximately by the number of beams) faster than traditional single-beam scanning devices, which is a crucial step forward in confocal live cell imaging. On the other hand, working at comparable standard frame rates, the laser power can be reduced by that factor compared to single-beam experiments, preventing photobleaching of the dyes or more serious phototoxic reactions in the sample. Additionally, back-illuminated EMCCD-cameras exhibit quantum efficiencies of 90% almost over the whole visible spectral range compared to less than 50% for photo multiplier tubes (PMT) used in single-beam scanning devices. Furthermore, there are no losses due to descanning the emission light since the EMCCD area detector can be directly placed in the image plane of the microscope tube lens.
In case of FCS, multi-spot measurements have been limited to special fixed excitation and detection setups. Two-beam cross-correlation was introduced by Brinkmeier et al.  for precise flow measurements in microstructured channels. Blom et al.  established four-spot FCS autocorrelation measurements by designing a diffractive optical element for excitation and using four optical fibers and individual avalanche photodiodes (APD) for detection. A complementary metal oxide semiconductor 2×2 array detector was presented by Gösch et al.  and compared to standard APD detection. However, a flexible multi-spot strategy has not been developed so far. Since multiplexed confocal excitation is not the limiting factor, the main problem seems to be the different detector needs for FCS and LSM. Whereas for LSM, detectors with linear gain and a large dynamic range are needed (PMTs operated in current mode or nowadays EMCCD-cameras), FCS is usually performed with single-photon counting devices such as APDs or PMTs operated in counting mode. One or two-dimensional photon counting APD or PMT arrays that qualify for FCS are so far not commercially available, although there is a large interest in developing such devices. Photon counting is the best recording method yielding a shot noise-limited signal. For single molecule techniques such as FCS, however, the essential requirement is single-photon sensitivity rather than photon counting. On-chip electron multiplication before digitalization suppresses the CCD readout noise to effectively less than 1 electron per pixel. The stochastic noise introduced by the multiplication process, the so-called excess noise, is lower than for intensified CCDs and together with the high quantum efficiency, EMCCD cameras should be applicable for FCS measurements. Clearly, a sequentially read-out CCD area detector will not reach the time resolution of a single photon counting point detector (e.g. dead time of an APD is 70 ns). For diffusion analysis however, µs-resolution is sufficient. In this work, we compare APD and EMCCD-based detection for different fluorescent molecules in solution.
The FCS setup (Fig. 1) is based upon an inverted microscope (IX71, Olympus, Germany) with a 60x water immersion objective, a 488 nm laser line (Stabilite, Spectra-Physics, Darmstadt, Germany), a dichroic mirror (LP 505) and a bandpass filter (HQ 535/70M, AHF, Tübingen, Germany). The laser beam is expanded two times and attenuated by neutral density filters (New Focus, San Jose, California) to 5 µW before entering the objective. Nanomolar concentrated sample solutions (200 µl) are pipetted in chambers (Nunc, Wiesbaden, Germany) to prevent evaporation. The detection method is chosen by means of the microscope switch for left or right side port. The output efficiency differs by less than 5%. “Left” means standard APD-detection (SPCM-AQR-13, PerkinElmer Optoelectronics, Fremont, California) with a multimode optical fiber of core diameter 50 µm (Thorlabs Europe, Karlsfeld, Germany), and “right” means EMCCD detection (iXon DV860, Andor Technology, Belfast, Northern Ireland). Optical fiber and EMCCD camera are mounted on µm-translational stages (Owis, Staufen, Germany, and Standa, Vilnius, Lithuania, respectively). The APD signal is fed into a digital multiple-tau correlator (ALV, Langen, Germany), displaying the instant count rate and correlation function. The CCD is operated by a home-written program based on the LabView (National Instruments, Austin, Texas) driver supplied by Andor, and image sequences are evaluated offline using Matlab (MathWorks, Natick, Massachusetts). Correlation curves are fit with Origin (OriginLab, Northhampton, Massachusetts) based on a Levenberg-Marquardt algorithm.
3. Kinetic mode
The camera is operated in the standard frame transfer mode (128×128 pixel imaging area plus same sized optically shielded storage area below, to which the image charges are shifted for later readout), enabling a shutterless operation. The shortest possible exposure time for a full frame acquisition at the fastest readout frequency (10 MHz) is 2 ms. Faster readout is possible, by only using a subregion of the CCD, for example, 5×5 pixels allow exposure times down to 0.26 ms. The fastest recommended vertical shift speed setting of 0.3 µs per line is chosen, yielding a frame transfer time of 38.4 µs. The thermoelectric cooling is set to -60° C and a full electron multiplying gain (software setting 255) is applied. The actual electron multiplication factor was determined to be 450 by comparing the zero gain and the full gain signal. The so called baseline clamp option is very important to assure a stable electronic offset for analog to digital conversion. First measurements were performed with a solution of CdSe/ZnS quantum dots (Evident Technologies, Troy, New York). A stack of 100 000 images (30 s) was recorded, together with a second dark measurement. The mean dark signal was subtracted pixel-wise from each image. Fig. 2 shows the integrated images over time for these two measurements (different lateral CCD positions relative to the emission spot maximum). The signal from the marked regions on the CCD is summed for each image in the stack, yielding signal-traces similar to the count rate trace of an APD detector (Fig. 2). The autocorrelation functions G(τ) of these traces are shown in Fig. 3, together with the curve for APD detection (measurement time was 30 s). Choosing differently sized regions of the CCD corresponds to changing the pinhole size, which can be performed conveniently by software.
Data were fit with the standard model function (Eq. 1) for free diffusion through a focal volume Veff with a three-dimensional Gaussian emission light intensity distribution . The fit parameters are the mean particle number N, the characteristic diffusion time τ D and the structure parameter SP=z0/r0, whereas z 0 and r 0 define the axial and lateral 1/e 2 radii of Veff . The diffusion constant D can be computed by D= /4τ D after the setup has been calibrated. This is usually done with a standard dye such as Alexa Fluor 488 (Molecular Probes Europe BV, Leiden, The Netherlands) with D=316 µm2/s, yielding r 0=0.27 µm, z 0=1.4 µm and Veff =π 3/2 z 0=0.6 fl.
The extracted fit parameters for the measurements with CCD detection are given in Fig. 3 for comparison with standard APD detection. Larger pinholes yield bigger focal volumes, hence increased particle numbers and diffusion times. The rectangular aperture, defined by 2×2 CCD pixels (48 µm×48 µm), yields a slightly larger focal volume compared to the one obtained by the circular fiber aperture of 50 µm diameter. The structure parameter only shows small changes for the different pinhole sizes. A significant increase of SP for larger pinholes due to the less axial confinement cannot be observed. The value of SP cannot strictly be related to the actual shape of the focal volume in these measurements since the autocorrelation function generally depends only weakly on this parameter and only part of the curve is accessible here due to the limited time resolution of 0.3 ms. Rather, stable fit results are obtained for all measurements, and the CCD detection with 2×2 pixels yields comparable results to the standard APD detection. The signal-to-noise ratio deduced from the signal trace is comparable to the one obtained from APD detection. Therefore, the kinetic mode can easily be used to perform FCS for slow diffusion (here the diffusion constant is D~15 µm2/s).
4. Fast kinetic mode
Obviously, a faster temporal resolution is needed to monitor smaller, commonly used fluorescent probes such as fluorescent proteins or organic dyes like Alexa Fluor 488 with a diffusion time of approximately 60 µs through the focal volume. The usual frame transfer mode of the CCD cannot achieve this time resolution, since the readout time of the chip is comparatively slow and even the frame transfer time itself (40 µs) would be comparable to the exposure time. We therefore adopted another mode called fast kinetic mode. The basic idea of this mode is to illuminate only several top rows of the image area of the chip and rapidly shift the charge down by several lines out of the illuminated region. Thus, a continuous acquisition is possible with exposure times down to 1 µs. Most of the chip is therefore used to store the images and consequently, part of the image area has to be used as an additional storage area. Because the out-of-focus light is not imaged into the pixels corresponding to the focus but spread to neighboring area, the bottom part of the image area, used now to store the previously illuminated (already shifted) lines, has to be shielded from the light in subsequent exposures. A direct shielding of the CCD chip is not possible since it is inside the camera housing. Therefore, an indirect shielding was applied as follows. The image of the focal spot was mapped 1:1 onto the CCD by an achromatic lens (see Fig. 1). A vertically adjustable razor blade was inserted in the original image plane of the microscope to block part of the out-of-focus light, thus effectively shielding the bottom part of the image area used now as a storage area. The edge steepness of this shielding was determined to be about two pixels, meaning that one row (and the part of the chip below) could be completely shielded and the signal two rows further up (and the part of the chip above) where not influenced. By using the top 6 CCD rows for illumination (as described later in this section) and the signal from 2×2 pixels (the middle 2 rows were used to acquire the signal), an efficient blocking could be achieved without reducing the signal in the favored pixels.
Once both chips are filled, they have to be read out and therefore, the measurement is finished, meaning that the fast kinetic mode is not a continuous mode in a strict sense. Rather, one can achieve a faster time resolution by renouncing a continuous acquisition, as employed in the kinetic mode, since the readout is still the speed-limiting process. Considering for example 6 lines at the top of the chip for illumination and a vertical shift time of 0.3 µs per line, the image transfer takes 1.8 µs. Therefore, exposure times down to 20 µs are sensible in order to have a ratio of shift to exposure time similar to the one in the kinetic mode. With these settings, 40 images are recorded in about 0.87 ms. The readout of both chips should take about 3.4 ms at a readout frequency of 10 MHz, which would be the minimal gap between two fast kinetic series. However, two problems arose, making the gap longer. First, the signal across the chip was strongly varying even for the dark measurement at 10 MHz, hence a readout rate of 3 MHz had to be used, yielding a minimal expected gap of 11 ms. Secondly, the CCD could not be programmed to acquire multiple fast kinetic series, but each of them had to be started by a separate command, enlarging the gap to 25 ms.
Nonetheless, measurements were carried out by starting 1000 fast kinetic series in a loop lasting 26 s, compared to an effective measurement time of 0.87 s. The data analysis was performed by correlating each single short fast kinetic trace and averaging over the 1000 repetitions. It has to be noted, that the simple subtraction of a time averaged reference dark measurement series was not sufficient in the fast kinetic mode, since the dark signal itself showed a non-zero correlation. This effect, which was not present in the kinetic mode, might originate from remaining varying offsets across the chip, i.e. the potential difficulty to stabilize the baseline in the fast kinetic mode. However, reliable results could be achieved by explicitly correcting for a correlated dark signal. Results for measuring enhanced Green Fluorescent Protein (eGFP; Clontech, Mountain View, California) and Alexa Fluor 488 are given in Fig. 4. Each fast kinetic measurement was complemented by a kinetic measurement (3 MHz readout, 0.82 ms exposure time per frame, 86 s total measurement time) to extend the time range for autocorrelation analysis. The signal-to-noise ratio is worse for the fast kinetic measurements due to the shorter effective acquisition time of less than 1 s. Therefore, the structure parameters had to be fixed while fitting the curves. The deduced particle numbers and diffusion times were in good agreement with the APD detection when taking the CCD signal from 2×2 pixels, as observed in the kinetic mode.
5. Two-spot measurement
A first simple proof-of-principle measurement was performed with two separate excitation volumes and a parallel detection along the horizontal axis of the CCD. The laser beam was split and reunited (see Fig. 1) by two pellicle beam splitters (Thorlabs), such that the reflected beam entered the objective back aperture at a slight angle. The two focal volumes are separated by 2 µm within the sample or 120 µm (5 pixels) in the image plane, respectively. For Alexa Fluor 488, parallel measurements in these two spots were performed with the same CCD settings as before. To compare with standard APD detection, two separate measurements were done by moving the fiber to the two detection positions. The resulting autocorrelation curves (Fig. 5) again show good agreement as discussed above. Although this parallel measurement could traditionally be done by a dual-core fiber and two APDs , the advantage of the CCD detection is the free choice of distance between the two detection volumes. Furthermore, this CCD setup is readily available to detect the signal from more than two excitation volumes at the same speed.
We have shown that FCS measurements are possible with an EMCCD camera, a flexible, binnable, two-dimensional detector. Although the CCD time resolution for full frame rate readout is limited to 2 ms, a quasi-continuous faster data acquisition (300 µs) is possible by choosing a subregion of the CCD in the kinetic mode. A better time resolution (20 µs) is achieved by repetitive starting a fast kinetic measurement, where only several top lines of the chip are illuminated and most of the chip is used to store the charge. The minimal sensible exposure time in this mode is only determined by the number of used CCD lines and the vertical shift speed. The finite readout speed and the size of the chip however lead to the main drawback of this mode, the non-continuous data acquisition. Nevertheless, reliable FCS curves can be acquired in this acquisition mode. This demonstrates the ability of the CCD to perform FCS on timescales interesting for biological applications.
With this short report we hope to contribute to the technique development for spatially resolved detection strategies for FCS. Today, EMCCD cameras exhibit the high sensitivity and speed needed to perform FCS. The simple two-spot measurement demonstrates, that a high temporal resolution is maintained independent of the used number of pixels in the horizontal direction of the CCD chip. A parallel detection along the vertical direction of the chip is possible in both acquisition modes. However, for the fast kinetic mode, the increased number of required CCD rows will reduce the time resolution and the number of images that can be stored. The time resolution for parallel detection in both dimensions of the CCD, over a large field of view, is therefore limited to the CCD frame rate in the kinetic mode.
As EMCCD technology development will continue, the speed of data readout is likely to be enhanced. This will enable a faster continuous data acquisition. Particular care has to be taken to ensure a stable detection efficiency for all detector elements across the chip, which is a crucial prerequisite for fluctuation analysis and single molecule investigation in general. If future developments fulfill these requirements, FCS might become as easily accessible as confocal imaging. These systems will then combine ultra-low light detection from wide-field excitation, and parallel confocal LSM and FCS by using one single detector.
We wish to thank Zdenĕk Petrášek for helpful discussion, Jonas Ries for an introduction to Matlab, and Jörg Mütze for a critical reading of the manuscript. This work was supported by grants from EFRE (4-0123.55-20-0370-03/3).
References and links
1. D. Magde, W. W. Webb, and E. Elson, “Thermodynamic fluctuations in a reacting system - measurement by fluorescence correlation spectroscopy,” Phys. Rev. Lett. 29(11), 705–708 (1972). [CrossRef]
2. R. Rigler, Ü. Mets, J. Widengren, and P. Kask, “Fluorescence correlation spectroscopy with high count rate and low background: analysis of translational diffusion,” Eur. Biophys. J. 22, 169–175 (1993). [CrossRef]
3. J. Widengren, R. Rigler, and Ü. Mets, “Triplet-state monitoring by fluorescence correlation spectroscopy,” J. Fluoresc. 4(3), 255–258 (1994). [CrossRef]
4. U. Haupts, S. Maiti, P. Schwille, and W. W. Webb, “Dynamics of fluorescence fluctuations in green fluorescent protein observed by fluorescence correlation spectroscopy,” Proc. Natl. Acad. Sci. U.S.A. 95(23), 13573–13578 (1998). [CrossRef]
5. P. Schwille, F. J. Meyer-Almes, and R. Rigler, “Dual-color fluorescence cross-correlation spectroscopy for multicomponent diffusional analysis in solution,” Biophys. J. 72, 1878–1886 (1997). [CrossRef] [PubMed]
6. M. Brinkmeier, K. Dörre, J. Stephan, and M. Eigen, “Two beam cross correlation: A method to characterize transport phenomena in micrometer-sized structures,” Anal. Chem. 71(3), 609–616 (1999). [CrossRef]
7. H. Blom, M. Johansson, A.-S. Hedman, L. Lundberg, A. Hanning, S. H°ard, and R. Rigler, “Parallel fluorescence detection of single biomolecules in microarrays by a diffractive-optical-designed 2×2 fan-out element,” Appl. Opt. 41(16), 3336–3342 (2002). [CrossRef]
8. M. Gösch, A. Serov, T. Anhut, T. Lasser, A. Rochas, P.-A. Besse, R. S. Popovic, H. Blom, and R. Rigler, “Parallel single molecule detection with a fully integrated single-photon 2×2 CMOS detector array,” J. Biomed. Opt. 9(5), 913–921 (2004). [CrossRef]