Anal Bioanal Chem DOI 10.1007/s00216-008-2591-x
SHORT COMMUNICATION
Rapid detection of Mycobacterium tuberculosis cells by using microtip-based immunoassay Woon-Hong Yeo & Shieng Liu & Jae-Hyun Chung & Yaling Liu & Kyong-Hoon Lee
Received: 27 October 2008 / Revised: 15 December 2008 / Accepted: 18 December 2008 # Springer-Verlag 2009
Abstract This paper describes a microtip-based approach of concentrating target analytes for a highly sensitive bioassay. As an example, rapid screening of bacterial whole cells is presented to detect Mycobacterium tuberculosis (MTB), a pathogenic bacterium for human tuberculosis (TB). The concentration and detection is performed with three sequential steps of (1) attracting bacterial whole cells in the vicinity of a microtip by alternating current electroosmotic flow; (2) capturing the cells onto the microtip by capillary action; (3) binding fluorophorelabeled polyclonal antibodies to the cells followed by fluorescence measurement (immunofluorescence). Through this mechanism, MTB cells have been detected to the concentration of 8,000 cells/mL within 10 min. This sensitivity is comparable to that of Ziehl–Neelsen smear microscopy, a common culture-free screening method for diagnosis of TB. For comparison, Escherichia coli O157: H7 cells have also been detected to the concentration of 30,000 cells/mL in the same way. Electronic supplementary material The online version of this article (doi:10.1007/s00216-008-2591-x) contains supplementary material, which is available to authorized users. W.-H. Yeo : S. Liu : J.-H. Chung Department of Mechanical Engineering, University of Washington, Campus Box 352600, Seattle, WA 98195, USA Y. Liu Department of Mechanical and Aerospace Engineering, University of Texas at Arlington, Box 19018, Arlington, TX 76019, USA K.-H. Lee (*) NanoFacture, Inc., 3983 Research Park Dr Suite 100, Ann Arbor, MI 48108, USA e-mail:
[email protected]
Keywords Immunoassay . Immunofluorescence . Mycobacterium tuberculosis . Biosensor . AC electroosmosis . Capillary action
Introduction Tuberculosis (TB), one of the most widely spread diseases, has infected one third of the world’s population, according to the report by WHO [1]. In 2006, 9.2 million new TB cases were reported with 1.7 million victims. Even in developed countries, the incidence of TB is a significant social concern because it may appear as a complication of acquired immune deficiency syndrome infection. For example, in the USA, there were 15,000 new patients in 2006. Since a single patient is known to transmit the disease to 12–15 people/year on average, the prompt diagnosis and treatment of a new patient is considered crucial to the effective control of the disease. To diagnose TB, Mycobacterium tuberculosis (MTB) is identified in clinical samples. The gold standard is cultural growth of the bacterial species [2], which requires more than 2 weeks for detection [3, 4]. For this reason, direct microscopy of a Ziehl–Neelsen smear using sputum (sensitivity, 104 cells/mL) [2] has traditionally been preferred as a relatively quick (culture-free) and inexpensive screening test; however, this method is labor-intensive and poses a large source of potential errors (the sensitivity is 20∼30%). To reduce the errors, the test should be performed three times for 3 days with a limited sample throughput (20 samples/technician/day). To overcome these issues, worldwide efforts have been made to develop a simple, rapid, yet highly sensitive assay available at an affordable cost to replace the century-old Ziehl–Neelsen smear technique. In the environment with limited infrastructure, an equipment-free test such as
W.-H. Yeo et al.
immunochromatic strip assay is preferred. Even in a wellequipped environment, simple instrumentation is expected to provide user-friendly interfaces and quick test results. For example, in the USA, the Center for Disease Control and Prevention (CDC) seeks to develop a new method for TB detection by 2010. The new method should be capable of diagnosing the disease within 48 hrs by nontrained personnel at public health laboratories. For more rapid and accurate detection, two major approaches are available without cell culture [3, 5]. One is to use nucleic acid testing with a target/signal amplification strategy, such as polymerase chain reaction. This method is generally sensitive and specific but entails cumbersome procedures run by skilled operators with expensive instrumentations and reagents. The other method is to employ an immunoassay based upon the detection of MTB-associated targets in clinical samples. The analytes include the whole cells of MTB [2, 5–7], antigens circulated in serum [8–11], antibodies forming against those antigens [10, 12], and antigen–antibody complexes [6, 10]. Due to the simple and straightforward working principle, immunoassays have been most extensively studied as a potential alternative for the smear method. However, immunoassays have failed to demonstrate enough sensitivity (comparable to or better than 104 cells/mL) to replace the smear method. It is fundamentally challenging to detect the lowabundance cells (104 cells/mL) by using an immunoassay. Recently, Sheehan et al. theoretically proved that the limit of detection (LOD) of a biosensor is associated with target material delivery to a sensing component rather than the detection capability [13]. This analytical prediction emphasizes the importance of preconcentration of analytes for highly sensitive detection. For instance, even with an extremely sensitive sensor capable of detecting a single molecule, the LOD cannot be lower than 105 molecules/mL, as far as a molecule is delivered to the sensor by only diffusion. In this regard, the concentration of TB cells to a sensing component is an important issue for a high sensitivity. This paper, for the first time, introduces a capillaryinduced concentration method of cells in conjunction with an alternating current (AC) electric field. When an AC field is applied to a tip, vortices due to electroosmosis are produced to concentrate cells in the vicinity of the tip. Unlike a previous concentration method [14], the concentrated cells are captured to a tip by capillary action, which enables rapid detection of the cells. In detail, a microtip is immersed in a sample solution containing analytes, either MTB or Escherichia coli O157:H7 cells. The tip sensor is operated to concentrate the analyte on its surface through the following sequences of (1) attraction of the cells in the vicinity of a tip by AC electroosmotic flow and (2) capture of the cells onto the high aspect ratio microtip by capillary action. The captured cells are then detected by using
fluorophore-labeled antibodies. Using the proposed method, both M. tuberculosis and E. coli O157:H7 are detected within 10 min. The concentration mechanism demonstrates a sensitivity comparable to or better than ∼104 bacterial cells/mL.
Experimental section In experiment, two kinds of electrodes were prepared as (1) microelectrodes and (2) a microtip and a coil. Microelectrodes were used to optimize the frequency of an AC field and validate the generation of an electroosmotic flow. The planar microelectrodes were convenient to observe the flow pattern rather than the three-dimensional tip electrodes. To fabricate the planar electrodes, 100-mm Si (silicon) wafers were oxidized to form a 300-nm-thick insulation layer. Gold electrodes having a 100-μm-wide gap were patterned by a conventional UV lithography technique. The planar electrodes were used to investigate the generation of an electroosmotic flow and the transport of MTB cells. For capturing MTB cells, a tip and coil-electrode configuration was used as illustrated in Fig. 1. A silver-coated copper (Cu) wire (250 μm in diameter, OK industries, Tuckahoe, NY, USA) was bent to form a ring-shaped electrode. The Cu ring electrode was aligned with a microtip on a homemade XYZ stage. The microtip was prepared by fracturing a gold-coated tungsten wire (50 μm in diameter, Sylvania, Towanda, PA, USA) by axial tension. Due to the necking of the wire, the end diameter of a prepared tip was reduced to 40 μm. For the MTB cells used in this experiment, a stock solution (M. tuberculosis, strain H37Ra, initial concentration: 8×1012 cells/mL) was obtained from National Institutes of Health (NIH) TB Vaccine Testing Materials Contract
Electroosmotic flow
Metal coil
Capillary action Microtip
Withdrawal direction
AC
Cells
Cell solution
Fig. 1 Concentration of target bacterial cells by electroosmotic flow (concentration of cells) and capillary action (capture of cells)
Rapid detection of Mycobacterium tuberculosis cells
(HHSN266200400091c, Colorado State University). Using a phosphate-buffered saline (PBS) buffer (1× PBS) from the original 10× concentrate (pH 7.4, Sigmal-Adrich, St. Louis, MO, USA), the prepared concentrations of MTB cells were 8×102, 8×103, 8×104, 8×105, 8×106, 8×107, 8×108, and 8×109 cells/mL. Note that the same 1× PBS buffer was used to prepare all MTB and E. coli cell suspensions, offering the consistent ionic concentration of the assay medium. Thereby, the electroosmotic flow was controlled by an AC field. The MTB cells were already treated with γ-irradiation by the provider. The cells were pathologically inactive but still biologically active by 93–95% according to the provider’s instruction. For fluorescent measurement, fluorescein-labeled polyclonal antibody was purchased from ViroStat Inc (#4603, Portland, ME, USA). Microelectrode experiment for frequency determination The prepared microelectrodes were used to demonstrate the generation of AC electroosmotic flow and to optimize the frequency of an AC field because the cell motion in the spherical drop of the ring-tip configuration was not clearly visible under an optical microscope. In spite of the different configurations, the underlying mechanism for generating AC electroosmotic flow (frequency dependency) is the same for both microelectrodes and a microtip. To investigate the concentration performance of MTB cells using an AC field, the frequencies tested were 10, 100, 1,000, and 5,000 kHz at 20 Vpp (Function Generator, Agilent 33220A, Santa Clara, CA, USA). Frequencies lower than 10 kHz were not attempted because the lower frequencies attracted ionic particles and contaminated electrodes. Microtip experiment for cell capturing To capture and detect MTB cells, a 2 μL MTB aliquot was held on the ring by surface tension in the XYZ stage. Then, the microtip was dipped in the aliquot with an AC field (immersion depth, 200 μm) as shown in Fig. 1. Using a selected frequency, an AC field of 20 Vpp (peak-to-peak voltage) was applied to the ring and the tip for the concentration of cells. This concentration process was also analyzed by our simulation. After 1 min immersion of a tip into a drop, it was withdrawn at a rate of ∼8 μm/s. The withdrawal rate did not significantly change the solution meniscus at the tip surface. Subsequently, the tip was immersed in a prepared antibody solution (3 μL, 4.5 mg/mL) for 5 min and then rinsed with a deionized water drop (10 μL) two times. As for the immersion time of the tip in the antibody solution, the interval from 1 to 5 min was varied by 1 min increments. It was found that the fluorescence intensity was saturated at 5 min. The rinsed tip was observed under an epifluorescence microscope (Olympus BX-41, Olympus
America, Melville, NY, USA). The schematic of the experimental procedure using a microtip is illustrated in the flow chart (Fig. 2). To obtain a background emission of immunofluorescence, a blank tip without MTB cells was immersed in the antibody solution and rinsed according to the same procedures. Subsequently, the intensity was detected by fluorescent microscopy. The intensity of fluorescence was measured by an imaging software (Image J, NIH) based on the area of 260× 95 μm from the tip end. All intensity measurements were performed for the same area throughout this paper, and all experiments were repeated three times. The captured cells onto the tip electrode were also investigated by a scanning electron microscope (FEI Sirion SEM, Hillsboro, OR, USA) for the validation of the fluorescent measurement. The same set of experiments was exactly repeated for E. coli O157:H7 positive control cells (heat treated; KPL, Inc., Fig. 2 Flow chart of the experimental procedure to capture and detect MTB cells
Device set-up (tip & coil preparation) Sample preparation (cell & antibody solution) Loading of a cell drop to a coil Switch on an AC field Tip immersion in the cell drop (1 min.) Tip withdrawal Switch off the AC field Coil replacement Loading of an antibody drop to the new coil Tip immersion in the antibody drop Rinsing with DI water Fluorescent detection
W.-H. Yeo et al.
Gaithersburg, MD, USA). The only difference was the prepared concentrations; 3×103, 3×104, 3×105, 3×106, 3× 107, 3×108, and 3×109 cells/mL. The difference between the concentration of MTB and E. coli cells was ascribed to the original concentration from the suppliers.
Results and discussion In the process of cell concentration, the frequency of an applied AC field is a crucial parameter to efficiently concentrate bacterial cells to a tip. The purpose of the experiment with planar electrodes is to demonstrate the transportation of cells due to electroosmotic flows because it is hard to observe the cell motion for a microtip electrode using an optical microscope. The frequency of an AC field was chosen on the basis of the observation of microelectrodes as explained in the “Experimental section.” Using the microelectrodes, cell motions upon application of an AC field could be observed and recorded by an optical microscope and recording equipments (Fig. 3). When a solution including MTB cells (8×107 cells/mL) was placed on microelectrodes, the cell behaviors were changed according to the frequency of an AC field. When an AC frequency was 10 kHz, MTB cells were attracted to the electrodes by electroosmotic flow. The AC field, however, partially damaged the Au electrode layer. At a 100-kHz frequency, electroosmotic flow was also generated in the vicinity of the electrode edges. The circulation flow transported and concentrated MTB cells to the electrodes with negligible damage to the electrodes. In particular, large clusters of cells were delivered to the edges of the electrodes. The white dots, as shown in Fig. 3 (movie captures), are colonies of cells transported by the electroosmotic flow. A cell colony, circled in the figure, is transported to the electrode edge (at 0 s) and then moved to the electrode surface (at 0.5 s). After a gentle blow dry of the solution by nitrogen gas, a viscous and thin film composed of cells was observed on the electrodes.
a
Without an AC field, such a film was not observed at the same concentration (8×107 cells/mL). When the frequency was changed to 1 MHz, the concentration efficiency was lowered such that the colony formation was rarely observed. At 5 MHz, an AC electroosmosis and the resultant cell concentration was not generated [15] because the electric polarity change was faster than the ion mobility. Regardless of the frequencies, dielectrophoresis was not induced in this experiment because the pathologically inactive MTB cells (dead cells) were less polarized than the buffer solution due to damaged extracellular matrix of the cells. This dielectrophoretic response of dead cells was investigated for separating live cells from dead cells [16]. Based on the results of the microelectrode experiment, an AC frequency was chosen as 100 kHz. The frequency at 20 Vpp was applied to capture target bacterial cells to a microtip throughout all experiments in this paper. The AC field was applied for 1 min between a microtip and a ring that was holding an aliquot of a cell solution. After the cell concentration by AC electroosmosis, the tip was withdrawn to capture cells. Note that the microtip, made of a goldplated tungsten wire, had been exposed to air prior to the capture. The passivated gold surface offered an adhesive layer to cells. When an AC field is applied to a tip, an ion layer forms on the surface of the tip. The sign of the charge of the electrode and the double layers changes according to the alternation of the potential. In such case, electrostatic force of the charged ions is generated in the tangential direction to the surface, which induces AC electroosmotic flow. As the electric field strength decreases with increasing distance from the tip end, the flow speed is maximal at the tip end and decreases rapidly toward the other end of the tip. Due to the non-uniform flow speed on the tip, vortices are produced to concentrate cells in the vicinity of the tip. This concentration mechanism was predicted by a modeling approach, the immersed electrokinetic finite
b
Fig. 3 Concentration of MTB cells on microelectrodes (cell concentration, 8×107 cells/mL). a A colony in the circle approaches to the electrodes by electroosmotic flows at 0 s; b the colony is transported to one electrode at 0.5 s (scale bars, 400 μm)
Rapid detection of Mycobacterium tuberculosis cells
a
b
c
d
Fig. 4 Simulation results of cell concentration by AC electroosmosis. a Cells positioned around a tip end located at the top center of a rectangular fluid domain (time=0 s); b–d cell concentrations at times 1, 2, and 10 s, respectively
a
b t’
t
c
d t
Fig. 5 Film formation during the withdrawal of a tip. The concentration of MTB cells for this experiment is 8×109 cells/mL. a, b A microtip with the application of an AC electric field; c, d microtip without an electric field. a Tip prior to complete withdrawal from a solution containing cells at 0 s; the diameter of the tip is about 55 μm (original diameter, 50 μm); b thickening of the film on the tip
t’
due to the evaporation and contraction of the solution at 4 s; the maximum thickness of the tip indicated by the arrows is 70 μm; c tip prior to complete withdrawal from a solution at 0 s; d after the tip withdrawing at 4 s. The thickness between t and t’ is negligibly changed (scale bars, 200 μm)
W.-H. Yeo et al.
a
b
50µm
50µm
Fig. 6 Results of capturing MTB cells at concentration of 8×106 cells/mL. a Optical microscope image of a tip having MTB cells bound with antibodies; b corresponding fluorescent microscope image
element method [17] (see video in Electronic Supplementary Material). In the simulation, the flow was generated by electroosmosis driven by the electrostatic force applied to the charged double layer on the tip surface. Because the Debye layer is only a few nanometers thick, the electroosmotic flow is considered as a slip boundary condition on the tip surface as: v ¼ "ym0 E in the simulation. Here, ψ0 is the zeta potential (2 mV [18]), μ is the medium viscosity (1 mPa.s [19, 20]), ɛ is the medium permittivity (7.1× 10−10 F/m [19, 20]), and E is the electric field. Figure 4 shows the simulation results that cells are concentrated in the vicinity of a microtip due to the electroosmotic flows. When the tip is withdrawn from the solution, capillary action between the tip and the solution forms a meniscus on the tip surface. Further, the concentrated cells increase the viscosity of the solution on the tip surface. As a result, a thin film forms on the tip as shown in Fig. 5a, b (see video in Electronic Supplementary Material). Without an AC electric field, the film formation is not observed (Fig. 5c, d), and cells are rarely captured to a microtip. The thin film formation is associated with the viscosity and the surface tension of a medium, which is described by the capillary number C ¼ mU g , where μ is the viscosity, U is
a
b
the relative velocity between a medium and a substrate, and γ is the fluid surface tension. When cells are concentrated to the tip, the medium becomes a non-Newtonian fluid with increased viscosity. This higher viscosity causes the formation of a film composed of cells [21]. Through this mechanism, cells are nonspecifically captured to a tip. Figure 6 represents the results of capturing MTB cells on a microtip. Figure 6a is an optical image of the tip capturing MTB cells, and Fig. 6b shows the corresponding fluorescent image. The captured cells can be observed by a SEM in Fig. 7. For comparison, Fig. 7a–c show an untreated tip immersed in a pure buffer solution, a tip having only MTB cells, and a tip having MTB cells bound with antibodies. When the cell concentration is varied between 800 (= ∼103) and 8×109 (= ∼1010) cells/mL, the microtip sensor was found to detect as low as 8,000 cells/mL based on the fluorescent measurement (Fig. 8). In the graph, the fluorescent intensity continuously increases in the range between 8×103 and 8×106 cells/mL. At the concentration of 800 cells/mL, cells are not captured to a microtip. The intensity of fluorescence is within the error range of the negative control signal. Note that all experiments were performed three times (n=3) for all concentrations.
c
Fig. 7 SEM images of microtips. a Untreated tip immersed in a pure buffer solution; b tip having only MTB cells without antibodies; c tip having MTB cells bound with antibodies (scale bars, 10 μm)
Rapid detection of Mycobacterium tuberculosis cells
a
Fig. 8 Fluorescent intensity at various cell concentrations (n=3), where negative control shows the background emissions. The error bars in the graph show maximum and minimum intensities
When the concentration is greater than 8×107 cells/mL, the fluorescent intensity fluctuates. The fluctuation of the intensity is ascribed to the formation of large colonies composed of bacterial cells. The colonies cannot be captured or may be released due to capillary action during the withdrawal of a microtip, which can be observed through a microscope during experiments. In spite of the fluctuation of the intensity, all fluorescent intensities at concentrations equal to or greater than 8,000 cells/mL are larger than the negative control signal. To test the performance of the microtip-based immunoassay, the same experiment was performed for capturing E. coli O157:H7 cells. Figure 9a shows the fluorescent intensity with respect to the number of cells. In this experiment, the sensitivity of a microtip for E. coli O157 cells is 30,000 cells/mL because the fluorescent emission of 3,000 cells/mL is in the error range of the negative control signal. Figure 9b, c show SEM images for the captured E. coli cells at 3×106 cells/mL. Overall, the normalized intensities of E. coli cells are much higher than those of TB cells, simply because the fluorescent intensity of the E. coli antibody is much higher than that of the TB antibody, both of which were obtained from the same commercial source. The intensity of the antibodies can be compared by dipping a bare microtip in the antibody solutions and measuring the fluorescent intensities without rinsing. The detection limit for both MTB and E. coli O157:H7 cells is on the order of ∼104 cells/mL (= 10 cells/μL). This sensitivity is one to two orders of magnitude higher than conventional immunoassays. The whole detection time was also significantly shortened to 10 min by directly detecting target cells captured on a microtip. In the proposed scheme, however, the assay appears to work only as a qualitative
b
50µm
c
5µm
Fig. 9 Results of capturing E. coli O157:H7 cells. a Fluorescent intensity at various cell concentrations (n=3), where negative control shows the background emissions. In the graph, error bars represent the maximum and minimum intensities; b SEM image of a tip electrode with E. coli cells captured on the surface (3×106 cells/mL); c Exploded image showing individual cells (rod and sausage shapes)
W.-H. Yeo et al.
true/false screening test according to the large error bars of the fluorescent-intensity graphs. To decrease the error bars, capillary action has to be controlled more precisely. Considering dependency of capillary action on tip geometry, this can be achieved through a more precise fabrication of a microtip. Also, automatic control over a uniform withdrawal rate of a microtip can reduce an abrupt change of capillary force, which can further decrease the error. It should also be noted that dead cells used in this paper are not affected by dielectrophoresis generated by an AC field. When this microtip-based sensor is applied to live cells, dielectrophoretic force can enhance the capturing performance. To enhance the sensitivity of the proposed microtip sensor, a sample volume should be increased. The sample volume used in this experiment is only 2 μL, which limits the number of cells in a sample drop. When the concentration of a solution is 104 cells/mL, only 20 cells are present in 2 μL. By increasing the sample volume to 100 μL or 1 mL, more cells can be contained in a sample aliquot. Considering the proposed electroosmotic concentration mechanism, the concentration at a larger volume should work, but the concentration time and the tip geometry should be optimized for handling such a greater volume. All these aspects considered here should be investigated further for a higher sensitivity of an immunoassay having a high throughput.
Conclusions An immunofluorescence tip sensor was developed for rapid detection of M. tuberculosis (MTB) and applied to detect E. coli O157:H7 cells. Using a microtip, an AC electric field was applied to generate AC electroosmotic flow that concentrated the bacteria in the vicinity of the microtip. The concentrated bacteria were captured by capillary action and detected by fluorophore-labeled polyclonal antibodies. Through this mechanism, MTB cells were detected to the concentration of 8,000 cells/mL within 10 min. In addition,
E. coli O157:H7 cells were detected to the concentration of 30,000 cells/mL in the same way. Acknowledgment The authors acknowledge the support of a CDC contract (200-2007-M-22794, Program Manager: David A. Wilson at Division of TB Elimination/Nat. Center for HIV, Viral Hepatitis, STD & TB/Centers for Disease Control and Prevention).
References 1. WHO (2006) Diagnostics for tuberculosis-global demand and market potential. Geneva, Switzerland 2. Chan ED, Reves R, Belisle JT, Brennan PJ, Hahn WE (2000) Am J Respir Crit Care Med 161:1713–1719 3. Cheng VCC, Yew WW, Yuen KY (2005) Eur J Clin Microbiol Infect Dis 24:711–720 4. Van Deun A, Portaels F (1998) Int J Tuberc Lung Dis 2:756–765 5. Sada E, Aguilar D, Torres M, Herrera TJ (1992) J Clin Microbiol 30:2415–2418 6. Arias-Bouda LMP, Nguyen LN, Ho LM, Kuijper S, Jansen HM, Kolk AHJ (2000) J Clin Microbiol 38:2278–2283 7. GlatmanFreedman A, Martin JM, Riska PF, Bloom BR, Casadevall A (1996) J Clin Microbiol 34:2795–2802 8. Araj GF, Fahmawi BH, Chugh TD, Abusalim M (1993) Diagn Microbiol Infect Dis 17:119–127 9. Attallah AM, Malak CAA, Ismail H, El-Saggan AH, Omran MM, Tabll AA (2003) J Immunoassay Immunochem 24:73–87 10. Diaz-Gonzalez M, Gonzalez-Garcia MB, Costa-Garcia A (2005) Biosens Bioelectron 20:2035–2043 11. Lyashchenko KP, Singh M, Colangeli R, Gennaro ML (2000) J Immunol Methods 242:91–100 12. Pottumarthy S, Wells VC, Morris AJ (2000) J Clin Microbiol 38:2227–2231 13. Sheehan PE, Whitman LJ (2005) Nano Lett 5:803–807 14. Wong PK, Chen CY, Wang TH, Ho CM (2004) Anal Chem 76:6908–6914 15. Trau M, Saville DA, Aksay IA (1997) Langmuir 13:6375–6381 16. Pohl HA, Hawk I (1966) Science 152:647–649 17. Liu Y, Liu WK, Belytschko T, Patankar N, To AC, Kopacz A, Chung JH (2007) Int J Numer Methods Eng 71:379–405 18. Pan B, Cui D, Xu P, Li Q, Huang T, He R, Gao F (2007) Colloid Surf A 295:217–222 19. Gray DS, Tan JL, Voldman J, Chen CS (2004) Biosens Bioelectron 19:1765–1774 20. Saha K (2003) Langmuir 19:1304–1311 21. Shen AQ, Gleason B, McKinley GH, Stone HA (2002) Phys Fluids 14:4055–4068
