Anal. Chem. 2004, 76, 3214-3221
Comparison of Hadamard Transform and Signal-Averaged Detection for Microchannel Electrophoresis Jennifer A. McReynolds and Scott A. Shippy*
Department of Chemistry (M/C 111), University of Illinois at Chicago, 845 West Taylor Street, Room 4500, Chicago, Illinois 60607-7061
A Hadamard transform (HT) detection method for microchip capillary electrophoresis with laser-induced fluorescence and a charge-coupled device (CCD) is described and compared to signal-averaged detection. A low-noise CCD camera is used to image a section of a separation channel where each camera pixel can be thought of as a unique detector. For signal averaging, electropherograms corresponding to individual pixels can be averaged for improved S/N. HT detection is performed on each pixel electropherogram to generate a contour plot electropherogram. The multiple injections required for HT provides an enhancement at the cost of longer times for the pseudorandom injection sequences. A short sample injection length of 0.25 s is used to reduce the overall analysis time and improve sensitivity compared to previously published results. An injection sequence is performed on the microchip that is based on a cyclic S-matrix of 513 elements that generates an 8-fold improvement in S/N compared to a single injection. This spatially resolved HT detection method is also capable of performing a multicomponent separation. Signal-averaged HT and singleinjection data are compared to experimental HT and single-injection results. The unique capabilities of each method are described. Microchip electrophoresis (MCE) has become an increasingly popular method in the field of separation science. Microfabricated devices are advantageous because of the capability to perform most chemical analysis functions on-chip. However, the ability to detect low-concentration components from low-volume samples remains problematic because of the short path lengths and small sample volumes associated with these miniature devices. Not surprisingly, the most popular detection method for MCE is laserinduced fluorescence (LIF) because of its high sensitivity.1-5 Yet, * Corresponding author. Phone: 312-355-2426. Fax: 312-996-0431. e-mail:
[email protected]. (1) Harrison, D. J.; Fluri, K.; Seiler, K.; Fan Z. H.; Effenhauser, C. S.; Manz, A. Science 1993, 261, 895-897. (2) Jiang, G. F.; Attiya, S.; Ocvirk, G.; Lee, W. E.; Harrison, D. J. Biosens. Bioelectron. 2000, 14, 861-869. (3) Liu, Y. J.; Foote, R. S.; Jacobson, S. C.; Ramsey, R. S.; Ramsey, J. M. Anal. Chem. 2000, 72, 4608-4613. (4) Ocvirk, G.; Tang, T.; Harrison, D. J. Analyst 1998, 123, 1429-1434. (5) Wallenborg, S. R.; Bailey, C. G. Anal. Chem. 2000, 72, 1872-1878.
3214 Analytical Chemistry, Vol. 76, No. 11, June 1, 2004
alternative detection formats that afford either an improvement in sensitivity over LIF or are based on a more common chemical property such as absorbance would be highly desirable. A class of methods that has the potential to enhance the ability to detect dilute analytes is known as multiplexed detection. These mathematical approaches have been applied toward analytical instrumentation such as mass spectrometry6-8 and nuclear magnetic resonance spectroscopy.9,10 In general, these methods are the simultaneous measurement of encoded analytical signals. These signals are subsequently decoded by mathematical processes that result in a sensitivity or resolution enhancement. Recently, these techniques have been demonstrated to improve detection in capillary electrophoresis (CE)11,12 and MCE.13-15 Cross-correlation is one such multiplexed technique that has been performed with both CE16-18 and MCE.19 In cross-correlation MCE, a sample is injected into the separation channel according to a pseudorandom binary sequence. When the summation of all electropherograms is cross-correlated with the input signal, the result is a correlogram with a S/N enhancement. A method that averaged consecutive correlograms was also reported to improve sensitivity in this report;19 however, long analysis times were required. Another form of multiplexed detection for MCE is the Shah convolution Fourier transform, or SCOFT.13,14,20 This method involves modulating the signal of fluorescent material as it flows through the separation channel. A Shah convolution is applied to (6) Marshall, A. G. Int. J. Mass. Spectrom. 2000, 200, 331-356. (7) Marshall, A. G.; Hendrickson, C. L. Int. J. Mass. Spectrom. 2002, 215, 5975. (8) Smith, R. D. Int. J. Mass. Spectrom. 2000, 200, 509-544. (9) Evilia, R. F. Anal. Lett. 2001, 34, 2227-2236. (10) Kupce, E.; Nishida, T.; Freeman, R. Prog. Nucl. Magn. Reson. Spectrosc. 2003, 42, 95-122. (11) Kaneta, T.; Yamaguchi, Y.; Imasaka, T. Anal. Chem. 1999, 71, 5444-5446. (12) Kaneta, T.; Kosai, K.; Imasaka, T. Anal. Chem. 2002, 74, 2257-2260. (13) Crabtree, H. J.; Kopp, M. U.; Manz, A. Anal. Chem. 1999, 71, 2130-2138. (14) Kwok, Y. C.; Manz, A. J. Chromatogr., A 2001, 924, 177-186. (15) McReynolds, J. A.; Edirisinghe, P.; Shippy, S. A. Anal. Chem. 2002, 74, 5063-5070. (16) Kuldvee, R.; Kaljurand, M.; Smit, H. C. J. High Resolut. Chromatogr. 1998, 21, 169-174. (17) van der Moolen, J. N.; Louwerse, D. J.; Poppe, H.; Smit, H. C. Chromatographia 1995, 40, 368-374. (18) van der Moolen, J. N.; Poppe, H.; Smit, H. C. Anal. Chem. 1997, 69, 42204225. (19) Fister, J. C.; Jacobson, S. C.; Ramsey, J. M. Anal. Chem. 1999, 71, 44604464. (20) Kwok, Y. C.; Jeffery, N. T.; Manz, A. Anal. Chem. 2001, 73, 1748-1753. 10.1021/ac035404z CCC: $27.50
© 2004 American Chemical Society Published on Web 05/04/2004
the fluorescence signal by use of a fixed optical mask with equally spaced chrome regions over a length of the channel illuminated lengthwise with a continuous wave laser. The Fourier transform of the collected fluorescence is used to generate a frequency spectrum of the fluorescent bands or microspheres that flowed down the channel. An improved format for SCOFT has been performed in our laboratory that utilizes the spatial resolution of a charge-coupled device (CCD) to modulate the fluorescent signal at different positions based on a sine wave.15 Each image of a section of the electrophoresis channel is multiplied by a sine wave and summed for Fourier analysis. With both the low-noise CCD detector and the sine wave convolution, a S/N improvement of >3 orders of magnitude was demonstrated for a plug of fluorescein dye and the ability to use this method for particle quantitation was also demonstrated. Recently, multiplexed detection for MCE has been performed with the Hadamard transform (HT) to generate a significant increase in S/N of the injected sample.21 HT detection is based on an injection sequence according to a row of a cyclic S-matrix of an order, n. An encoded electropherogram results from the multiple sample and buffer injections. Multiplication with the inverse S-matrix decodes the data to generate a standard electropherogram. There is increased sensitivity over a single-injection electropherogram as the peak signals from all sample injections correlate and add faster than detector noise. As the injection sequence length increases, so will the S/N of the signal. Unfortunately, multiple injections require time, and the length of the increasing injection sequence to obtain high S/N values is limited by the overall analysis time. In the previously published paper reporting microchannel HT detection, an injection time of 1 s was used for each element in a cyclic S-matrix.21 These long injection times led to long analysis times where the highest order reported of 255 elements required over 4 min. Another promising method demonstrated for CE is to average a number of electropherograms. Rather than performing multiple injections, 1500 diodes in a linear array collect absorbance signal along a length of a capillary. Averaging electropherograms from individual diodes has been demonstrated to lower the UV detection limit in capillary electrophoresis.22,23 This is accomplished over the time of a single run where peak migration is accounted for by simply time shifting the diodes with respect to each other. The theoretical S/N enhancement equals the square root of the number of diodes, and an 85-fold improvement in S/N was reported over a single electropherogram. Notably, the noise decreased 1.7 times more than expected through this averaging method. This approach, alone or in combination with other detection formats, may provide a significant enhancement in S/N for MCE. In this paper, we compare signal-averaged and HT detection with a spatially resolved detector for MCE. The performance was optimized for the HT detection method with a 0.25-s injection scheme that greatly reduces the overall time of analysis. An injection sequence based on a 513-element S-matrix and multicomponent separation was demonstrated for the HT. Significant (21) Hata, K.; Kichise, Y.; Kaneta, T.; Imasaka, T. Anal. Chem. 2003, 75, 17651768. (22) Culbertson, C. T.; Jorgenson, J. W. Anal. Chem. 1998, 70, 2629-2638. (23) Culbertson, C. T.; Jorgenson, J. W. J. Microcolumn Sep. 1999, 11, 652662.
S/N enhancements and demonstrations of unique capabilities for each method are shown. EXPERIMENTAL SECTION Chemicals and Reagents. All chemicals were purchased from Sigma-Aldrich unless specified otherwise. Boric acid and NaOH were obtained from Fisher-Scientific (Fair Lawn, NJ). The run buffer used for all experiments was a 20 mM borate buffer. The borate buffer was made by dissolving boric acid in ultrafiltered, deionized water (US Filter, Warrendale, PA), and the pH was adjusted to 9.2 with 1 M NaOH. Fluorescein was dissolved in a minimal amount of N,N-dimethyl formamide (EM Science, Cherry Hill, NJ) and diluted with borate buffer to the desired concentration. For the multicomponent separation fluorescein, 5-[(4,6dichlorotriazin-2-yl)amino]fluorescein (DTAF), and fluorescein 5-isothiocyanate (FITC) were diluted in the same manner as the single-component fluorescein sample to the desired concentration. Chip Electrophoresis. A description of the microchip (Micralyne, Alberta, Canada), sample loading, and detection system is described elsewhere.15 An in-house-created LabView program (National Instruments, Austin, TX) was used to control the highvoltage power supply (Spellman, CZE 1000R, Huntsville, NY) and a high-voltage reed relay switching system built by the UIC Chemistry electronics shop. The microchip was filled with 1 µM fluorescein and the laser aligned lengthwise along the channel nearest to the double-T injection region. The power of the Innova 90C argon ion laser (Coherent, Santa Clara, CA) before beam expansion was 10 mW. Detection was performed with a CCD (Roper Scientific, Trenton, NJ) with an array of 1340 × 100 elements. The 2-cm detection length of the channel formed an image on the CCD with pixel dimensions of 1340 × 15. To increase S/N, the pixels across the channel width were binned into a single row before readout to produce a data array size of 1340 × 1. An effective length of 6 cm was measured from the injection region of the separation channel to the first pixel of the detection window. The fluorescent data were collected with the software WinVIEW (Roper Scientific, Trenton, NJ) that was designed to operate with the CCD. The CCD was operated without a shutter and was maintained at -85.0 °C with liquid nitrogen to minimize detector noise. The LabVIEW program created to control the power supply and switching system was designed with a 3-s delay after manually beginning the program. This was created so that the data collection software could be synchronized with the beginning of the experiment. The CCD was connected to an external TTL pulse generator (Datapulse, Culver City, CA) calibrated daily to a rate of 3.44 Hz (0.29 s between pulses) to exactly match the sampling rate. HT Multiple Injections. During sample injection, the buffer waste was grounded, the sample reservoir was held at 6.1 kV (693 V/cm) and the buffer reservoir held at 0 kV. During buffer injection, the sample reservoir potential was held at 0 kV, while 6.1 kV (718 V/cm) was applied at the buffer reservoir. Sample was injected into the separation channel according to a sequence of 1’s and 0’s of a cyclic simplex matrix obtained from the corresponding Hadamard matrix. Optimized injection of the sample was 0.25 s for every “1” represented in the sequence, and optimized injection of the buffer was 0.29 s for every “0” represented in the sequence. Because shorter injection sequences required only a few seconds to inject the entire sequence, they Analytical Chemistry, Vol. 76, No. 11, June 1, 2004
3215
Figure 1. Diagram of HT process performed via laser-induced fluorescence with a CCD. (A) After performing the cyclic S-matrix-derived injection sequence, the data are reshaped and plotted. (B) An extracted pixel row from the contour plot is shown. (C) The inverse HT is applied to a region on each extracted pixel row and plotted as individual electropherograms. (D) Each of the 1340 electropherograms is stacked to illustrate the migration of the sample via CCD pixel number, time (s), and relative fluorescence intensity.
were repeated several times until sample was simultaneously visible in the detection region. A time range corresponding to one of the injection sequence repetitions was chosen for data analysis. To perform the longer injection sequences corresponding to 255and 513-element cyclic S-matrices, it was necessary to optimize the longer segments of repeated sample or buffer injections. The timing of segments of repeated 0’s or 1’s in shorter injection sequences was adjusted to match the sampling rate and correct the overall timing of the run. These values were then extrapolated linearly to fit the longer repeat sample or buffer injections found with the longer injection sequences of higher-order S-matrices. Data Processing. The CCD data were converted to text for all data sets and exported to the Matlab software package (The Mathworks Inc., Natick, MA) for further analysis by the methods described below. S/N measurements are performed by calculating the maximum intensity of the peak and dividing this value by the standard deviation of a section of the baseline near to the sample peak. The S/Ns of the single injections were acquired in the same manner. Measurements of full width at half-maximum (fwhm) were performed by counting the number of data points of the peak at half of the maximum intensity of that peak. HT Detection. Hadamard transform detection for CE has been previously described.11,12,21,24 Our approach is similar except as noted below. Figure 1 represents the schematic of the HT detection procedure for our system. Briefly, the CCD has 1340 pixel elements that can be thought of as single-channel detectors. (24) Kaneta, T. Anal. Chem. 2001, 73, 540A-547A.
3216
Analytical Chemistry, Vol. 76, No. 11, June 1, 2004
Each pixel/detector is focused on a unique section along the length of the electrophoresis channel and can be monitored through time. An in-house-developed Matlab routine represents that collected data in the form of an array of intensities for pixel numbers (1-1340) versus time (Figure 1A). For a given pixel’s intensity versus time row, an element range such as the 2n - 1th to the 3n - 1th data points (according to n of the S-matrix) were selected for inverse transform (Figure 1B). The encoded element range, [η], is decoded to the electropherogram, [E], by this process: [E] ) [S]-1 × [η], where [S]-1 is the inverse simplex matrix. Then, the array [E] was plotted versus time starting with the last point of the matrix to the first (Figure 1C). This process generates an electropherogram for each pixel. Finally, electropherograms from each pixel were stacked to make a contour plot of sample intensities versus CCD pixel number and time (Figure 1D). To display a standard two-dimensional electropherogram, a single pixel row is plotted that contains the maximum fluorescence intensity of the contour plot. Signal Averaging. The CCD camera contains 1340 individual detector elements so the possibility of higher S/N via signal averaging was explored. Data obtained from the above method for 83 and 513 encoded injections and a single injection of 20 nM fluorescein were analyzed for possible sensitivity enhancements via signal averaging. The contour plot represented in Figure 1D was reprocessed in Matlab such that the signal was aligned into a single column. This was required to account for the migration of the band through the channel and was performed by shifting
individual CCD pixel rows separately until the signal was vertically arranged. For a given time (or frame), all intensities (the entire Y-axis) were summed giving a fluorescence intensity versus time data set. Each summed intensity data point was then divided by the number of points averaged. The process of alignment and signal averaging generated a two-dimensional electropherogram from which S/N was determined as described above. RESULTS AND DISCUSSION Limits of detection for microchannel electrophoresis can be hindered by the short optical path length available due to channel geometry. In this paper, we explore two methods for improving the sensitivity obtained for a dilute analyte during an electrophoretic separation. A section of the electrophoretic channel is imaged in these studies for multiple points of detection. The use of a spatially resolved detector allows the comparison of signal averaging over multiple detection points with the HT method of multiple injections and a combination of the two. The combination of HT detection followed by signal averaging gave an improvement over HT alone for the injection sequence length of 83 and remained the same for the longer injection sequence length. Yet, a greater S/N enhancement was seen with signal averaging a single injection alone. The HT method involved greater complexity and was optimized for comparison with signal averaging. HT detection has been used for a number of spectroscopic and imaging applications25 and, recently, for MCE.21 This multiplexed detection provides a sensitivity enhancement because it is based on the measurement of multiple injections rather than a single injection. The sample intensities of these multiple injections are combined, while noise is averaged over the multiple measurements of signal. A considerable increase in S/N results, which is dependent on the number of injections. Unfortunately, one drawback to this technique is the longer analysis time required for the repeated injection sequences. In this paper, we describe rapid LIF HT detection with a CCD for MCE with shorter injection times. Additionally, the two-dimensional format of the CCD detector allows multiple measurements of a sample peak as it electrophoreses through the channel. These measurements can be used to further improve the S/N of electrophoretic peaks. In HTCE, the sample is injected into the separation channel in a pseudorandom sequence based on a cyclic S-matrix of order, n. Data are collected at a rate equal to the sampling rate. Each image of the channel measures 1340 × 15 pixels. To optimize signal intensity and timing control, the 15-pixel axis of the CCD camera that spans the width of the microchannel is binned as previously described.15 The dimension of each data frame then becomes 1340 × 1 pixels with resolution only along the length of the channel. The number of data frames acquired during a particular experiment is dependent upon the length of the injection sequence and the number of times that sequence is repeated. The data are collected as relative fluorescence intensity versus pixel versus frame number. Illustrated in Figure 1C, the inverse S-matrix is multiplied across a region of intensities from a single pixel to generate an electropherogram. Because each pixel generates a string of intensities through time that can be decoded, 1340 single pixel electropherograms are generated by this process. The (25) Harwit, M.; Sloane, N. J. A. Hadamard Transform Optics; Academic Press: New York, 1979.
Figure 2. (A) HT three-dimensional plot of 500 nM fluorescein performed with an 83-element injection sequence. Red represents the intensity of the signal as it passes through the detection region versus time. (B) Pixel row 700 is extracted from (A) and plotted as relative fluorescence intensity versus time.
electropherograms can then be stacked to create a contour plot as illustrated in Figure 2A. Notably, rather than multiplying each pixel electropherogram by the inverse matrix in sequence, the encoded intensity versus pixel number versus time array for the selected time region (such as 2n - 1 to 3n - 1 elements) can be multiplied in one short step. This single-step multiplication saves a tremendous amount of computational time as it is easily completed in
