Continuous Separation of High Molecular Weight Compounds Using a

A microliter volume free-flow electrophoresis microstruc- ture (µ-FFE) was used to ... high field strengths and high flow rates were required. Howeve...
0 downloads 0 Views 366KB Size
Download PDF
Anal. Chem. 1996, 68, 2515-2522

Continuous Separation of High Molecular Weight Compounds Using a Microliter Volume Free-Flow Electrophoresis Microstructure Daniel E. Raymond,*,† Andreas Manz,‡ and H. Michael Widmer

Ciba-Geigy Ltd., Corporate Analytical Research, CH-4002 Basel, Switzerland

A microliter volume free-flow electrophoresis microstructure (µ-FFE) was used to perform a continuous separation of high molecular weight compounds. The µ-FFE microstructure had a separation bed volume of 25 µL and was fabricated from silicon using standard micromachining technology. Laser-induced fluorescence was used to detect the sample components, which were labeled with fluorescein isothiocyanate (FITC) prior to analysis. The continuous separation of human serum albumin (HSA), bradykinin, and ribonuclease A demonstrated that only 25 V/cm was required to isolate HSA from bradykinin and ribonuclease A, while 100 V/cm was needed for the separation of bradykinin from ribonuclease A. Comparison of the observed band broadening with the theoretical variance indicated that dispersion due to the initial bandwidth, diffusion, and hydrodynamic broadening were the main contributors to the band broadening of HSA and bradykinin. However, the band broadening for ribonuclease A could not be sufficiently accounted for using the above contributors. Adsorption was found to be a possible contributor to the overall variance for ribonuclease A. Calculation of the theoretical variance due to Joule heating indicated that broadening due to Joule heating effects was insignificant. This was likely due to the narrow cross-sectional area of the device, which facilitated efficient cooling. Electrohydrodynamic distortion was observed for HSA as it migrated toward the side bed. Studies of the resolution of bradykinin and ribonuclease A as a function of field strength at various sample and carrier flow rates indicated that, for maximum throughput, high field strengths and high flow rates were required. However, no optimal conditions were found. The µ-FFE device has a peak capacity of ∼8 bands/cm, while for a typical separation of proteins using a commercial system, a peak capacity of 10 bands/cm is obtained. Thus, the resolving power of the µ-FFE device is similar to those of conventional systems. The continuous separation of tryptic digests of mellitin and cytochrome c demonstrated the ability to continuously separate more complex mixtures. Finally, modifications were made to the microstructure to facilitate fraction collection, and the fractionation of whole rat plasma was performed. Off-line analysis of the resulting fractions indicated that the complete isolation of serum albumin and globulins was possible using a field strength of 25 V/cm. S0003-2700(95)00766-9 CCC: $12.00

© 1996 American Chemical Society

For the purification or isolation of specific chemical compounds, filtration, ultrafiltration, cross-flow filtration, dialysis, and microdialysis are powerful tools. These methods can be performed as either batch processes or continuous flow techniques. Batch systems, or injection of sample into a flowing stream, allows for the purification of small quantities of sample, whereas the continuous flow approach serves to increase the quantity of sample purified over a given period of time. Regardless of the approach employed, these methods are widely used as preparative or micropreparative tools in sample preparation for analytical applications. Furthermore, a number of separation techniques have been developed for application as preparative methods. Although some methods have been developed solely for purification purposes, many are simply modified analytical techniques.1 For instance, gas chromatography, supercritical fluid chromatography, and liquid chromatography are common preparative methods employed for the purification of a wide variety of materials, ranging from petroleum products to proteins. Although the above techniques are all based on chromatography, various preparative methods also utilize electrophoresis as the separation process. Isoelectric focusing, isotachophoresis, and free zone electrophoresis have been used as purification methods in gel and free solution.1,2 Preparative methods can be either one- (1D) or two-dimensional (2D). Two-dimensional methods are generated from the combination of various 1D methods in such a manner that the sample is subject to two orthogonal displacement processes. The large variety of 1D displacement processes can be divided into two categories: bulk flow effects and systems that discriminate molecular transport. The former includes pressure-induced flow or electroosmotic flow, while the latter includes chromatography (differential migration due to interaction with stationary phase), field flow fractionation variants, electrophoretic migration (isoelectric focusing and isotachophoresis), sedimentation, magnetic gradients, molecular sieving, thermal diffusion, and thermogravitation.3,4 This results in an immense array of possible 2D methods. In most cases, the 2D methods obtained are inherently more powerful than the 1D methods since the resulting peak capacity † Present address: Nanogen Inc., 10398 Pacific Center Ct., San Diego, CA 92121. ‡ Present address: Department of Chemistry, Imperial College of Science, Technology and Medicine, London SW7 2AY, U.K. (1) Proceedings of the 10th Symposium on Preparative Chromatography. J. Chromatogr., A 1994, 648. (2) Catsimpoolas, N. Methods of Protein Separation; Plenum Press: New York, 1975, 1976; Vols. 1 and 2. (3) Giddings, J. C. Anal. Chem. 1984, 56, 1258A-1270A. (4) Wankat, P. C. Sep. Sci. Technol. 1984-85, 19, 801-829.

Analytical Chemistry, Vol. 68, No. 15, August 1, 1996 2515

is increased.3 Two-dimensional methods can also be used as batch processes or as continuous flow techniques. Batch processes include 2D TLC and slab gel methods, while the continuous methods include annular chromatography and the variants of free flow electrophoresis (FFE).2,4 FFE has been used primarily for the purification of biological materials,2 and recently the use of annular chromatography for protein purification was reported.5 The initial concept of continuous free solution zone electrophoresis presented by Philpot6 in 1940 has developed into a variety of working models,6-15 including the recent development of a miniaturized free-flow electrophoretic system (µ-FFE) fabricated out of silicon and glass.16 Since FFE does not use a supporting medium (silica or gel), organic solvents, or high salt concentrations the method is gentle enough for separations involving cells and proteins.17 It is this inherent gentleness and the continuous nature of the method that have encouraged the development of FFE as a preparative technique for biotechnology. However, problems associated with dispersion and sample throughput have limited its use. A recent review discussed the variety of commercial FFE systems available.17 In particular, a “miniature” FFE system, with a cross-sectional area of 60 mm2 and a separation bed length of 4 cm, has been commercialized. Typical residence times for this device were reported to be 30-90 s. Nice features of this system are the on-line UV monitoring at 225 nm and the presence of an autosampling system. Due to the narrow separation bed depth, this system has two drawbacks: the hydrodynamic and electrodynamic distortions and the inability to scale up the sample throughput. In comparison, the micromachined µ-FFE device16 has a cross-sectional area of 50 µm2 and a bed length of 5 cm, with typical residence times of 30-300 s. This structure is unable to incorporate an on-line UV monitor because the bottom of the separation bed is made from silicon and is therefore opaque. For this reason, laser-induced fluorescence was the optical detection method used. Also, because of the narrow separation bed depth, this structure is expected to have the same drawbacks as those for the “miniature” system discussed above. The miniaturized free-flow electrophoretic microstructure (µFFE)16 was developed as a sample pretreatment module for incorporation into a miniaturized total analysis system (µ-TAS).18 The motivation for the development of a µ-TAS originated from the need to carry out on-line monitoring of a chemical process in real time. It is clear that there is the need to incorporate a sample pretreatment step prior to analytical evaluation. For real time online monitoring, it would be ideal to have a continuous sample pretreatment method from which samples could be repetitively (5) Bloomingburg, G. F.; Bauer, J. S.; Carta, G.; Byers, C. H. Ind. Eng. Chem. Res. 1991, 30, 1061-1067. (6) Philpot, J. St. L. Trans. Faraday Soc. 1940, 36, 38. (7) Dobry, R.; Finn, R. K. Chem. Eng. Progr. 1958, 54, 59. (8) Hannig, K. J. Chromatogr. 1978, 159, 183. (9) Barrollier, J.; Watzke, E.; Gibian, H., Z. Naturforch., B 1958, 13, 754. (10) Strickler, A. Sep. Sci. 1967, 2, 335-355. (11) Huebner, V. R.; Lawson, R. H. Sep. Sci. 1968, 3, 265. (12) Shevedov, V. P.; Stepanov, A. V. Russ. J. Phys. Chem. 1961, 35, 102. (13) Hanning. K.; Wirth, H.; Meyer, B. H.; Zeiller, K. Hoppe-Seyler’s Z. Physiol. Chem. 1975, 356S, 1209-1223. (14) Zeiller, K.; Loser, R.; Pascher, G.; Hanning., K. Hoppe-Seyler’s Z. Physiol. Chem. 1975, 356S, 1225-1244. (15) Hanning, K.; Wirth, H.; Schindler, R. K.; Spiegel, K. Hoppe-Seyler’s Z. Physiol. Chem. 1977, 358S, 753-763. (16) Raymond, D. E.; Manz, A.; Widmer, H. M. Anal. Chem. 1994, 66, 28582865. (17) Roman, M. C.; Brown, P. R. Anal. Chem. 1994, 66, 86A-94A. (18) Harrison, D. J.; Flury, K.; Seiler, K.; Fan, Z.; Effenhauser, C. S.; Manz, A. Science 1993, 261, 895-897.

2516

Analytical Chemistry, Vol. 68, No. 15, August 1, 1996

injected into some type of analysis system. Also, it would be ideal to have a system which samples only a small quantity of a crude mixture. The use of FFE as a sample pretreatment method allows for continuous sample pretreatment, while the small dimensions of the µ-FFE device allow for smaller sample volumes to be processed. Thus the motivation for the development of the µ-FFE was not for high-volume purification but rather as a small-volume pretreatment method. Small samples can be processed on conventional systems; however, it defeats the purpose of having a large preparative system which processes only 25 µL of sample. Also, it is not practical to have such a system used for on-line monitoring. It is clear that the µ-FFE device is unable to have the same sample throughput as conventional systems, simply because of the significantly smaller separation bed volume. However, the µ-FFE system would complement the use of commercial instruments, since it could be used to find the optimum conditions for a particular separation, a process which can waste valuable sample and time. The basic FFE characteristics of the µ-FFE microstructure are comparable to those of the conventional systems.16 The initial report on the µ-FFE device described the device characteristics in terms of fluid flow, current-voltage behavior, and separation of small ionic species. Although FFE can be used to separate simple ions, its power and potential are in the separation of macromolecules ranging in size from 100 Å for the smallest viruses to tens of micrometers or larger for many tissue cells. There are many examples of the use of FFE for the isolation of cells, cell organelles, cell membranes, bacteria, viruses, proteins, enzymes, and nucleic acids.17,19-23 A recent book detailing FFE theory and application in cell separation is particularly useful.24 Many of the reported separations of biomolecules were carried out using commercial instruments which, in general, have a separation bed volume of 25 mL. In contrast, the µ-FFE device16 has a separation bed volume of 25 µL (5 cm length, 1 cm width, and 50 µm depth). The large open cross sections of these systems, relative to the size of a typical cell, make the systems ideal for the separation of large particles in a continuous manner. Standard filtration methods often require replacement of the filter or flushing to clear the filter surface. The continuous separation of particles via FFE alleviates these tasks. In this report, the use of the small-volume µ-FFE device for the continuous separation of large biomolecules and the evaluation of the observed band broadening and resolution are presented. Also discussed is the use of the device to separate complex mixtures, such as protein tryptic digests. Finally, modification of the device to facilitate fraction collection for the purification of whole rat plasma is reported. EXPERIMENTAL SECTION Silicon Devices and Equipment. The miniaturized free-flow electrophoretic device, equipment, and experimental setup used were identical to those previously reported.16 The miniaturized (19) Hanning, K.; Heidrich, H. G. Free-Flow Electrophoresis; GIT Verlag GMBH: Darmstadt, Germany, 1990; and references therein. (20) Nath, S.; Schutte, H.; Weber, G.; Hustedt, H.; Deckwer, W.-D. Electrophoresis 1990, 11, 937-941. (21) Clifton, M. J.; Jouve, N.; de Balmann, H.; Sanchez, V. Electrophoresis 1990, 11, 913-919. (22) Hoffstetter-Kuhn, S.; Kuhn, R.; Wagner, H. Electrophoresis 1990, 11, 304309. (23) Kessler, R.; Manz, H.-J. Electrophoresis 1990, 11, 979. (24) Bauer, J. Cell Electrophoresis; CRC Press: Boca Raton, FL, 1994.

Figure 1. µ-FFE detection schemes. (A) Analytical evaluation: Detection occurs by scanning across the separation bed at some fixed detection position. In the approach, all the sample recombines and exits the device via one outlet hole. (B) Fraction collection approach: Fractions are collected and analyzed off-line using capillary electrophoresis. Any sample not collected via the fraction collection holes recombines and exits the device via one outlet hole (as in A above). (a) Left carrier inlet, (b) sample inlet, (c) right carrier inlet, and (d) main outlet.

free-flow electrophoretic devices were fabricated at IC Sensors (Milpitas, CA). The channel systems were first etched into silicon, after which a glass cover plate was anodically bonded to form the channels. The device has been discussed in detail elsewhere,16 but briefly it consists of a separation bed (10 mm wide, 50 mm long, and 50 µm in depth), which is isolated from the electrodes via two arrays of 2500 v-groove channels. Platinum wire (30 µm) was used as electrodes, which were placed into the device via holes in the cover plate. The beds containing the electrodes are 2 mm wide, 50 mm long, and 50 µm deep. The detection system used for the µ-FFE device was laser-induced fluorescence, although off-line capillary electrophoresis was used for analysis of samples obtained via fraction collection. The concept of free-flow electrophoresis (FFE) is shown schematically in Figure 1. In FFE, a narrow sample stream is continuously fed into a carrier solution which flows perpendicular to an applied electric field. Charged species are then deflected from the direction of flow at an angle determined by a combination

of the carrier flow velocity and the respective electrophoretic mobilities of the sample components. In general, the large-scale preparative FFE systems either collect the individual sample bands and analyze these fractions off-line by UV/visible spectroscopy or incorporate some type of on-line monitoring. As with the commercial systems, the µ-FFE device can be used in two modes: analytical evaluation or fraction collection. Figure 1A illustrates the approach used for purely analytical evaluation. In this mode the device is horizontally translated relative to a fixed detection system,16 and in this manner it is possible to obtain a scan across the separation bed width. To simplify fabrication, the µ-FFE device was fabricated with only one exit hole; consequently, the sample components recombine before leaving the device (see Figure 1A). The fraction collection mode is shown in Figure 1B. To achieve this, four holes were drilled through the cover plate near the exit region of the device. The specifics of these holes will be discussed later. This fraction collection approach is identical to conventional FFE; however, in this case the collected fraction were analyzed off-line using capillary electrophoresis. Note that, in this approach, any material that does not exit via one of the fraction collection holes exits via the main outlet hole. Chemicals. Human serum albumin (HSA), bradykinin, bovine ribonuclease A, bovine cytochrome c, mellitin, and trypsin were purchased from Sigma (CH-9470 Buchs, Switzerland). Fluorescein isothiocyanate (FITC), fluorescein (sodium), tris(hydroxymethyl)aminomethane (Tris), and boric acid were purchased from Fluka AG (CH-9470 Buchs, Switzerland). The tryptic digests were performed using a previously reported method.25 Labeling with FITC was performed using the previously reported method.16 The use of FITC and isothiocyanatosubstituted analogues of polymethine as fluorescent tags for bovine serum albumin and human serum albumin have been reported.26,27 These dyes were covalently attached to the primary amine groups (mainly lysine residues) of the protein, and the molar ratio of dye to protein was found to be approximately 1. In our case, the ratio of dye to protein was not known, although matrix-assisted laser desorption/ionization mass spectrometry (MALDI-MS) was used to obtain the molecular weight of the labeled samples. Using the mass-to-charge ratio obtained for unlabeled and labeled protein, it was found that, on average, the numbers of FITC molecules per molecule of HSA, bradykinin, and ribonuclease A were 2, 1, and 2, respectively. The molecular weight values for the HSA, bradykinin, and ribonuclease A were then modified to 69 252, 1436, and 14 453, respectively. It should be noted that labeling of biomolecules can alter their bioactivity; however, this aspect is not addressed here. RESULTS AND DISCUSSION Separation of Large Biomolecules: Mobility Evaluation. In the initial report on the µ-FFE device, a baseline separation of three rhodamine-B isothiocyanate-labeled amino acids was presented.16 The purpose of this report was to illustrate the ability of the device to continuously separate small, low molecular weight ions which differ from each other by one charge unit. This separation should not be understated since various samples of (25) Schar, M.; Bornsen, K. O.; Gassman, E. Rapid Commun. Mass Spectrom. 1991, 2, 319-326. (26) Williams, R. J.; Lipowska, M.; Patonay, G.; Strekowski, L. Anal. Chem. 1993, 65, 601-605. (27) Cheng, Y. L.; Darst, S. A.; Robertson, C., J. Colloid Interface Sci. 1987, 118, 212-223.

Analytical Chemistry, Vol. 68, No. 15, August 1, 1996

2517

Figure 2. Separation of large biomolecules at various field strengths. Sample: (a) FITC, (b) human serum albumin, (c) bradykinin, and (d) ribonuclease A. Flow rate, 0.015 µL/min; separation bed buffer, 20 mM borate/100 mM tris buffer (pH 9); left flow rate, 15 µL/min; right flow rate, 6 µL/min; side bed buffer, 20 mM borate/100 mM tris buffer (pH 9) plus 0.25 M Na2SO4; flow rate 25 µL/min. Detection at 3.1 cm along the separation bed length, with a corresponding residence time of 62 s.

interest will consist of simple ions. However, for biological samples, it is clear that sample components with large molecular weights will be encountered. If the µ-FFE device is to function as a sample pretreatment system for biological samples, it must be capable of separating large biomolecules. Figure 2 shows the separation of FITC-labeled HSA (MW 69 526; pI ) 4.9), bradykinin (MW 1436; pI ) 12.5), and ribonuclease A (MW 14 453; pI ) 9.6) at various field strengths in a 20 mM borate/100 mM Tris buffer (pH 9). It is evident that HSA (peak b) can be completely isolated from the other components using only 25 V/cm, while 100 V/cm is required to separate bradykinin (peak c) from ribonuclease A (peak d). The small peak (peak a) is due to free FITC. The order of separation is independent of the molecular weight but depends on variations in the mobility (mass/charge ratio). In FFE, the relationship between the migration distance of a species and the field strength is described by eq 1,15 where µ is

d ) µIt/Aκ ) (µep + µeo)Et

(1)

the apparent mobility (cm2/V‚s), µep is the electrophoretic mobility (cm2/V‚s), µeo is the electroosmotic mobility (cm2/V‚s), I is the current (A), t is the residence time (s), A is the cross-sectional area across the separation bed (cm2), κ is the carrier buffer conductivity (Ω-1 cm-1), and E is the electric field strength (V/ cm). Figure 3 is a plot of migration distance as a function of field strength for HSA, bradykinin, and ribonuclease A. The magnitude of µep for the respective proteins was calculated from the slope using the known µeo value. Analysis of the migration distance of fluorescein as a function of the field strength yielded a µeo of 9.5 × 10-5 cm2/V‚s for the buffer system used. The electrophoretic mobility of fluorescein used to calculate the electroosmotic mobility was assumed to be -2.6 × 10-4 cm2/V‚s.16 Measurement of fluorescein migration without the presence of protein yielded 2518 Analytical Chemistry, Vol. 68, No. 15, August 1, 1996

Figure 3. Plot of migration distance versus field strength. Side channel array indicates the distance from the position at 0 V/cm to the side of the separation bed. Electrohydrodynamic distortion boundary indicates the distance from the position at 0 V/cm to the point where focusing occurs (see text).

a µeo of 2.1 × 10-4 cm2/V‚s. Thus, the above value for µeo is lower than anticipated. This is likely caused by adsorption of the sample to the device walls during optimization of the detection signal.27 To optimize the fluorescence signal, the whole separation bed is filled with sample. The subsequent adsorption of the sample proteins would result in a decrease in the ζ potential and consequent decreases the electroosmotic mobility. The electrophoretic mobilities obtained for HSA, bradykinin, and ribonuclease A are -1.99 × 10-4, -1.39 × 10-4, and 1.16 × 10-4 cm2/V‚s, respectively. Band Broadening. Band broadening in FFE is largely attributed to five major sources: diffusion, hydrodynamic broadening, electrodynamic broadening, electrohydrodynamic distortion, and Joule heating.17 Hydrodynamic broadening, also known as the “crescent effect”, results from the inherent distribution of flow velocities present in a parabolic flow profile.10,13,17 Since the maximum velocity occurs at the center of the bed, any species in the central region will move faster than those near the walls. Consequently, the faster-moving species experience the electric field for a shorter length of time and do not migrate as far as those near the walls. Thus, for a specific residence time, the species in the different flow velocities will have different migration distances, and this results in the formation of the crescent shape. Hydrodynamic distortion can be minimized by either reducing the initial sample bandwidth or increasing the bed depth. However, reducing the initial bandwidth will decrease the sample throughput, and increasing the bed depth will increase Joule heating effects. Electrodynamic broadening is caused by electroosmotic flow within the separation bed. If the FFE apparatus used is an open system, then the fluid flow is in the direction of the applied field, and the flow profile is plug-shaped. However, if the system is closed, as is the case with most commercial systems, the electroosmotic flow at the chamber walls must be balanced by return flow in the opposite direction along the center of the bed. This results in a parabolic flow profile, and the observed broadening is also crescent-shaped.19,24 If the migration of sample

To evaluate the performance of the µ-FFE device, it was necessary to investigate the band broadening observed for HSA, bradykinin, and ribonuclease A. Figure 4 is a plot of the experimental variances for HSA, bradykinin, and ribonuclease A as a function of the field strength. Included in the plot are the theoretical variances (solid lines) calculated using eq 2, assuming band broadening contributions from the initial bandwidth, diffusion, hydrodynamic broadening, Joule heating distortion, and adsorption.19,29-32

σT2 ) σINJ2 + σD2 + σHB2 + σJH2 + σADS2 σINJ2 ) wi2/12 σD2 ) 2Dt

injection plug width diffusion

(2) (a) (b)

σHB2 ) [(xcenter - xmarginal)/4]2 hydrodynamic broadening19 (c) xcenter ) 2/3Et(µep - µeo/2) Figure 4. Plot of the experimental variances for ribonuclease A, bradykinin, and human serum albumin at various field strengths. Experimental variance values were obtained from the experimental baseline widths, i.e., wb ) 4σ. The error bars represent 1σ for four measurements, and the solid lines are the theoretical variances.

components is against the electroosmotic flow, then the electrodynamic crescent shape is in the reversed direction from that obtained for hydrodynamic broadening, and the two effects cancel each other. Therefore, to minimize electrodynamic distortion, it is desirable to have an open system such that there is no return flow in the central region of the separation bed. Electrohydrodynamic distortion results from differences in the conductivity between the sample and the carrier buffers and is independent of hydrodynamic and electrodynamic broadening.17,28 When a sample band migrates into a region of higher conductivity, the bandwidth decreases, while it is observed to broaden as the sample band migrates into a region of lower conductivity.28 This can be reduced or eliminated by choosing buffer systems that minimize conductivity differences between the sample and carrier. However, the observation that the sample bandwidth decreases when the sample migrates into a region of higher conductivity can be exploited for use as a mechanism for concentrating sample in FFE. The approach has been used in a variant of FFE known as field step FFE. Joule heating occurs when current is passed through a conducting medium. Joule heating distortion arises from convective motion caused by density gradients that form as the temperature increases within the separation bed. The heat generated is dependent upon a number of factors.17 It is proportional to the square of the applied voltage, the specific conductance of the buffer, the residence time of the buffer in the applied field, and the area of the electrodes (which depend on the separation bed thickness). It is evident that, in order to minimize Joule heating effects, the applied voltages must be as small as possible, the buffer must have a low conductivity, the flow rate must be high to minimize the residence time, and the separation bed thickness must be narrow to facilitate efficient thermal dissipation. (28) Rhodes, P. H.; Snyder, R. S.; Roberts, G. O. J. Colloid Interface Sci. 1989, 129, 78-90.

xmarginal ) 2/3Et(µep - µeo/2) + 2

σJH2 ) (wb/4)2

z2/r2 /3Etµep(1 - µeo/µep) (1 - z2/r2)

Joule heating broadening29,30

(d)

wb ) [t(µep + µeo)]1/2/[κ/2λ∆T]1/4; ∆T ) r2E2κλ-1/2 σADS2 )

[

][

]

2 k′ r2ν + νL 2 4D K (1 + k) d

adsorption31,32

(e)

In eqs 2a-e, wi is the initial plug width (cm) [wi ) 0.038 cm], D is the solute diffusion coefficient (cm2/s) [DHSA) 6.1 × 10-7, DBrad ) 1.0 × 10-6, and DRibo A ) 1.07 × 10-6 cm2/s], xcenter is the migration distance for a species at the center of the separation bed (cm), xmarginal is the migration distance for a species at some radial position, z, from the center of the separation bed (cm), r is half the bed depth (cm), ∆T is the temperature difference between the buffer solution and the separation bed wall (°C), κ is the carrier buffer conductivity (Ω-1 cm-1), λ is the thermal conductivity (W cm-1 °C-1), ν is the solute velocity (cm/s), Kd is the first-order dissociation constant (s-1), and k′ is the capacity factor. The remaining terms have their usual meanings. To obtain the best fit between the theoretical and experimental variance, a value of 0.47r was used for z in the calculation of the hydrodynamic broadening contribution. Values for z used to calculate hydrodynamic broadening have been reported to range from r/5 to 2r/ 3.10 However, it is unclear which is the appropriate value to use, since z actually represents all radial positions across the separation bed depth. For analysis of the variance and resolution, the experimental baseline widths were assumed to be equal to 4 times the standard deviation (4σ). (29) Ravoo, E.; Gellings, P. J.; Vermeulen, Th. Anal. Chim. Acta 1967, 38, 219232. (30) Ostrach, S. J. Chromatogr. 1977, 140, 187-195. (31) Schure, M. R.; Lenhoff, A. M. Anal. Chem. 1993, 65, 3024-3037. (32) Swedberg, S.; McManigill, D. In Techniques in Protein Chemistry; Hugli, T., Ed.; Academi Press: San Diego, CA, 1989.

Analytical Chemistry, Vol. 68, No. 15, August 1, 1996

2519

Joule heating distortions were found to have an insignificant contribution to the theoretical variance. The µ-FFE device has a very narrow separation bed depth (50 µm), which facilitates efficient cooling and therefore minimizes any Joule heating effects. Evaluation of the theoretical variance for HSA and bradykinin indicated that dispersion due to the initial bandwidth, diffusion, and hydrodynamic broadening are the main contributors to the band broadening. Broadening due to electrohydrodynamic distortion is not expected since the sample buffer and carrier buffer used in the separation of HSA, bradykinin, and ribonuclease A were identical. However, the side bed buffer contained 0.25 M Na2SO4 and consequently had a higher conductivity. If any of the side bed buffer were to enter the separation bed, then there is potential for electrohydrodynamic distortion. In Figure 4, the HSA variance initially increases but then decreases at large field strength. This is further illustrated in Figure 2, where the baseline widths of fluorescein (peak a) and HSA (peak b) are observed to increase as the field strength increases. However, at a field strength of 100 V/cm, the peaks appear to sharpen as they migrate closer to the edge of the separation bed. This does not appear to occur to the baseline widths of bradykinin (peak c) and ribonuclease A (peak d). Similarly, in Figure 3, the migration distance of HSA at field strengths greater than 80 V/cm is no longer a linear function of field strength, as expected from eq 1. The decrease in HSA bandwidth and the apparent decrease in electrophoretic mobility of HSA are likely caused by electrohydrodynamic distortion. The presence of electrohydrodynamic distortion effects results from the design of the µ-FFE device. As reported earlier,16 this device uses two arrays of 2500 v-groove channels (each groove is 12 µm wide, 10 µm deep, and 1 mm long) to isolate the separation bed from the beds containing the electrodes. In comparison to conventional FFE, the use of these channels results in a more open system, and consequently flow from the side beds into the separation bed occurs at high side bed buffer flow rates.16 When fluorescein was present in the side bed buffers as a marker, and the experimental conditions were identical to those used in Figure 2 (side bed flow rate larger than the separation bed flow rate), fluorescein was observed to enter 1-2 mm into the separation bed. Later, when the separation bed flow rate was equal to or greater than the side bed flow rate, the fluorescein did not enter the separation bed. This indicated that the fluorescein entered the separation bed because of pressure effects rather than by electrophoresis, electroosmosis, or diffusion. If fluorescein enters the separation bed at high side bed flow rates, then so must the side bed buffer. This results in a zone of higher conductivity buffer, which extends from the 2500 v-groove channel array ∼1-2 mm into the separation bed. Since the sample components are moving from a zone of low conductivity into a zone of higher conductivity, the bandwidths decrease,28 and the peaks appear to be focused. Only the sample components that migrate to within 1-2 mm of the side channel array will experience this focusing effect. It is for this reason that the focusing is not observed (Figure 2) for bradykinin or ribonuclease A. This focusing effect has been utilized in a variant of FFE known as field step FFE.33 This approach uses a carrier buffer that has a higher conductivity, generally 20 times higher, than that of the (33) Kuhn, R.; Hoffstetter-Kuhn, S.; Wagner, H. Electrophoresis 1990, 11, 942947.

2520

Analytical Chemistry, Vol. 68, No. 15, August 1, 1996

sample solution. The sample is bracketed by the carrier buffer, and, as a result of these differences in conductivities, the sample has a higher migration velocity in the low-conductivity region than in the region of higher conductivity. This decrease in mobility in the higher conductivity buffer is due to a decreased voltage drop relative to that for the lower conductivity buffer. Therefore, at the boundary between the carrier buffer and the sample buffer, the mobility of the sample decreases dramatically,33 and consequently the sample components are concentrated as they focus at the boundary between the solutions. The experimental conditions used to separate the biomolecules in the µ-FFE device involved using a buffer in the side beds that had a conductivity 42 times that of the sample and separation bed buffers. Thus, the focusing effect observed for HSA and FITC is identical to that observed in field step FFE, and it is the change in the mobility of HSA in the region of higher conductivity that causes the curvature observed in Figure 3. The electrohydrodynamic distortion boundary and the separation bed edge are indicated in Figure 3. Note that the separation bed is actually 10 mm wide, but the edge distance shown in Figure 3 is given as 6.8 mm. This is because the migration distances are measured relative to the position of the sample band at zero applied field, which occurs at 3.2 mm. The positions for bradykinin and ribonuclease A do not approach these boundaries, and the focusing effect is not observed. The sources of ribonuclease A band broadening also include the initial bandwidth, diffusion, and hydrodynamic broadening. However, the band broadening for ribonuclease A could not be sufficiently accounted for using the above contributors. Adsorption was thought to be a possible contributor to the overall variance for ribonuclease A. Shown in Figure 4 is the theoretical variance for ribonuclease A with (dashed line) and without (solid line) consideration of adsorption. It is clear that there must be another source of band broadening. In the calculation of the contribution from adsorption values, the capacity factor, k′, and the first-order dissociation constant , Kd, for ribonuclease A were required. A value of 0.013 was used for k′,32 and Kd was then adjusted to 0.35 s-1 to obtain the best fit. The capacity factor, k′, is the ratio of the total time the solute spends adsorbed onto the wall to the total time spent in free solution and is therefore a measure of the extent of retention of the solute on the walls of the device. k′ is not measurable in an experiment involving electrophoretic separations,31 since it is unlikely that the unretained compound will have the same mobility as the retained compound. However, k′ is used here to express the relative degree of retention31 and to demonstrate that the increase in total variance for ribonuclease A may be due to adsorption. The solute velocity used in the above equations was obtained from the vector sum of the hydrodynamic velocity (vh ) Q/A) and the electrokinetic velocity (vek ) (µeof + µel)E), where Q is the volumetric flow rate. Ribonuclease A has a pI of 9.6 and should be positively charged at the pH of the buffer used. It is possible that adsorption of ribonuclease A occurs as a result of the interaction of the negatively charge wall with regions of positive charge on ribonuclease A. However, other possible sources that might contribute to the overall variance of ribonuclease A are electrodynamic dispersion and electrohydrodynamic distortion. As discussed above, the µ-FFE device is an open system, and therefore it is unlikely that electrodynamic dispersion is present. Also, electrohydrodynamic distortion should result in peak sharpening; instead, broadening is observed for ribonuclease A.

Figure 5. Comparison of the large biomolecules separation profiles obtained under different flow rates and applied fields. (A) Flow rate, 15 µL/min, 50 V/cm; (B) flow rate, 30 µL/min, 100 V/cm. Note: the side bed flow rate was constant for A and B at 25 µL/min. Other experimental conditions are as in Figure 2.

Resolution. For the µ-FFE device to function as a sample pretreatment system, it is necessary to have a high sample throughput. Thus, the residence time (i.e., flow rate) and the field strength must be manipulated to maximize the resolution at the desired throughput. If the residence time is halved, then the field strength must be doubled such that the same migration distance is obtained (eq 1). Shown in Figure 5 is a comparison of the separation profiles of HSA, bradykinin, and ribonuclease A for the case where the residence time is halved (flow rate is doubled) and the field strength is doubled. From this figure, it is clear that the migration distances are not the same, and the shorter residence time and higher field strength yield better resolution (Figure 5B). This behavior is unexpected based on eq 1. However, as mentioned above, when the separation bed flow rate, was equal to or greater than the side bed flow rate no side buffer enters the separation bed. The lower separation profile (Figure 5B) was obtained under these conditions, which result in little or no electrohydrodynamic distortion, and consequently improved resolution. Figure 6 is a plot of the resolution of ribonuclease A and bradykinin as a function of field strength at various carrier flow rates. Included in the figure are the theoretical resolution values (R ) ∆µ/σ2) for each flow rate. Note that ∆µ is the difference in the experimental mobilities, and σ2 is the theoretical variance. From this plot, it is evident that the resolution increases with field strength and that slower flow rates give better resolution at specific field strengths. However, no optimal conditions were observed. Therefore, in order to obtain a high sample throughput, large field strength and high flow rate are required. Finally, to evaluate the separation performance of the µ-FFE device relative to that of a conventional system, the peak capacities per centimeter of separation bed are compared. The µ-FFE device has a peak capacity of ∼8 bands/cm, while for a typical separation of proteins using a commercial system,21,34 a peak capacity of 10 bands/cm

Figure 6. Resolution of bradykinin and ribonuclease A at various field strengths and flow rates. Other experimental conditions are as in Figure 2. The error bars represent 1σ for four measurements, and the solid lines are the theoretical values.

Figure 7. Separation profiles of tryptic digests of (A) bovine cytochrome c and (B) mellitin. (A) 20 mM borate/100 mM tris buffer (pH 9); sample flow rate, 0.015 µL/min; left carrier flow rate, 15 µL/ min; right carrier flow rate, 6 µL/min; 75 V/cm. (B) 20 mM borate/100 mM tris buffer (pH 9); sample flow rate, 0.015 µL/min; left carrier flow rate, 15 µL/min; right carrier flow rate, 6 µL/min; 75 V/cm. See text for designation of peaks. Detection at 3.1 cm along the separation bed length with a corresponding residence time of 62 s.

was observed. Thus, the resolving power of the µ-FFE device is similar to those of conventional systems. Separation of More Complex Samples. Figure 7 shows the separation profiles of tryptic digests of bovine cytochrome c (A) and mellitin (B). The purpose of these separation was to investigate whether the µ-FFE can be used for more complex separations. Electropherograms of these digests (not shown) obtained using capillary electrophoresis indicated the presence of six peaks for the mellitin digest and nine peaks for the (34) Nath, S.; Schutte, H.; Weber, G.; Hustedt, H.; Deckwer, W.-D. Electrophoresis 1990, 11, 612-616.

Analytical Chemistry, Vol. 68, No. 15, August 1, 1996

2521

cytochrome c digest. The separation using the bare µ-FFE device gave five bands for the mellitin digest and six bands for the cytochrome c digest. The bands obtained for cytochrome c were not identified; however, some of the bands for the mellitin digest have been identified as (a) mixture of arginine and lysine residues, (b) glutamine-glutamine, and (e) undigested mellitin. Band c is thought to be due to the seven amino acid residue (Gly-IleGly-Ala-Val-Leu-Lys), while band d is from the 14 amino acid residue (Val-Leu-Thr-Thr-Gly-Leu-Pro-Ala-Leu-Ile-Ser-Trp-Ile-Lys). Further improvements are needed to enhance these separations; however, they demonstrate that the device is capable of separating more complex samples. The separation profile observed in Figure 7 is not unlike that obtained for separations of proteins from cell extracts using conventional systems.38 Therefore, although the resolution obtained for the tryptic digests requires further improvement, the µ-FFE device appears to function as well as its larger counterpart. Fraction Collection. The device used in the separation described above possessed only one outlet hole. As reported earlier,16 an outlet array of 125 v-groove channels was incorporated into the structure for use as a fraction collector. However, to simplify fabrication for initial experimentation, these channels were connected to an outlet system with only one outlet hole, and therefore it was impossible to collect fraction from this device. To use the device to collect fractions, outlet holes were drilled through the glass cover plate. These holes were placed at nearly equal spacings along the array. In this manner, access was made to channels 14-16, 43-45, 74-76, and 102-104. The device was then used to separate whole rat plasma, and the subsequent fractions were then analyzed off-line using capillary electrophoresis. Figure 8 shows the electropherograms of the fractions obtained compared to an electropherogram of diluted rat plasma. It is quite clear that channels 74-76 isolate only albumin, and channels 43-45 isolate only globulins. The outlet that connects all channels shows essentially the same electropherogram as the diluted rat plasma, but with fractions cut out. Finally, the channels 14-16 and 102-104 did not show the presence of any species and were identical to the electropherogram of the running buffer used for the µ-FFE separations. This result demonstrates the feasibility of the system for use as a fraction collection system for complex biological samples. In conclusion, for analysis of biological samples, it is desirable to have some appropriate method for sample cleanup prior to the actual analytical step. The µ-FFE device has proven to be a viable method for this. It is capable of separating large biomolecules in a continuous fashion. It has been demonstrated that the µ-FFE device performs as well as the large commercial systems. The peak capacities per centimeter of separation bed width are, in fact, comparable for both systems: 8 for the µ-FFE device and 10 for a typical separation on a commercial system. However, it is unlikely that the µ-FFE device could outperform these system as a purification system, simply because the sample throughput is too low. A typical throughput of proteins obtained for commercial systems is 82 mg/h.22 In contrast, the µ-FFE throughput for HSA is decreased by over a factor of 5000 to 15 µg/h. It must be emphasized that the µ-FFE device was not developed as a large-

2522 Analytical Chemistry, Vol. 68, No. 15, August 1, 1996

Figure 8. Electropherograms of the fractions collected from the µ-FFE microstructure during purification of undiluted rat plasma. The electropherograms were obtained using off-line capillary electrophoresis. The fraction collection holes were drilled through the coverplate of the device, and the channels accessed in the 125 v-groove outlet array are indicated in the figure. The electropherogram of diluted rat plasma shown at the top is for comparison purposes. The carrier buffer was 0.3 M Tris/0.1 M boric acid (pH 8.8), and the side bed buffer was 1.16 M Tris/0.57 M boric acid (pH 8.8) containing 0.25 M Na2SO4. The sample, carrier buffer, and side bed buffer flow rates were 0.2, 5, and 20 µL/min, respectively. Total residence time was 300 s, and the fractions were collected for 2 h. The field strength used was 25 V/cm.

scale purification system. The microstructure was developed as a sample pretreatment system for a miniaturized analysis system that could ultimately be used for on-line monitoring or in a small, portable device. The device does function as a micropurification system for both simple ionic samples and biomolecules and is ideal for use as a sample pretreatment method in a miniaturized total analysis system. ACKNOWLEDGMENT The authors thank Olaf Bo¨rsen for the MALDI analyses of the labeled proteins and Antje Klockow-Beck for assistance with the off-line CE analyses. Also, part of this work has been supported by the European Research Project (BCR, Measurement and Testing). Finally, D.E.R. thanks Dr. J. Michael Ramsey (Oak Ridge National Laboratory) for his patience while this manuscript was prepared during D.E.R.’s postdoctoral work in Dr. Ramsey’s laboratory. Received for review August 1, 1995. Accepted May 6, 1996.X AC950766V X

Abstract published in Advance ACS Abstracts, June 15, 1996.