Microchip Laser-Induced Fluorescence Detection of Proteins at

Anal. Chem. 2004, 76, 4705-4714

Microchip Laser-Induced Fluorescence Detection of Proteins at Submicrogram per Milliliter Levels Mediated by Dynamic Labeling under Pseudonative Conditions Braden C. Giordano,†,‡ Lianji Jin,† Abigail J. Couch,† Jerome P. Ferrance,† and James P. Landers*,†,§

Department of Chemistry, University of Virginia, McCormick Road, Charlottesville, Virginia 22901, and Department of Pathology, University of Virginia Health Science Center, Charlottesville, Virginia 22901

We have previously demonstrated on-column dynamic labeling of protein-SDS complexes on capillaries and microchips for laser-induced fluorescence (LIF) detection using both a commercially available fluor and a protein separation buffer. Upon binding to hydrophobic moieties (of the analyte or separation buffer), the fluor undergoes a conformational change allowing fluorescence detection at 590 nm following excitation with 488-nm light. Our original work showed on-chip limits of detection (LOD) comparable with those using UV detection (1 × 10-5 M) on capillariessfalling significantly short of the detection limits expected for LIF. This was largely a function of the physicochemical characteristics of the separation buffer components, which provided significant background fluorescence. Having defined the contributing factors involved, a new separation buffer was produced which reduced the background fluorescence and, consequently, increased the available dye for binding to protein-SDS complexes, improving the sensitivity in both capillaries and microchips by at least 2 orders of magnitude. The outcome is a rapid, sensitive method for protein sizing and quantitation applicable to both capillary and microchip separations with a LOD of 500 ng/mL for bovine serum albumin. Interestingly, sensitivity on microdevices was improved by inclusion of the dye in the sample matrix, while addition of dye to samples in conventional CE resulted in a drastic reduction in sensitivity and resolution. This can be explained by the differences in the injection schemes used in the two systems. The linear range for protein quantitation covered at least 2 orders of magnitude in microchip applications. On-chip analysis of human sera allowed abnormalities, specifically the presence of elevated levels of γ-globulins, to be determined. Sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) is the workhorse of clinical laboratories requiring * To whom correspondence should be addressed. Phone: 804-243-8658. Fax: 804-924-3048. E-mail: [email protected]. † Department of Chemistry. ‡ Current address: Naval Research Laboratory, 4555 Overlook Ave., S.W., Chemistry Division, Code 6112, Washington, DC 20375. § Department of Pathology. 10.1021/ac030349f CCC: $27.50 Published on Web 07/03/2004

© 2004 American Chemical Society

separations of proteins based on size. SDS-PAGE offers simple methodologies, parallel processing capabilities, and accurate mass determination of complex protein samples.1-3 Over the past decade, capillary electrophoresis (CE) has begun to supplant the more traditional gel electrophoresis platforms in the biomedical and clinical settings.4-7 This can be attributed to order-ofmagnitude decreases in separation times, on-line detection (for quantitation), automation, low sample volumes, and the potential ability to parallel process in 96-capillary array instruments. In fact, the replacement of SDS-PAGE with a capillary counterpart was addressed early in the genesis of CE technology by Cohen and Karger8 by simply cross-linking polyacrylamide within the capillary (as an alternative to slab gels). SDS-PAGE is labor-intensive, requiring staining with Coomassie Blue (or other stains), followed by destaining for peak visualization and densitometry for protein quantitation. In contrast, UV detection in CE using on-line detection at 200 or 214 nm allows for direct protein detection and quantitation. Many groups have further developed CE-SDS analysis by moving away from a “gel-in-a-capillary” approach and toward nongel sieving polymers. Dextran,9 non-cross-linked polyacrylamide (PAA),10-14 poly(ethylene oxide) (PEO),11-18 and hydroxy(1) Hames, B. D., Ed. Gel Electrophoresis of Proteins, A Practical Approach, 2nd ed.; Oxford University Press: New York, 1998. (2) O’Farrell, P. H. J. Biol. Chem. 1975, 250, 4007-4021 (3) O’Farrell, P. Z.; Goodman, H. M.; O’Farrell, P. H. Cell 1977, 12, 11331141. (4) Jollif, C. R.; Blessum, C. R. Electrophoresis 1997, 18, 1781-1784. (5) Liebich, H. M.; Lehmann, R.; Xu, G.; Wahl, H. G.; Haring, H. U. J. Chromatogr., B 2000, 745, 189-196. (6) Jenkins, M. A.; Ratnaike, S. Clin. Chim. Acta 1999, 289, 121-132. (7) Pancholi, P.; Oda, R. P.; Mitchell, P. S.; Landers, J. P.; Persing, D. H. Mol. Diagn. 1997, 2, 27-37. (8) Cohen, A. S.; Karger, B. L. J. Chromatogr. 1987, 397, 409-417. (9) Takagi, T.; Karim, M. R. Electrophoresis 1995, 16, 1463-1467. (10) Manabe, T.; Oota, H.; Mukai, J. Electrophoresis 1998, 19, 2308-2316. (11) Wu, D.; Regnier, F. E. J. Chromatogr. 1992, 608, 349-356. (12) Widhalm, A.; Schewer, C.; Blaas, D.; Kenndler, E. J. Chromatogr. 1991, 549, 446-451. (13) Werner, W. E.; Demorest, D. M.; Stevens, J.; Wiktorowicz, J. E. Ananl. Biochem. 1993, 212, 253-258. (14) Benedek, K.; Theides, S. J. Chromatogr., A 1994, 676, 209-217. (15) Guttman, A.; Shieh, P.; Hoang, D.; Horvath, J.; Cooke, N. Electrophoresis 1994, 15, 221-224. (16) Benedek, K.; Guttman, A. J. Chromatogr., A. 1994, 680, 375-381. (17) Guttman, A.; Shieh, P.; Lindahl, J.; Cooke, N. J. Chromatogr., A 1994, 676, 227-231.

Analytical Chemistry, Vol. 76, No. 16, August 15, 2004 4705

propyl cellulose (HPC)19 have all been used in CE-SDS analysis with the focus being on ease of use, sensitivity, and large range for accurate protein sizing. Cooke and co-workers18 showed that dextran and PEO could produce relatively low viscosity solutions even at high w/v percents and were useful in sizing proteins in the 14-200 kDa mass range. The PEO and dextran-based CESDS buffer systems both reduced the background noise in UV absorbance seen with other sieving polymers such as PAA and demonstrated detection limits as low as 500 ng/mL (10% dextran; 2 000 000 molecular weight; S/N ∼2) with a myoglobin standard.18 While ease of use was certainly improved by the dextran- and PEO-based sieving matrixes, covalent coating of the capillary surface was necessary for reducing electroosmotic flow (EOF), which works against the electrophoretic mobility of protein-SDS complexes. Bean and Lookhart20 continued to improve on CESDS separations by utilizing preseparation conditioning steps that allowed for capillary self-coating by PAA and PEO. Rinsing with either strong acid or base prior to introducing the sieving polymer enhanced polymer interaction with the capillary surface, reducing EOF enough for successful protein sizing. While there has been much success with respect to protein analysis on capillaries, translating these separations to the microchip platform has been difficult. This can be attributed to the inability to attain sensitive UV detection on microchips, due to limitations associated with path length, the cost of quartz to produce chips with low background UV absorbance, and the reliance on nontrivial labeling procedures for laser-induced fluorescence (LIF) detection. Until recently, protein analysis on chips relied on either detection using natural fluorescence or time-consuming precolumn tagging with fluorescent labels. Work by Ramsey and co-workers21 demonstrated that simple postcolumn labeling with a commercially available dye could be achieved with a unique cross-tee mixing design positioned prior to the detection point. Conversely, sizebased analysis of DNA on microchips is commonplace due to the ease with which DNA can be labeled with intercalating dyes during separation.22-24 Unlike DNA analysis, protein detection on microchips has not had an analogous labeling approach until recent works published by our group25 and Bousse et al. from Caliper.26 Our initial work described the use of dynamic labeling of protein-SDS complexes utilizing a commercially available buffer system for protein sizing.25 With the simple addition of a commercial fluor to the CE-SDS separation buffer, proteins could be visualized using LIF detection (488 excitation; 590 emission). While this dynamic labeling procedure enabled the detection of (18) Ganzler, K.; Greve, K. S.; Cohen, A. S.; Karger, B. L.; Guttman, A.; Cooke, N. C. Anal. Chem. 1992, 64, 2665-2671. (19) Hu, S.; Zhang, Z.; Cook, L. M.; Carpenter, E. J.; Dovichi, N. J. J. Chromatogr., A 2000, 894, 291-296. (20) Bean, S. R.; Lookhart, G. L. J. Agric. Food Chem. 1999, 47, 4246-4255. (21) Liu, Y.; Foote, R. S.; Jacobson, S. C.; Ramsey, R. S.; Ramsey, J. M. Anal. Chem. 2000, 72, 4606-4613. (22) Ulfelder, K. J.; Schwartz, H. H.; Hal, J. M.; Sunezi, F. J. Anal. Biochem. 1992, 200, 260-267. (23) McCord, B. R.; McClure, D. L.; Jung, J. M. J. Chromatogr., A 1993, 652, 75-82. (24) Rampal, S.; Lui, M. S.; Chen, F. T. A. J. Chromatogr., A 1997, 781, 357365. (25) Jin, L. J.; Giordano, B. C.; Landers, J. P. Anal. Chem. 2001, 73, 49944999. (26) Bousse, L.; Mouradian, S.; Minalla, A.; Yee, H.; Williams, K.; Dubrow, R. Anal. Chem. 2001, 73, 1207-1212.

4706

Analytical Chemistry, Vol. 76, No. 16, August 15, 2004

proteins on a microchip, the commercial buffer system had a high viscosity, making it difficult to handle, and the detection sensitivity was only comparable to UV detection on a capillary system. In contrast, Skinner and co-workers27 using NanoOrange as an oncolumn dynamic labeling reagent for proteins in CE, although under nondenaturing conditions, achieved a limit of detection of 212 ng/mL for human serum albumin. Ramsey et al.21 utilized postcolumn labeling of nondenatured proteins on a microchip with a commercially available dye, with sensitivity as low as 1.21 µg/ mL. Bousse et al. included dye in both the sample matrix and separation buffer for labeling protein-SDS complexes for sizing on a microchip for use with a commercial microchip device.26 The authors found it necessary to incorporate a dilution step prior to detection. They observed high background fluorescence, due to dye interaction with SDS micelles in the separation buffer. The dilution step reduced the SDS concentration below the critical micelle concentration (CMC), thus improving detection limits to the low-micrograms per milliliter range. The work presented in this report describes the development of a refined buffer system, which is more amenable for dynamic labeling with a commercially available dye in capillary and microchip electrophoresis-SDS analysis. Buffer components that were investigated included the sieving polymer, ionic strength, and SDS concentration with a goal of increasing the detection sensitivity without loss of resolution. Rapid protein sizing and quantitation performed on a microchip platform using this new buffer were compared with traditional analyses and extended to conditions that separated proteins in their pseudonative (pseudodenatured) state. EXPERIMENTAL SECTION Apparatus. Beckman P/ACE 5510 (Beckman Instruments, Inc.) with LIF detection was used to obtain capillary electrophoresis data. The microchip electrophoresis system was assembled in-house. A 488-nm laser beam emitted from an argon ion laser source (Laser Physics) was reflected to a dichroic beam splitter (505DRLP02; Omega Optical) set 45° to the incident beam and then focused onto the channel of an electrophoretic microchip by a microscope objective (16×/numerical aperture, 0.32; Melles Griot). Two mirrors (Thor Labs, Inc.) positioned parallel to each other were set after the laser source to guide the laser beam to the beam splitter. Fluorescence emitted by the sample was collected by the same microscope objective, and detected by a Photosensor Module H5784-01 PMT (Hamamatsu) (filtered through a 590-nm band-pass filter (590DF30, Omega Optical)). Data collection and application of potentials to the microchip were processed via a program written in Labview. Reagents. Tris(hydroxymethyl)aminomethane (Tris), 2-(Ncyclohexylamino)ethanesulfonic acid (CHES), acrylamide, bovine serum albumin (BSA), phosphorylase b, myosin, dextran (2 000 000 MW), hydroxypropylcellulose (100 000 MW), and hydroxyethylcellulose (250 000 MW) were purchased from Sigma-Aldrich (St. Louis, MO). Sodium dodecyl sulfate, bisacrylamide, and an eightcomponent protein ladder were obtained from Bio-Rad (Hercules, CA). Poly(ethylene oxide) (100 000, 200 000, and 600 000 MW) was purchased from Acros (Pittsburgh, PA). Methanol, glacial (27) Harvey, M. D.; Bablekis, V.; Banks, P. R.; Skinner, C. D. J. Chromatogr., B 2001, 754, 345-356.

acetic acid, Coomassie Brilliant Blue R-250, and Bromophenol Blue were purchased from Fisher Scientific (Pittsburgh, PA). NanoOrange dye was purchased from Molecular Probe and used according to the product instruction manual. NanoOrange stock solution (500X) was added to the separation buffer at 0.2% v/v concentration unless otherwise specified. Microchip Fabrication. The microchip was fabricated using standard photolithography and wet chemical etching techniques. A film mask containing the microchip design was prepared on a negative film using an image setter. The microchip design, a traditional cross-tee type, was transferred onto the glass wafer (Nanofilm) with positive photoresist by UV exposure. The channels were etched with HF solution. The etched plate was thermally bonded to a drilled cover plate in a programmable furnace (Ney Dental Inc.). The configuration of a single-channel microchip is 7.5 cm from injection cross to the outlet, 0.5 cm from injection cross to the inlet, sample, and sample waste reservoirs. Capillary and Microchip Separations. Unless otherwise specified, capillaries and microchips were flushed for 10 min with 1 M HNO3, followed by a 10-min rinse with NanoOrange containing run buffer. Proteins (dissolved in 25 mM Tris-CHES, 0.1% SDS, and 1 mM DTT, heated at 94 °C for 5 min, and then cooled to room temperature) were electrokinetically injected into the capillary or microchip. Separations were performed under reverse polarity (inlet was cathode) with a field strength of 370 V/cm unless otherwise specified. Partial Purification of Plasma. Poly(ethylene glycol) (average MW 3400; Aldrich Chemical Co., Milwaukee, WI) was added to recovered human plasma while stirring at 4 °C, to a final concentration of 20% w/v. After 3 h, plasma was centrifuged at 1800g for 45 min. The supernatant was clarified by filtration (0.22 µm) and applied to a Q Sephadex FF 10/30 column (Amersham, Piscataway, NJ) equilibrated in 100 mM potassium phosphate buffer, pH 7.4. The nonbinding plasma component was used for further analysis. Serum Sample Analysis. Prior to analysis, all serum samples were stored at -20 °C. For CZE analysis, samples were diluted 1 in 20 in water. For pseudodenaturing microchip analysis, samples were diluted 1 in 500 in 0.5% SDS with 1% v/v dye included. SDS-PAGE. A separation gel containing 11% acrylamide gel was prepared in 375 mM Tris, pH 8.8, containing 0.1% w/v SDS. The stacking gel consisted of 5% acrylamide in 12.5 mM Tris, pH 6.8, containing 0.1% SDS. Samples were dissolved in 110 mM Tris containing 3.6% w/v SDS, 18% glycerol w/v, and 9% v/v β-mecaptoethanol. Bromophenol Blue was included in the sample as a marker. Gels were run on a model P81 vertical gel separation system from Owl Separation Systems, Inc. and prepared according to manufacturer’s recommendations. Briefly, the samples were loaded onto the gel and voltage was applied to produce a current of 15 mA. After samples completely entered the stacking gel, the voltage was increased until the current was between 35 and 45 mA. Voltage ceased when the Bromophenol Blue was ∼2 cm from the bottom of the gel. Gels were stained with a Coomassie Blue solution (30% v/v methanol, 10% v/v glacial acetic acid, 1.2 mM Coomassie Brilliant Blue R-250) for ∼1 h at room temperature. Gels were destained with several washes of a destain solution (30% v/v methanol, 10% v/v glacial acetic acid).

RESULTS AND DISCUSSION In our previous work, we described use of a commercially available protein sieving buffer system for CE-SDS analysis in conjunction with LIF detection.25 While functional, the lack of sensitivity was particularly problematic and was attributed to hydrophobic components of that buffer system. In addition, the high viscosity of this buffer made handling difficult and was especially complicating when attempting to translate to the microchip. In developing a new buffer system amenable to dynamic labeling, a number of criteria were established for evaluation of success. With respect to the polymer chosen as a sieving matrix, it must be relatively hydrophilic in character, allow for self-coating (i.e., no covalent modification of capillary or microchannel surfaces necessary for electroosmotic flow suppression), and be easy to handle (e.g., low viscosity) for both capillaries and microchips. I. Evaluation of Sieving Polymers. A candidate list of polymers that have been successfully utilized in CE-SDS analysis was derived from the literature; these included PEO,20 HPC,19 linear polyacrylamide (LPA),20 and dextran.20 The performance of each of these was evaluated by the separation of an eightcomponent protein sizing ladder, with each polymer used at a select molecular mass [PEO (100 kDa), PEO (200 kDa), HPC (100 kDa), hydroxyethylcellulose (HEC) (250 kDa), and dextran (2000 kDa)]. These results (not shown) determined that PEO (100 kDa) and HEC were the most viable candidates. Dextran required covalent modification of the capillary surface for EOF suppression, PEO (200 kDa) was difficult to handle due to viscosity, and HPC proved to be incompatible with this method of labeling, likely due to its hydrophobic nature relative to HEC and PEO. HEC and PEO (100 kDa) provided sufficient resolving power (baseline resolution of the standards with the exception of the 14.4 and 21.5 kDa proteins) and both appeared to have minimal contributions to background fluorescence. While equally suited as sieving polymers, PEO (100 kDa) was ultimately chosen as the sieving matrix of choice for continuing the evaluation and improving dynamic labeling. Figure 1A illustrates a capillary separation of an eight-component protein sizing ladder (32 µg/ mL total protein concentrations∼4 µg/mL for each protein) using 5% PEO (100 kDa), 250 mM Tris-CHES, 0.1% SDS separation buffer with 0.2% dye included in the separation buffer; conditions similar to those described here have been utilized by several groups for CE-SDS analysis using UV detection (e.g., refs 14 and 18). This separation buffer did not improve upon the detection sensitivity observed using the commercial CE-SDS separation buffer; however, issues with respect to polymer hydrophobicity, self-coating of the capillary for EOF suppression, and viscosity have been addressed. II. Evaluation of Dye Interaction with Sample and Separation Buffer Components. Addition of Dye to Sample Matrix. To achieve low detection limits, it is necessary to understand and manipulate the interaction of dye with the sample and separation buffers so that background fluorescence is reduced. As stated previously, it was hypothesized that the hydrophobicity of the commercially available sieving buffer was the source of this high background and, consequently, the poor sensitivity when dye was included only in the separation buffer. We also speculated that binding of the dye to components in the buffer would reduce the Analytical Chemistry, Vol. 76, No. 16, August 15, 2004

4707

Figure 1. Separation of protein sizing standards in a capillary using dynamic labeling. Protein is at a total concentration of 32 µg/mL. Separation buffer was 250 mM Tris, 250 mM CHES, pH 8.7, with 0.1% w/v SDS, 0.2% dye, and 5% w/v 100 000 MW PEO. Capillary was 50-µm diameter, 27 cm long (20 cm to detection window) fused silica. Sample was injected for 10 s at a field strength of 370 V/cm. Separation field strength was 370 V/cm. (A) Separation of sample with 0.2% v/v dye included in the separation buffer. (B) separation of 100 µg/mL total protein with dye included in sample matrix. (C) dye included in both 100 µg/mL total protein sample and separation buffer. Peaks: 1, lysozyme, 14.4 kDa; 2, trypsin inhibitor, 21.5 kDa; 3, carbonic anhydrase, 31.0 kDa; 4, ovalbumin, 45.0 kDa; 5, serum albumin, 66.2 kDa; 6, phosphorylase b, 97.0 kDa; 7, β-glactosidase, 116.0 kDa; 8, myosin, 200.0 kDa.

dye available for binding to the protein analytes; thus, the effect of inclusion of dye in the sample matrix (25 mM Tris-CHES, 0.1% SDS, 1 mM DTT) was explored to increase the amount of dye bound to protein. Figure 1B shows the electropherogram for a 100 µg/mL (total protein) ladder with 0.2% dye added to the sample matrix after heat denaturing, with no dye in the separation buffer. A single peak is observed that is attributed to dye-saturated 4708 Analytical Chemistry, Vol. 76, No. 16, August 15, 2004

SDS micelles. Note that no peaks corresponding to the protein components of the ladder are observed. Including dye in both the protein sample and the separation buffer results in the same system peak and poorly resolved protein mix components (Figure 1C). A larger standard concentration (100 µg/mL versus 32 µg/ mL) was used in Figure 1B and C to demonstrate the extent of sensitivity lost upon addition of dye to the sample matrix. This indicates that addition of dye in the sample matrix is actually detrimental to resolution and detection of the ladder components in capillary separations. These results are contrary to those made by Banks and co-workers28 when using Sypro Red for on-column labeling of proteins. They found that inclusion of dye in the separation buffer did not allow for good sensitivity as dye preferred binding to SDS micelles. They were, however, able to achieve 100 ng/mL detection limits for BSA on capillaries by including dye only in the sample matrixsbut the authors noted poor efficiencies due to dispersion, with peak widths on the order of 0.5-1 min. SDS Concentration. SDS is included in the separation buffer to maintain the charge-to-size ratio stability of protein-SDS complexes during separation. With SDS-PAGE and CE-SDS analyses, the concentration of SDS is not often considered, as its presence at high concentrations (0.1-1.0%) does not significantly affect the sensitivity or resolving power of these systems adversely (i.e., 2, 8, 14, 20). However, the hydrophobic nature of SDS, particularly as micelles, could affect the sensitivity of the labeling system. To test the effect of SDS on background fluorescence, dye was added to various combinations of components (PEO, SDS, Tris-CHES) for the new separation buffer to determine contributions to the background fluorescence. As reported in Table 1, addition of SDS always increases the background fluorescence, due to the increased hydrophobic character of the buffer. Interestingly, the background fluorescence of 1% SDS is ∼20 times higher than that of 0.1% SDS (3.4 mM)sexceeding the 10-fold increase that would be expected if SDS concentration was the sole contributor. The increase observed was rationalized as resulting from micellization of SDS, creating stable hydrophobic pockets for binding the dye. This is supported by the fact that inclusion of SDS at 0.1% to 250 mM Tris-CHES resulted in the same 20fold increase in background fluorescence as observed with 1% SDS in water. It is known that SDS micellization can be reduced to between 0.03 and 0.1% (1-3 mM) in high ionic strength buffer, as anticipated for the system used here.18 Titration of the SDS concentration from 0.1 to 0% in 250 mM Tris-CHES containing 5% PEO resulted in no appreciable change in background fluorescence until 0.04% was reached. Here, the signal dropped 12% from ∼200 relative fluorescence units (RFU) to ∼175 RFU. At this concentration, it is likely that SDS is not completely micellized, thus reducing the background fluorescence and making more dye available for binding to protein-SDS complexes (as it is not bound to SDS micelles in the separation buffer). Using a new buffer system (250 mM Tris-CHES, 0.04% SDS, 5% PEO), the protein standard mixture was analyzed by CE and the results were compared to the separation obtained using the separation buffer containing 0.1% SDS. Resolving power was only marginally affected, there was no obvious effect on separation (28) Harvey, M. D.; Bandilla, D.; Banks, P. R. Electrophoresis 1998, 19, 21692174.

Table 1. Contribution to Background Fluorescence Made by Various Combinations of Buffer Components All Containing 0.2% Dyea

a

solution

RFU

solution

RFU

250 mM Tris-CHES 0.1% SDS 1.0% SDS 250 mM Tris-CHES, 0.1% SDS 250 mM Tris-CHES, 5.0% PEO

1 9.5 195.4 196.0 13.6

250 mM Tris-CHES, 0.10% SDS, 5.0% PEO 250 mM Tris-CHES, 0.08% SDS, 5.0% PEO 250 mM Tris-CHES, 0.06% SDS, 5.0% PEO 250 mM Tris-CHES, 0.04% SDS, 5.0% PEO 250 mM Tris-CHES, 0.02% SDS, 5.0% PEO

201.1 194.7 201.3 176.0 105.1

Measurements were made on a Beckman 5510 with an LIF detector by flowing the solution by the detector window.

Figure 2. (A) Log(MW) versus 1/MT for the eight-component protein ladder on the microchip using 5% PEO, 250 mM Tris-CHES, 0.04% SDS separation buffer. (B) Relative peak area of BSA normalized against a fixed concentration of myosin versus concentration of BSA from 500 ng/mL to 50 µg/mL.

time, and most importantly, the signal was increased 2.5-fold, with no appreciable change in the noise level. A limit of detection of 500 ng/mL (S/N ) 3) was determined for BSA using these separation conditions. III. Microchip Electrophoresis: Detection Sensitivity and Range. The 250 mM Tris-CHES, 0.04% SDS, 5% PEO separation buffer system established in capillary separations was utilized in a traditional cross-tee injection configuration microchip for protein analysis. When equivalent results were not obtained (data not shown), the fluorescent dye concentration was increased to improve sensitivity. A dye concentration of 0.8% provided the best sensitivity on the microchip, with a detection limit of 1 µg/mL for BSA (S/N ) 3). Interestingly, use of this dye concentration in capillary separations resulted in increased noise levels with no appreciable increase in signal (data not shown). The reasons for the enhanced sensitivity by addition of dye to the separation buffer are not immediately understood; however, it was possible that equilibrium between dye and sample proteins was not established during the few hundred seconds of separation. Conversely, in capillaries the separation time was doubled, possibly allowing equilibrium to be achievedsthis may explain why adding more dye in capillaries does not improve the signal-to-noise ratio. In addition to a sensitive detection limit, an effective protein separation method should allow protein sizing and have a detector response that is linear with respect to concentration. A log molecular weight versus 1/migration time plot yielded an R2 value of 0.9957, indicating the ability to size proteins on the microchip (Figure 2A). To test the linearity of this method with concentration, BSA at various dilutions was prepared in sample matrix and electrophoresed. Initially, detector response was not linear with respect to concentration due to run-to-run differences in chip

alignment. Myosin from rat muscle was, therefore, added as an internal standard to all samples at a fixed concentration (myosin, 20 µg/mL). Myosin dissociates into three peaksstwo light chains (∼20 000 Da), which are baseline resolved under the electrophoresis separation conditions used here, and one heavy chain (∼200 000 Da). The peak area of BSA was normalized to the peak area of the larger myosin light chain and correlated to BSA concentration. The normalized peak area is linear with concentration between 500 ng/mL and 50 µg/mL for BSA, with an R2 value of 0.9955 (Figure 2B). If the dye binds to protein-SDS complexes in a nonspecific fashion, similar to the binding of SDS to proteins, then binding should be “per mass” in nature. This implies that the limit of detection should not be thought of in terms of molarity, but as a concentration limit of detection (CLOD) in terms of micrograms per milliliter. Thus, a per-mass LOD of 500 ng/mL yields a molar LOD of 35 nM for a 14 400 Da protein but 2.5 nM for a 200 000 Da protein. To verify per-mass binding as not being protein-specific, the peak area response for BSA was used to determine the concentration of phosphorylase b in solution. A 25 µg/mL sample of phosphorylase b was initially electrophoresed by itself to verify peak migration at the expected protein size. The protein was found to be contaminated with several unknown species with the expected peak accounting for ∼70% of the total peak area. Phosphorylase b was then coinjected with myosin and the relative peak area used to determine the concentration of the main phosphorylase b peak. Normalized peak reproducibility had a standard deviation of 10%, comparable to the reproducibility using the Agilent BioAnalyzer 2100.26 For a particular experiment, the main phosphorylase b peak concentration was calculated to be 18.4 µg/mL. When corrected to account for the sample impurities, the total concentration was 26.3 µg/mL, or a 4.9% error Analytical Chemistry, Vol. 76, No. 16, August 15, 2004

4709

Figure 3. Sample matrix components comparison between capillary and microchip. (A) 25 mM Tris-CHES, 1 mM DTT on capillary. (B) 25 mM Tris-CHES, 1 mM DTT on microchip. (C) 1.0% SDS 25 mM Tris-CHES, 1 mM DTT on capillary. (D) 1.0% SDS, 25 mM Tris-CHES, 1 mM DTT on microchip. (E) 1.0% SDS, 25 mM Tris-CHES, 1 mM DTT with 1.6% dye included on capillary. (F) 1.0% SDS, 25 mM Tris-CHES, 1 mM DTT with 1.6% dye included on microchip. (G) 25 mM Tris-CHES, 1 mM DTT with 1.6% dye included on capillary. (H) 25 mM Tris-CHES, 1 mM DTT with 1.6% dye included on microchip.

in concentration determination for the prepared 25 µg/mL sample. IV. Microchip Electrophoresis: Effect of Dye in Sample Buffer. In contrast to the detrimental effect observed in capillaries, increasing the dye concentration increased the sensitivity of the microchip system. The dye concentration was, therefore, titrated from 0 to 2.0% in the sample matrix to explore the effect of this variable. It was found that addition of 1.6% dye to the sample matrix in combination with 0.8% dye in the separation buffer allowed for a CLOD of 500 ng/mL BSA with a signal/noise ratio of 8. No improvements in S/N were observed with dye concentration in excess of 1.6% v/v. Accepting a minimal S/N limit of 3, one can calculate a CLOD of less than 200 ng/mL under these conditions. 4710

Analytical Chemistry, Vol. 76, No. 16, August 15, 2004

This represents a ∼5-fold improvement in microchip limits of detection as reported by Bousse et al.26 Electropherograms from the microchip and capillary systems were further compared in an attempt to understand the differences between the separations. Injection of 25 mM Tris-CHES, 1 mM DTT in the absence of SDS produces the same profile in capillaries and in microchips. A baseline drop, always observed upon injection and separation of a sample matrix, is followed by a peak that can be attributed to dye-bound SDS micelles (Figure 3A,B). Injection of the normal sample matrix, which has a higher SDS concentration (0.1% SDS) than the separation buffer (0.04%) results in two system peaks in both the capillary and microchip separations. One

Figure 4. Pictorial representation of capillary and microchip electrokinetic injection.

of these peaks is again attributed to the dye-bound SDS-micelles (MC-dye), the other is attributed to monomeric SDS-bound dye (SDS-dye). Panels C and D of Figure 3 illustrate, however, that the appearance of the system peaks varies drastically between the two systems. On capillaries, the SDS-dye peak dominates and there are two baseline perturbations, the first between the SDS-dye and MC-dye peaks (due to the sample matrix), while the second occurs after the MC-dye peak and is attributed to a region that lacks dye trailing the MC-dye peak. On the microchip, the MC-dye and the SDS-dye peak are similar in magnitude, the sample matrix-induced drop in baseline is not resolved (Figure 3D), but the dip associated with the absence of dye following the MC-dye peak is observed. The less obvious nature of this drop can be attributed to the higher concentration of dye used in the microchip experiments. Conducting these same experiments with dye in the sample matrix shows markedly different profiles for the capillary and microchip systems (Figure 3E-H) Repeating the experiments with SDS-containing sample matrix, now also containing 1.6% dye, the capillary separation showed only the MC-dye peak and the slight dip expected for the sample matrix (Figure 3E). As expected, the dip attributed to absence of dye following the MCdye peak is not present due to the excess of dye in the sample matrix, but there is also no SDS-dye peak. This is ascribed to the induced micellization of SDS in the sample matrix in the presence of the dye. The corresponding microchip experiment shows an increase in the MC-dye peak but maintains the SDSdye peak (Figure 3F). This is not unexpected due to the differences between the electrokinetic injections seen in capillaries and microchips (Figure 4sdiscussion to follow). Injection of 25 mM Tris-CHES, 1 mM DTT containing 1.6% dye in the absence of SDS produced two peaks with the microchip and only one with the capillary. The peaks observed on the microchip are attributed to SDS- and MC-dye, respectively (Figure 3H). Inclusion of dye

Figure 5. Comparison of 11% acrylamide SDS-PAGE gel separation of partially purified human plasma with microchip-based CE-SDS analysis. Microchip separations as described in Figure 4B.

in 25 mM Tris-CHES, 1 mM DTT with no SDS yields an anomalous peak tentatively identified as SDS-dye on the capillary (Figure 3G). The dye is net neutral at the pH of the sample and Analytical Chemistry, Vol. 76, No. 16, August 15, 2004

4711

Figure 6. Comparison of CZE analysis of human sera to partially denatured sera using the 0.04% dye containing run buffer. CZE analysis used a 67.5-cm capillary (8 cm effective length) with 100 mM Borate, 3 mM diaminobutane, pH 8.5; 30 kV separations. Sera were diluted 1:20 in water. For microchip analysis, sera are diluted 1:500 in 0.5% SDS, with 1% dye included.

separation buffers (pH 8.7), thus, if included in sample matrix, should not be injected in the absence of a carrier anion, such as SDS (demonstrated in Figure 3E). We speculate that dye may itself micellize causing charge condensation to a net negative species, which is capable of being injected. Upon injection, the dye-micelle would bind free SDS in the separation buffer to produce the observed SDS-dye peak. Note, however, that this is only speculation made in the absence of any structural and concentration details for the dye. This separation system exhibits the fundamental difference in capillary versus microchip electrokinetic injectionsswith capillaries, sample is injected into the separation channel with partial separation of sample components during the injection, while on microchips, sample is injected across the separation channel with a truly representative sample present in the cross-t (after a particular time) (Figure 4). The microchip injection appears to 4712

Analytical Chemistry, Vol. 76, No. 16, August 15, 2004

result in a concentrating of SDS-dye and MC-dye (as seen in Figure 3E, F, and H) on both sides of the sample matrix (the side toward the inlet and the side toward the outlet), whereas in capillaries, the side of the sample matrix facing the outlet is disrupted during injection and the concentrating effect is seen only on the side toward the inlet. Ideally, one would compare a capillary pressure injection with the microchip electrokinetic injection; unfortunately, due to the viscosity of the separation buffer, pressure injection results are not reproducible. What this means is that, for LIF detection of proteins using dye, the conditions for achieving maximum sensitivity are system dependent. V. Microchip Electrophoresis: Clinical Application. The utility of this labeling method under CE-SDS separation conditions was tested using partially purified human plasma that was diluted in sample buffer and electrophoresed on a microchip. Figure 5

Figure 7. Separation of human serum under CZE conditions: 150 mM borate, 3 mM diaminobutane, pH 8.5. Inset: an overlay of 100 µg/mL BSA and transferrin, 150 mM borate, pH 9.5.

shows the separation of a sample diluted 200× with protein peaks sized at 18.3, 55.1, and 84.3 kDa. For comparison, the sample (20× dilution in SDS-PAGE sample buffer) electrophoresed on a 10% SDS-PAGE gel yielded protein sizes of 15.1, 50.8, 68.5, and 85.0 kDa. The estimated sizes are in good agreement; however, the resolution in the SDS-PAGE separation is better than that observed on the microchip. This is attributed to a better than 10fold increase in resolving power of the SDS-PAGE gel for proteins smaller than 50 kDa (data not shown). The inherent limitation of a universally wide separation range, as depicted in the linearity of the Log(MW) versus 1/MT plot (Figure 2A), is observed. In allowing for a wide sizing range, we have sacrificed resolving power of midsized fragments. This resolution issue may be addressed by altering the concentration and molecular weight of PEO for the particular separation required. As stated previously, 200K PEO was not utilized due to viscosity issues; however, its use may be warranted under certain circumstances. One application of particular interest is the identification of gammopathies by human serum protein profiling.29-31 Figure 6A-C shows traditional CZE separations of human sera in a high pH buffer: (A) is that of normal serum, (B) demonstrates a spike in the β-region, and (C) shows a spike in the γ-region protein concentration. It would be ideal if these CZE separation conditions could be linked with a dynamic labeling approach27ssimple, sensitive LIF detection would result as would the potential for translating to the microchip for on-chip diagnostics. While addition of 0.4% dye did not affect the mobilities of the serum components (UV detection)sLIF detection with a diaminobutane-containing borate buffer,32 showed that the γ-region was not detected, the β-region was only partially apparent, and the albumin peak was so intense that it made detection of the R-region impossible (Figure 7). In addition, rather than maintaining the relative peak intensities observed via UV detection, the albumin peak is nearly (29) Clark, R.; Katzmann, J. A.; Kyle, R.; Fleisher, M.; Landers, J. P. Electrophoresis 1998, 19, 2479-2484. (30) Roche, M. E.; Oda, R. P.; Landers, J. P. Biotechnol. Prog. 1997, 13, 659668. (31) Katzmann, J. A.; Clark, R.; Sanders, E.; Landers, J. P.; Kyle, R. A. Am. J. Clin. Pathol. 1998, 110, 503-509. (32) Landers, J. P.; Oda, R. P.; Madden, B.; Spelsberg, T. C. Anal. Biochem. 1992, 205, 115-124.

100 times more intense than the β-region peak. This is likely due to a limitation in the mechanism of binding in a non-SDScontaining separation systems, where the dye binds to exposed hydrophobic surfaces on proteins. Hrkal33 showed, using hydrophobic interaction chromatography, that albumin is the most retained of the serum proteins, indicating a high degree of surface hydrophobicity, casting the above results in a light that is not surprising. The inset in Figure 7 further emphasizes this, showing an overlay of 100 µg/mL BSA with 100 µg/mL transferrin (a β-region protein). Ultimately, the limited dye binding to γ-, β-, and R-region proteins in CZE separations makes it difficult to assign any diagnostic potential to this technique. Interestingly, the separation conditions described in this paper are not limited to purely sizing purposes. A sample protein dissolved in SDS-containing sample matrix without heat denaturing does not unfold fully, but enough SDS binds to the protein surface to impart a negative charge to provide it with adequate electrophoretic mobility to mobilize it past the detector. This does not provide an accurate sized-based separation, but the surfacebound SDS does allow dye labeling for LIF detection. We applied the (pseudo) partially denatured conditions (dissolved in SDS, but no heat denaturing) in the analysis of the samples used in Figure 6A-C. Figure 6D-E shows microchip analysis of three human serum samples under these conditions. (D) shows a normal serum profile, (E) shows the profile of a sample with a spike in the β-region, and (F) contains a spike in the γ-region (peak migration time variability in (E) and (F) was a result of run-to-run variance in chip alignment). Diagnostic capabilities for disorders that are revealed by serum protein profile abnormalities are maintained using the partially denaturing microchip separationsin contrast to CZE conditions where diagnosis is not possible. This method offers an alternative to detection of serum proteins that does not suffer from preferential labeling as demonstrated by Figure 7 or work by Colyer et al.34 where a postcolumn reaction with 2-toluidinonaphthalene-6-sulfonate is used to detect a “synthetic serum” (a mixture of purified serum proteins) sample. They noted that dye bound preferentially to albumin, requiring a distorted sample composition to visualize all components.34 Further studies need to be performed in order to determine whether the microchip separation is sensitive enough to pick up subtle differences between normal and abnormal sera, such as minor γ-region elevations or spikes. The above analysis, however, illustrated the versatility of dynamic labeling with dye using samples not fully denatured. CONCLUSIONS On-chip protein analysis with limits of detection in the lownanograms per milliliter range have been demonstrated. The addition of dye to the separation buffer and sample matrix allows for a simple method for noncovalent labeling of protein-SDS complexes for sizing. The detection limit has been improved since our original efforts by buffer optimization to reduce overall hydrophobicity. The method is amenable to both the capillary and microchip format, with limits of detection superior on the microchip. We determined that dye binds in a per-mass fashion and detector response is linear with respect to protein concentra(33) Hrkal, Z.; Rejnkova, J. J. Chromatogr. 1982, 242, 385-388. (34) Colyer, C. L.; Tang, T.; Chiem, N.; Harrison, D. J. Electrophoresis 1997, 18, 1733-1741.

Analytical Chemistry, Vol. 76, No. 16, August 15, 2004

4713

tion over 2 orders of magnitude. The utility of this labeling technique was tested in the analysis of human plasma and sera. Diagnostic capabilities were maintained for the detection of elevated β-region proteins and γ-region spikes using a partially denaturing separation, where sample was dissolved in SDScontaining sample matrix, but not heated. Future work will focus on developing clinical applications on microchips for denaturing

4714

Analytical Chemistry, Vol. 76, No. 16, August 15, 2004

and partial-denaturing conditions, as well as developing a fundamental understanding of on-chip injection phenomenon, which allows for the sensitive on-chip detection. Received for review October 2, 2003. Accepted April 6, 2004. AC030349F