Anal. Chem. 2003, 75, 6761-6768
Articles
On-Chip Hydrodynamic Chromatography Separation and Detection of Nanoparticles and Biomolecules Marko T. Blom,† Emil Chmela,‡ R. Edwin Oosterbroek,*,† Rob Tijssen,‡ and Albert van den Berg†
MESA+ Research Institute, P.O. Box 217, 7500 AE Enschede, The Netherlands, and Department of Chemical Engineering, University of Amsterdam, Nieuwe Achtergracht 166, 1018 WV Amsterdam, The Netherlands
For the first time, on-chip planar hydrodynamic chromatography is combined with UV absorption detection. This technique is suitable for size characterization of synthetic polymers, biopolymers, and particles. Possible advantages of an on-chip hydrodynamic chromatography system over conventional techniques, such as size exclusion chromatography, and field-flow fractionation are fast analysis, high efficiency, reduced solvent consumption, and easy temperature control. The hydrodynamic separations are performed in a planar configuration realized in fused silica using a mixture of fluorescent and nonfluorescent polystyrene particles with sizes ranging from 26 to 155 nm. The planar chip configuration consists of a 1-µm-high, 0.5-mm-wide, and 69-mm-long channel, an integrated 150-pL injection structure, and a 30-µm-deep and 30-µm-wide detection cell, suitable for UV absorption detection. By combination of the separation data obtained in the new fused-silica chip with those obtained using a previously presented planar hydrodynamic chromatography chip, which was realized using silicon and glass microtechnology, a description of the retention and dispersion behavior of planar hydrodynamic chromatography is obtained. Especially the influence of the sidewalls on the dispersion is investigated. Furthermore a hydrodynamic separation within 70 s of several biopolymers is shown in the glass-silicon chip. Size characterization of synthetic polymers, biopolymers, and nanoparticles is an important tool in chemical analysis. Conven* Corresponding author: (e-mail)
[email protected]; (fax) +31 53 489 2287; (phone) +31 53 489 1074. † MESA+ Research Institute. ‡ University of Amsterdam. 10.1021/ac034663l CCC: $25.00 Published on Web 11/08/2003
© 2003 American Chemical Society
Figure 1. Separation principle of hydrodynamic chromatography in which a particle with effective radius reff is excluded from the wall in a channel with diameter h in which a parabolic velocity profile u(y) is applied giving a maximum velocity umax.
tional methods that are applied for determination of the size distribution are light scattering, size exclusion chromatography, flow field flow fractionation, thermal field flow fractionation (ThFFF), and hydrodynamic chromatography (HDC). In this paper, we focus on miniaturization of chromatographic methods and in particular HDC.3-9 The HDC separation principle (shown in Figure 1) relies on the size-dependent exclusion from the wall in a channel in which a pressure-driven flow is applied. The corresponding parabolic flow profile results in a larger velocity for larger analytes, since they will on the average spend more time in the faster flowing regions (1) Chmela, E.; Tijssen, R.; Blom, M. T.; Gardeniers, J. G. E.; van den Berg, A. Anal. Chem. 2002, 74, 3470-3475. (2) Blom, M. T.; Chmela, E.; Gardeniers, J. G. E.; Tijssen, R.; Elwenspoek, M.; van den Berg, A. Sens. Actuators, B 2002, 82, 111-116. (3) Pedersen, K. O. Arch. Biochem. Biophys. Suppl. 1962, 1, 157-168. (4) DiMarzio, E. A.; Guttman, C. M. Macromolecules 1970, 3, 131-146. (5) Small, H. J. Colloid Interface Sci. 1974, 48, 147-161. (6) Brenner, H.; Gaydos, L. J. J. Colloid Interface Sci. 1977, 58 (2), 312-356. (7) Tijssen, R.; Bleumer, J. P. A.; van Kreveld, M. E. J. Chromatogr. 1983, 260, 297-304. (8) Tijssen, R.; Bos, J.; van Kreveld, M. E. Anal. Chem. 1986, 58, 3036-3044. (9) Venema, E.; Kraak, J. C.; Poppe, H.; Tijssen, R. J. Chromatogr., A 1996, 740, 159-167.
Analytical Chemistry, Vol. 75, No. 24, December 15, 2003 6761
of the channel. Classically, HDC is performed in columns packed with nonporous particles giving a relatively small efficiency especially for small packing particle sizes.9 Miniaturization of analytical separation methods can be advantageous because of reduction of solvent consumption and possible improvements in efficiency and analysis time. The first step toward miniaturization was done by using smaller separation columns such as fused-silica capillaries for capillary electrophoresis (CE), size exclusion electrochromatography, and HDC. However, the use of external injectors and detectors results in undesired extracolumn peak broadening. For capillary HDC, for instance, detection proved to be rather difficult because of the small detection volume in the micrometer-sized capillaries.8 Microtechnology offers the possibility of integrated injection and detection, thus reducing extracolumn peak broadening. A thorough overview of analytical separation chips can be found in refs 10 and 11. Microtechnology has predominantly been focused on CE. The first on-chip CE device, presented by Harrison and Manz,12 was followed by numerous on-chip CE separations of peptides, proteins, and amino acids on glass or fused-silica chips. Some examples of on-chip LC can be found as well. However, the first functional LC chip, presented by Ocvirk,13 showed a limited efficiency. Only recently, Ericson14 and McEnery15 have shown more promising results. Microtechnology also enables the development of novel separation techniques such as shear-driven chromatography,16 allows DNA separations by using a microfabricated array of entropic traps,17 or enables the definition of microfabricated structures inside a channel that can be used to attach a stationary phase.18 For on-chip separation of large noncharged analytes, so far only DEP-FFF19 and ThFFF20 were available. These are suitable for a limited group of analytes. However, recently a Pyrex-siliconbased planar HDC chip has been presented1,2 that is capable of performing analytical separations of large noncharged analytes as well. It consists of a 1-µm-high separation channel that is as wide as possible in order to increase detection volume. Furthermore, a structure is integrated, capable of injecting a subnanoliter sample volume, thus reducing extracolumn peak broadening. In this configuration, however, the increased detection volume of the planar configuration was not used for a larger optical path length, thereby limiting detection to external fluorescence detection. Furthermore, the difference between the thermal expansion of Pyrex and silicon created a slight deformation of the channel, (10) Reyes, D. R.; Iossifidis, D.; Auroux, P.-A.; Manz, A. Anal. Chem. 2002, 74, 2623-2636. (11) Auroux, P.-A.; Iossifidis, D.; Reyes, D. R.; Manz, A. Anal. Chem. 2002, 74, 2637-2652. (12) Harrison, D. J.; Manz, A.; Fan, Z.; Lu ¨ di, H.; Widmer, H. M. Anal. Chem. 1992, 64, 1926-1932. (13) Ocvirk, G.; Verpoorte, E.; Manz, A.; Grasserbauer, M.; Widmer, H. M. Anal. Methods Instrum. 1995, 2 (2), 74-82. (14) Ericson, C.; Holm, J.; Ericson, T.; Hjerten, S. Anal. Chem. 2000, 72, 8187. (15) McEnery, M.; Tan, A.; Alderman, J.; Patterson, J.; O’Mathuna, S. C.; Glennon, J. D. Analyst 2000, 125, 25-27. (16) Desmet, G.; Vervoort, N.; Clicq, D.; Huau, A.; Gzil, P.; Baron, G. V. J. Chromatogr., A 2002, 948 (1-2), 19-34. (17) Han, J.; Craighead, H. G. Science 2000, 288, 1026-1029. (18) He, B.; Ji, J.; Regnier, F. E. J. Chromatogr., A 1999, 853, 257-262. (19) Wang, X.; Yang, J.; Huang, Y.; Vykoukal, J.; Becker, F. F.; Gascoyne, P. R. C. Anal. Chem. 2000, 72, 832-839. (20) Edwards, T. L.; Gale, B. K.; Frazier, A. B. Anal. Chem. 2002, 74, 12111216.
6762
Analytical Chemistry, Vol. 75, No. 24, December 15, 2003
resulting in a deformed peak shape. The separation performance of the chip was demonstrated by separation of fluorescent polystyrene nanoparticles.1 This paper focuses on two related subjects. First, separations are presented of fluorescent polystyrene nanoparticles and of biomolecules, performed in the Pyrex-silicon HDC chip. Observation of the peak shape for two different channel widths illustrates the influence of the channel width on the sidewall induced dispersion. Furthermore, the capability of on-chip HDC separation of biomolecules is shown. Second, an improved version of the HDC chip is presented that is fabricated out of fused silica. The design incorporates an improved injection system and an integrated optical detection cell that gives a 30-fold increase in optical path length. Because of the large optical transmission window of fused silica, UV absorption detection can be used in this detection cell. This is used for detection of a mixture of HDC-separated fluorescent and nonfluorescent particles. An additional advantage is the reduction of thermal deformation effects by using fused silica for both top and bottom wafers. Combination of the separation data from both chip types gives a description of the retention behavior of planar HDC. Additionally, the UV-detected particle separations enable a calculation of the separation efficiency. THEORY Retention. For the theoretical description of the retention and dispersion in planar HDC, refer to Figure 1. The retention behavior of planar hydrodynamic chromatography has been described in ref 1. For the sake of completeness, the different interactions governing HDC retention will be briefly discussed here as well. For most models that account for interactions beside the normal exclusion from the wall, the equations for the relative residence time of an analyte can be unified according to
τ ) 1/(1 - λ - Cλ2)
(1)
in which λ is the relative analyte size defined as
λ ) 2reff/h
(2)
For the simple exclusion in a parabolic flow profile C ) 1/2.21 The most important additional interactions are as follows. Friction. DiMarzio and Guttman4 expressed friction as a lower constant component in the total average analyte velocity. The constant changes to CDG ) 1/2 + 3/4γ. For particles γ ) 2/3, for polymers γ ) 2π/27. Hydrodynamic Interactions. In ref 6, Brenner and Gaydos took into account both hydrodynamic interactions with the wall and the slip velocity of point-size particles. The slip velocity expresses the fact that the particle center lags the local fluid velocity. Lift forces, caused by fluid inertia, are not incorporated in their model, so that the particle velocity does not have a transverse component. Thus, any transverse transport is caused by diffusion. Unfortunately, the model is only valid for a circular geometry. Comparison with DiMarzio and Guttman shows that the constant C increases (21) Giddings, J. C. Sep. Sci. Technol. 1978, 13 (3), 241-254.
if these effects are accounted for. It is doubtful, however, if this model can be applied to polymers, as for hydrodynamic interactions polymers cannot be considered to be spherical particles. Inertia (or Tubular Pinch Effects). In ref 22 Ploehn used an adapted form of the theory developed by Brenner and Gaydos in order to incorporate this effect. Fluid inertia adds a transverse component to the particle velocity that depends on the particle position in the channel. For neutrally buoyant spherical particles near the wall, the velocity is directed away from the wall while particles near the channel center are transported outward. This means that an equilibrium transverse position exists (independent of particle size) at which the particles are focused. Such an effect severely decreases the selectivity of the HDC separation. The relative importance of this effect can be expressed by a Peclet number that represents the ratio of inertial to diffusive forces acting on a particle:22 3
Pe ) (6πF 〈u〉2 reff λ2)/kT
(4)
Peak broadening is caused by concentration gradients in the channel. For analytes that are transported through a column and do not interact with the column, the Taylor-Aris dispersion can be derived from these concentration gradients.23,24 The effective diffusion constant includes a contribution due to stationary (22) Ploehn, H. J. Int. J. Multiphase Flow 1987, 13 (6), 773-784. (23) Taylor, G. I. Proc. R. Soc. 1953, A219, 186-203. (24) Aris, R. Proc. R. Soc. 1956, A235, 67-77.
Deff ) D12(1 + R0Pe2)
(5)
In this equation, D12 is the binary diffusion coefficient. For particles, this can be related to the particle radius reff through the Stokes-Einstein equation:
D12 ) kT/6reff πη
(6)
in which η is the dynamic viscosity. In eq 5, R0 is the so-called dispersion constant and Pe the Peclet number, which is a measure of the relative influence of convectional and diffusional forces:
Pe ) 〈u〉Lch/D12
(7)
(3)
in which k is the Boltzmann constant, T the absolute temperature, and F the fluid density. It is clear that the influence of the inertial forces decreases strongly with decreasing particle size and velocity. For chip-HDC, it turns out that Peinertia , 1, which means that the inertial forces have a negligible impact on the HDC effect. Electrostatic Wall Interactions. This reduces the effective channel height for charged particles as described by Ploehn.22 Tijssen8 compared the DiMarzio-Guttman and the BrennerGaydos models to experimental residence time data obtained for polystyrene separations in microcapillaries. It turns out that for these polymers the DiMarzio-Guttman model performs best. This is probably because the hydrodynamic and inertial interactions described in the Brenner-Gaydos and Ploehn models are based on spherical particles and polymers do not have a well-defined spherical shape. Furthermore, electrostatic interactions are small for the nonaqueous buffers used in polymer separations. Dispersion. The two major sources of dispersion in planar HDC are nonuniformity of the channel cross section, resulting in a nonuniform average velocity across the channel width and the normal Taylor-Aris dispersion for a rectangular channel. Since the exact nonuniformity is often unknown, its dispersive effect will only become clear experimentally. For the Taylor-Aris dispersion in a rectangular channel, however, a theoretical description can be given. The efficiency described by plate height H can be expressed as a function of the average velocity 〈u〉 and an effective TaylorAris diffusion coefficient Deff :
H ) 2Deff/〈u〉
diffusion effects and a convective component:
Here Lch is a characteristic length. In a circular tube Lch will be equal to the tube radius and in a planar geometry to half of the channel height. It is clear that for smaller dimensions the diffusion dominates the dispersion. The dispersion constant depends on the geometry and on the transport mechanism that is used. It can therefore be influenced by adapting the geometry. In various publications,25,26 a dispersion constant of 2/105 is derived for a parallel plate configuration. However, this only takes into account concentration gradients in the y-direction (across the channel height) and thus only applies to a situation in which gradients in the x-direction (across the channel width), caused by the presence of the sidewalls, do not play a role. Fully developed concentration gradients in the y-direction are a prerequisite for the HDC separation: the analyte must have been able to sample all velocity streamlines. In other words, the concentration must be constant across the channel height for each position inside the channel, including the sample plug. The time scale at which the gradients in the y-direction can be considered fully developed can be expressed by
D12t/h2 . 1
(8)
For larger time scales, when both gradients are fully developed, a considerable increase in dispersion is found. This can be expressed by (w is the channel width)
D12t/w2 . 1
(9)
This situation is considered in ref 27 giving a dispersion coefficient that is ∼8 times larger than the coefficient for a planar configuration. However, this does not cover the intermediate time given by
h2/D12 , t , w2/D12
(10)
(25) Golay, M. J. E. In Gas Chromatography; Desty, D. H., Ed.; Butterworths: London, 1958; pp 36-55. (26) Dill, L. H.; Brenner, H. J. Colloid Interface Sci. 1983, 93 (2), 343-365. (27) Pagitsas, M.; Nadim, A.; Brenner, H. Physica A 1986, 135, 533-550.
Analytical Chemistry, Vol. 75, No. 24, December 15, 2003
6763
In ref 28 Doshi gave an approximation, reproduced below, for the dispersion coefficient in a rectangular channel for all times. This, therefore, also covers the transient dispersion in the time frame given by eq 10. It should be noted that the sidewall-induced dispersion creates a tailing effect along the channel sidewalls, which gives rise to an increased background signal that does not necessarily contribute to the measured dispersion in the center of the channel. Therefore, the effective diffusivity, calculated in this way, gives a somewhat negative picture of the dispersion, which is aggravated by the averaging of the concentration profile across the channel cross section employed in ref 28. In ref 29 a numerical model is employed to estimate the complete transient-state dispersion. However, this analysis is performed for aspect ratios (defined by w/h) up to 20, which is well below the aspect ratio used in chipHDC. Consequently, the dispersion calculated here according to ref 28 must be seen as an upper limit to the Taylor-Aris dispersion in a rectangular channel. We can define a dimensionless time t* according to
t* ) 4D12t/w2
(11)
It turns out that for large aspect ratios the effective diffusivity according to Doshi can be written as a sum of the separate contributions of the gradients in the x-direction and the ydirection.28 Using eq 11, the solution found by Doshi for R0 is
R0 )
∞
9
∑B 8
2 2/
(1 - e-m π t ) +
2 m
m)1
9
∞
∑B 8
2
n
2 2/
(1 - e-n π t AR ) 2
n)1
(12)
an increase in dispersion with reduced time is found for channels with aspect ratios larger than 100. Since this is exactly the region in which HDC separations are performed, reduction of the channel width (affecting t*) leads to an increase in dispersion. The dispersion according to eq 12 is given for instantaneous dispersion constants. The dispersion coefficient averaged over a complete analysis would be
R0 )
256(-1)m
π3n3
∑
6
4(-1)n
∞
j)0 (2j
+
tanh [(2j + 1)πAR/2] 3
2
2
-2
(13)
+ 1) ((2j + 1) + 4m (AR) )
0
Lch ) (1/2)(h - 2reff) ) (1/2)h(1 - λ)
256(-1)n+1
∞
nπ6AR
j)0
(2j + 1)3(2n + 2j + 1)(2n - 2j - 1)
In these equations, AR is the aspect ratio. Thus, the dispersion at a certain reduced time t* is specified by the aspect ratio. It turns out that first the dispersion develops, caused by the gradients in the y-direction. For later times, the dispersion caused by gradients in the x-direction is added which for t* f ∞ leads to a constant dispersion:
R0 ≈ 8(2/105)
(15)
This result is also obtained in ref 27, For aspect ratios larger than 100, a temporal region of constant dispersion is present for which R0 ) 2/105. For intermediate time scales, 10-4 < t* < 10-1, (28) Doshi, M. R.; Daiya, P. M.; Gill, W. N. Chem. Eng. Sci. 1978, 33, 795-804. (29) Dutta, D.; Leighton, Jr., D. T. Anal. Chem. 2001, 73, 504-513.
Analytical Chemistry, Vol. 75, No. 24, December 15, 2003
(17)
(18)
For the Pe number, this leads to a correction factor:
∑× (14)
(16)
The second correction can be made by observing that the dispersion is based on velocity differences. The HDC velocity consists of a constant part and a part that depends on the particle position in the channel. For the dispersion, only the nonconstant term 〈u〉rel,HDC is important. This is related to the average velocity of the solvent as
Pe )
tanh[(2j + 1)πAR/2]
6764
t
0
〈u〉rel,HDC ) 〈u〉(1 - λ)2
mπ Bn )
∫ R (t) dt
The analysis presented above results in a dispersion in a rectangular channel for point-size particles that do not have any interaction with the channel walls. Neglecting the additional interactions (hydrodynamic, inertial, electrostatic) mentioned in the previous section, a correction for the finite particle size can be derived. First, the characteristic length Lch used in the definition of the Peclet number from eq 7 can be corrected by taking into account the effective channel cross section the particle can sample (refer to Figure 1):
in which
Bm )
1 t
〈u〉rel,HDCLch 〈u〉h ) (1/2)(1 - λ)3 D12 D12
(19)
This correction factor corresponds to refs 4 and 21. The correction factor calculated by Brenner and Gaydos6 is more complicated, but corresponds to eq 19 in a first-order approximation. Combination of eqs 4, 5, 7, 12, 16, and 19 finally leads to an effective, time-averaged effective diffusivity of
H)
2D12 〈u〉
+ (1/2)(1 - λ)6
〈u〉h2 R (AR,t,t*) D12 0
(20)
Thus, we see that the dispersion for a planar HDC configuration cannot be characterized by a dispersion constant, but that the transient dispersion for time scales given by eq 10 leads to a time-dependent increase in dispersion. DESIGN AND FABRICATION Design. The design and fabrication sequence of the HDC chip fabricated using Pyrex and silicon has been described in refs 1 and 2. It incorporates a 1-µm-high, 0.5- or 1.0-mm-wide separation
Figure 2. Transition structures guiding the plug from the shallow, wide separation channel to a deeper, less wide detection cell.
channel and an integrated injection structure that is more thoroughly described in ref 30. In the fused-silica HDC chip, a detection cell is integrated with a transition structure that guides the plug from the wide, shallow separation channel to a deeper, less wide detection cell.31 For that purpose, a structure as shown in Figure 2A is used to guide the plug from the separation channel to a 30-µm-wide channel. Subsequently, the structure from Figure 2B provides a transition from the shallow 1-µm depth to the 30-µm-deep detection channel. The “bird beaks” are added for a smooth depth transition by employing the so-called RIE lag.32 This RIE-lag causes a higher aspect ratio structure to be etched less deep. A more thorough description of the complete structure is given in ref 31. In this optical detection cell, UV absorption detection can be performed since the used UV grade fused silica has a transmission region of 0.18-2.5 µm.33 Fused-Silica Chip Fabrication. A thorough description of the complete fabrication procedure is given in ref 35. Throughholes in the top wafer were created using powder blasting.34 The bottom fused-silica wafer was etched to a depth of 1 µm using a buffered HF etching solution. The separation channel was 0.5 mm wide and 69 mm long. For the injection structure and detection cell, deeper 30-µm structures were defined using deep reactive ion etching (DRIE) with a Cr/Ni masking layer. SEM pictures after the DRIE are shown in Figure 3. From Figure 3A it is clear that the effect of the RIE lag is not as prominent as was assumed. Apart from some scattered pinholes, the surface of the bottom of the optical detection slit shown in Figure 3B is smooth. This is an important result since this reduces scattering losses. After cleaning, the top and bottom wafers were contacted and a prebond was formed. Subsequently, the wafer pair was annealed at 700 °C for 2 h. EXPERIMENTAL SECTION Apparatus. HDC chips were attached to a macro-flowactuating system equipped with a fluorescence microscope as described in ref 1. In the setup for the UV detection, the microscope was replaced by an external optical fiber and a photocell, (30) Chmela, E.; Blom, M. T.; Gardeniers, J. G. E.; van den Berg, A.; Tijssen, R. Lab Chip 2002, 2 (4), 235-241. (31) Chmela, E.; Blom, M. T.; Oosterbroek, R. E.; van den Berg, A.; Tijssen, R., manuscript in preparation. (32) de Boer, M.; Gardeniers, H.; Jansen, H.; Smulders, E.; Gilde, M. J.; Roelofs, G.; Sasserath, J. N.; Elwenspoek, M. J. Microelectromech. Syst. 2002, 11 (4), 385-401. (33) http://www.almazoptics.com. (34) Wensink, H.; Jansen, H. V.; Berenschot, J. W.; Elwenspoek, M. C. J. Micromech. Microeng. 2000, 10, 175-180. (35) Blom, M. T., Ph.D. Thesis, University of Twente, 2002.
connected to a standard HPLC detector system (type Spectra 100, Spectra Physics). A rectangular aperture of 100 × 200 µm, defined in a 100-µm-thick steel blade, was placed between the chip and the photocell and aligned with the longer side parallel to the outlet slit, directly after the depth transition structure. All tubing and valves were stainless steel standard HPLC 1/16in. components (Valco). For the connections of the solvent and sample inlet and outlet tubing, a stainless steel connection piece with Kalrez O-rings was used. This provided a chemically resistant connection even after immersion in THF for several months. Chemicals. Polystyrene latex particle standards of nominal sizes 61 (Polysciences) and 155 nm (Duke Scientific), fluorescently labeled polystyrene latex nanoparticles (carboxylate-modified FluoroSpheres yellow-green fluorescent) of nominal diameters 26, 44, and 110 nm and fluorescently labeled bovine serum albumin (BSA) and bovine eye R-crystallin (Molecular Probes, Leiden, The Netherlands) were used as test analytes, together with t0 markers 4(5)-carboxyfluorescein (Sigma-Aldrich) and sodium benzoate. Several aqueous buffers were used as eluents: 10 mM phosphate buffer pH 7.0, 10 and 1 mM MOPS buffer pH 7.0, and 10 and 0.1 mM borate buffer pH 9.2. All water was deionized and subboil-distilled. Procedures. Before the first elution experiments, the new chips were flushed with a buffer of pH 9 in order to maximize the number of silanol groups on the channel surface. These groups should reduce the possible interaction of a negatively charged marker with a surface, because they are dissociated at the pH 7 and 9 used in the study. All eluents were filtered prior to use through a 0.02-µm membrane filter (Anotop 25, Whatman). Upstream from each reservoir, a 0.5-µm in-line filter (Upchurch) was placed. The aqueous samples with particles were sonicated and filtered through a 0.45-µm hydrophilic syringe filter (type Millex LH, Fisher Scientific) before injection. When UV detection was used instead of fluorescence imaging, a visual control of the injection process on the chip was not possible. In that case, the injection steps were controlled by timed switching of the valves. Safety. The use of pressurized gas in the homemade solventdelivery system (see ref 1) requires extra caution. This holds particulary during depressurizing, when the escaping gas is saturated with the vapors of the solvent, possibly causing it to be toxic or explosive. The working pressure should not exceed 10 bar unless additional safety tests are performed; conform to general rules for the work with pressurized gases. RESULTS AND DISCUSSION Particle Separations. Separation of 26-, 44-, and 110-nmdiameter fluorescent particles was performed in 0.5- and 1.0-mmwide Pyrex-silicon channels. Contrary to previous measurements, presented in ref 1, dicarboxyfluorescein was used as a t0 marker. This compound is more negatively charged than the previously used fluorescein and should thus interact less with the negatively charged surface. In this situation, only hydrodynamic and diffusion effects should contribute to its dispersion. The largest sources of dispersion, next to the “normal” TaylorAris dispersion, caused by concentration gradients across the channel height, are the previously mentioned nonuniformity of the channel cross section and the retentive effect of the sidewalls. Analytical Chemistry, Vol. 75, No. 24, December 15, 2003
6765
Figure 3. Etching results in fused silica of the transition region between the nozzle at the end of the separation channel and the optical detection slit. From (A) it is clear that the effect of the RIE lag is not as prominent as desired. In (B) an enlargement of the bottom of the optical detection slit is shown. Apart from some scattered pinholes, the surface appears to be smooth.
Figure 5. Similar separation as in Figure 4, performed in a fusedsilica chip. Buffer was 10 mM borate at pH 9.2; the working pressure was 6.5 bar. In the upper panel, the zone shapes appear somewhat deformed due to nonuniformity of the illumination.
Figure 6. Separation of two proteins and the t0 marker CFL in a Pyrex-silicon chip with a 0.5-mm-wide separation channel. Conditions: 10 mM borate buffer pH 9.2; working pressure 8 bar. Figure 4. Top view of the HDC separation of fluorescent nanoparticles and a marker (carboxyfluorescein, CFL) in a 1- (above) and 0.5-mm-wide (below) channel in the silicon-glass chip. Buffer was 10 mM borate at pH 9.2; the working pressure was 3.5 bar.
The zone profiles in the 0.5-mm-wide channel in Figure 4 clearly show the transient form of the sidewall effect, which applies for h2 , D12t , w2 and illustrate the limitations of its theoretical description. The expression for the effective diffusivity, derived in the Theory section, is not very accurate in the earlier part of the transient period (Figure 4 below, 25 mm), because it is largely formed by the contribution of the long, narrow tail along the sidewalls (Figure 4, 75 mm). In practice, this form of tailing would rather increase the baseline than broaden the peaks. Nevertheless, further down the channel, the sidewall effect becomes significant across the complete channel width and thus reduces the final efficiency. Figure 4 shows that a narrower channel is affected less by deformation of the wafers, since the lateral profiles of the zones of the separated particles as well as the marker are much less deformed in the 0.5-mm channel. Therefore, the largest dispersion source in the 1.0-mm-wide channels is the zone deformation by the channel nonuniformity, while in the 0.5-mm-channel, the wall effect dominates. As a comparison, a similar separation, performed in a fusedsilica chip, is shown in Figure 5. In this chip, the zone distortion 6766 Analytical Chemistry, Vol. 75, No. 24, December 15, 2003
is much smaller than in the 1.0-mm-wide channel in the Pyrexsilicon chip. Unfortunately, no 1.0-mm-wide fused-silica channels were available so that a good comparison of the cross section nonuniformity is difficult. It can be expected, however, that the influence of the thermal deformation is reduced as top and bottom wafers are thermally matched. Biopolymer Separations. Attempts to apply HDC chromatography to separation of biomacromolecules such as proteins, DNA, and TMV or even cells can be found in the literature.3,36-38 In the work of Kraak et al.,36 a preliminary separation of proteins was obtained in packed column HDC (PCHDC) with 2-µm packing size. We have performed a comparable separation under similar conditions in the 1-µm-deep, 0.5-mm-wide channel of the Pyrexsilicon HDC chip (Figure 6). In both cases, the effective conduit size was in the same order of magnitude (∼1-2 µm), the conduit walls were from silica, a buffer of pH 9 and ionic strength of ∼10-3 M was used, and in both cases, the smaller protein (BSA, respectively. γ-globulin) had a size of ∼10 nm. In PCHDC,36 the small protein was still only partly separated from the t0 marker after 8 min, while in our HDC chip it was almost completely separated in ∼1 min, i.e., 1 order of magnitude faster. (36) Kraak, J. C.; Oostervink, R.; Poppe, H.; Esser, U.; Unger, K. K. Chromatographia 1989, 27, 585-590. (37) Hoagland, D. A.; Prud’homme, R. K. Macromolecules 1989, 22, 775-781. (38) Noel, R. J.; Gooding, K. M.; Regnier, F. E.; Ball, D. M.; Orr, C.; Mullins, M. E. J. Chromatogr. 1978, 166, 373-382.
Figure 8. Relative residence time of colloidal particles in HDC chip separations in aqueous eluents under different conditions. Data from silicon-glass chips (+, X) and a fused-silica chip (4, O, 0). Ionic strength of the eluent is indicated in the figure (measurement in water is assigned I < 10-5 M); pH is ∼9 (gray filling) or ∼7. Linear flow velocity is ∼0.21 (O), ∼0.46 (0, +, ×), or ∼0.72 mm/s (∆). In the measurements with UV detection (solid shapes), benzoate was used as t0 marker. Carboxyfluorescein was used as a marker in all other measurements. Drawn lines indicate the theoretical retention eq 1 with C ) 0.5, 1, and 2.
Figure 7. Chromatograms obtained in the fused-silica HDC chip using UV detection. Separation of polystyrene latex standards of 155 (1) and 61 nm (2), the previously used fluorescent particles 26 nm (3), and benzoate (marker, 4) in 1 mM MOPS buffer, pH 7. Detection at 230 nm and working pressure 6.1 bar (A, above), 1.7 (A, below) and 4.6 bar (B). The three curves in (B) illustrate the variation in the noise level between three comparable separation runs.
UV Absorption Separations. The fused-silica chip that was used for separation and subsequent fluorescence detection of particles, shown in Figure 5, was used in a UV absorption detection setup. Chromatograms showing separation of polystyrene latex particle standards and benzoate, used as t0 marker, are presented in Figure 7. The 26-nm fluorescently labeled polystyrene particles were added for comparison with the previous fluorescence measurements. As can be seen, reproducible chromatograms at different flow velocities can be obtained with this basic setup. However, the noise level of the detection is high and unstable during the experiments. This is because the optical components in the prototype are not mounted rigidly to the chip but to the common base, which also supports the multicomponent positioning system of the chip. The sensitivity is also limited by stray light bypassing the cell because of the larger aperture and by the fact that no focusing is used for the excitation beam. However, considering the simplicity of the system, the obtained results are very promising and motivate further developments. Retention Curve for Different Ionic Strengths. As already noted in the Theory section, the elution behavior of particles in aqueous buffers is much more complicated than the simplest HDC theories predict. This is illustrated in Figure 8, where the experimental data obtained from various chips under various conditions are plotted together with the simple theoretical predictions.
The strong influence of ionic strength can be explained by the repulsion of the expanded electrical double layers at the channel wall and the particle surface. This repulsion restricts the particles to a narrower inner core of the channel than the steric exclusion from the wall would do and effectively reduces the channel dimensions. Although this increases the selectivity of the method, the results in very weak buffers (I < 1 mM) are less reproducible than in stronger buffers. There is also an effect of pH visible, probably caused by the influence of the pH on the ζ potential. The fact that this is most apparent for the 44-nm particles could be due to different surface properties of those particles compared to the 26- and 110-nm particles from the same manufacturer. The influence of the flow velocity is almost negligible in the studied region of conditions. This suggests that radial hydrodynamic forces, such as the “tubular pinch”, are small. However, lateral forces such as the retarding particle-wall hydrodynamic interaction (see ref 6) may be present. In fact, the slope of trendlines of the experimental data is higher than the slope of the basic model with C ) 1 (which corresponds to an absence of hydrodynamic particle-wall interaction). In the plot, a value C ) 2 is used for comparison. Clearly, a better theoretical description of elution of particles in aqueous buffers is needed for the HDC chip. Dispersion: Sidewall Effect. From the chromatograms obtained using UV absorption detection, a preliminary quantification of the dispersion in the fused-silica chip is possible. For determination of the peak widths, Gaussian fits were used. The results are shown in Figure 9 for the two sizes (61 and 155 nm) of nonfluorescent particles, for 26-nm fluorescent particles, and for benzoate, which was used as a t0 marker. The data are compared to theoretical values, calculated using the expression for the effective diffusivity given in the Theory section. The particle diffusion constants are determined according to formula 6. Considering the noise level of the UV measurements, a reasonable correspondence between theoretical and experimental dispersion is found for both nonfluorescent particles and the Analytical Chemistry, Vol. 75, No. 24, December 15, 2003
6767
cies of ∼10 000 plates obtained for the smaller analytes are too low. The relative influence of the Taylor-Aris dispersion, compared to other dispersion sources, suggests that improvements can be found in both areas. Thus an improvement in channel uniformity will be beneficial. A reduction of the contribution of the sidewalls to the dispersion can be achieved by using wider channels.
Figure 9. Dispersion expressed in the number of theoretical plates for the 0.5-mm-wide separation channel defined in the fused-silica HDC chip. The measured values are obtained under the same conditions as in Figure 7. Plate numbers are obtained at different velocities: 0.72 (O), 0.46 (0), and 0.21 mm/s (4). The theoretical dispersion is indicated by dotted lines.
marker. However, there is a large difference between theory and reality for the fluorescent 26-nm particles for which no satisfactory explanation can be found. The theoretical expression for the dispersion describes the transient situation in which vertical concentration gradients are fully leveled out and horizontal gradients are starting to add to the dispersion. It turns out that neglecting these horizontal contributions leads to even higher theoretical plate numbers. The same applies if the correction for finite particle sizes is not implemented. It can therefore be concluded that the incorporated correction terms are qualitatively correct. The theoretical dispersion model only takes Taylor-Aris dispersion into account. However, injection, detection, and nonuniformity of the separation channel increase the total peak broadening as well. Provided the theory gives an accurate description, the remaining difference between theoretical and actual plate numbers can be attributed to these dispersion sources. The highest achieved plate numbers in the experiments with 61- and 155-nm particles were ∼80 000. Nevertheless, the efficien-
6768
Analytical Chemistry, Vol. 75, No. 24, December 15, 2003
CONCLUSIONS For the first time, on-chip planar hydrodynamic chromatography was combined with UV absorption detection. The hydrodynamic separations were performed in a planar configuration, realized in fused silica, using a mixture of fluorescent and nonfluorescent polystyrene particles. The planar chip configuration consisted of a 1-µm-high, 0.5-mm-wide, and 69-mm-long channel, an integrated 150-pL injection structure, and a 30-µm-deep and 30-µm-wide detection cell, suitable for UV absorption detection. Combination of these separation data and those obtained using a previously presented planar silicon-glass hydrodynamic chromatography chip leads to a description of the retention and dispersion behavior of planar hydrodynamic chromatography. Especially the influence of the sidewalls on the dispersion was investigated. Next to the nanoparticle separations, a hydrodynamic separation within 70 s of several biopolymers was shown in the glass-silicon chip. ACKNOWLEDGMENT The presented work was a part of the project “Hydrodynamic Chromatography in Integrated Micromachined Separation Systems” carried out by cooperation of the University of Amsterdam and the University of Twente and funded by the Dutch Technology Foundation STW-NWO as Project AAC4556.
Received for review June 19, 2003. Accepted September 19, 2003. AC034663L
