Anal. Chem. 2007, 79, 3509-3513
Increasing the Resolution of Single Pair Fluorescence Resonance Energy Transfer Measurements in Solution via Molecular Cytometry James H. Werner,*,† Evan R. McCarney,‡ Richard A. Keller,§ Kevin W. Plaxco,‡ and Peter M. Goodwin†
Center for Integrated Nanotechnologies, and Bioscience Division, Los Alamos National Laboratory, Los Alamos, New Mexico 87545, and Department of Chemistry and Biochemistry, University of California Santa Barbara, Santa Barbara, California 93106
We report a method to increase the resolution of single pair fluorescence resonance energy transfer (spFRET) measurements in aqueous solutions. Solution-based spFRET measurements of fluorescently labeled biological molecules (proteins, RNA, DNA) are often used to obtain histograms of molecular conformation without resorting to sample immobilization. However, for solution-phase spFRET studies, the number of photons detected from a single molecule as it diffuses through an open confocal volume element are quite limited. An “average” transit may yield on the order of 40 photons. Shot noise on the number of detected photons substantially limits the resolution of the measurement. The method reported here uses a hydrodynamically focused sample stream to ensure molecules traverse the full width of an excitation laser beam. This substantially increases the average number of photons detected per molecular transit (∼85 photons/ molecule), which increases measurement precision. In addition, this method minimizes another source of heterogeneity present in diffusive measures of spFRET: the distribution of paths taken through the excitation laser beam. We demonstrate here using a FRET labeled protein sample (a FynSH3 domain) that superior resolution (a factor of ∼2) can be obtained via molecular cytometry compared to spFRET measurements based upon diffusion through an open confocal volume element. The study of single molecules by laser-induced fluorescence has emerged as a powerful tool to explore complex systems.1,2 As opposed to ensemble measurements, single molecule methods facilitate the study of rare events, direct characterization of molecular heterogeneity, and enable the observation of dynamic, stochastic processes without the need to synchronize molecules via a rapid perturbation. As such, single molecule fluorescence measurements have seen increased use, with examples ranging * To whom correspondence should be addressed. E-mail:
[email protected]. Phone: (505)-667-8842. † Center for Integrated Nanotechnologies, Los Alamos National Laboratory. ‡ University of California Santa Barbara. § Bioscience Division, Los Alamos National Laboratory. (1) Ambrose, W. P.; Goodwin, P. M.; Jett, J. H.; Van Orden, A.; Werner, J. H.; Keller, R. A. Chem. Rev. 1999, 99, 2929-2956. (2) Weiss, S. Science 1999, 283, 1676-1683. 10.1021/ac070142c CCC: $37.00 Published on Web 03/27/2007
© 2007 American Chemical Society
from studying single molecule enzymatic turnovers,3 the kinetics of molecular motors,4,5 or the conformation of DNA,6-8 RNA,9 or proteins.10-14 While a number of single molecule spectroscopies are available to study biomolecules, thus far, single pair fluorescence resonance energy transfer (spFRET)15 has emerged as the primary method of choice for monitoring conformation. FRET involves the nonradiative transfer of energy from a fluorescence donor fluorophore to a fluorescence acceptor, with the acceptor generally possessing a red-shifted emission and absorption spectrum from that of the donor. The energy transfer rate is inversely proportional to the separation distance between the molecules to the sixth power.16 Due to the strong dependence of energy transfer rate on separation distance, energy transfer can be used as a “spectroscopic ruler” to measure distances over a range of 20-80 Å.17 While several methods of measuring energy transfer efficiency are available,18 perhaps the most straightforward and commonly employed method is to choose an excitation wavelength that (3) Lu, H. P.; Xun, L.; Xie, X. S. Science 1998, 282, 1877-1882. (4) Funatsu, T.; Harada, Y.; Tokunaga, M.; Saito, K.; Yanagida, T. Nature 1995, 374, 555-559. (5) Yildiz, A.; Forkey, J. N.; Mckinney, S. A.; Ha, T.; Goldman, Y. E.; Selvin, P. R. Science 2003, 300, 2061-2065. (6) Deniz, A. A.; Dahan, M.; Grunwell, J. R.; Ha, T. J.; Faulhaber, A. E.; Chemla, D. S.; Weiss, S.; Schultz, P. G. Proc. Natl. Acad. Sci. U.S.A. 1999, 96, 36703675. (7) Werner, J. H.; Larson, E. J.; Goodwin, P. M.; Ambrose, W. P.; Keller, R. A. Appl. Opt. 2000, 39, 2831-2839. (8) Mckinney, S. A.; Declais, A. C.; Lilley, D. M. J.; Ha, T. Nat. Struct. Biol. 2003, 10, 93-97. (9) Zhuang, X. W.; Bartley, L. E.; Babcock, H. P.; Russell, R.; Ha, T. J.; Herschlag, D.; Chu, S. Science 2000, 288, 2048-2052. (10) Talaga, D. S.; Lau, W. L.; Roder, H.; Tang, J. Y.; Jia, Y. W.; Degrado, W. F.; Hochstrasser, R. M. Proc. Natl. Acad. Sci. U.S.A. 2000, 97, 13021-13026. (11) Schuler, B.; Lipman, E. A.; Eaton, W. A. Nature 2002, 419, 743-747. (12) Deniz, A. A.; Laurence, T. A.; Beligere, G. S.; Dahan, M.; Martin, A. B.; Chemla, D. S.; Dawson, P. E.; Schultz, P. G.; Weiss, S. Proc. Natl. Acad. Sci. U.S.A. 2000, 97, 5179-5184. (13) Rhoades, E.; Gussakovsky, E.; Haran, G. Proc. Natl. Acad. Sci. U.S.A. 2003, 100, 3197-3202. (14) McCarney, E. R.; Werner, J. H.; Bernstein, S. L.; Ruczinski, I.; Makarov, D. E.; Goodwin, P. M.; Plaxco, K. W. J. Mol. Biol. 2005, 352, 672-682. (15) Ha, T.; Enderle, T.; Ogletree, D. F.; Chemla, D. S.; Selvin, P. R.; Weiss, S. Proc. Natl. Acad. Sci. U.S.A. 1996, 93, 6264-6268. (16) Fo ¨rster, T. Ann. Phys. 1948, 2, 55-75. (17) Stryer, L.; Haugland, R. P. Proc. Natl. Acad. Sci. U.S.A. 1967, 58, 719-&. (18) Clegg, R. M. In DNA structures. Part A. Synthesis and physical analysis of DNA; Lilley, D. M. J., Dahlber, J. E., Eds.; Methods in Enzymology 211; Academic Press, Inc.: San Diego, CA, 1992.
Analytical Chemistry, Vol. 79, No. 9, May 1, 2007 3509
preferentially excites the donor fluorophore and measure the enhanced fluorescence emission of the acceptor. For spFRET measurements in solution, individual molecules diffuse into and out of an open confocal excitation volume element.10-12,14 The transit of individual fluorescently labeled molecules through the optical probe volume leads to “bursts” of photons that are easily distinguishable from the background counts. The collected photons are split spectrally into donor and acceptor channels and are usually analyzed in discrete time intervals larger than the average burst duration. For a selected burst, if Na photons are measured in the acceptor channel and Nd photons are measured in the donor channel, one can obtain an “apparent” energy transfer efficiency for that molecule as
Eapp ) Na/(Na + Nd)
(1)
One can compensate for counts on the detectors due to background, differences in the detection efficiencies in the two detection channels, and differing fluorescence quantum yields of the fluorophores to arrive at a corrected transfer efficiency.6 For this work, we have chosen to follow the convention of Schuler et al.11 and simply report the apparent transfer efficiency in the absence of any such corrections. Typically, an energy transfer efficiency is calculated for each molecular transit using eq 1; the Eapp values measured from several thousand single molecules are then combined to build a histogram of conformations. The contribution of shot noise to the width of the spFRET histogram has been considered on a number of occasions.11,19,20 If one assumes the donor and acceptor emissions are governed by Poisson statistics, then if N photons are detected on average from a single molecule, the standard deviation of the number of photons detected is N1/2. Using standard error propagation, one can arrive at an expected standard deviation of the spFRET histogram as
σEapp )
x{
} {
∂Eapp σ ∂Na Na
2
+
} x
∂Eapp σ ∂Nd Nd
2
)
( - 1) NT
(2)
In the above equation, σEapp denotes the standard deviation of the Eapp distribution, σNa is the uncertainty on the number of acceptor photons (σNa ) Na1/2), σNd is the uncertainty on the number of donor photons (σNd ) Nd1/2), NT is the threshold number of photons used to construct the spFRET histogram (NT ) Na + Nd), and is the mean value of the spFRET efficiency ( ) Na/(Na + Nd)). The above expression (eq 2) is identical to that derived through more rigorous methods by Gopich and Szabo20 (square root of eq 2.25 of that work), and as such, we have adopted much of that work’s nomenclature. In addition to the contribution to the spFRET histogram width due to shot noise, Gopich and Szabo20 also considered other sources of noise that can broaden the distribution. In particular, two additional broadening mechanisms for solution-based spFRET measurements are the diffusion of the molecule in the optical probe volume and triplet shelving of the acceptor fluorophore.20 (19) Dahan, M.; Deniz, A.; Ha, T.; Chemla, D.; Schultz, P.; Weiss, S. Chem. Phys. 1999, 247, 85-106. (20) Gopich, I.; Szabo, A. J. Chem. Phys. 2005, 122, 014707.
3510
Analytical Chemistry, Vol. 79, No. 9, May 1, 2007
Figure 1. Schematic representation of a flow cytometer. Single protein molecules are introduced by a tapered capillary into a fast moving sheath flow. An excitation laser beam, focused by a lens, intersects the sample stream downstream of the capillary orifice. Fluorescence is collected out of the page by a high NA microscope objective (not shown).
The method used here keeps the advantages of spFRET measurements via diffusion through an open confocal volume element while eliminating some of its drawbacks. In particular, examining molecules in a well-defined sample stream minimizes one source of measurement noise, namely, the distribution of paths taken through the excitation laser beam for different molecules. If performed properly, measurements in flowing sample streams can have each molecule take a similar spatial trajectory down the center of an optical probe volume,21 thus minimizing heterogeneity resulting from differing diffusion paths across the excitation probe volume. As noted previously,19 the fact spFRET is a ratiometric measurement inherently minimizes this contribution to the measurement uncertainty. However, the ratiometric nature of spFRET does not totally eliminate this source of measurement noise, as discussed in ref 20. The technique described here suppresses this noise source and also reduces the measurement uncertainty introduced by shot noise (Poisson statistics) on the number of photons detected, as the technique enables more photons to be detected from an “average” molecule. As demonstrated here, this increase in photon number, as well as the reduction in heterogeneity due to different molecules taking different trajectories through the optical probe volume, leads to increased spFRET resolution. MATERIALS AND METHODS Protein Expression and Labeling. Protein mutation, expression, labeling (donor, Alexa 488; acceptor, Alexa 594), and purification were performed as described previously.14 The double FynSH3 mutant used for this work was K105C/S115C. spFRET via Diffusion through an Open Confocal Volume Element. The experimental apparatus and excitation conditions were as described previously.14 spFRET in Flowing Sample Streams. These measurements were performed on a home-built single molecule fluorescence flow system modified from previous versions21,22 for two-color detection. In brief, the sample was introduced at a low volumetric flow rate into a larger, faster moving sheath flow stream from a small orifice (see Figure 1 for a schematic of the apparatus). For these (21) Keller, R. A.; Ambrose, W. P.; Goodwin, P. M.; Jett, J. H.; Martin, J. C.; Wu, M. Appl. Spectrosc. 1996, 50, A12-A32. (22) Werner, J. H.; Cai, H.; Jett, J. H.; Reha-Krantz, L.; Keller, R. A.; Goodwin, P. M. J. Biotechnol. 2003, 102, 1-14.
Figure 2. Scatter plot of burst transit time versus burst size. Each dot is a detected single molecule of FynSH3. The red dots represent data taken by diffusion through a confocal volume element, whereas the green dots represent proteins detected by hydrodynamically focused flow through a larger, picoliter-sized optical probe volume. A probability histogram of burst sizes is shown in the bottom of the figure, and a probability histogram of transit times is shown to the left. The data collected in flow lead to longer average transit times and, consequently, more photons detected per single molecule.
experiments, the sample is introduced through a fused-silica capillary ground to a conical tip (New Objective, Cambridge, MA) ∼40-µm i.d., and the sheath channel is a flow cytometry cuvette that has a 250 × 250 µm square bore. One side of the cuvette is ground down to coverslip thickness to enable the collection of fluorescence emission using a high NA objective (NSG Precision Cells, White Plains, NY). Fluorescence excitation is provided by 10 mW, average power, from a mode-locked argon ion laser (496 nm, ∼100-ps pulse width, 82-MHz pulse repetition rate; model 2080, Spectra Physics) focused with a 50-mm-focal length lens to a ∼10 µm (1/e2 diameter) spot 100 µm downstream from the capillary orifice. The excitation laser beam was orthogonal to both the sample stream and the optical collection axis of the system, with the laser polarization pointing down the optical collection axis, which reduces artifacts caused by chromophore orientation and minimizes Raman scatter.7 A volumetric sheath flow rate of 5 µL/min yielded a linear flow velocity of ∼2.7 mm/s in the center of the flow channel and a transit time of 3.7 ms across the 1/e2 width of the optical probe volume. Fluorescence from the proteins was collected and by a 1.2 NA, 60× water immersion microscope objective (Nikon CFN Plan Apochromat). The collected fluorescence passes through a spatial filter in the image plane of the objective (a slit of dimensions 1 mm by 0.5 mm, long axis oriented parallel to the flow) and is split spectrally between two beam paths
by a 560-nm dichroic reflector, (560DRLP, Omega Optical). Fluorescence detection employed two single-photon counting avalanche photodiodes (Perkin-Elmer Optoelectronics) masked using band-pass filters appropriate for the donor (30-nm bandpass centered at 530 nm, 530DF30, Omega Optical) and acceptor (30nm bandpass, centered at 630 nm, Omega Optical) emission. Raw photon counts were recorded using a Becker & Hickl SPC630 photon counting card and router. Time Gating. For data collected in flow, time gating was used to reduce the contribution of Raman scatter as described previously.1,21,22 Only photons that arrived >1 ns after the mode-locked excitation laser pulse were used for data analysis. We note that while time gating reduces the contribution of Raman scattering to these measurements, the gate also reduces the number of donor and acceptor photons measured during a single molecule fluorescence burst. In particular, for fluorophores with a short fluorescent lifetime (e.g., a donor undergoing efficient energy transfer to an acceptor), the time gate can substantially reduce the number of photons measured in this channel for a given molecule. This reduction in measured photons was not accounted for here, but in principle can be, using similar methods to account for the difference in detection efficiencies for the two detection channels. We note that the average Eapp for the data collected in flow (0.85), which uses a time gate, is similar to that collected in Analytical Chemistry, Vol. 79, No. 9, May 1, 2007
3511
Figure 4. Comparison of measured energy transfer distribution widths versus the shot-noise prediction. While the measured distribution narrows with increasing threshold in roughly the manner expected, the width of the distribution is broader than the shot-noise prediction. At low thresholds (∼20), most of the width of the spFRET distribution can be attributed to shot noise. For measurements at large photon thresholds (>100), the dominant broadening mechanism of the distribution is no longer shot noise on the number of detected photons. Residual width remains due to photophysical mechanisms of broadening (e.g., triplet states) or due to other technical noise sources. Figure 3. Single molecule energy transfer distributions measured by diffusion through an open confocal volume element (A,B) and via focused flow (C,D). Raising the threshold used to generate the energy transfer distribution substantially cuts into the number of molecules for the data collected by diffusion through an open confocal volume element. In contrast, meaningful energy transfer distributions can be generated from the flow data using thresholds as high as 150 photoelectrons. The width of the histograms in (D) is roughly half that of (A).
the confocal apparatus (0.83), which was collected without a time gate. Burst Sifting and Energy Transfer Efficiency Calculations. We search for fluorescence bursts in the raw photon data using algorithms described in detail elsewhere.23 In brief, the interphoton arrival times are smoothed (5-point Fourier smoothing), and runs of consecutive photons that arrive below a threshold value of clock ticks are identified as photon bursts. The threshold used in these experiments (both for the confocal and the time-gated data taken in flow) was 3000 ticks of a 20-MHz clock (i.e., less than 150 µs must elapse between successively detected photons to be considered a burst). We record the number of photons (burst size) as well as the total time the burst was below our sift threshold (the transit time) for each single molecule detected. Photon bursts were located in the acceptor and the donor channel in an independent fashion, and the apparent energy transfer was calculated for each burst via eq 1. Background counts were ∼4800 (donor) and ∼2500 Hz (acceptor) for the time-gated data collected in flow, while the confocal data had backgrounds of 1700 (donor) and 2000 Hz (acceptor). These counts were not subtracted or corrected for in construction of the Eapp histogram. RESULTS AND DISCUSSION The data collected in flow lead to longer transit times and consequently more detected photons. Figure 2 shows a scatter (23) Goodwin, P. M.; Affleck, R. L.; Ambrose, W. P.; Jett, J. H.; Johnson, M. J.; Martin, J. C.; Petty, J. T.; Schecker, J. A.; Wu, M.; Keller, R. A. In Computer Assisted Analytic Spectroscopy; Brown, S., Ed.; John Wiley & Sons: New York, 1996.
3512 Analytical Chemistry, Vol. 79, No. 9, May 1, 2007
plot of the transit time versus burst size (donor + acceptor photons) for single molecules detected in a hydrodynamically focused sample stream (green) or via diffusion through an open confocal volume element (red). Each dot in the scatter plot represents a single detected molecule. The bottom of Figure 2 shows histograms ofthe burst size for the two data sets, while the left-hand side of this figure shows histograms of the transit times of the single molecules for the two experimental methods. The data collected in flow yield a longer transit time (2.2 vs 1.2 ms) and consequently more detected photons for an “average” burst (85 vs 42). This increase in the number of photons detected from an “average” molecule enables the construction of energy transfer efficiency histograms using larger photon thresholds than is possible with the data collected via diffusion through an open confocal volume element. This is demonstrated in Figure 3, which shows Eapp histograms obtained for an increasing value of the photon threshold. For the data collected via diffusion through an open confocal volume element, raising the threshold from 25 to 50 photoelectrons drops the number of detected single molecules by a factor of 5 (Figure 3A). Increasing the threshold further to 100 photons totally eliminates the heterolabeled peak from the Eapp distribution (Figure 3B). For the confocal data, shot noise on the number of molecules examined, not only on the number of photoelectrons measured per molecule, starts to contribute to the uncertainty in the measured distribution for thresholds of >50 PE. For this reason, most studies that use diffusion through an open confocal volume element use modest thresholds to construct the spFRET histogram, such as 25 11,14 or 30.12 In contrast to the diffusion through an open confocal volume element, for the data collected in flow, raising the threshold from 25 to 50 photoelectrons barely limits the number of molecules used to construct the spFRET distribution (Figure 3C). Moreover, there are plenty of molecules left in the tail of the distribution to
construct meaningful energy transfer efficiency distributions for thresholds as large as 100-150 photoelectrons (Figure 3D). The increase in photon number substantially narrows the width of the spFRET distribution. Figure 4 compares the width of the energy transfer efficiency distribution measured in flow versus the width of the distribution measured via diffusion through a confocal volume element. These widths were determined by Gaussian fits to the Eapp histograms. Overlaid with these experimentally determined standard deviations is the shot-noise-limited standard deviation (eq 2). As expected, as the threshold used to generate the energy transfer distribution is increased, the width of the energy transfer distribution decreases with a 1/NT1/2 dependence. Note that, at all values of the threshold employed, the flow data have a narrower energy transfer distribution than that obtained via diffusion through an open confocal volume element. Moreover, there appears to be an almost constant offset between the shot-noise predicted width and the measured spFRET histogram width for the data collected in flow (Figure 4). Some other source of noise, either technical or photophysical in nature, is the reason for this offset. In particular, as discussed by Gopich and Szabo,20 additional broadening mechanisms include triplet shelving of the acceptor and the heterogeneity introduced by random diffusion through the excitation probe volume. While the method used here minimizes the heterogeneity due to different paths through the excitation laser, some heterogeneity still exists, due to either diffusional broadening of the hydrodynamically focused sample stream,24 sample stream drift, or incomplete hydrodynamic focusing of the analyte.21 While this source of noise may not be thus entirely eliminated in the focused flow methods, it is substantially suppressed in comparison to the random trajectories through the probe volume that result from diffusion alone. (24) Demas, J. N.; Wu, M.; Goodwin, P. M.; Affleck, R. L.; Keller, R. A. Appl. Spectrosc. 1998, 52, 755-762. (25) Deniz, A. A.; Laurence, T. A.; Dahan, M.; Chemla, D. S.; Schultz, P. G.; Weiss, S. Ann. Rev. Phys. Chem. 2001, 52, 233-253. (26) Lipman, E. A.; Schuler, B.; Bakajin, O.; Eaton, W. A. Science 2003, 301, 1233-1235. (27) Regenfuss, P.; Clegg, R. M.; Fulwyler, M. J.; Barrantes, F. J.; Jovin, T. M. Rev. Sci. Instrum. 1985, 56, 283-290. (28) Knight, J. B.; Vishwanath, A.; Brody, J. P.; Austin, R. H. Phys. Rev. Lett. 1998, 80, 3863-3866. (29) Werner, J. H.; Cai, H.; Keller, R. A.; Goodwin, P. M. Biophys. J. 2005, 88, 1403-1412. (30) Pollack, L.; Tate, M. W.; Darnton, N. C.; Knight, J. B.; Gruner, S. M.; Eaton, W. A.; Austin, R. H. Proc. Natl. Acad. Sci. U.S.A. 1999, 96, 10115-10117.
CONCLUSIONS This work has primarily emphasized that focused flow methods can increase the observation time and consequently the number of photons detected from single FRET pair labeled protein samples, leading to more precise spFRET measurements. However, there are other advantages to using single molecule flow “cytometry” to observe single molecule biomolecular conformation. First, as with diffusional studies,25 this method does not require immobilization of the species under investigation. Second, in contrast to diffusion-based methods, this technique enables direct comparisons between different molecules, as every molecule is probed in a nearly identical fashion. Third, though not exploited here, the experimental equipment and methods are well suited to explore single molecule dynamics and kinetics.26 In particular, the small orifice of the sample introduction capillary enables rapid mixing between the sample and sheath fluid using diffusional, rather than turbulent, mixing.27,28 One can put a chemical denaturant26 or cofactor29 in the sheath stream that is absent in the sample stream. The denaturant concentration is then “jumped” at the sample orifice, and the time after the jump is reflected in the distance downstream from the orifice.27,28,30 While such rapid mixing experiments have been combined with spFRET before,26 they lacked the resolution provided by the current focused flow approach, as the optical probe volume was a near-diffractionlimited confocal volume element and most molecules skirt the edge, rather than traverse the full width, of the probe volume.21 Finally, we note that energy transfer is often referred to as a “spectroscopic ruler”.17 In this metaphor, the width of the single molecule energy transfer distribution reflects the width of the tick marks of the ruler. This work demonstrates that narrower tick marks on this ruler and better measurement precision can be obtained by using flow-based methods (Figure 3C,D) rather than by diffusion through an open confocal volume element (Figure 3A,B). This increase in resolution has promise in the exploration of the finer details of the heterogeneity in structure of biological molecules. ACKNOWLEDGMENT This work was supported by the Los Alamos National Laboratory LDRD program (J.H.W., R.A.K., P.M.G.) and by CULAR and NIH (RO1GM62868-01A2) funds (K.W.P., E.R.M.). Received for review January 24, 2007. Accepted February 19, 2007. AC070142C
Analytical Chemistry, Vol. 79, No. 9, May 1, 2007
3513
