Anal. Chem. 2003, 75, 986-994
Hydroxypropyl Cellulose as an Adsorptive Coating Sieving Matrix for DNA Separations: Artificial Neural Network Optimization for Microchip Analysis Joshua C. Sanders,† Michael C. Breadmore,† Yien C. Kwok,† Katie M. Horsman,† and James P. Landers*,†,‡
Departments of Chemistry and Pathology, University of Virginia, Charlottesville, Virginia 22904
Effective DNA separations in microelectrophoretic systems are complicated by the need to passivate the surface dynamically or covalently. We describe the optimization and utilization of a novel buffer system for fast DNA separations by capillary and microchip electrophoresis without the need for any surface modification or conditioning prior to separation. At concentrations as high as 5%, hydroxypropyl cellulose (HPC) has a relatively low viscosity, allowing for microchip channel filling to be performed with ease. A MES/TRIS buffer system at pH 6.1 eliminates the need for surface preconditioning procedures due to the promotion of hydrogen bonding of HPC with the wall. An additional benefit with this buffer system is the low current observed at high fields when compared to other common DNA separation buffers. An artificial neural network (ANN) was used to model the data and to predict the optimum conditions. Utility of the ANN-optimized system for molecular diagnostic testing was demonstrated by performing microchip separations on DNA samples from patients suspected of having genetic mutations associated with Duchenne muscular dystrophy (DMD). Microchip analysis easily allowed for the patient samples positive for DMD mutations to be distinguished from patient samples negative for the disease. Over the past decade, extensive efforts have been directed toward the miniaturization of analytical systems. The many advantages include high-throughput capabilities, reduced sample and solvent consumption, increased portability, and reduction in analysis time.1 In the area of clinical and molecular diagnostics, the latter two are highly significant because they create the opportunity to develop a portable point-of-care (POC) device that could allow rapid diagnosis of various disease states. This was recognized early in the development of the microchip platform, and as such, considerable attention has been paid to the separation of PCR-amplified DNA fragments. Accordingly, numerous ex* Corresponding author. Phone: 434-243-8658. Fax: 434-243-8852. E-mail:
[email protected]. † Department of Chemistry. ‡ Department of Pathology. (1) Dolnik, V.; Liu, S.; Jovanovich, S. Electrophoresis 2000, 21, 41-54.
986 Analytical Chemistry, Vol. 75, No. 4, February 15, 2003
amples can be found in the literature illustrating the benefits of microchips for detecting genetic mutations.2-5 The electrophoretic separation of DNA fragments is based on size, and while this has traditionally been performed using rigid slab gel (acrylamide or agarose) systems, a much simpler approach has been provided by “sieving matrixes” consisting of a soluble polymer added to a background electrolyte.6 While there has been extensive research into the appropriate choice of sieving matrix (recently reviewed in ref 7), selection of the “best” system is not a trivial matter and depends on the application. As such, several different polymers [e.g., hydroxyethyl cellulose (HEC), hydroxypropyl cellulose (HPC), agarose, linear polyacrylamide (LPA), and poly(ethylene oxide) (PEO)] have all been employed effectively for DNA separations.8-21 One significant drawback to this approach, however, is the need to control the magnitude of the electroosmotic flow (EOF). Various methods have been developed to eliminate the influence of EOF on DNA size separations. Early approaches involved (2) Munro, N. J.; Snow, K.; Kant, J. A.; Landers, J. P. Clin. Chem. 1999, 45, 1906-1917. (3) Tian, H.; Jacquins-Gerstl, A.; Munro, N.; Truco, M.; Brody, L.; Landers, J. P. Genomics 2000, 63, 25-34. (4) Sung, W. C.; Lee, G. B.; Tzeng, C. C.; Chen, S. H. Electrophoresis 2001, 22, 1188-1193. (5) Cantafora, A.; Blotta, I.; Bruzzese, N.; Calandra, S.; Bertolini, S. Electrophoresis 2001, 22, 4012-4015. (6) Grossman, P. D.; Soane, D. S. J. Chromatogr. 1991, 559, 257-266. (7) Albarghouthi, M. N.; Barron, A. E. Electrophoresis 2000, 21, 4096-4111. (8) Barron, A. E.; Sunada, W. M.; Blanch, H. W. Electrophoresis 1995, 16, 6474. (9) Shihabi, Z. K. J. Chromatogr., A 1999, 853, 349-354. (10) Ren, J.; Ulvik, A.; Refsum, H.; Ueland, P. M. Anal. Biochem. 1999, 276, 188-194. (11) Madabhushi, R. S.; Vainer, M.; Dolnik, V.; Enad, S.; Barker, D. L.; Harris, D. W.; Mansfield, E. S. Electrophoresis 1997, 18, 104-111. (12) Madabhushi, R. S. Electrophoresis 1998, 19, 224-230. (13) Kim, Y.; Yeung, E. S. Electrophoresis 1998, 18, 2901-2908. (14) Gao, Q.; Yeung. E. S. Anal. Chem. 1998, 70, 1382-1388. (15) Liang, D.; Song, L.; Zhou, S.; Zaitsev, V. S.; Chu, B. Electrophoresis 1999, 20, 2856-2863. (16) Song, L.; Fang, D.; Kobos, R. K.; Pace, S. J.; Chu, B. Electrophoresis 1999, 20, 2847-2855. (17) Carrilho, E. Electrophoresis 2000, 21, 55-65. (18) Chiari, M.; Cretich, M.; Horvath, J. Electrophoresis 2000, 21, 1521-1526. (19) Ren, J.; Fang, Z. F. J. Chromatogr., B 2001, 761, 139-145. (20) Tseng, W. L.; Hsieh, M. M.; Wang, S. J.; Huang, C. C.; Lin, Y. C.; Chang, P. L.; Chang, H. T. J. Chromatogr., A 2001, 927, 179-190. (21) Huang, M. F.; Hsu, C. E.; Tseng, W. L.; Lin, Y. C.; Chang, H. T. Electrophoresis 2001, 22, 2281-2290. 10.1021/ac020425z CCC: $25.00
© 2003 American Chemical Society Published on Web 01/17/2003
covalent modification of the silica surface23-26 to reduce the influence of EOF, and while this has been applied to microchips,27-29 the process is often laborious and time-consuming. For example, channel surface modification with poly(vinylpyrrolidone) (PVP) yields a robust microchip for DNA separations, but the coating procedure involves a multistep, 18-h procedure.29 In this light, noncovalent coatings provide an attractive and potentially simpler approach to EOF reduction.9-16,19 Since these coatings are not covalently attached to the surface of the capillary wall, they can easily be removed. Moreover, selection of the appropriate polymer enables the same substance to function as both the “coater” and the sieving matrix, greatly simplifying the system.9,13,14 However, even with the advantage of being able to remove the surface coating, surface preparation is still required in many cases to successfully coat the capillary or microchannel. Recently, an approach has been presented that maximizes EOF to separate DNA fragments in a counter-EOF mode. This contrasts the more traditional methods since the fragments migrate in reverse order, i.e., largest to smallest.20,21 Although adequate separations could be obtained, the technique is not robust due to a need for stringent control of EOF. Because of this, there was a need to flush with NaOH to replenish the capillary surface prior to each separation. Recent work in our laboratory has been directed toward the reduction of capillary/microchannel surface preparation prior to analysis by the use of adsorbed-coating sieving matrixes. DNA separations in an unmodified capillary or microchip have been accomplished following extensive flushing with acid (HCl). Fung and Yeung were the first to show that this was feasible using PEO for DNA sequencing.30 More recently, Shihabi9 demonstrated this phenomenon using a cellulosic polymer (hydroxypropylmethyl cellulose; HPMC), as did Tian and Landers,31 who used HEC. While this was a significant step toward simplification of DNA separations on a microchip platform, the procedure had two flaws. First, the lengthy preconditioning step reduced the overall benefit gained from performing microchip separations. Second, the concentration of polymer required for adequate sieving resulted in a highly viscous solution, which created difficulties with microchip filling. Here, we improve on the previously developed buffer systems to allow for rapid DNA separations on an unmodified capillary or microchip. The use of hydroxypropyl cellulose (instead of HEC or HPMC) allowed for more concentrated solutions to be used without the concomitant increase in solution viscosity, and slightly acidic buffer conditions avoided the lengthy preconditioning step necessary with more basic buffers. Optimization of the buffer system ionic strength, polymer concentration, and separation (22) Dougherty, A. M.; Cooke, N.; Shieh, P. In Handbook of Capillary Electrophoresis, 2nd ed.; Landers, J. P., Ed.; CRC Press: New York, 1997. (23) Gelfi, C.; Curcio, M.; Righetti, P. G.; Sebastiano, R.; Citterio, A.; Ahmadzadeh, H.; Dovichi, N. J. Electrophoresis 1998, 19, 1677-1682. (24) Srinivasan, K.; Pohl, C.; Avdalovic, N. Anal. Chem. 1997, 69, 9, 27982805. (25) Hjerten, S. J. J. Chromatogr. 1985, 14, 191-198. (26) Tian, H.; Brody, L. C.; Mao, D.; Landers, J. P. Anal. Chem. 2000, 72, 54835492. (27) Wooley, A. T.; Mathies, R. A. Proc. Natl. Acad. Sci. U.S.A. 1994, 91, 1134811352. (28) Wooley, A. T.; Mathies, R. A. Anal. Chem. 1995, 67, 3676-3680. (29) Munro, N. J.; Huhmer, A. F. R.; Landers, J. P. Anal. Chem. 2001, 73, 17841794. (30) Fung, E. N.; Yeung, E. S. Anal. Chem. 1995, 67, 1913-1919. (31) Tian, H.; Landers, J. P. Anal. Biochem. 2002, 309, 212-223.
voltage using artificial neural networks (ANNs), and a new criterion based on the normalized resolution and the normalized time, allowed for fast DNA separations to be achieved at reasonably high fields with low system current. Furthermore, the use of this polymer network for clinical analysis is demonstrated by detecting genetic mutations consistent with Duchenne muscular dystrophy (DMD). MATERIALS AND METHODS Materials. Hydroxypropylcellulose (HPC; 100 000 MW), HEC (250 000 MW), tris(hydroxymethylamino)methane (TRIS), and 2-(4-morpholino)ethanesulfonic acid (MES) were all purchased from Acros Organics. Sodium tetraborate and EDTA were purchased from Sigma (St. Louis, MO). YOPRO-1 was purchased from Molecular Probes (Eugene, OR). Polymer Network/Buffer Solutions. Polymer solutions were prepared incrementally, increasing polymer (HPC) concentration by 0.5% per increment (w/v) between 1.0% and 5.0%. A stock solution containing 1 M MES and 0.5 M TRIS (pH 6.1) was prepared and utilized for all buffers throughout the experiments. For each polymer concentration, five different ionic strengths of MES/TRIS were utilized ranging from 5/2.5 to 105/52.5 mM in steps of 25/12.5 mM. The appropriate mass of polymer was weighed and transferred to a 25-mL volumetric flask to which the appropriate amount of buffer was added from the stock solution. Water was added to the flask, and the solution was stirred overnight. Following stirring, the solution was filled to the mark and subsequently stirred (∼10 min) to allow for sufficient mixing. Upon mixing, the solution was degassed and filtered through a 5-µm filter. Intercalating dye (YOPRO-1) was added to a concentration of 1 µL/mL. Kinematic polymer viscosities were measured using a VI-VG-9100 bubble viscometer (Gardner, Pompano Beach, FL). Capillary Electrophoresis. For capillary electrophoresis experiments, a Beckman P/ACE 2100 (Fullerton, CA) was utilized with a 27-cm bare silica capillary (50-µm i.d.) (Polymicro Technologies, Phoenix, AZ) with an effective length of 7 cm to the detector window. Sequences were designed to evaluate each polymer concentration with the aforementioned ionic strengths. Each concentration/ionic strength combination was evaluated under field strengths ranging from 150 to 595 V/cm consecutively in steps of 75 V/cm. Prior to each polymer/ionic strength solution, the capillary was rinsed for 10 min. Samples consisted of 2.5 µg/ mL φX-174-HaeIII dsDNA digest dissolved in 1/10 the ionic strength of the separation buffer. This allowed for an equal amount of sample injected for the different ionic strengths. Electrokinetic injections were performed by applying 3 kV (110 V/cm) for 6 s to the sample vial. For comparison, a previously published protocol based on 2.5% HEC/1× TBE (89 mM TRIS/89 mM Borate/2 mM EDTA) was utilized. All other conditions were as previously mentioned, and separation occurred with an optimized field strength of 370 V/cm. Following this experiment, 2.5% HEC was dissolved in the optimized ionic strength MES/TRIS buffer (80 mM/40 mM). A fresh capillary was utilized, and buffer was flushed for 10 min. For separation, a field strength of 370 V/cm was utilized. Optimization Using an Artificial Neural Network. The artificial neural network software used here was Trajan (Trajan House, Durham, U.K.) installed on a Sony laptop computer. Analytical Chemistry, Vol. 75, No. 4, February 15, 2003
987
Training was performed using standard back-propagation methods with a total of 5000 epochs. Microchip Fabrication and System Setup. Microchips were fabricated by using standard photolithography and wet chemical etching techniques. The template was designed with AutoCAD and printed on a transparent film with 3600 dpi resolution. The design was transferred onto a glass wafer (Nanofilm, Westlake Village, CA) with positive photoresist using UV exposure. The channels were etched with concentrated HF solution (HF:HNO3: H2O ) 20:14:66). Small holes (1.1 mm) coinciding with the ends of the channels were drilled with diamond drilling on a cover plate. The etched plate and the cover plate were bonded together via thermal bonding. Each microchip contained an 8-cm separation channel and a 1-cm sample injection channel, which was positioned 0.5 cm from the buffer inlet reservoir. All channel widths were 60 µm at half-height and 30 µm deep. Microchip electrophoresis with laser-induced fluorescence (LIF) detection utilizing a 488-nm argon ion laser (Laser Physics, Salt Lake City, UT) was performed in a fashion similar to that previously described.29,32-34 For excitation, the laser beam passed through a beam expander (Spindler & Hoyer, Milford, MA)and a dichroic beam splitter with a cut on 520 nm (Optometrics, Nayer, MA) and into a 16× microscope objective (Melles Griot, Rochester, NY), which focused the laser onto the separation channel 1 cm from the separation channel outlet giving the microchannel utilized in this study an effective length of 6.5 cm. Emitted fluorescence was collected through the same microscope objective and back through the beam splitter. The light was then filtered through a 560-nm (40-nm band-pass) filter (Optometrics, Nayer, MA), passed through a pinhole (Spindler & Hoyer, Milford, MA) and onto a photomultiplier tube (PMT) (Hamamatsu, Bridgewater, NJ). Signal processing was performed by an independent microprocessor (SBC2000-332, Vesta, Wheat Ridge, CO), which gathered the signal from the PMT and transferred this signal to a computer (AMD K6) via a RS232 communication port. Data were collected and displayed using an in-house Labview (National Instruments, Austin, TX) program allowing for multiple continual separations and automatic data storage. This Labview software also allowed for control of an in-house designed and built power supply. Microchip Separations. For microchip separation of DNA, buffer was flushed through the channels (volume ∼120 nL) via vacuum (600 Torr) for 10 min following a 1-min flush with deionized water. φX174-HaeIII dsDNA digest dissolved in water to a final concentration of 2.5 µg/mL was added to the sample inlet reservoir. Amplified DMD sample DNA was diluted 40-fold in deionized water from the original concentration (as supplied by Dr. Karen Snow, Mayo Clinic, Rochester, MN). For sample injection, a potential of 300 V (300 V/cm) was applied to the sample waste reservoir, while the sample inlet reservoir was held at ground and both the buffer inlet and outlet reservoirs allowed to float. Injection occurred for 70 s. Following sample injection, subsequent separation occurred by applying a potential of -500 (32) Huang, Z.; Munro, N.; Huhmer, A. F. R.; Landers, J. P. Anal. Chem. 1999, 71, 5309-5314. (33) Huang, Z.; Jin, L.; Sanders, J. C.; Dunsmoor, C.; Zheng, Y.; Tian, H.; Landers, J. P. IEEE BME Trans. 2002, 49 (8), 859-866. (34) Huang, Z.; Sanders, J. C.; Dunsmoor, C. Ahmadzadeh, H.; Landers, J. P. Electrophoresis 2001, 22, 3924-3929.
988
Analytical Chemistry, Vol. 75, No. 4, February 15, 2003
V to the buffer inlet and 2500 V to the buffer outlet while the sample inlet and waste were held at ground. This resulted in a field strength of 375 V/cm, consistent with the optimized separation voltage of 370 V/cm for the capillary separations. RESULTS AND DISCUSSION In a continued effort to examine suitable buffer systems capable of providing simple and rapid size-based separations of DNA in microchip systems, a significant reduction in viscosity associated with the use of HPC solutions in comparison with the HEC buffer typically used for DNA separations was noticed. Using comparable polymer concentrations of each, it was determined that the HPC solution is almost 4-fold less viscous than HEC, based upon the experimentally obtained dynamic viscosities of the solutions [50 cP for HPC and 230 cP for HEC (water ∼1 cP)]. The differences in viscosities, which can also be intuitively determined from the differences in molecular weight, should make HPC more amenable to microchip systems than HEC. Tian et al. had shown that, in addition to being effective as a sieving matrix for DNA separations, HEC in 1× TBE buffer at pH 8.9 functioned to coat the microchannel.31 Surface conditioning with HCl was found to be critical to the ability of HEC to passivate the surface, suggesting that hydrogen bonding is the main force driving this “adsorbed” interaction. Similar observations were made with HPC following extensive surface conditioning with HCl and support this as a possible mechanism. However, it was speculated that the coupling of a lower viscosity polymer (HPC) with reduction of the buffer pH into the acidic range might provide an enhanced separation system. The lower viscosity of HPC would allow for higher concentration polymer solutions to be used without the concordant problems with microchip filling, while the lower pH may enhance surface coating. Since the pKas of silanol groups cover a broad range between 3 and 8, reducing the buffer pH to 295 V/cm (0.538) > 445 V/cm (0.434). While this illustrates the effect of field strength on the separation, it should be noted that these comparisons are made at conditions that are not necessarily optimum at 295 and 445 V/cm and, as such, provide a somewhat unfair comparison. Nevertheless, they do show the effect of increasing field strength on the separation and again the complexity of the DNA migration method.6 By examining separation conditions around the optimum as selected by the criterion NormR,t, plus considering buffer 992 Analytical Chemistry, Vol. 75, No. 4, February 15, 2003
viscosity and the potential for Joule heating, the optimum conditions are defined as 3.5% HPC, 80 mM MES/40 mM TRIS, and 370 V/cm. Comparison of Optimized HPC and HEC Buffer Systems. Since the development of the HPC/MES/TRIS buffer system is an improvement upon the previously developed HEC/1× TBE buffer system, with respect to solution viscosity and preconditioning time, it was important to determine whether this new system was significantly better or just marginally so. For comparison, a separation was performed using the HEC/1× TBE buffer, under optimized conditions as per Tian and Landers,31 without the HCl rinse (Figure 4A). From this separation, resolution of the DNA fragments was quite poor and substantially inferior to that obtained using HPC, giving merit to the need for the copious flushing. The Figure 4A inset shows the same separation but under the conditioning procedures as demonstrated in ref 31. When compared to the optimized conditions utilizing HPC as the polymer (Figure 4B), it is seen that separations obtained with HEC and HPC are comparable. For interest, an additional separation using HEC at the previously optimized concentration of 2.5% was performed using the optimized HPC conditions (80 mM MES/40 mM TRIS). This separation is shown in the Figure 4B inset, where it can be seen that superior resolution is obtained compared to the 1× TBE buffer. Application to Microchip Separations of Molecular Diagnostic Importance. The optimized conditions for the HPC buffer system determined from the CE optimization (3.5% HPC, 80 mM MES, and 370 V/cm) were utilized for the microchip separations, with the separation of the φX174-HaeIII digest dsDNA ladder on a microchip shown in Figure 5A. Resolution of fragments 271 and 281 on the microchip (0.82) was comparable to that in the
Figure 4. Comparison of HPC and HEC buffer systems. A comparison of previously reported buffer system31 without the prescribed 1-h preconditioning with 1 M HCl to the new HPC buffer system was performed. (A) 2.5% HEC with 1× TBE without preconditioning. A separation voltage of 370 V/cm was utilized. Inset shows the same conditions but with 1-h preconditioning with 1 M HCl prior to separation. (B) Optimized separation utilizing the HPC, MES/TRIS buffer system without preconditioning. Separation conditions consisted of 3.5% HPC, 80 mM MES/40 mM TRIS, 370 V/cm. Inset shows separation under same conditions, but with 2.5% HEC as the polymer.
capillary (0.88), illustrating the successful transferal of the method from the capillary to microchip format. This minor loss in resolution can be attributed to a decrease in effective length in the microchannel by 0.5 cm. As a demonstration of the potential of this new buffer system in molecular diagnostics, PCR-amplified DNA samples from patients tested for mutations correlative with DMD were evaluated. Figure 5B shows the separated fragments from amplified DNA of a patient negative for DMD. The multiple peaks represent multiplex PCR-amplified DNA fragments from select exons of the dystrophin gene (Kunkel series).45 These exons were chosen due to their propensities to contain a deletion or duplication mutation causative of DMD. These two mutations occur when a fragment of an exon has either been removed or repeated within that exon. For diagnostic purposes, the peak of interest with this particular sample is exon 47 which, when compared to exons 6 and 13, is smaller. Figure 5C shows a patient with a positive diagnosis for DMD. This is determined from the increase in relative peak height of the 181-bp fragment (exon 47) with respect to the other peaks. In this particular case, the resulting mutation is due to a duplication mutation in exon 47 of the patient, thus doubling the peak intensity of exon 47 with respect to the other PCR-amplified exons in the electropherogram.46 Utilizing this novel buffer system, (45) Beggs, A. H.; Koenig, M.; Boyce, F. M.; Kunkel, L. M. Hum. Genet. 1990, 86, 45-48.
Figure 5. Microchip separations employing the optimized buffer system. (A) Separation of DNA ladder with 3.5% HPC, 80 mM MES/ 40 mM TRIS buffer system with a field strength of 375 V/cm. (B) Separation of a mulitplex PCR sample from a patient negative for DMD utilizing the optimized buffer system. Field strength of 375 V/cm. (C) Separation of multiplex PCR sample of a patient diagnosed with DMD utilizing the optimized buffer system. Field strength of 375 V/cm.
a clinical diagnosis of a relevant disease has been performed in under 250 s under optimized separation conditions in a microchip without the need for surface modification or lengthy conditioning. This is a significant step toward the development of a portable clinical diagnostic device. CONCLUSIONS We have demonstrated the use of a novel buffer system for the separation of DNA fragments in both capillary and microchip format. Parameters such as polymer concentration, ionic strength, and applied voltage were optimized using an artificial neural network. The benefits of this buffer system include no need for surface modification, low current in comparison to other buffer systems for DNA separations, and low viscosity. These advantages make this buffer system ideal for microchip separations. Indeed, with no need for covalent surface modifications or lengthy (46) Ferrance, J.; Snow, K.; Landers, J. P. Clin. Chem. 2002, 48, 380-383.
Analytical Chemistry, Vol. 75, No. 4, February 15, 2003
993
preparation steps, the speed of analysis is increased for microchip DNA analysis. The feasibility of this system for microchip clinical diagnostics was also demonstrated by separating PCR product of samples for DMD detection with resolution comparable to previously reported results. ACKNOWLEDGMENT The authors thank Dr. Karen Snow (Mayo Clinic-Rochester) for samples and Dr. Steve Weber (Chemistry, University of Pittsburgh) for the original suggestion on optimization using ANN. The authors also acknowledge funding supporting this work from
994
Analytical Chemistry, Vol. 75, No. 4, February 15, 2003
the National Institute of Environmental Health Sciences (NIEHS), Grant 1 R24 ES102229-01. SUPPORTING INFORMATION AVAILABLE Additional information as noted in the text. This material is available free of charge via the Internet at http://pubs.acs.org. Received for review June 29, 2002. Revised manuscript received November 20, 2002. Accepted November 30, 2002. AC020425Z
