Imaging Diffusion in Living Cells Using Time-Correlated Single

Sep 6, 2007 - we introduce a new method termed diffusion imaging microscopy that combines ... to living cells and recorded a diffusion map of a living...
0 downloads 0 Views 656KB Size
Download PDF
Anal. Chem. 2007, 79, 7340-7345

Imaging Diffusion in Living Cells Using Time-Correlated Single-Photon Counting Christian M. Roth,† Pia I. Heinlein,† Mike Heilemann,*,†,‡ and Dirk-Peter Herten*,†

Physikalisch-Chemisches Insitut, Universita¨t Heidelberg, Im Neuenheimer Feld 253, 69120 Heidelberg, Germany, and Angewandte Laserphysik und Laserspektroskopie, Universita¨t Bielefeld, Universita¨tsstrasse 25, 33615 Bielefeld, Germany

Current efforts to monitor the diffusion of proteins in living cells are based on either fluorescence correlation spectroscopy (FCS), fluorescence recovery after photobleaching, or image correlation spectroscopy. However, these methods cannot generate a map of diffusion times. Here, we introduce a new method termed diffusion imaging microscopy that combines scanning confocal microscopy, time-correlated single-photon counting, and FCS and thus allows us to measure spatially resolved diffusion times. In our approach, we record scan images with timeresolved photon streams within each individual pixel. By extending the pixel dwell time to 25-100 ms, a software correlation of individual photons within each pixel yields the average diffusion time. Additionally, information on fluorescence intensity (number of photons) and fluorescence lifetime is available and can be used to sort fluorescence photons and to discriminate from autofluorescence. We evaluated our method by measuring diffusion times of dT20-TMR in solutions of different viscosity. We further demonstrate the applicability of the method to living cells and recorded a diffusion map of a living 3T3 mouse fibroblast incubated with dT20-ATTO488. Systems-level understanding of cellular behavior demands novel methods exploring biomolecular processes in the heterogeneous environment of a living cell. Powerful methods based on fluorescence microscopy were developed to measure mobility and directed movement of proteins and protein complexes. Image correlation spectroscopy and related methods have the advantage that the average mobility of proteins is determined by correlating the spatial and temporal fluorescence fluctuations of fluorescence intensity images showing the spatial distribution of the labeled proteins or protein complexes.1-3 Long dwell times and frame rates limit their application to slower motional processes, such as the diffusion of membrane proteins. Faster processes, such as free * To whom correspondence should be addressed. E-mail: mike.heilemann@ urz.uni-heidelberg.de. E-mail: [email protected]. Fax: +49-6221544255. † Universita ¨t Heidelberg. ‡ Universita ¨t Bielefeld. (1) Hebert, B.; Costantino, S.; Wiseman, P. W. Biophys. J. 2005, 88, 36013614. (2) Petersen, N. O.; Hoddelius, P. L.; Wiseman, P. W.; Seger, O.; Magnusson, K. E. Biophys. J. 1993, 65, 1135-1146. (3) Wiseman, P. W.; Brown, C. M.; Webb, D. J.; Hebert, B.; Johnson, N. L.; Squier, J. A.; Ellisman, M. H.; Horwitz, A. F. J. Cell Sci. 2004, 117, 55215534.

7340 Analytical Chemistry, Vol. 79, No. 19, October 1, 2007

diffusion in the cytoplasm or translocation in and out of the nucleus, are frequently studied by fluorescence recovery after photobleaching (FRAP).4 Here a region of interest within the cell is illuminated at high laser power to bleach the fluorescent labels. Subsequently, the increase in fluorescence intensity within the bleached region is monitored to determine the mobility of the mobile fraction of labeled protein. In addition, techniques from single-molecule fluorescence spectroscopy were developed in the past decade and applied to resolve spatial and temporal heterogeneities in various molecular processes.5-8 More recently, fluorescence correlation spectroscopy (FCS) and fluorescence cross-correlation spectroscopy (FCCS) were applied to study diffusional motion and protein/protein interactions in living cells.9 FCS measures and analyzes temporal fluctuations in fluorescence intensity that are due to variations in the number of labeled molecules diffusing through the tiny observation volume defined by the diffraction-limited focal volume of a confocal microscope. Due to the high time resolution (1 h, thus preventing its application to living cells. In contrast to the previous work, we demonstrate for the first time, that a similar approach can be used to record diffusion maps in living cells. We show, that acquisition times can be reduced by 1 order of magnitude by use of the time-tag-to correlate algorithm and a simplified model fit to sparse correlation data to obtain contrast in microscopic images based on diffusion times.17 Within a single measurement of a few minutes, we simultaneously obtain images of the fluorescence intensity and fluorescence lifetime as well as correlation curves holding information about diffusion time and concentration within living cells by using fluorescence lifetime imaging microscopy (FLIM).18,19 The advantage of the simultaneous acquisition of multiple parameters is that the fluorescence intensity allows discriminating between immobile and mobile fractions in the cell, while fluorescence lifetime can be used to sort labeled biomolecules, unspecific background, or autofluorescent moieties. EXPERIMENTAL SECTION Synthesis of Dye-Labeled Oligonucleotides. DNA oligonucleotides were ordered from IBA GmbH (Go¨ttingen, Germany). Single-stranded dT20 carrying a C6-amino modification at the 5′end was labeled with ATTO488 (ATTO-TEC GmbH, Siegen, Germany) NHS ester according to manufacturer’s instruction. Labeled oligonucleotides were purified by HPLC (Agilent, Santa Clara, CA) on a reversed-phase column (Knauer, Berlin, Germany) packed with octadecylsilane-Hypersil C18. (13) Bates, I. R.; Wiseman, P. W.; Hanrahan, J. W. Biochem. Cell Biol. 2006, 84, 825-831. (14) Kannan, B.; Guo, L.; Sudhaharan, T.; Ahmed, S.; Maruyama, I.; Wohland, T. Anal. Chem. 2007, 79, 4460-4470. (15) Xiao, Y.; Buschmann, V.; Weston, K. D. Anal. Chem. 2005, 77, 36-46. (16) Lenne, P. F.; Colombo, D.; Giovannini, H.; Rigneault, H. Single Mol. 2002, 4, 194-200. (17) Wahl, M. I. G.; Patting, M.; and Enderlein, J. Opt. Express 2003, 11, 35833591. (18) Knemeyer, J. P.; Herten, D. P.; Sauer, M. Anal. Chem. 2003, 75, 21472153. (19) Tinnefeld, P.; Buschmann, V.; Herten, D. P.; Han, K.-T.; Sauer, M. Single Mol. 2000, 1, 215-223.

Figure 1. Schematic of confocal FLIM setup.

Cell Culture. 3T3 mouse fibroblasts (DSMZ) were cultured in Dulbecco’s modified Eagle medium (DMEM) (Biochrom AG, Berlin, Germany) supplemented with 10% (v/v) bovine serum (Invitrogen, Karlsruhe, Germany) and 1% (v/v) penicillin/streptomycin (Biochrom AG). The cells were grown in cell culture flasks in an atmosphere of 5% CO2 at 37 °C. Incubation of the 3T3 Cells. Before the experiments, the cells were transferred into eight-well chambers (Nunc, Wiesbaden, Germany) and grown overnight with DMEM and 10% (v/v) bovine serum. The cells were incubated with 10-6-10-8 M dT20-ATTO488 for 2 h. After incubation, the medium was removed and replaced by the growth medium. All cell measurements were done at room temperature. Experimental Setup. The experimental principle of FLIM is shown in Figure 1 and has been described in detail elsewhere.19 Briefly, a pulsed laser diode emitting at 470 nm (PicoQuant GmbH, Berlin, Germany) with a repetition rate of 40 MHz was driven by a pulsed laser driver (PDL 808 “Sepia”, PicoQuant GmbH) and coupled into an inverted microscope (Axiovert 100TV, Zeiss, Germany) equipped with a three-axis piezoscanning stage (PI Physik Instrumente, Karlsruhe, Germany). The collimated laser beam was directed into an oil immersion microscope lens (100×, NA ) 1.45, Olympus) using a dichroic mirror (z488/633 RPC, AHF Analysentechnik, Tu¨bingen, Germany) and focused onto the sample. Fluorescence emission was collected through the same microscope lens, focused onto a 120-µm pinhole, filtered with an emission bandpass filter (HQ515/70M, AHF Analysentechnik), split by a nonpolarizing beam splitter (Thorlabs, Newton, MA), and focused onto two avalanche photodiodes (APD, SPCM-AQR15, Perkin-Elmer). A PC card for TCSPC (SPC-132, Becker&Hickl, Berlin, Germany) and custom-written LabVIEW software was used for data recording and analysis. Analytical Chemistry, Vol. 79, No. 19, October 1, 2007

7341

For experiments with dT20-TMR, a continuous-wave and frequency-doubled NdYAG laser (Crystalaser, Lasergate, Germany) with a wavelength of 532 nm was used, together with appropriate laser cleanup filter (z532/10), excitation dichroic beamsplitters (z532/633) and bandpass filter (HQ560/50, all AHF Analysentechnik, Tu¨bingen, Germany). Data Analysis. The correlation curve is generated from FIFO data using custom-written analysis software written in LabVIEW. The correlation function Gij(τ) is calculated according to (1):

Gij(τ) )

〈δFi(t) δFj(t + τ)〉 〈Fi(t )〉〈Fj(t )〉

(1)

Here i * j represents the cross-correlation and i ) j the autocorrelation function. The typical algorithms for analyzing FIFO data are fast Fourier transformation or the “bin and correlate” method. These methods, which bin the data using the smallest desired bin width that corresponds to the smallest lag time in the correlation function, are time and memory consuming when used for software correlation because of their linearity in the lag time τ. Therefore, we used the time-tag-to-correlate algorithm for an asynchronous correlation of our data taken with our TCSPC system.17 With this method, the first lag time of the correlation is the so-called macrotime, i.e., the photon arrival time with respect to the start of the measurement. After certain repetitions of the correlation procedure, the lag time width is doubled. This leads to the known multiple-τ correlation with quasi-logarithmic distributed data points. To improve the quality of the fitting results, an error was estimated for each data point of the correlation function, according to Wohland et al.20 To generate a diffusion image from a fluorescence intensity image, the FIFO data from four neighboring pixels (2 × 2) were correlated separately and averaged according to (2):

〈G(τ)〉 )

1 N

N

∑ G (τ) i

(2)

i)1

The error of the averaged correlation curve is calculated from the estimation of the error of a single pixel, following (3):

〈σ(τ)〉 )

x∑ 1

N

N

i)1

(σi(τ))2

(3)

The resulting correlation curves are fed into a nonlinear leastsquares fit (Levenberg-Marquardt algorithm) for estimating the diffusion time τDiff and the amplitude A according to a simple twodimensional diffusion model, as described in eq 4:

1 G(τ) ) A (1 + τ/τDiff) The amplitude A is reciprocal to the average number of molecules that are detected in the confocal volume at the same time. (20) Wohland, T.; Rigler, R.; Vogel, H. Biophys. J 2001, 80, 2987-2999.

7342

Analytical Chemistry, Vol. 79, No. 19, October 1, 2007

For the data analysis of an image, the photons of the pixels are correlated individually as described above. With this correlation method, the computational time for the analysis of a whole image of 20 × 20 µm2 is reduced to a few minutes only. RESULTS AND DISCUSSION To demonstrate the capability of our method to determine diffusion times from confocal scanning data, we recorded scan images of 1 nM dT20-tetramethylrhodamine (dT20-TMR) in phosphate-buffered saline (PBS) buffer with varying glycerol concentrations from 0 to 71% (v/v) (Figure 2A). The 4 × 4 µm2 scan images were recorded with a pixel resolution of 50 nm and an integration time of 25 ms/pixel. The diffusion time images (Figure 2B) were generated from cross-correlation of photon arrival times within each pixel and subsequent fit of the correlation curves to the simple two-dimensional diffusion model. Exemplary crosscorrelation curves together with fit functions are shown in Figure 2C for dT20-TMR in PBS (black squares, red line) and 71% glycerol (white circles, green line). It is an advantage of our mathematical procedure that reliable fitting is possible with such short acquisition times, even for correlation functions that do not converge to zero (corresponding data are shown in Figure S-1 of the Supporting Information). From the two-dimensional diffusion model, we obtained diffusion times of 0.21 (PBS) and 0.80 ms (PBS with 71% glycerol). To demonstrate the resolution of diffusion times, corresponding values from all pixels from one image were summed up in histograms. Histograms of diffusion time distributions for various viscosities are shown in Figure 2D, and distinct populations for different glycerol concentrations are obtained. The mean values of the distributions of diffusion times (Figure 2D) clearly show an increase from 0.21 (PBS) to 0.80 ms (PBS with 71% glycerol). Upon addition of glycerol, an increase of fluorescence intensity is observed initially, which is explained by an increase of the fluorescence quantum yield of TMR in a medium of higher viscosity.21 At higher concentrations of glycerol, the width of the diffusion time distribution increases noticeably, a fact that can be explained by a reduced flux of molecules into and out of the focal volume and by an increased shot noise. Nevertheless, the data demonstrate that our method is capable of measuring diffusion times over a range of 1 order of magnitude with integration times as short as 25 ms/pixel and provides sufficiently high contrast to discriminate between species of different diffusion constants. Mapping an area of 4 × 4 µm2 with a resolution of 50 nm and an integration time of 25 ms/pixel requires an acquisition time of 160 s, which makes this method potentially interesting for various applications interested in liquid flow at a microscopic scale, e.g., in microchannel arrays15 or nanopipets. As diffusion imaging microscopy (DIFIM) proved its suitability to measure diffusion times on a large scale and with high contrast while scanning a sample, we extended our approach to living cells. Our aim was to demonstrate that DIFIM can generate a map of diffusion coefficients of a living cell, which will allow visualizing transport pathways through and preferred targets within a cell. However, the degree of heterogeneity in a cell is clearly different from a homogeneous solution, and complex diffusion processes (21) Arden, J.; Deltau, G.; Huth, V.; Kringer, U.; Peros, D.; Drexhage, K. H. J. Lumin. 1991, 48&49, 352.

Figure 2. dT20-TMR diluted in PBS with glycerol added to various final concentrations (v/v). (A) Fluorescence intensity images, scan size 4 × 4 µm2, 50 nm resolution per pixel and 25 ms dwell time. (B) The corresponding diffusion time images were obtained by correlating the arrival times of photons within each pixel and fitting to a two-dimensional diffusion model function. (C) Typical correlation curves and fits for dT20-TMR in PBS buffer (black squares, red line) and in PBS with 71% (v/v) glycerol (circles, green line), yielding diffusion times of 0.21 (PBS) and 0.80 ms (PBS and 71% glycerol). (D) Histograms of diffusion times, acquired from summing up diffusion times from all pixels within one image.

can be observed. Additionally, different compartments of a cell might differ in refractive index, which has an influence on the shape of the laser focus and thus on diffusion times extracted from fluorescence fluctuations. It is clear that, with the current method, the measurement of diffusion times in cells thus remains qualitative, but our method does serve as a valuable tool to generate quantitative diffusion maps of cells or cell compartments and hereby introduces additional contrast. To evaluate the applicability of our method to living cells, we used living 3T3 mouse fibroblasts as a model system. The cells were adsorbed to a glass surface and were incubated with dT20ATTO488, which we expected to randomly diffuse within a cell. The cells were scanned at a resolution of 100 nm/pixel, with a dwell time of 100 ms/pixel. A typical fluorescence intensity scan image with an image size of 5 × 10 µm2 is shown in Figure 3A. For this particular cell, we observed higher fluorescence intensity in the nucleus while the fluorescence intensity within the cytosol is lower and more heterogeneous. Almost no fluorescence was observed in the medium. As explained in the calibration experiment with dT20-TMR and demonstrated in Figure 2C, single cross-correlation curves can be calculated for each pixel of the image. Exemplary correlation functions derived from two distinct positions within the nucleus of the cell in Figure 3A, together with a fit to a two-dimensional diffusion model function, are shown in Figure 3B. Diffusion times of 0.85 (blue curve) and 4.4 ms (red curve) were determined.

A diffusion map of the cell was generated pixelwise, and the corresponding diffusion time image is shown in Figure 3C. With the color coding chosen for the diffusion times, one clearly can distinguish regional heterogeneity within the nucleus of the cell, which is evident from neither the fluorescence intensity nor the fluorescence lifetime image. As the photon statistics within one pixel is an important parameter, we normalized the diffusion time image with the fluorescence intensity, yielding a weighted diffusion time image in Figure 3D. The intensity-weighted diffusion maps allowed us to increase the contrast of the diffusion image. Bright-colored regions in Figure 3D indicate good photon statistics with confident fits to the correlation function, whereas dark-colored regions indicate low photon counts and substantial contribution of noise, where as a consequence the algorithm does not produce a confident fit to the correlation function of a pixel. Additionally to diffusion time images, we show images of the amplitudes of the cross-correlation function in Figure 3E and of intensity-weighted amplitudes in Figure 3F. As a measure of concentration, the amplitude reflects in average a similar concentration of dT20-ATTO488 throughout the nucleus and lower concentration in the cytosol or surrounding medium, which is in good agreement with the intensity image in Figure 3A. As the amplitudes of the correlation function appear rather constant throughout the nucleus, the observed contrast in the diffusion time image have to be associated with a change in mobility of the dT20-ATTO488 probe. Currently, we can only speculate about Analytical Chemistry, Vol. 79, No. 19, October 1, 2007

7343

Figure 3. 3T3 cell incubated with dT20-ATTO488. (A) Fluorescence intensity image, 5 × 10 µm2. (B) Correlation curves measured at two different positions within the cell nucleus with fit functions, yielding diffusion times of 0.85 (white circles, blue line) and 4.4 ms (black squares, red line). (C) Diffusion time image and (D) intensity-weighted diffusion time image. (E) Amplitude image and (F) intensity-weighted amplitude image. (G) Fluorescence lifetime image and (H) intensity-weighted fluorescence lifetime image. Intensity-weighted images were generated with a color code ranging from 0 (black) to 1750 photons (full color).

possible reasons leading to the observed contrast in the diffusion time images, reflecting regions of different viscosities within the nucleus. The dT20-ATTO488 probe can hybridize to polyA tails of nascent mRNA, which would lead to a 100-1000-fold change in mass and thus indicate regions with high mRNA concentration by an apparent decrease in mobility.22 By using pulsed laser diodes as excitation sources and thus a FLIM setup, additional information comes with fluorescence lifetime, a unique property of fluorophores. The fluorescence lifetime image of the cell in Figure 3A is shown in Figure 3G; the intensity-weighted fluorescence lifetime image is shown in Figure 3H. Here, we use a threshold to avoid artifacts appearing for low photon counts per pixel. In Figure 3G and H, we can now distinguish between the lifetime of ATTO488, 4.0 ns, and autofluorescence, ∼3.2 ns measured in untreated cells. Evidently, dT20-ATTO488 has accumulated in the nucleus of the cell, whereas the fluorescence signal in the cytosol can be

attributed to mainly autofluorescence. Due to very low fluorescence intensity in the medium, no fluorescence lifetime was determined. Furthermore, the diffusion pattern of dT20-ATTO488 in the nucleus exhibits clear contrasts and reveals heterogeneities, which indicates that the distribution of dT20-ATTO488 depends on its localization within the nucleus. More cell scans and DIFIM images are shown in Figures S-2, S-3, and S-4 in the Supporting Information. Our method of extracting diffusion times from confocal singlemolecule scan images can be extended and strengthened by recording both spectral information of fluorescence emission and fluorescence lifetime, i.e., additional parameters that can be derived from spectrally resolved FLIM (SFLIM) techniques.23 The advantage of additional parameters lies in the possibility to sort photons, e.g., by emission wavelength and fluorescence lifetime, and thus discriminate between the diffusion of differently labeled species, as well as to eliminate autofluorescence. The set of

(22) Politz, J. C.; Browne, E. S.; Wolf, D. E.; Pederson, T. Proc. Natl. Acad. Sci. U.S.A. 1998, 95, 6043-6048.

(23) Tinnefeld, P.; Herten, D. P.; Sauer, M. J. Phys. Chem. A 2001, 105, 79898003.

7344 Analytical Chemistry, Vol. 79, No. 19, October 1, 2007

information that becomes available can help to shed light on dynamic processes in a complex environment, as is the case for diffusion of biomolecules or signaling pathways in living cells. CONCLUSION We demonstrated that our method can be used to measure diffusion times ranging from 200 µs to 5 ms in the concentration range of 10-8-10-10 M. Spatial heterogeneities can be resolved by object scanning, and the measured diffusion times can be normalized by fluorescence intensity to suppress artifacts. In addition, the fluorescence lifetime can be used to discriminate the fluorescence emission of the probe from an autofluorescent background. It is thus possible to sort photons upon their origin, as well as monitor interactions between molecules with different labels. This principle can be further improved by introducing a spectral separation in the emission pathway of the setup, i.e., by applying SFLIM. We demonstrated that our method can be used to measure diffusion times in living cells and that image analysis allows generating a diffusion map of a cell. However, a current limitation of the method is the relatively long time of data acquisition. Although the step size can be increased and the integration time reduced (see Figure S-3 in the Supporting Information for the effect of reduced integration time), care should be taken to find the right balance between acquisition time and dynamic changes in the cell, like cell motility and cell cycle, which have to be monitored prior and after the measurement (see Figure S-3). Nevertheless the presented method seems promising to investigate differences in nucleocytoplasmic cycling of cell signaling proteins before and after receptor activation, like STAT-5 or members of the Smad protein family.24,25 As maintenance of receptor activity by phosphorylated signaling proteins was re(24) Inman, G. J.; Nicolas, F. J.; Hill, C. S. Mol. Cell 2002, 10, 283-294. (25) Swameye, I.; Muller, T. G.; Timmer, J.; Sandra, O.; Klingmuller, U. Proc. Natl. Acad. Sci. U.S.A. 2003, 100, 1028-1033.

ported to take place in the order of 30 min up to several hours, acquisition of the underlying diffusion dynamic lies in range of the presented method. Furthermore, photostability of the probes is an important task, as photobleaching does deteriorate the data quality. Measurements of diffusion in cells can be expected to suffer from a significant distortion of the laser beam profile that might influence the observed diffusion kinetics.26 Beyond this, the internal structure of the cells themselves, e.g., differences in the refractive index of various organelles, will influence the quality of the data. Finally, the method cannot be used to determine directed transport, as it currently operates one laser focus only. A solution to these problems lies in multifocal techniques that have already been proposed and developed by several groups27 and could be used for the acquisition of directed flow by FCCS. ACKNOWLEDGMENT We gratefully acknowledge the financial support by the Deutsche Forschungsgemeinschaft (DFG, HE 4559/1-1). We thank Jo¨rg Enderlein (University of Tu¨bingen), Johan Hofkens (Katholieke Universiteit Leuven), Ursula Klingmu¨ller (DKFZ Heidelberg), Markus Sauer (University of Bielefeld), and Ju¨rgen Wolfrum (University of Heidelberg) for fruitful discussion and critical comments on the manuscript. SUPPORTING INFORMATION AVAILABLE Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review May 21, 2007. Accepted July 16, 2007. AC071039Q (26) von der Hocht, I.; Enderlein, J. Exp. Mol. Pathol. 2007, 82, 142-146. (27) Dertinger, T.; Pacheco, V.; von der Hocht, I.; Hartmann, R.; Gregor, I.; Enderlein, J. Chemphyschem 2007, 8, 433-443.

Analytical Chemistry, Vol. 79, No. 19, October 1, 2007

7345