On-the-Fly Fluorescence Lifetime Detection of Dye-Labeled DNA

In the other dye system, addition of 10% DMSO to the run buffer changed the ... Fluorescence Lifetime for On-the-Fly Multiplex Detection of DNA Restri...
0 downloads 0 Views 166KB Size
Anal. Chem. 1998, 70, 3413-3418

On-the-Fly Fluorescence Lifetime Detection of Dye-Labeled DNA Primers for Multiplex Analysis Hui He, Brian K. Nunnally, Liang-Chi Li, and Linda B. McGown*

Department of Chemistry, P. M. Gross Chemical Laboratory, Duke University, Box 90346, Durham, North Carolina 27708-0346

Mixtures of dye-labeled, M13-forward DNA primers were separated by capillary gel electrophoresis and detected on-the-fly, using fluorescence lifetime measurements, to evaluate four-decay detection for multiplex DNA sequencing. Three different four-dye systems were used, two that were excited at 488 nm and one that was excited at 514 nm. Each dye-labeled primer was identified on the basis of the lifetime of the conjugated dye using nonlinear least squares or the maximum entropy method to analyze the lifetime data. Overlapping electrophoretic peaks were generated by making multiple injections of mixtures of the dye-labeled primers. The overlapping peaks were resolved by fitting the data to two-, three- or four-component lifetime models used in nonlinear least-squares analysis in which each lifetime component was fixed to the predetermined lifetime of the corresponding dye-labeled primer. In two of the dye systems, the lifetimes of the four dye-labeled primers were sufficiently different to allow peak resolution. In the other dye system, addition of 10% DMSO to the run buffer changed the lifetime of one dye-labeled primer, allowing it to be resolved from another dye-labeled primer with similar lifetime. Fluorescence-based detection for DNA sequencing has enjoyed widespread use in recent years due to its high sensitivity, ease of automation, and ability to provide real-time detection of DNA sequencing products during electrophoretic separation. A particular advantage of fluorescence detection is the ease with which it can be adapted to multiplex analysis for identification of the four DNA bases in a single lane or column. Most commonly, multiplex fluorescence detection is implemented in a four-color scheme in which the terminal base is identified by the color of the dye that is attached to the fragment.1,2 To discriminate among the four dyes, it is necessary to achieve wavelength selectivity in the emission spectrum, sometimes in combination with excitation wavelength. Detection using only filter selection of color may be inadequate to resolve overlapping peaks, while the use of array detection to collect a spectrum in order to better identify and resolve colors can decrease sensitivity as a result of dispersion. (1) Smith, L. M.; Sander, J. Z.; Kaiser, R. J.; Hughes, P.; Dodd, C.; Connell, C. R.; Heiner, C.; Kent, S. B. H.; Hood, L. E. Nature 1986, 321, 674. (2) Connell, C.; Fung, S.; Heiner, C.; Bridgham, J.; Chakerian, V.; Heron, E.; Jones, B.; Menchen, S.; Mordan, W.; Raff, M.; Recknor, M.; Smith, L. M.; Springer, J.; Woo, S.; Hunkapiller, M. W. BioTechniques 1987, 5, 342. S0003-2700(98)00170-X CCC: $15.00 Published on Web 07/08/1998

© 1998 American Chemical Society

An alternative to four-color detection is four-decay detection, in which fluorescence lifetime replaces fluorescence color as the discriminating characteristic. The potential use of fluorescence lifetime detection has been discussed elsewhere, in the development of near-infrared dyes and time domain detection strategies.3,4 Real-time, on-the-fly measurement of fluorescence lifetime has previously been demonstrated5 and applied to the detection and resolution of mixtures of DNA primers labeled with four different visible dyes and separated by CE.6,7 In this paper, we describe investigations of four-decay detection of three different dye-labeled primer systems, two that are excited at 488 nm and one that is excited at 514 nm. Overlapping electrophoretic peaks were artificially generated using sequential injections, to simulate actual sequencing results. Both the maximum entropy method (MEM) and nonlinear least-squares (NLLS) data analysis were used to recover lifetime profiles at 1or 2-s intervals across the electropherogram. Overlapping electrophoretic peaks were resolved using NLLS in which the lifetimes of the dye-labeled primers were fixed to values recovered from individual injections of each dye-labeled primer. EXPERIMENTAL SECTION Sample Preparation. Dye-labeled M13-forward primers d(5′TGTAAAACGACGGCCAGT-3′) were custom-synthesized by Midland Certified Reagent using succimidyl esters of the dyes. The dyes included NBD-aminohexanoic acid (NBD), tetramethylrhodamine (TMR), rhodamine green (RG), and BODIPY-FL (BOD), BODIPY-FL-Br2 (BBr), and BODIPY 530/550 (B530), all from Molecular Probes, Eugene, OR, and Cy3 from Amersham Life Science, Inc., Cleveland, OH. CE separations were performed using a Beckman P/ACE 5000 which was equipped with a CE-MS interface to provide an external power supply. A ssDNA separation kit (eCAP ssDNA 100-R, Beckman, Fullerton, CA) provided the DNA capillary column, a replaceable, modified polyacrylamide gel, and a Trisborate buffer with 7 M urea (TBU). Electrokinetic injection was (3) Seeger, S.; Bahteler, G.; Drexhage, K. H.; Arden-Jacob, J.; Deltau, G.; Galla, K.; Han, K. T.; Mueller, R.; Koellner, R. A.; Sauer, M.; Schulz, A.; Wolfrum, J. Ber. Bunsen-Ges. Phys. Chem. 1993, 97, 1542. (4) Soper, S. A.; LeGendre, B. L., Jr.; Williams, D. C. Anal. Chem. 1995, 67, 4358. (5) Li, L.; McGown, L. B. Anal. Chem. 1996, 68, 2737. (6) Li, L.; He, H.; Nunnally, B. K.; McGown, L. B. J. Chomatogr. 1997, 695, 85. (7) Nunnally, B. K.; He, H.; Li, L.; Tucker, S. A.; McGown, L. B. Anal. Chem. 1997, 69, 2392.

Analytical Chemistry, Vol. 70, No. 16, August 15, 1998 3413

used to introduce samples into the gel-filled capillary column (100mm i.d.). The total column length was typically 46 cm, and the distance from inlet to detection window was 38 cm. The separation voltage was approximately -15 kV to generate an electric field of 300 V/cm. A phase-modulation (frequency-domain) fluorescence lifetime instrument (model 4850 MHF, Spectronics Instruments, Rochester, NY) was interfaced to the CE for on-the-fly fluorescence lifetime measurements.5 An air-cooled argon ion laser (model 543R-AP-A01, Omnichrome, Chino, CA) provided 100-mW excitation at 488 or 514 nm. The emission signal was collected by a 10× microscope objective. The capillary was tilted approximately 30° relative to the vertical axis of the microscope objective to reduce scattered light. For 488-nm excitation, incident laser light was purified by a 488-nm bp filter and emission was selected through a 515-nm long-pass filter combined with a 488-nm holographic filter (Notch-Plus, Kaiser Optical Systems, Ann Arbor, MI) to further reduce the contribution from scattered laser light. For 514-nm excitation, emission was selected through a 550-nm long-pass filter. A cross-correlation frequency of 10 Hz was used in the lifetime measurements, resulting in 10 phase-modulation measurements per second. Scattered light provided the lifetime reference. Each electrophoretic run results in a huge amount of data. For example, a 20-min run may result in as many as 12 000 phasemodulation multifrequency profiles. Each profile is a Fouriertransformed, binary computer file containing phase and modulation data for each of 20 or more modulation frequencies. An inhouse software program was written to process the multitude of data. The software includes several modules to perform various tasks, including extraction of the fluorescence intensity from the dynamic data to produce an intensity electropherogram ,which can be imported into plotting software, and analysis of the lifetime data using nonlinear least-squares analysis software (Globals Unlimited, Urbana, IL). The NLLS package was designed for userinteractive computer data analysis. However, for on-the-fly CE data, other modules of the in-house software served to interface between the computer files and the NLLS software. These modules take each binary data file and convert it into a text file that is reformatted for entry into the NLLS software. Since reading/writing data from the computer hard disk is slow, the in-house software was written to instruct the computer to use 32 Mb of its random access memory as a virtual hard disk for temporary read/write function. In this research, the in-house software also included a module to average 10 or 20 successive measurements prior to data analysis to yield a lifetime profile every 1 or 2 s. Using NLLS, lifetime-intensity electropherograms were obtained by fitting the data to a priori models of one, two, three, or four discrete lifetime components. The lifetimes of the components were allowed to float. In some cases, an additional lifetime component of very short, fixed lifetime was added to the fitting model to account for scattered light. Using the maximum entropy method, a unique lifetime profile is recovered by imposing dual constraints of minimum χ2 and maximum statistical entropy without a priori modeling or assumptions about the fluorescence decay of the sample. MEM provides evidence of impurities, scatter, and background through the 3414 Analytical Chemistry, Vol. 70, No. 16, August 15, 1998

unbiased recovery of a complete lifetime profile.8-12 Its effectiveness for analysis of on-the-fly data in CE has been demonstrated.5,6 In this work, MEM was performed using commercial software from Maximum Entropy Consultants, Cambridge, U.K. In the MEM lifetime electropherograms, each lifetime profile was multiplied by the fluorescence intensity at that point as previously described.5 The effect of this weighting scheme is to emphasize lifetimes in proportion to signal strength and reduce the appearance of background contributions in baseline regions of the electropherogram. Since lifetime is undefined in the absence of detectable signal, the lifetimes that are recovered at baseline regions of low signal are meaningless manifestations of random noise and systematic error. Resolution of overlapping electrophoretic peaks was achieved using NLLS in which the lifetime components in the fitting models were fixed to values that were recovered for the individual dyelabeled primers under identical conditions. The recovered fractional intensities at each point in the lifetime electropherogram were then multiplied by the total intensity at that point to reconstruct the peaks of each labeled primer. RESULTS AND DISCUSSION Dye-Labeled M13-Forward Primers Excited at 488 nm. The first four-dye system for excitation at 488 nm has been previously described7 and includes NBD, TMR, RG, and BOD. Figure 1 shows the intensity and lifetime electropherograms of the individual labeled M13 primers recovered from both NLLS and MEM analysis. Lifetimes and standard deviations (in parentheses) recovered by NLLS from individual injections of the labeled primers were 1.8 ((0.06), 3.7 ((0.07), 3.7 ((0.04), and 5.3 ((0.02) ns for M13-NBD, M13-TMR, M13-RG, and M13-BOD, respectively. Lifetimes recovered by MEM analysis were within 0.1 ns of these values. It is not possible to obtain a meaningful uncertainty for each lifetime in the electropherogram since, in on-the-fly detection, each is based on a single measurement of phase and of modulation. Instead, the standard deviations reported above are calculated from the average of the lifetimes across the peak of each individual component. There are three significantly different lifetimes among the four dye-labeled primers. Unfortunately, M13-TMR has a lifetime of 3.7 ns, which is different from the lifetime of 2.2 ns recovered for the free dye and indistinguishable from the lifetime of M13-RG. The lifetimes of the other dye-labeled primers are similar to those of the free dyes. Using these dye-labeled primers, we performed experiments on two- and three-component mixtures. For a two-component mixture of M13-BOD and M13-RG (Figure 2), five sequential injections were timed to artificially generate four overlapping peaks. The first and the last peaks were pure M13-BOD and M13RG, respectively. The remaining peaks were resolved from the total intensity electropherogram using NLLS analysis in which the lifetimes were fixed to the lifetimes recovered for the peaks of the individual labeled primers. A third lifetime component was (8) Livesey, A. K.; Brochon, J. C. Biophys. J. 1987, 52, 693. (9) Brochon, J. C.; Livesey, A. K. Chem. Phys. Lett. 1990, 174, 517. (10) Shaver, J. M.; McGown, L. B. Anal. Chem. 1996, 68, 9. (11) Shaver, J. M.; McGown, L. B. Anal. Chem. 1996, 68, 611. (12) Swaminathan, R.; Periasamy, N. Proc. Indian. Acad. Sci (Chem. Sci.) 1996, 108, 39.

Figure 1. Fluorescence intensity (solid line) and lifetime (shaded or dotted line) electropherograms of dye-labeled M13 primers recovered from MEM (left) and NLLS (right) analysis. The MEM lifetime distribution at each point in the electropherogram is weighted by the total intensity at that point. The NLLS lifetimes were recovered by a two-component model in which the second lifetime component (not shown) was fixed to a very short lifetime to account for scattered light. The concentrations of M13-NBD, M13-TMR, M13-RG, and M13-BOD are 19, 53, 13, and 23 µM, respectively; the samples were injected at 10 kV for 2 s, 5 kV for 2 s, 10 kV for 2 s, and 1 kV for 2 s, respectively. Excitation, 488 nm.

Figure 2. Lifetime-resolved intensity peaks of individual dye-labeled M13 primers from sequential injections (each for 2 s at 10 kV) of a mixture of 1.9 µM M13-BOD and 2.1 µM M13-RG. Fine solid line, total intensity; dotted line, intensity of M13-BOD; thick solid line, M13-RG. Peaks were recovered from three-component NLLS analysis in which the lifetimes were fixed to predetermined values for the individual dyelabeled primers under identical conditions and a third lifetime component (not shown) was fixed to a very short lifetime to account for scattered light. Excitation, 488 nm.

Analytical Chemistry, Vol. 70, No. 16, August 15, 1998

3415

Figure 3. Lifetime-resolved intensity peaks of individual dye-labeled M13 primers from five sequential injections (each for 5 s at 10 kV) of a mixture of 2.8 µM M13-BOD, 2.1 µM M13-RG, 8.5 µM M13-TMR, and 18.7 µM M13-NBD. Fine solid line, total intensity; dotted line, intensity of M13-BOD; dashed line, overlapping intensities of M13-RG and M13-TMR; thick solid line, intensity of M13-NBD. Peaks were recovered from four-component NLLS analysis in which the lifetimes were fixed to predetermined values for the individual dye-labeled primers under identical conditions (TMR and RG labeled primers constituted one component since their lifetimes were indistinguishable). A fourth lifetime component (not shown) was fixed to a very short lifetime to account for scattered light. Excitation, 488 nm.

Figure 4. Electropherogram of mixture of (in order of migration) M13-BOD, M13-NBD, M13-RG and M13-TMR (concentrations as in Figure 3) using 90% TBU/10% DMSO as the run buffer (and to prepare the gel). Solid circles are lifetimes of major component (fractional contribution >90%) and open circles are lifetimes of minor components, recovered from NLLS fits to a two-component model. Excitation, 488 nm.

fixed to zero to account for scattered light. Addition of a fourth, floating lifetime component to the model caused distortions of the recovered lifetimes for the pure peaks, and this model was therefore rejected. Lifetime resolution was able to recover the individual contributions of both labeled primers even when overlap was almost complete. The recovered peak-to-height ratio of the two labeled primers in each of the sequential injections was 3416 Analytical Chemistry, Vol. 70, No. 16, August 15, 1998

Figure 5. Fluorescence intensity (solid line) and lifetime (dotted line) electropherograms of M13-Cy3. Lifetimes were recovered from NLLS fits to a one-component model. A 16 µM sample was injected at 10 kV for 1.5 s. Excitation, 488 nm.

consistent with the ratio for a single injection of the same mixture, for which electrophoretic separation of the components was complete. The χ2 goodness-of-fit parameter was not used to select the best model for NLLS analysis since noise in each individual measurement across a peak could result in multiple models being used across a peak. Moreover, due to the nature of on-the-fly detection, the χ2 associated with a single point on the electropherogram (which is a single value of phase and of modulation)

Figure 6. MEM recovered lifetime (top) and intensity (bottom) electropherograms of a mixture of (in order of migration) 3.7 µM M13BOD, 19 µM M13-NBD, 2.6 µM M13-Cy3, and 2.1 µM M13-RG. The lifetime distribution at each point in the electropherogram is weighted by the total intensity at that point. The mixture was injected at 10 kV for 2 s. Excitation, 488 nm.

is not a reliable indicator. Instead, we used the number of components that provides the best recovery of peaks that we know to contain only a single labeled primer. For example, a twocomponent fit may give good recovery of the lifetime and fractional intensity across such a “pure” peak while a three-component fit may result in poor recovery of the lifetime, splitting a lifetime into two similar lifetime components or yielding negative fractional intensity for a component. In such a case, we would use the twocomponent fit. MEM is a further guide to selection of the best model. It has been reported that dyes may exhibit biexponential decay upon tethering to DNA, which is attributed to electron transfer between the dye and guanosine residues in the oligonucleotide. This results in two lifetimes, one longer and one shorter than the lifetime of the free dye in solution.13,14 We have observed such behavior in batch solution but not in the CE experiments, from either MEM analysis or NLLS analysis in which the lifetimes are allowed to float. Any short components that we observe are much shorter than those reported for tethered dyes (tens of picoseconds vs ∼0.7 ns) and are therefore attributed to scattered light. The absence of the interaction may be due to the application of the strong electric field in the CE experiment, which could overcome the attraction between a positively charged dye such as TMR and the negatively charged oligonucleotide and cause the dye to stretch away from the oligonucleotide and toward the opposite electrode. (13) Edman, L.; Mets, U.; Rigler, R. Proc. Natl. Acad. Sci. U.S.A. 1996, 93, 6710. (14) Vamosi, G.; Gohlke, C.; Clegg, R. M. Biophys. J. 1996, 71, 972.

Figure 7. MEM recovered lifetime (top) and intensity (bottom) electropherograms of dye-labeled M13 primers. Injection sequence: 0.26 µM M13-Cy3, 3.0 µM M13-B530, and three injections of a mixture of 0.26 µM M13-Cy3, 10 µM M13-BBr, 0.53 µM M13-TMR, and 3.0 mM M13-B530. The lifetime distribution at each point in the electropherogram is weighted by the total intensity at that point. Each injection was for 8 s at 10 kV. Excitation, 514 nm.

A three-component experiment was performed using the same approach as for the two-component system. In this case, five sequential injections were made of a mixture of M13-BOD, M13NBD, M13-TMR, and M13-RG. The order of migration is M13BOD, M13-NBD, and comigrating M13-RG and M13-TMR. M13RG and M13-TMR are treated as a single lifetime component in the NLLS analysis because they have unresolvable lifetimes and migrate together. Results are shown in Figure 3 for identification and resolution of peaks using NLLS with three fixed lifetime components for the dye-labeled primers and a fourth component fixed to zero to account for scattered light. The fifth electrophoretic peak contains all three lifetime components. These components were successfully resolved by NLLS. Not surprisingly, the noise level is higher than in the two-component experiment. The effect of run buffer composition on fluorescence lifetimes of the labeled primers was investigated in an attempt to resolve M13-RG and M13-TMR. Addition of 10% DMSO to the run buffer decreases the observed lifetimes of M13-NBD and M13-BOD to 0.90 ((0.06) and 5.0 ((0.04) ns, respectively. Most importantly, the lifetimes of M13-TMR and M13-RG both decrease but to different extents (M13-TMR decreases to 3.5 ((0.03) ns, and M13-RG decreases to 3.3 ((0.03) ns), allowing them to be distinguished by a separation of more than three standard deviations (Figure 4). In the second system for excitation at 488 nm, TMR was replaced by Cy3. Intensity and lifetime electropherograms are Analytical Chemistry, Vol. 70, No. 16, August 15, 1998

3417

Figure 8. Lifetime-resolved intensity peaks of dye-labeled M13 primers from sequential injections of a four-component mixture (same injection sequence and concentrations as in Figure 7). Resolution was performed using NLLS fits to a fixed lifetime, four-component model across the electropherogram. Fine solid line, total intensity; thick solid line, intensity of M13-Cy3; dash-dot-dot line, M13-BBr; dotted line, M13-TMR; dashed line, M13-B530. Each injection was for 6s at 10 kV. Excitation, 514 nm.

shown in Figure 5 for M13-Cy3 and in Figure 6 for a mixture of M13-Cy3, M13-NBD, M13-RG, and M13-BOD. Although the lifetimes of M13-Cy3 and M13-NBD are similar 1.4 (( 0.007) and 1.8 ns ((0.06), respectively, they are resolvable by more than three standard deviations. The lifetimes recovered in baseline regions of the electropherograms, as in the region before the M13-Cy3 peak in Figure 5, are due to as yet unidentified background noise. It must be emphasized that these lifetimes correspond to regions of very low signal and are not a significant contribution to the signal in peak regions. Lifetime is undefined in the absence of signal, resulting in recovery of noisy, meaningless values. These contributions are a negligible fraction of the total intensity in peak regions and have little impact on the results. Dye-Labeled M13-Forward Primers Excited at 514 nm. The 514-nm excitation system includes M13 forward primers labeled with Cy3, BBr, TMR, and B530. Sequential injections of the mixture of the four labeled primers were performed to artificially create pure peaks and overlapping peaks. Figure 7 shows the lifetime and intensity electropherograms recovered by MEM analysis of the sequential injections. The first electrophoretic peak is M13-Cy3, the second peak is M13-B530, the third, fourth, and fifth peaks are due to the M13-BBr preparation, and the ninth peak is M13-TMR. The three other peaks contain overlapping labeled primers. Analysis of these peaks using MEM shows the leading portions of the seventh and eighth peaks to be M13-TMR but is unable to resolve the latter portions of the these peaks and also the sixth peak. Resolution of the overlapping peaks was achieved using fixed lifetime NLLS analysis. For peaks 1, 2, and 7-10, the lifetimes were fixed to 1.2, 3.0, and 4.5 ns in a three-component model for NLLS analysis for recovery of the intensity peaks of M13-Cy3, M13-TMR, and M13-B530, respectively. For peaks 3-6, the lifetimes were fixed to 2.0 and 4.5 ns in a two-component model 3418 Analytical Chemistry, Vol. 70, No. 16, August 15, 1998

for NLLS analysis for recovery of the intensity peaks of M13-BBr and M13-B530, respectively. The sixth, seventh, and eighth peaks were nicely resolved. Though none of the electrophoretic peaks contain all four labeled primers, we tried fixing all of the lifetimes in a fourcomponent model for NLLS analysis to recover the intensity peaks of each labeled primer. Results are shown in Figure 8. The interference between each lifetime channel is larger, but results still indicate that the sixth peak contains two components and the seventh and eighth peaks each contain three components. CONCLUSIONS Identification and resolution of CE peaks using fluorescence lifetime detection was achieved for dye-labeled DNA primers. The three four-dye sets were all shown to be feasible for application to four-decay DNA sequencing; in one case, addition of DMSO to the run buffer provided sufficient lifetime perturbation to distinguish between two dye-labeled primers with similar lifetimes. Peak resolution, even in the case of highly overlapping peaks, can be achieved using NLLS analysis in which the lifetimes are fixed to predetermined values for the labeled primers. MEM serves as a useful guide to choosing and/or verifying a particular model in NLLS in order to maximize the accuracy of the results. ACKNOWLEDGMENT This work was supported by the National Institutes of Health (Grant R01HG01161).

Received for review February 13, 1998. Accepted June 4, 1998. AC980170I