Two-Dimensional Fluorescence Correlation in Capillary

Anal. Chem. 2000, 72, 4531-4542

Accelerated Articles

Two-Dimensional Fluorescence Correlation in Capillary Electrophoresis for Peak Resolution and Species Identification Gufeng Wang and Lei Geng*

Department of Chemistry, University of Iowa, Iowa City, Iowa 52242

A new spectroscopic dimensionsfluorescence intensity correlationsis introduced to enhance peak resolution and species identification in capillary electrophoresis. In twodimensional correlation CE, a conventional electropherogram is spread into two dimensions through crosscorrelation analysis of fluorescence time response. A laser that is sinusoidally modulated in intensity is used as the excitation source. Three channels of information are collected during a CE run: the steady-state intensity, the ac amplitude, and the phase-resolved fluorescence intensity. The correlation between two chosen channels is then evaluated. A two-dimensional correlation electropherogram consists of a plot of the correlation intensity versus two axes of migration time. Through correlation analysis, species discrimination and peak resolution are significantly enhanced without having to physically separate the solutes. Two-dimensional correlation CE showed complete resolution between two overlapping sample peaks with a resolution of 0.28 in the conventional onedimensional electropherogram. In separations of polycyclic aromatic hydrocarbons by micellar electrokinetic chromatography (MEKC), two-dimensional correlation analysis resolved all overlapping elution peaks unseparable by one-dimensional MEKC, demonstrating the utility of 2D correlation in separation method development. The capability of 2D correlation CE in species identification is demonstrated with a sequence of 39 consecutively injected peaks containing four fluorescent dyes. Species identification in sequencing is achieved without complex data treatment in two-dimensional correlation CE. Two constant motivations in the development of detection techniques for chemical separations are resolution enhancement and species identification.1-18 Peak overlap is common in the initial 10.1021/ac000534i CCC: $19.00 Published on Web 09/01/2000

© 2000 American Chemical Society

stages of method development in capillary electrophoresis and chromatography. Even after optimization of the separation conditions, electropherograms or chromatograms of multicomponent mixtures may still contain overlapped peaks. In these situations, the detector can provide additional molecular information that aids the resolution of unseparated solutes. Species identification is another motivation that drives the development of detection (1) Ding, J.; Vouros, P. Anal. Chem. 1999, 71, 378A-385A. (2) Bayer, E.; Albert, K.; Nieder, M.; Grom, E.; Keller, T. J. Chromatogr. 1979, 186, 497-507. (3) Wu, N.; Peck, T. L.; Webb, A. G.; Magin, R. L.; Sweedler, J. V. J. Am. Chem. Soc. 1994, 116, 7929-7930. (4) Olson, D. L.; Peck, T. L.; Webb, A. G.; Magin, R. L.; Sweedler, J. V. Science 1995, 270, 1967-1970. (5) Lacey, M. E.; Subramanian, R.; Olson, D. L.; Webb, A. G.; Sweedler, J. V. Chem. Rev. 1999, 99, 3133-3152. (6) Walker, P. A., III; Morris, M. D. J. Chromatogr., A 1998, 805, 269-275. (7) Walker, P. A., III; Morris, M. D.; Burns, M. A.; Johnson, B. N. Anal. Chem. 1998, 70, 3766-3769. (8) Sweedler, J. V.; Shear, J. B.; Fishman, H. A.; Zare, R. N.; Scheller, R. H. Anal. Chem. 1991, 63, 496-502. (9) Cheng, Y.-F.; Piccard, R. D.; Vo-Dinh, T. Appl. Spectrosc. 1990, 44, 755765. (10) Karger, A. E.; Harris, J. M.; Gesteland, R. F. Nucleic Acids Res. 1991, 19, 4955-4962. (11) Timperman, A. T.; Khatib, K.; Sweedler, J. V. Anal. Chem. 1995, 67, 139144. (12) Oldenburg, K. E.; Xi, X.; Sweedler, J. V. J. Chromatogr., A 1997, 788, 173183. (13) Carson, S.; Cohen, A. S.; Belenkii, A.; Ruin-Martinez, M. C.; Berka, B. L.; Karger, B. L. Anal. Chem. 1993, 65, 3219-3226. (14) Desilets, D. J.; Kissinger, P. T.; Lytle, F. E. Anal. Chem. 1987, 59, 18301834. (15) He, H.; Nunnally, B. K.; Li, L.-C.; McGown, L. B. Anal. Chem. 1998, 70, 3413-3418. (16) Soper, S. A.; Legendre, B. L., Jr.; Willams, D. C. Anal. Chem. 1995, 67, 4358-4365. (17) Lieberwirth, U.; Arden-Jacob, J.; Drexhage, K. H.; Herten, D. P.; Muller, R.; Neumann, M.; Schulz, A.; Siebert, S.; Sagner, G.; Klingel, S.; Sauer, M.; Wolfrum, J. Anal. Chem. 1998, 70, 4771-4779. (18) Dvorak, M. A.; Oswald, G. A.; Van Benthem, M. H.; Gillispie, G. D. Anal. Chem. 1997, 69, 3458-3464.

Analytical Chemistry, Vol. 72, No. 19, October 1, 2000 4531

techniques. The most common method for species identification in chemical separations is based on elution or migration times. Comparison of migration times of the mixture peaks with those of solute standards provides identities of species in the mixture. This method requires strict control of experimental conditions to ensure reproducibility of migration times between the standard and sample runs. In addition, multiple runs of the single solutes are necessary to generate migration time standards. The capability of identifying the eluting components with molecular properties other than the migration time can significantly relax the requirement of reproducibility in migration times, since the chemical identities of the solutes can be recovered in a single electrophoretic or chromatographic run. In other applications such as DNA sequencing, the species identification or base calling has to be accomplished with spectroscopic properties of the species since the migration time can no longer be used for this purpose.19, 20 Various detection methods have been devised for the identification of eluting species in chemical separations. Mass spectrometry1 and nuclear magnetic resonance2-5 detectors provide detailed information about the structure of the eluting molecule and can be ideal detectors for species identification. The limitations of MS and NMR detectors include complex and expensive instrumentation and extensive, time-consuming postexperimental data manipulation. In addition, despite significant advances made in recent years to improve the sensitivity,3-5 NMR detectors are still limited to fairly high concentrations and low time resolution as the result of the long measurement time necessary to attain an adequate signal-to-noise ratio. Among optically based detectors, Raman detection would be uniquely suited for species identification since the group frequencies for vibrational modes obtained in a Raman spectrum can aid the structural elucidation of solute molecules.6,7 The low Raman cross sections, however, make it difficult to detect small quantities of solutes present in capillary electrophoresis that are often below femtomole levels. Spectrally resolved fluorescence detection also offers the capability of species identification.8-13 Extensive spectrum fitting is required to match the sample spectra to standard spectra for peak resolution and species identification. Time-resolved fluorescence has been developed by several groups to facilitate peak resolution and species identification in chromatography and capillary electrophoresis.14-18 Early experiments aimed at background reduction with gated integration of fluorescence.21-24 Since then, the fluorescence lifetimes for elution peaks have been measured in both the time domain14,16-18,25-30 (19) Smith, L. M.; Sanders, J. Z.; Kaiser, R. J.; Hughes, P.; Dodd, C.; Connell, C. R.; Heiner, C.; Kent, S. B. H.; Hood, L. E. Nature 1986, 321, 674679. (20) Prober, J. M.; Trainor, G. L.; Dam, R. J.; Hobbs, F. W.; Robertson, C. W.; Zagursky, R. J.; Cocuzza, A. J.; Jensen, M. A.; Baumeister, K. Science 1987, 238, 336-341. (21) Richardson, J. H.; Larson, K. M.; Haugen, G. R.; Johnson, D. C.; Clarkson, J. E. Anal. Chim. Acta 1980, 116, 407-411. (22) Imasaka, T.; Ishibashi, K.; Ishibashi, N. Anal. Chim. Acta 1982, 142, 1-12. (23) Furura, N; Otsuki, A. Anal. Chem. 1983, 55, 2407-2413. (24) Iwata, T.; Senda, M.; Kurosu, Y.; Tsuji, A.; Maeda, M. Anal. Chem. 1997, 69, 1861-1865. (25) Ishibashi, K.; Imasaka, T.; Ishibashi, N. Anal. Chim. Acta 1985, 173, 165175. (26) Kawabata, Y.; Sauda, K.; Imasaka, T.; Ishibashi, N. Anal. Chim. Acta 1988, 208, 255-262. (27) Burt, J. A.; Dvorak, M. A.; Gillispie, G. D.; Oswald, G. A. Appl. Spectrosc. 1999, 53, 1496-1501.

4532

Analytical Chemistry, Vol. 72, No. 19, October 1, 2000

and the frequency domain.15,31-34 The solutes are identified by their corresponding fluorescence lifetimes. An inhomogeneous peak in the chromatogram or electropherogram is indicated by varying fluorescence lifetime across the peak. The overlapping peaks can be resolved with multiexponential fit of the fluorescence decay profile. DNA sequencing is one of the most important applications requiring species identification. Since its introduction into the application,19,20 laser-induced fluorescence has become the most widely used detection method for DNA sequencing.10,13,35-40 Four fluorescent tags of different colors were used for the identification of the four nucleotide bases A, C, T, and G. Base calling was achieved by monitoring the colors of dye fluorescence with four detection channels.19 Several encoding schemes have been devised since with the objective of simplifying instrumentation and reducing the number of detection channels, including a four-dyetwo-channel ratiometric method,20,35 binary encoding,36 and fluorescence resonance energy transfer (FRET)39 method. All these methods require more than two dyes and more than two detection channels, or a multichannel detector can be used to collect the entire emission spectrum. Any method that uses more than two dyes can suffer from differential mobility shifts caused by size differences between the dyes.40 In addition, the overlap of fluorescence emission spectra of the dyes can introduce cross talk between different detection channels and pose a problem in base calling. By carefully choosing the dyes and linkers, differential mobility shifts were minimized in FRET detection39 but cross talk between channels persists since the base calling is still based on differentiation in the emission spectra of the four FRET acceptors. A method requiring a single dye, single lane, and single detection channel would be an ideal detector for DNA sequencing. Besides being immune to the problems of differential mobility shifts and spectral cross talk, such a method provides much desired simplification of instrumentation and chemical processingsonly one sequencing reaction needs to be run instead of four that are required in current methods of detection. A single-dye method has been devised that distinguishes between A, C, T, and G with different intensity levels controlled by concentration levels in the enzymatic sequencing reactions.41-43 This method has a low accuracy in base calling due to the variations in the efficiency of enzymatic reactions. (28) Miller, K. J.; Lytle, F. E. J. Chromatogr. 1993, 648, 245-250. (29) Zhang, Y.; Soper, S. A.; Middendorf, L. R.; Wurm, J. A.; Erdmann, R.; Wahl, M. Appl. Spectrosc. 1999, 53, 497-504. (30) Latva, M.; Ala-Kleme, T.; Bjennes, H.; Kankare, J.; Haapakka, K. Analyst 1995, 120, 367-372. (31) Cobb, W. T.; McGown, L. B. Appl. Spectrosc. 1987, 41, 1275-1279. (32) Smalley, M. B.; Shaver, J. M.; McGown, L. B. Anal. Chem. 1993, 65, 34663472. (33) Smalley, M. B.; McGown, L. B. Anal. Chem. 1995, 67, 1371-1376. (34) Li, L.-C.; McGown, L. B. Anal. Chem. 1996, 68, 2737-2743. (35) Li, Q.; Yeung, E. S. Appl. Spectrosc. 1995, 49, 1528-1533. (36) Huang, X. C.; Quesada, M. A.; Mathies, R. A. Anal. Chem. 1992, 64, 967972. (37) Ueno, K.; Yeung, E. S. Anal. Chem. 1994, 66, 1424-1431. (38) Swerdlow, H.; Zhang, J. Z.; Chen, D. Y.; Harke, H. R.; Grey, R.; Wu, S.; Dovichi, N. J.; Fuller, C. Anal. Chem. 1991, 63, 2835-2841. (39) Ju, J.; Ruan, C.; Fuller, C. W.; Glazer, A. N.; Mathies, R. A. Proc. Natl. Acad. Sci. U.S.A. 1995, 92, 4347-4351. (40) Tan, H.; Yeung, E. S. Electrophoresis 1997, 18, 2893-2900. (41) Tabor, S.; Richardson, C. C. Proc. Natl. Acad. Sci. U.S.A. 1987, 84, 47674771. (42) Tabor, S.; Richardson, C. C. J. Biol. Chem. 1990, 265, 8322-8328. (43) Willams, D. C.; Soper, S. A. Anal. Chem. 1995, 67, 3427-3432.

On-line fluorescence lifetime detection has been actively explored as a method for base calling in DNA sequencing15,17,44-48 with the eventual objective of developing a single-dye, single-lane, single-detection channel method.47 The technique relies upon different fluorescence lifetimes for the four tags used to label bases A, C, T, and G. The first example of fluorescence lifetime detection in a DNA sequencing run has been reported recently that demonstrated over 90% accuracy for a capillary gel electrophoresis of 660 bp.17 The disadvantage of current approaches of timeresolved fluorescence detection is the extensive, time-consuming postexperimental data analysis. The required data treatment steps of setting a threshold in integrated intensity, introducing a suitable delay time according to the properties of each peak, and choosing an appropriate decay model all necessitate the judicial judgment of an experienced spectroscopist. The data treatment is followed by time-consuming point-by-point fitting of the fluorescence decay profiles with nonlinear least-squares (NLLS) or maximum likelihood estimator (MLE) methods. In a general application such as DNA sequencing that will be routinely performed in hospitals and clinics in the future, a simple method that does not require timeconsuming data analysis and specialized spectroscopic training of the operator is desired. In this paper, we introduce a new fluorescence detection principle for capillary electrophoresis based on generalized fluorescence correlation between migration times. The correlation function gives rise to a two-dimensional electropherogram that is a plot of the correlation intensity versus two axes of migration times. Using two-dimensional correlation capillary electrophoresis, baseline resolution has been achieved for overlapping peaks with a resolution of 0.28 in conventional onedimensional capillary electrophoresis. Species identification is accomplished by examining peaks with vanishing asynchronous cross-correlation intensity. The potential of two-dimensional fluorescence correlation capillary electrophoresis (2D CCE) in DNA sequencing application is demonstrated by “base calling” of synthetic “sequences” made up of four fluorescent dyes of different lifetimes. Peak resolution and species identification are achieved in 2D CCE through very simple data analysis without any special data treatment. THEORY 2D CCE enhances resolution by spreading a conventional electropherogram into two dimensions using generalized fluorescence correlation. The method originates from the time correlation function, which can be evaluated in either the time or the frequency domain. We will construct the framework of 2D CCE with an impulse excitation and then discuss the correlation functions with a sinusoidal excitation in the frequency domain. A generalized form will then be introduced for convenient implementation in capillary electrophoresis and better signal-to-noise ratio. To evaluate the time correlation function of fluorescence intensity between two migration times, an external field is (44) Chang, K.; Force, R. K. Appl. Spectrosc. 1993, 47, 24-29. (45) Nunnally, B. K.; He, H.; Li, L.-C.; Tucker, S. A.; McGown, L. B. Anal. Chem. 1997, 69, 2392-2397. (46) Li, L.-C.; He, H.; Nunnally, B. K.; McGown, L. B. J. Chromatogr., B 1997, 695, 85-92. (47) Flanagan, J. H., Jr.; Owens, C. V.; Romero, S. E.; Waddell, E.; Kahn, S. H.; Hammer, R. P.; Soper, S. A. Anal. Chem. 1998, 70, 2676-2684. (48) Li, L.; McGown, L. B. J. Chromatogr. A. 1999, 841, 95-103.

introduced to perturb the molecules being separated in capillary electrophoresis. The temporal response of the fluorescence intensity at migration time t is I(τ;t); τ is the time axis on which intensity fluctuation occurs. Compared to the migration times, which are on the order of seconds and minutes in common capillary electrophoresis, the time scale of the intensity fluctuation is negligibly short, on the order of picoseconds to microseconds. The source of fluorescence fluctuation can be quite general. Any kinetic process the molecule is undergoing can serve as the source of fluctuation. For example, the fluctuation can be caused by molecules diffusing out of and back into the detection volume, by the rotational reorientation of the molecules when linearly polarized emission is detected or by the fluorescence decay after excitation. The parameters that control these fluctuations are the diffusion constants, the rotational correlation times, and the fluorescence lifetimes, respectively. In this paper, we will focus our attention on fluorescence fluctuation caused by the natural deactivation of excited-state fluorescent molecules. The measured signal at each migration time in this case will be the fluorescence decay. Specifically, I(τ;t) is the fluorescence intensity at time τ after an impulse excitation, monitored at migration time t in capillary electrophoresis. The two-dimensional time correlation function between migration times t1 and t2 can be evaluated by49

1 C(τ;t1,t2) ) lim Tf∞2T

∫

T

-T

I*(τ;t1)I(τ + δt2) dδ

(1)

where the asterisk denotes complex conjugate of the fluorescence intensity response. When the fluorescence decays with similar rates at times t1 and t2, a strong correlation will be observed between these two migration times. On the other hand, if the fluorescence decays at t1 and t2 are not related, the amplitude of the time correlation function will be small. In the extreme case where fluorescence decays with a lifetime of zero at time t1 and a lifetime of infinity at time t2, the correlation between t1 and t2 approaches zero. Experimentally, the correlation function can be obtained by monitoring fluorescence intensity decays across the entire time axis of an electropherogram in the time domain. A pulsed laser is used for excitation, and the fluorescence intensity decay can be collected with time-correlated single photon counting, or with a gigahertz oscilloscope. A more convenient way to evaluate the two-dimensional time correlation function is to excite the solute molecules with a sinusoidally modulated light beam A:

A ) A0eiωτ

(2)

where A0 is the amplitude of the exciting light and ω is the angular modulation frequency. The fluorescence response,

I ) I0ei(ωτ-φ)

(3)

which is also sinusoidally modulated at the same frequency, but is delayed in phase by an angle of φ, is measured. Substitution of this expression of fluorescence intensity into eq 1 leads to the time correlation function at frequency ω: (49) Geng, L.; He, Y.; Cox, J., submitted.


4533

C(ω;t1,t2) ) [Φ(ω;t1,t2) - iΨ(ω;t2,t2)]eiωτ

(4)

The real part of the correlation function Φ(ω;t1,t2) is defined as the synchronous correlation intensity:

1 Φ(ω;t1,t2) ) [IR(t1)IR(t2) + II(t1)II(t2)] 2

(5)

and the imaginary part Ψ(ω;t1,t2) is defined as the asynchronous correlation intensity:

1 Ψ(ω;t1,t2) ) [IR(t1)II(t2) - II(t1)IR(t2)] 2

(6)

between migration times t1 and t2 at frequency ω. IR(t) and II(t) are the in-phase and quadrature fluorescence intensities, or the real and imaginary parts of the fluorescence response, at migration time t in capillary electrophoresis. They are related to the steadystate fluorescence intensity I(t) by

IR(t) ) KI(t)m cos φ

(7)

II(t) ) KI(t)m sin φ

(8)

and

where K is an instrument factor and m is the modulation ratio of the fluorescence emission. A two-dimensional correlation electropherogram is a plot of the correlation intensity versus two axes of migration time t1 and t2. The plots of Φ(t1,t2) and Ψ(t1,t2) are termed synchronous and asynchronous electropherograms, respectively. EXPERIMENTAL SECTION Chemicals. All chemicals were obtained at the highest purity available and used as received. Anthracene was purchased from Sigma (St. Louis, MO). Benzo[k]fluoranthene (B(k)F) was obtained from Fluka (Milwaukee, WI). Pyrene, benzo[a]pyrene (B(a)P) and 1,4-bis(5-phenyloxazol-2-yl)benzene (POPOP) were purchased from Aldrich (Milwaukee, WI). Buffer solutions were prepared with ultrapure water generated by a MilliQ system (MilliQ-Plus, MilliPore Corp., Bedford, MA). Capillary Electrophoresis. Capillary electrophoresis was performed with an in-house CE instrument that is described elsewhere.50 A high-voltage power supply (CZE1000PN30, Spellman, Hauppauge, NY) was controlled with a program written inhouse in Component-Works (National Instruments, Austin, TX). The current was constantly monitored with a picoammeter (model 486, Keithley Instruments Inc., Cleveland, OH), and the values were transferred to the instrument control program through a data acquisition board purchased from National Instruments. Samples were electrokinetically injected. The total length of the capillary was 50 cm, with a distance of 32 cm between the inlet and the detection window. The inner diameter of the capillary was 50 µm. These conditions were used in all experiments. (50) He, Y.; Geng, L., manuscript in preparation.

4534


Phase-Resolved Fluorescence Detection. A HeCd laser provided the excitation of fluorescence at 325 nm. A spectrofluorometer (SLM 48000MHF, Spectronic Instruments, Inc., Rochester, NY) was used for the phase-resolved fluorescence detection. Laser line mirrors were used to direct the beam and to block the plasma emission. The excitation laser beam was sinusoidally modulated with a Pockels cell and focused into the separation capillary mounted on a three-dimensional translation stage. A modulation frequency of 17 MHz was used throughout the experiments. Fluorescence emission was collected with an f/0.8 fused-silica lens. Scattered laser light was isolated with a laser line filter centered at 325 nm, with a bandwidth of 10 nm (10LF10325, Newport Corp., Irvine, CA) and was used as the reference to set the 0° phase angle. During electrophoresis runs, the collected emission was passed through a 345-nm long pass filter to minimize the scattered laser light in fluorescence measurement. The filtered emission was detected by a photomultiplier tube that was synchronously modulated at the same frequency as the laser. The phase angle between the excitation and the detector was varied to optimize the signal-to-noise ratio. The signal from the PMT was integrated to yield the phase-resolved fluorescence intensity at a data acquisition rate of 1 Hz. The ac amplitude and the dc intensity were also collected during the CE run, with an ac tuned amplification and an integration circuit, respectively.51 Three one-dimensional electropherograms, plots of the dc intensity, the ac amplitude and the phase-resolved intensity as a function of migration time, were collected in this detection scheme and will be referred to as the DC, AC, and PR electropherograms. The dc intensity-based electropherogram is identical to one that is obtained in conventional one-dimensional capillary electrophoresis. The two-dimensional correlation electropherogram was calculated from one-dimensional electropherograms in MatLab (MathWorks, Inc., Natick, MA) according to eqs 5 and 6. RESULTS Two-Dimensional Synchronous and Asynchronous Correlation Electropherograms. The DC, AC, and PR electropherograms measured for a consecutive injection of pyrene and anthracene are shown in Figure 1. The DC electropherogram is identical to the conventional one-dimensional electropherogram collected with fluorescence intensity detection. The intensity ratio between the pyrene and the anthracene peaks in the AC electropherogram is different from that in the DC electropherogram due to different fluorescence lifetimes of the two solutes. The ac signal is weighted by the modulation ratio of fluorescence that is determined by the fluorescence lifetime of the solute. The PR signal is weighted by a factor that depends on both the lifetime of the solute and the detector phase angle or the phase delay of the detector from the exciting beam in the heterodyne detection. This detector phase angle can be varied to enhance signal-to-noise ratio of the measurement. The phase-resolved fluorescence intensity, 52

IPR ) KI0m cos(φ - φD)

(9)

is maximized when the detector phase angle φD is equal to the (51) Jameson, D. M.; Gratton, E.; Hall, R. D. Appl. Spectrosc. Rev. 1984, 20, 55-106.

Figure 1. Three electropherograms collected in phase-resolved fluorescence detection: DC, dc intensity; AC, ac amplitude; PR(315°), phase-resolved fluorescence intensity measured at a detector phase angle of 315°. PR(45°), or the phase-resolved fluorescence intensity measured at a detector phase angle of 45°, was calculated from the AC and the PR(315°) electropherograms. Pyrene (first peak at 90 s) and anthracene (second peak at 110 s) were electrokinetically injected for 5 s at 25 kV, with 20 s of buffer injection in between. Electrophoresis was driven at 25 kV. The solvent was 50% acetonitrile and 50% 4 mM borate buffer at pH 9.0. The concentrations for pyrene and anthracene were 20 and 40 µM, respectively. Electropherograms were displaced on the intensity axis for the clarity of display.

phase delay of the fluorescent molecule and is nulled when the detector is 90° out of phase from the solute. Strictly, the in-phase and quadrature fluorescence intensities should be measured at detector phase angles of 0° and 90°, respectively. In practice, the phase angle of the detector is determined by considering the signal-to-noise ratio of the measurement and experimental convenience. In all experiments reported in this paper, the detector was set to 315° phase-delayed from the excitation in data collection during a CE run. At the modulation frequency of 17 MHz, the phase delay is 25.3° and 79.5° for anthracene and pyrene, respectively. The PR electropherogram thus shows negative intensity for pyrene and a positive peak for anthracene. This electropherogram is defined as the in-phase fluorescence intensity. The quadrature electropherogram was calculated according to 2

2

IR + II ) AC2

(10)

where AC is the ac amplitude. The electropherogram, a plot of the quadrature phase-resolved fluorescence intensity versus time, is displayed in Figure 1. The synchronous and asynchronous correlation intensities were evaluated according to eqs 5 and 6 and stored in two matrices with t1 and t2 as the row and column indices. Plotting the correlation intensity against these two indices yields threedimensional graphs, which we will call the synchronous (Figure 2A) and the asynchronous (Figure 2B) correlation electropherograms. In both correlation electropherograms, the correlation (52) Lakowicz, J. R. Principles of Fluorescence Spectroscopy; Kluwer/Plenum: New York, 1999.

intensity is proportional to the concentrations of both solutes that elute at t1 and t2 in the one-dimensional electropherogram. As indicated by eq 5, the synchronous correlation electropherogram is symmetric with respect to the diagonal and shows both autocorrelation and cross-correlation peaks. The asynchronous correlation electropherogram, on the other hand, is antisymmetric. For each positive correlation peak there must exist a negative peak at the reflection point with respect to the diagonal of the 2D electropherogram. More importantly, the definition of asynchronous correlation in eq 6 predicts that asynchronous correlation vanishes between two migration times where the fluorescence decays at identical rates. In other words, asynchronous correlation electropherogram shows a zero value between peaks of the same chemical species, including both autocorrelation and cross correlation. Asynchronous cross correlation only appears between species with different fluorescence lifetimes. The one-dimensional electropherograms of individual species are obtained by drawing cross sections in the two-dimensional asynchronous correlation spectrum. These are one-dimensional slices in the 2D correlation spectrum. Specifically, the cross section at 90 s is the one-dimensional electropherogram of anthracene. Pyrene electropherogram is obtained by cutting a cross section at 110 s of migration time. Both recovered component electropherograms are depicted in Figure 3, together with the one-dimensional electropherogram of the mixture. Peak tailing as appeared in the conventional electropherogram also appeared in the correlation spectra. Clearly, electropherograms of individual solutes can be resolved from the mixture electropherogram through two-dimensional correlation analysis. Generalized Correlation Electropherograms. Correlation electropherograms calculated according to eqs 5 and 6 are based on the time correlation function in fluorescence intensity responses between two migration times t1 and t2 in capillary electrophoresis. Experimentally measured signals are the in-phase and quadrature electropherograms. We can extend the definition of correlation intensities with two electropherograms that are quite general in nature. Assuming that I1 and I2 are two signals measured during a CE run, the generalized synchronous and asynchronous correlation intensities are expressed as

1 Φ(t1,t2) ) [I1(t1)I1(t2) + I2(t2)I2(t2)] 2

(11)

1 Ψ(t1,t2) ) [I1(t1)I2(t2) - I2(t1)I1(t2)] 2

(12)

and

The generalized synchronous and asynchronous correlation electropherograms will have the same properties as discussed above. Although not directly correlated to the time correlation function, the generalized synchronous and asynchronous correlation intensities contain information on fluorescence lifetimes and are effective in peak resolution and species identification. The flexibility of choosing two functions to calculate correlation electropherograms provides the opportunity to enhance the signal-tonoise ratio and allows a variety of different CE experiments to be analyzed by two-dimensional correlation. Analytical Chemistry, Vol. 72, No. 19, October 1, 2000

4535

Figure 2. Surface and contour plots of the two-dimensional synchronous (A) and asynchronous (B) fluorescence correlation electropherograms of the CE run in Figure 1. The conventional one-dimensional electropherogram is plotted on the sidebars of the contour plots to demonstrate the cross correlation peaks.

Among the three signals collected in phase-resolved fluorescence detection, the PR signal generally has the lowest signal-tonoise ratio since it is only a part of the AC signal. Correlation electropherograms calculated with the DC and the AC signals have the best S/N allowed by the experiments. The generalized correlation intensities will be used in all the 2D correlation electropherograms presented in the rest of the paper, calculated from the DC intensity and the AC amplitude measured during CE runs. Resolution of Coeluting Peaks. The ability of two-dimensional correlation electropherograms to resolve coeluting solutes is demonstrated by consecutive injections of pyrene, buffer, and anthracene. Since both pyrene and anthracene are uncharged, they migrate at the rate of electroosmotic flow and cannot be separated in normal capillary electrophoresis. Buffer was injected for 0, 3, 5, and 20 s between pyrene and anthracene to generate different lengths of buffer plug and thus different degrees of overlap between the pyrene and the anthracene peaks. The conventional one-dimensional electropherogramssplots of the dc fluorescence intensity versus the migration timesare shown in Figure 4. Pyrene and anthracene were electrokinetically injected 4536 Analytical Chemistry, Vol. 72, No. 19, October 1, 2000

Figure 3. Resolution of the mixture electropherogram into individual electropherograms of pyrene (cross section at 110 s) and anthracene (cross section at 90 s).

for 5 s at 25 kV in each run. In our instrument setup, a height differential exists between the inlet and the highest position of

Figure 4. Electropherograms (black) of consecutive injections of pyrene, buffer, and anthracene. Pyrene and anthracene were injected for 5 s except one injection of 4 s as labeled. Buffer injection times are indicated in the plot. Experimental conditions are identical to those in Figure 1. The individual electropherograms for pyrene (red) and anthracene (green) were recovered from cross sections in the twodimensional correlation spectra. Table 1. Capillary Electrophoresis of Consecutive Injections of Pyrene and Anthracene buffer injection time (s)

t1 (s)

t2 (s)

∆t (s)

width (s)

R

20 5 3 0 0 (4-s sample injection)

89 105 107 108 112

111 111 111 111 113

22 6 4 3 1

5.3 5.3 5.3 5.3 3.6

4.15 1.13 0.75 0.57 0.28

the capillary. The high potential of injection was necessary to overcome this height differential. To generate an even higher degree of peak overlap, a CE run was conducted with 4 s of pyrene and anthracene injected consecutively without a buffer plug. With 5 s of solute injection, the two species are completely separated with a 20-s buffer plug between them but are partially overlapped when the buffer was injected for 3 and 5 s. The two peaks are significantly overlapped in the one-dimensional electropherogram when pyrene and anthracene were injected without a buffer plug between them. Anthracene appeared as a shoulder in the pyrene peak. At 4-s solute injection, a single, seemingly homogeneous peak appeared, indicating a complete overlap between pyrene and anthracene. The migration times of pyrene and anthracene and the corresponding resolution R values are listed in Table 1 for all electropherograms. The resolution was evaluated with

R ) 2(t2 - t1)/(w2 + w1)

(13)

where t1 and t2 are the migration times at the center of peaks; w1 and w2 are the widths of the pyrene and the anthracene peaks at the base.

The corresponding two-dimensional asynchronous correlation electropherograms (Figure 5) exhibited nonzero cross-correlation intensities between pyrene and anthracene. Even in the apparently homogeneous peak in one-dimensional electropherogram for consecutive injection of pyrene and anthracene without buffer, the existence of two chemical species is clearly evidenced by two antisymmetric correlation peaks in the two-dimensional correlation electropherogram. For overlapping peaks, the one-dimensional electropherograms of individual solute species are found by cutting cross sections in the two-dimensional correlation plot. The recovered one-dimensional electropherograms are depicted in Figure 4. Fluorescence correlation clearly shows that the apparently homogeneous peak in one-dimensional electropherogram for 0-s buffer injection is composed of two different molecular species, whose electropherograms can be recovered by twodimensional fluorescence correlation. The resolution value between pyrene and anthracene in this CE run is 0.28 in the onedimensional electropherogram. The fact that they can be completely resolved demonstrates the resolving power of 2D correlation analysis. PAH Separation. The resolution enhancement in the twodimensional correlation capillary electrophoresis can be quite useful in the method development of CE separations. In the initial separation runs of a complex multispecies mixture, the conditions are not optimized and it is not uncommon to have more than one species coelute as one peak. Two-dimensional correlation analysis will detect these unresolved species and provide guidance for the optimization of separation conditions. An attempt to separate four polycyclic aromatic hydrocarbons (PAHs), anthracene, pyrene, B(a)P, and B(k)F, by micellar electrokinetic chromatography (MEKC) is shown in Figure 6A. Sodium dodecyl sulfate (SDS) micelles were added as the pseudostationary phase to facilitate the separation of these uncharged solutes, which would comigrate at the electroosmotic flow rate without SDS micelles. The interaction of PAH molecules with the pseudostationary phase provided the basis of separation of these neutral molecules. The one-dimensional electropherogram shows two features at 236 and 252 s (Figure 6). The PR signal disclosed that the first feature in the electropherogram is the result of two overlapping species, splitting them into two peaks of opposite signs. The second feature at 252 s stayed as a symmetric homogeneous peak, seemingly consisting of a single solute species. The two-dimensional asynchronous correlation electropherogram displays many cross-correlation peaks (Figure 6B), but it is clear that there are four solute species in the CE run, upon close examination of the correlation pattern. A cross section at the migration time of a certain correlation peak gives a onedimensional electropherogram of the other three species. For example, the cross section at 231 s shows a 1D electropherogram consisting of a single peak at 236 s, for pyrene, and a peak at 252 s, which is generated by the overlap of B(a)P and B(k)F migration peaks. The cross section at 252 s displays three peaks at 231, 236, and 246 s, corresponding to the migration of anthracene, pyrene, and B(a)P (Figure 6C). With this information as guidance, the next step in CE development is to change the separation conditions to resolve B(a)P and B(k)F and further separate anthracene and pyrene. With higher concentrations of micelles and the addition of β-cyclodextrin into the buffer, complete Analytical Chemistry, Vol. 72, No. 19, October 1, 2000

4537

Figure 5. Two-dimensional asynchronous correlation electropherograms for CE runs in Figure 4: (A-D) 5-s injections of pyrene and anthracene, with 20, 5, 3, and 0 s buffer plugs between the two solutes. (E) 4-s solute injections without a buffer plug. The conventional one-dimensional electropherograms are plotted on the sidebars of the contour plots to demonstrate the cross-correlation peaks.

separation of these PAHs can be achieved with micellar electrokinetic chromatography.53,54 4538


Species Identification. The two-dimensional correlation described here is based on the different fluorescence decay rates

Table 2. Fluorescent Dyes Used in the Sequencing Run

pyrene B(a)P B(k)F POPOP

lifetimea (ns)

phase delay (deg)

50.7 24.0 8.9 1.6

79.5 68.7 43.7 9.6

a Fluorescence lifetimes were measured in 50% acetonitrile and 50% 4 mM borate buffer at pH 9.0.

Figure 7. DC, AC, and PR electropherograms of a sequence of 39 peaks. The four fluorescent dyes used are POPOP, pyrene, B(a)P, and B(k)F. The solvent was composed of 50% acetonitrile and 50% 4 mM borate buffer at pH 9.0. The sample was injected for 5 s at 15 kV. Electrophoresis was run under a voltage of 15 kV. The concentrations were 1.33, 20, 40, and 40 µM for POPOP, pyrene, B(a)P, and B(k)F, respectively.

Figure 6. Separation of four PAHs by micellar electrokinetic chromatography. (A) Three electropherograms collected during the CE run. (B) Two-dimensional asynchronous correlation electropherogram. The conventional one-dimensional electropherogram is plotted on the sidebars of the contour plots to demonstrate cross correlation peaks. The positive and negative correlation peaks are shown by “+” and “-” signs, respectively. (C) Resolution of four PAH peaks. SDS concentration, 50 mM. Solvent, 50% acetonitrile and 50% 8.5 mM borate buffer at pH 9.4. The sample was injected for 5 s at 25 kV. Electrophoresis was run under a voltage of 25 kV. The concentrations were 20, 40, 40, and 40 µM for pyrene, anthracene, B(a)P, and B(k)F, respectively.

of the solute species. Fluorescence lifetimes can be further used to provide solute identification in a CE separation. Especially of (53) Jimenez, B.; Patterson, D. G.; Grainger, J.; Liu, Z.; Gonzalez, M. J.; Marina, M. L. J. Chromatogr., A 1997, 792, 411-418. (54) Terabe, S.; Miyashita, Y.; Ishihama, Y.; Shibata, O. J. Chromatogr. 1993, 636, 47-55.

interest is the species identification, or base calling, in DNA sequencing. We generated a sequence of 39 peaks by consecutively injecting four fluorescent dyes POPOP, pyrene, B(a)P, and B(k)F at random order. To assist species identification, four different dyes were injected at the beginning as the first four “bases”. The fluorescence lifetimes of these dyes are listed in Table 2, along with the expected phase angles at the modulation frequency of 17 MHz. The three electropherograms directly measured in the CE run are depicted in Figure 7. Differing fluorescence lifetimes of the four fluorescent dyes resulted in different peak intensity patterns in the DC and the AC plots. The phase-resolved intensity is positive for some peaks and negative for the others, depending on the phase delay of the corresponding solute species. The two-dimensional asynchronous correlation electropherogram shows zero correlation intensity between peaks containing the same fluorescent dye (Figure 8). Blank areas in the correlation electropherogram are manifestations of cross positions between the same fluorescent species in the sequence and are symmetrically distributed with respect to the diagonal of the plot. The correlation pattern can be understood by examining the correlation between all 39 peaks with the first peak of the electropherogram (the last row or the first column of the 2D correlation electropherogram) as an example. The asynchronous correlation vanishes between peak 1 and peaks 1, 9, 10, 12, 16, 17, 18, 22, 30, 31, 35, and 36, indicating that all these 12 peaks are composed of the same solute species. Examination of the last four rows of the 2D correlation plot reveals the entire sequence. The one-dimensional electropherograms obtained from cross sections through the last four rows are depicted in Figure 9. Theoretically, the asynchronous correlation intensity should be Analytical Chemistry, Vol. 72, No. 19, October 1, 2000

4539

Figure 8. Two-dimensional asynchronous correlation electropherogram of the sequencing run in Figure 7. The conventional one-dimensional electropherogram is plotted on the sidebars of the contour plots to demonstrate cross-correlation peaks.

zero between two peaks of identical fluorescent species, displaying flat baseline at such locations. Experimentally, it was observed that, in general a small, polarized correlation peak appears between two peaks of identical species. Such a peak displays positive correlation intensity on one side and negative correlation intensity on the other side. This peak shape also appeared in the cross sections in the experiments of pyrene and anthracene and in the separation of PAHs. The origin of this polarized correlation shape is currently under investigation. The dc intensity and the ac amplitude were collected with two different electronic circuits, which may have generated a displacement between the two electropherograms, resulting in the polarized correlation pattern. In fact, a close examination of the DC and the AC electropherograms revealed that the AC signal is slightly delayed with respect to the DC signal, by a fraction of a second. A delay in ac amplitude measurement has also been observed in on-line fluorescence lifetime detection.55 Since the data acquisition rate in our experiments was 1 Hz, limited by the current instrumentation, exact compensation of this effect will require modification of the data 4540


acquisition system to allow higher rates. However, the existence of a polarized correlation peak is sufficient for the purpose of species identification in sequencing. The polarized correlation is thus taken to be indicative of a cross point between two peaks of identical species. Using this criterion, the polarized peaks across the last four rows of the 2D correlation reveal the complete sequence of 39 peaks, as indicated in Figure 9. In a sequencing application, if the four fluorescent tags are introduced as the first four peaks (fragments), the entire sequence can be obtained with two-dimensional fluorescence correlation analysis, by examining the correlation patterns between the first four peaks and the whole electropherogram. The peaks are identified by the vanishing asynchronous correlation intensities. Practically, the first four peaks can be four oligonucleotides labeled with the four fluorescent tags. The oligonucleotides need to be shorter than the primer used in the sequencing reaction to ensure their elution before all DNA fragments. (55) Cobb, W. T.; McGown, L. B. Appl. Spectrosc. 1989, 43, 1363-1367.

DISCUSSION Time-resolved fluorescence detection is a powerful technique for the detection and resolution of overlapping peaks and for species identification in chromatography and electrophoresis, due to its high sensitivity and relatively simple instrumentation. The resolution enhancement is important because complete separation of multicomponent mixtures can be difficult even after the optimization of separation conditions.56 A few groups have pursued fluorescence lifetime detection as a method for species identification or base calling in DNA sequencing. Many dye molecules have been evaluated in both the visible15,44-46 and NIR regions17,47 to formulate suitable dye sets for the labeling of the four nucleotide bases. Some dyes whose fluorescence decays single-exponentially in free form have shown double-exponential decays of fluorescence upon bonding to oligonucleotides.17,45 This behavior of dye molecules complicates base calling that relies on the recovery of four different lifetimes for the four dyes. A limitation for timeresolved fluorescence to become a standard detection principle for everyday DNA sequencing is the extensive, time-consuming data treatment and analysis that require the utmost attention of experienced spectroscopists. A 3-h sequencing run can generate over 10 000 fluorescence decay curves at 1-Hz data acquisition

rate. If a higher acquisition rate is used, as often needed in the resolution of overlapping peaks, a 3-h sequencing experiment can easily collect close to 100 000 fluorescence decays, each to be analyzed by NLLS or MLE fitting. Furthermore, the raw timeresolved data files need to be treated before NLLS or MLE fitting. For example, in the time domain, an intensity threshold is first determined. Decay profiles with integrated intensities higher than the threshold will be summed within electrophoretic peaks to generate decay curves for the peaks. These decay curves are then examined to determine whether a time delay should be introduced to truncate the time range of data sets in MLE analysis. After the data treatment, several hundred peak decay curves are then subjected to MLE fitting to yield the average lifetimes for the sequencing peaks. The lifetimes will then be used to compare with standards for base calling. Considering a future where DNA sequencing will be a common technology used in hospitals and clinics, in the hands of nonspecialized operators, simplicity of instrumentation and data analysis is essential. The two-dimensional correlation approach introduced in this paper couples the simplicity of instrumentation for phase-resolved fluorescence detection and the simplicity of data analysis with twodimensional correlation. These are suitable traits of a detection technique for species identification. In addition, our technique has the capability of resolving overlapping peaks and has demonstrated better detection sensitivity than existing detection techniques in both the time and the frequency domain in the UV/ visible region.50 Non-monoexponential decays of some dye molecules tagged to DNA sequencing fragments pose a complicating problem in existing time-resolved fluorescence detection schemes but will not affect two-dimensional fluorescence correlation analysis. We have demonstrated in two-dimensional fluorescence correlation spectroscopy that component spectra can be completely resolved from the mixture even when they display multiexponential decays.57 Since both signals used to calculate the correlation electropherograms are collected simultaneously during a CE run, the correlation technique is not affected by the lack of migration time reproducibility in capillary electrophoresis. Although demonstrated in the frequency domain in this paper, the two-dimensional correlation principle can be applied to the data analysis in the time domain. The two-dimensional correlation analysis in this paper is based on different fluorescence lifetimes of the solutes and requires a displacement between the elution peaks. Solutes that have identical lifetimes and elution behavior (identical elution times and peak widths) cannot be resolved by two-dimensional correlation analysis or by the existing time-resolved fluorescence detection methods. The limit where two overlapping peaks can be resolved by 2D CCE depends on many factors including the ratio of fluorescence lifetimes of the solutes and their intensity ratios. The capability of 2D CCE to resolve pyrene and anthracene peaks with an R value of 0.28 is encouraging. For chromatographic separations where the peak widths are broader than in CE, a larger separation between the peak centers will be necessary for the same R. In the events of one peak containing many overlapping solutes, the correlation pattern may become very complex and difficult to interpret.

(56) Davis, J. M.; Samuel, C. J. High Resolut. Chromatogr. 2000, 23, 235-244.

(57) Cox, J.; Geng, L., manuscript in preparation.

Figure 9. Species identification of the sequencing run in Figure 7. The four lanes are correlation peaks with POPOP (A), pyrene (B), B(k)F (C), and B(a)P (D).

We examined experimental variations between runs to assess their impact on the accuracy of species identification. We found excellent reproducibility across runs and between days. In fact, the 39-peak sequence shown in Figure 9 was constructed by four shorter sequences run on four separate days. The principle of two-dimensional correlation worked well for data between days. For example, peaks 1, 16, 30, and 36 in Figure 7 were collected in four separate electrophoresis runs performed on four different days. The cross-correlation analysis between the four peaks showed that they contain the identical solute (Figures 8 and 9). Experimental variations and noises did not seem to affect the cross-correlation analysis significantly. The entire sequence of 39 peaks were recovered correctly, demonstrating the robustness of the approach.


4541

The generalized correlation principle is not restricted to fluorescence lifetime-based detection. In fact, it is not limited to fluorescence detection methods. If two signals are observed during a CE run that will have different responses to different solute species, two-dimensional correlation analysis can be done on these two signals to enhance peak resolution and aid species recognition. For example, with an instrument that has both absorption and fluorescence detectors, correlation analysis can be performed between the absorbance electropherogram and the fluorescence intensity electropherogram. Solutes will be differentiated by their characteristic molar absorptivities and fluorescence quantum yields. CONCLUSION We have introduced a new detection principle for capillary electrophoresis based on generalized two-dimensional fluorescence correlation analysis. By spreading an electropherogram into

4542


two dimensions using the fluorescence decay constants, we have achieved baseline resolution for two overlapped peaks that have a resolution of 0.28. Species identification has been accomplished by examining the two-dimensional correlation pattern, and a simple, reliable identification of a sequence of 39 peaks has been demonstrated. ACKNOWLEDGMENT The work was partially supported by the National Institute of Health (Grant CA82706) and the University of Iowa through a Central Investment Fund for Research Enhancement grant. We thank the Center of Biocatalysis and Bioprocessing of the University of Iowa for a fellowship awarded to G.W. Received for review May 11, 2000. Accepted July 24, 2000. AC000534I

Two-Dimensional Fluorescence Correlation in Capillary

Recommend Documents