E

Ra

b

a

ARRAA

KADIMAS

ftfgfl

1

0d

Sensors and Actuators B 147 (2010) 366–375

Contents lists available at ScienceDirect

Sensors and Actuators B: Chemical

journa l homepage: www.e lsev ier .com/ locate /snb

nhancement of heterogeneous immunoassays using AC electroosmosis

obert Harta, Ryszard Leca, Hongseok “Moses” Nohb,∗

School of Biomedical Engineering, Science & Health Systems, Drexel University, United StatesDepartment of Mechanical Engineering & Mechanics, Drexel University, United States

r t i c l e i n f o

rticle history:eceived 10 December 2009eceived in revised form 26 January 2010ccepted 7 February 2010vailable online 23 February 2010

eywords:C electroosmosisielectrophoresis

mmunoassayicrofluidics

C electrokineticsimulation

a b s t r a c t

Enhancing the transport of analyte to a transducer surface is of critical importance to many biosen-sors, especially for those designed to capture and detect large biomolecules (nucleic acids, proteins, etc.)which diffuse slowly and therefore cause inordinately long detection times. For microfluidic-based sen-sors, which are characterized by low Reynolds number flow, the mass transport is purely diffusive innature, which eliminates the possibility of turbulent mixing. In this article, enhancement of heteroge-neous immunoassays using AC electroosmosis was studied through the use of finite element modelingand fluorescent immunoassays. Simulations of interdigitated planar electrodes indicate that optimumelectrode geometry (electrode width and spacing) scales with the chamber height and that binding timescan be reduced by up to a factor of 6. Optimum frequency of electrical excitation of the electrodes wasbetween 100 and 200 Hz, which was confirmed by fluorescent immunoassays. Immunoassays also showthat binding of secondary antibody on a functionalized surface can be significantly enhanced, especiallyat the ends of each electrode without noticeable nonspecific binding. Unexpectedly, the antibody bind-ing was not uniformly distributed across the surface but patterned preferentially on the edges of theelectrode, which indicates the presence of dielectrophoresis. Analysis of fluorescent intensity shows anenhancement of about 1.9 times at the center of the device and an enhancement of about 6.7 times at the

eno

edges of the best device typfield of electrokinetic enhaor near the binding region

1. Introduction

The current trend towards assay miniaturization on microflu-idic devices creates unique challenges for rapid detection ofbiomolecules. Specifically, detection times for heterogeneousimmunoassays can be inordinately long (several hours [1–3])because of the time it takes for analytes to travel to the devicesurface or transducer, where they then must bind or react withsurface-bound receptors. This detection time limitation is a criticalbarrier that needs to be overcome for a wide variety of importantbiosensor applications, such as rapid bedside diagnostics, environ-mental testing, counterterrorism, etc. [4–7]. Common conditionssuch as low diffusion coefficients and fast surface reactions causethe system to be mass-transport limited (the reaction rate is much

aster than the transport of reactants to the surface) [8,9]. In ordero alleviate this condition, AC electroosmosis (ACEO) as a methodor creating advective mixing is investigated. ACEO is commonlyenerated using simple planar interdigitated electrodes (IDEs). Theuid motion caused by such devices is the result of movement of

∗ Corresponding author at: 3141 Chestnut Street, Rm. 2-115, Philadelphia, PA9104, United States. Tel.: +1 215 895 2273; fax: +1 215 895 1478.

E-mail address: mosesnoh@coe.drexel.edu (H. Noh).

925-4005/$ – see front matter © 2010 Elsevier B.V. All rights reserved.oi:10.1016/j.snb.2010.02.027

compared to a non-mixed counterpart. This work contributes to the growingcement of biosensors through the use of microelectrodes built directly onto,f a transducer.

© 2010 Elsevier B.V. All rights reserved.

induced charges at the surface of the electrodes and creates a rota-tional velocity pattern that can be envisioned as a set of vorticesthat exist on either side of each electrode finger. These flow pat-terns will bring reactants to the surface, thus helping to overcomethe slow diffusional transport that would otherwise dominate.

The very low Reynolds numbers in microscale flows prevent tur-bulent mixing and designs typically utilize channel flow to improvemass transport instead [10]. In this case, the reaction rate can beimproved, but at the cost of much larger sample volumes andthrough the use of bulky pumping equipment or high voltage powersupplies. By contrast, ACEO can be generated using simple planarmicroelectrodes with low voltage and power requirements and canbe microfabricated cheaply with established techniques.

ACEO is one of three AC electrokinetic phenomena. The othertwo are the electrothermal effect (ETE) and dielectrophoresis (DEP).Like ACEO, ETE is a hydrodynamic force. ACEO arises from themovement of induced charges in the double layer on the electrodesurface, whereas ETE arises from the movement of induced chargesgenerated from joule heating within the bulk of the solution. BothETE and ACEO can generate rotational velocity patterns near the

electrodes (Fig. 1). In general, ETE dominates in higher conductivitysolutions and at higher electrical frequencies (typically >1 mS/cmand around 100 kHz) [11] while ACEO exists at lower frequenciesand conductivities (typically <10 �S/cm and less than 10 kHz [12]).

R. Hart et al. / Sensors and Actuators B 147 (2010) 366–375 367

on afluiiesto t

Fig. 1. Diagram of an immunoassay performed with AC electrokinetic enhancementand the concentration field (red: high concentration, blue: low concentration). Theconcentration field image. Red arrows indicate the force due to DEP and the antibodinterpretation of the references to color in this figure caption, the reader is referred

The capability for ETE to improve heterogeneous assays by stirringhas been investigated already [11,13]. Though ETE can operate inphysiologically relevant fluids, voltages of an order of magnitudelarger are typically required in order to attain useful ETE velocities.The other AC electrokinetic force, DEP is caused by a difference inpolarizability between a particle and the medium in which it is sus-pended. In a non-uniform AC field, this difference results in a forceon the particle which causes it to move either towards high electricfield strength (positive DEP—shown in Fig. 1) or towards low elec-tric field strength (negative DEP—not shown). DEP is often used tomanipulate particles larger than several hundred nanometers butis not generally used for molecules. When particles are of molecu-lar length scales the DEP force becomes non-deterministic due tothe dominance of thermal randomization [14,15]. Though DEP hasnot yet been applied to improving molecule detection, it has beenshown to improve particle detection [16,17].

In this article, the use of AC electroosmosis to improve the rateof antibody binding to a sensor surface was investigated using finiteelement modeling and fluorescent immunoassays. The aim of thispaper is to understand how the geometry of the electrodes andfluidic channel affect assay improvement, what optimum voltageand frequency to use, what types of reactions will benefit fromthe use of ACEO and finally to show proof of concept by perform-ing fluorescent immunoassays. These assays will also demonstratethe feasibility of applying these techniques to relevant biologicaltests performed on biosensors. As is the case with many sensorsystems, the more rapid and sensitive the technology is, the moreapplications become available. Currently there are no widely usedimmunosensors in diagnostics or clinical research. This study istargeted to help translate these technologies.

1.1. AC electroosmosis

ACEO is caused by an interaction between a non-uniform electricfield and the electric double layer that is induced at the surface of asubmerged electrode. This electric double layer consists of counter

microelectrode. The concentration is indicated both by the antigen particles (orange)d motion caused by ACEO is shown as blue arrows as well as the streamlines in theare bound both on the surface of the electrode as well as on the glass substrate. (Forhe web version of the article.)

ions that accumulate near the surface of the polarized electrode.The free charges in the double layer experience a lateral coulombforce due to the electric field and translate across the electrodesurface. The mass movement of these charges drags the fluid fromthe edge of the electrode towards the center. This driving force,which exists only very near the surface (a few nanometers) [18],is transferred to the bulk solution through viscous drag ultimatelyforming a rotational fluid pattern (Fig. 1). A more detailed descrip-tion of ACEO, including equations describing the fluid motion nearthe electrode surface is described in an earlier work [19].

1.2. Immunoassay surface reaction

Immunoassays were performed in order to provide empiri-cal validation of simulation results and as a way of observingthe pattern of antibody binding on interdigitated electrodes.Immunoassays are ubiquitous and rely on the excellent speci-ficity of antibody–antigen reactions to capture target proteins (alsoviruses, bacteria, etc.). They are currently a mainstay of medicaldiagnostics and clinical research. Heterogeneous immunoassaysconsist of a surface-based attachment between an antibody and anantigen. Either species may be first immobilized on the solid phasewhile the other is contained in the sample volume and allowed toreact with the surface. The use of a tracer, which may be radioac-tive or fluorescent, is required to generate the signal. In biosensorapplications, the tracer is usually eliminated and the presence ofanalyte is detected through mass, optical, electrical or other phys-ical change on the surface of the transducer.

The chemical reaction that describes a surface binding event isas follows:

c + �kads�kdes

cs (1)

where c is concentration of the species, � is concentration of bind-ing sites at the surface, cs is concentration of bound species at thesurface, kads is the forward binding constant and kdes is the reverseconstant. The rate of change of the bound species at the surface,

ctu

conditions is shown in Fig. 2 (top left).

368 R. Hart et al. / Sensors and A

including surface diffusivity is [20]:

∂cs

∂t+ ∇ · (−Ds∇cs) = kadsc� − kdescs (2)

The concentration of remaining binding sites, �, can also bedescribed as the difference between the total number of binding�0 sites and the occupied sites cs. This equation can therefore beexpressed as follows:

∂cs

∂t+ ∇ · (−Ds∇cs) = kadsc(�0 − cs) − kdescs (3)

Since this reaction removes species from the bulk solution nearthe surface, concentration gradients will form and, in a flow field,interact with fluid flow. This makes the determination of c nontriv-ial and requires concurrent solving of the mass transport and flowfields in the bulk solution.

One way to estimate whether or not a reaction is transportlimited or reaction-rate limited is to calculate the Damkohlernumber (Da). This is a ratio of the reaction velocity (kads�0) tothe rate of transport (D/h) where h is chamber height (Fig. 2).

Da � 1 indicates a transport-limited system while Da � 1 indi-cates a reaction-rate limited system. A range of chemical reactionswith varying transport-limited Da numbers was employed for thenumerical simulations. If the analyte is a large molecule, a protein

Fig. 2. Top left and bottom: simulation geometry and boundary conditions. The numerside of a microfluidic chamber. The model takes advantage of symmetry and repeatingheight, electrode width and gap width respectively. x is the cross-sectional position stin the table to the right. Note that the reaction occurs only on the bottom surface (3 adetails were published earlier [19]. Top right: flow chart describing simulation schemviscosity and density of the fluid are inputs for the fluid dynamics simulation but not s

ators B 147 (2010) 366–375

for example, Da will usually be high due to its small diffusion coef-ficient, especially if the reaction at the surface is a simple, rapidbinding event.

2. Materials and methods

2.1. Numerical simulation

The numerical simulation describes the behavior of a parallelinterdigitated electrode (IDE) placed at the bottom of a microchan-nel. The model assumes a location in the center of the IDE whichis free of edge effects. Furthermore, since the fluid behavior is thesame along the length of an electrode, a 2D simulation is sufficient.(IDEs used in the experimental section consist of 40–200 fingers,each 1 cm long.) As the structure consists of repeating units of elec-trodes with each unit behaving identically, it was only necessary tomodel a single electrode. Moreover, the simulation can take advan-tage of the symmetry of each electrode finger and the generatedfluid pattern. A schematic of the simulation geometry and boundary

The overall model consists of three linked simulations, which arediagrammed in Fig. 2 (top right). (1) A steady-state fluid dynamicssimulation generates a fluid field produced by ACEO. Parametersfor this simulation include (among others) the geometry of the

ical simulation consists of a cross-section of an interdigitated electrode placed on oneunits and only considers one half of an electrode and gap. h, a and b are the channel

arting from the center of the electrode gap. Boundaries are numbered and explainednd 4). The reaction simulation is therefore inactive on all other boundaries. Equatione. The input parameters shown are important but not a complete set. For example,

hown in the diagram.

ctuators B 147 (2010) 366–375 369

2.1.3. Surface reaction simulationSince there is a balance between the bound surface mate-

rial and the material contained within the bulk solution, theconvection–diffusion simulation must be coupled to the surfacereaction simulation. COMSOL’s weak boundary form [23], which

Fig. 3. Two fluorescent assays were developed to test proof of concept for the ACEOmixer device. (a) Shows a nonspecific adsorption assay with FITC-IgG. (b) Shows animmunoassay with a functionalized surface.

provides the ability to implement PDEs on a boundary and couplethem to the physics in other simulations is needed in order to modelthe reactive boundary. This both allows for the tracking of boundmaterial and provides for a means of simulating surface diffusion.The condition at the surface is a PDE (Eq. (3)) which describes theaccumulation of mass from the solution as well as surface diffusionalong the boundary.

2.2. Fluorescent immunoassays

A heterogeneous fluorescent immunoassay can reveal the pres-ence of a biological analyte by binding it to a surface and labelingit with a fluorescent antibody. Using this technique can revealwhether ACEO increases the accumulation of fluorescently labeledantibodies on the surface. These assays were performed on the sur-face of IDEs and imaged using fluorescence microscopy. In orderto test the simplest transport-limited case, a direct, nonspecificadsorption assay using fluorescent IgG was employed. This assay, adiagram of which is shown in Fig. 3 (left), consisted of incubating asolution of purified fluorescein-labeled IgG on the IDE surface for anallotted time period. The second assay tests binding on a function-alized surface. In this case, the functionalized surface is prepared byincubating whole rabbit IgG on the surface of the device, blockingwith BSA, and then reacting with fluorescein-labeled goat anti-rabbit IgG. A diagram of this assay is shown in Fig. 3 (right).

2.2.1. Electrode fabrication and cleaning procedureIDEs consisting of chromium/gold electrodes on a glass substrate

were fabricated using the lift-off technique, which is described in anearlier work [19]. IDE devices were immersed sequentially in thefollowing solutions for 10 min each: methanol, acetone and iso-propanol. After drying under N2 gas, the devices were placed inpiranha solution (7:3, H2SO4:H2O2) for 1 min at 80 ◦C, followed bya 10 min immersion in water. The devices were then washed in anethanol stream, dried under N2 gas, and then baked dry in an 80 ◦Coven for 30 min.

2.2.2. Antibody solution preparationAntibody solution was prepared as follows. 2 �S/cm water

was prepared using a Millipore filtration system and the con-ductivity was adjusted with potassium chloride. Conductivitywas tested with an Orion 105A+ conductivity meter (ThermoElectron Corp.). In order to accurately control the final conduc-tivity of the antibody solution, which was originally preparedin PBS, dialysis was performed. A 20 kDa cutoff dialysis car-tridge (Thermo Scientific) was immersed in the 2 �S/cm waterfor 1 min in order to hydrate the membrane. Next, 300 �l of0.1 mg/ml FITC goat anti-rabbit IgG (Sigma–Aldrich) in 1× PBS

R. Hart et al. / Sensors and A

electrode and chamber, the frequency and voltage of the excitationsignal and viscosity and density of the fluid. Next, the fluid velocityfield is used by a pair of coupled time dependant simulations—(2) aconvection and diffusion simulation and (3) a surface reaction sim-ulation. These two simulations are coupled since mass is removedfrom the 2D space and deposited on the boundary where thereaction takes place. The rate of reaction will depend on the concen-tration near the boundary as well as the amount of material alreadybound. These simulations depend on the concentration and diffu-sion coefficient of the free analyte as well as certain properties ofthe chemical reaction (rate constants and density of surface bind-ing sites). The result of these three coupled simulations consists ofconcentration maps for the time span used as well as a record ofthe bound material over time. These simulations were all carriedout using COMSOL MULTIPHYSICSTM version 3.5a finite elementsoftware with Matlab support.

2.1.1. Fluid dynamics simulationThe 2D fluid dynamics simulation uses the Navier–Stokes equa-

tions for an incompressible Newtonian fluid. The temperature ofthe liquid is assumed to be constant because the expected jouleheating will be insignificant due to the low solution conductivity(2 �S/cm) and the voltages applied will be low [21]. Steady stateis achieved in an insignificant time (less than a second) comparedto the time scales of the binding reaction (minutes or hours) [22].Furthermore, the fluid behavior is the same along the length of theelectrode, except near the end of the electrode finger. The Reynoldsnumbers are well below 0.1, allowing for a laminar flow assumptionas well.

The two equations governing fluid flow are the mass conserva-tion equation:

∇ · U = 0 (4)

where U is the velocity vector, and the Navier–Stokes equation:

�U · ∇U = −∇P + �∇2U (5)

where � is the fluid density, P is the pressure, and � is the dynamicviscosity. Boundary conditions (Fig. 2) are as follows. A no slipboundary condition exists on the top chamber wall and the glassportion of the chamber bottom. Symmetry boundary conditionsare placed on the two side walls. The electrode generates the fluidmotion as a boundary condition according to the ACEO equationsdescribed in an earlier work [19].

2.1.2. Convection–diffusion simulationAt the start of the simulation, the analyte is evenly distributed

within the bulk solution and the surface receptors are completelyfree of bound material. As the time increases, the reaction at the sur-face removes material from the solution till equilibrium betweenthe two domains is reached. The movement of mass within thedefined geometry is governed by the convection–diffusion equa-tion:

∂c

∂t+ ∇ · (−D∇c + cu) = 0 (6)

where c is concentration, t is time, D is the diffusion coefficient andu is the fluid velocity vector. Boundary conditions (Fig. 2) consistof insulation on all boundaries except the bottom surface wheremass is removed from the subdomain according to the following:flux = kadsc(�0 − cs) − kdescs.

was injected into the cartridge and allowed to incubate in abeaker of 2 �S/cm water for 2 h. The conductivity of the waterin the beaker will increase slightly due to the dialysis and soit was replaced with fresh 2 �S/cm water half way through theincubation. The IgG was then removed from the cartridge and

tu

370 R. Hart et al. / Sensors and Ac

diluted to 0.01 mg/ml using the same 2 �S/cm water as dilu-ent.

2.2.3. Nonspecific binding assayThe nonspecific binding assay was performed by placing the

antibody solution in direct contact with the IDE. IDEs werecleaned immediately prior to each experiment. The antibodysolution was incubated for 15 min on an IDE while the elec-trodes were supplied with a signal from a function generator(HP 33120A). After incubation, the electrodes were washed inDI water and imaged using fluorescence microscopy (Nikoneclipse ME600 upright microscope). Primarily two geometry elec-trodes were used: (1) a 100 �m electrode with a 25 �m gap(edge-to-edge) and (2) a 20 �m electrode with a 20 �m gap.An AC signal (2 Vpp and 100 Hz) was applied to each electrode.Other electrodes designs were used and described in Section3.

Image analysis was performed using NIH ImageJ on the 20 �melectrode experiment by measuring the fluorescent intensity inseveral regions throughout each image. Background fluorescencewas removed by subtracting the intensity from an unreacted bareelectrode surface (a blank). Each measurement region was takenwith a ratio of glass and gold equal to that of the general deviceratio. For 20 �m IDEs, this ratio is 1:1.

2.2.4. Heterogeneous immunoassayThe heterogeneous immunoassay was performed by first func-

tionalizing the surface of the IDE and then reacting with afluorescent antibody under ACEO mixing conditions. The surfacewas functionalized by direct adsorption of 0.1 mg/ml whole rab-bit antibodies (Sigma–Aldrich) in PBS for 2 h, after which, thesubstrate was washed with DI water 3 times. Next, the surfacewas blocked with 1 mg/ml BSA in PBS overnight. Negative con-trols were prepared with BSA alone. To test the effectiveness ofACEO for improving detection, 0.01 mg/ml FITC-anti-rabbit anti-bodies (Sigma–Aldrich) were incubated on the surface for 15 minwhile a 100 Hz signal was supplied to the devices. Devices with the20 �m gap and those with a 25 �m gap were excited with a 2 Vpp

and 2.5 Vpp signal respectively. Fluorescent microscopy was per-formed to visualize distribution of IgG binding. Image analysis wasperformed as in Section 2.2.3.

3. Results and discussion

3.1. Simulation results

For a given parameter set (geometry, electrical, chemical, etc.)the simulation produces a series of time-dependent concentration

Fig. 4. Concentration maps of a 100 �m wide electrode and 30 �m wide gap operating aby the gray line under each map. Flow across the surface of this half of the electrode wi

ators B 147 (2010) 366–375

maps. An example of concentration maps produced by the simula-tion can be seen in Fig. 4.

3.1.1. Optimization of electrode geometry and excitationfrequency

Electrode geometry and excitation frequency will both affectfluid velocity and binding rates. To find the most efficient elec-trode geometry a range of electrode sizes and gap sizes were testedfor several chamber heights. For each chamber height, electrodewidth, a, and gap width, b, (Fig. 2 (top left)) were varied within1–200% of the channel height. An initial scan of these geometrieswas performed with 5% increments followed by a finer scan at1% increments near the optimum region. These simulations used100 kV/m maximum electric field (voltage was made proportion-ate to the gap size—for example, an electrode with a 20 �m gapwould receive a 2 Vpp signal). An improvement factor was calcu-lated by dividing the time to reach equilibrium in the unmixedcase (teq-unmixed) by the time for the mixed case (teq-mixed). Equilib-rium was considered to be achieved when the adsorption rate wasa factor of a thousand less than the initial rate. Fig. 5 (left) showsa surface map of the improvement factors for a fine pass using a100 �m chamber height. The dimensions that produced the lowesttime for each chamber height were selected and plotted in Fig. 5(right).

Simulations to determine the optimum frequency were per-formed as well by varying frequency (from 40 Hz to 1 MHz) onthe ideal electrode geometry. It was observed that the optimumfrequency was independent of the electrode size. The optimumelectrode width and gap distance is ∼70% and ∼30% of the chamberheight respectively. The reason for this optimum is likely a balancebetween two geometrically dependant properties. Decreasing boththe electrode and gap size too far contains the vortex within a nar-row geometry and consequently causes most of the flow velocity tobe constrained to a shallow depth. Increasing the electrode too fardecreases the efficiency as well. The highest fluid velocity occursat the electrode edge and decreases towards the electrode center.Electrodes that are too wide will therefore cause slower overallfluid velocities. A balance, therefore, of electrode size and gap sizewith the chamber height will cause the most efficient mixing.

Optimum frequencies are in the kHz or lower range, with thebest time near 100–200 Hz. (The binding time, as a function of fre-quency, is shown in Fig. 6 (left) for a 100 �m chamber height.) This

corresponds to the range of highest generated velocity. The low-est frequency simulated was 40 Hz, which is slightly below whatis safe to use in the lab due to hydrolysis and electrode damage.This prediction was corroborated by fluorescent assays and will bediscussed below.

t 2 Vpp. The scale shows the relative concentration c/c0. Electrode position is indicatedll be from right to left. The last concentration map is for an inactive electrode.

R. Hart et al. / Sensors and Actuators B 147 (2010) 366–375 371

Fig. 5. Geometry optimization. Left: contour plot showing the improvement factor as ageometry as a function of chamber height. For each chamber height, the ideal geomeadjusted for each geometry so that the maximum electric field was the same regardles

Table 1Binding constants for the simulation. George et al. [24] used immobilized saporinand free IgG as the analyte. Hu et al. [20] used generic immunoassay values. Huanget al. [13] used immobilized C reactive protein (CRP) and free IgG as the analyte.Vijayendran et al. [10] used immobilized anti –TNT antibodies with free trinitroben-zene (340 Da molecular weight).

Parameter George Hu Huang Vijayendran

kads (mol m3 s−1) 503 100 1E4 320kdes (s−1) 1.83E−4 1E−3 2.6E−2 1.5E−2

−1

3

fstsl(mtF

F1aa

c0 (mol m ) 5E−5 1E−4 6.4E−6 3.3E−8D (mol m2 s−1) 1E−11* 1E−11 2.18E−11 19E−10�0 (mol m2) 1.35E−7 1E−7 1.4E−8 4.3E−9

.1.2. Improvement for different chemistriesIn order to determine how much improvement can be gained

rom using ACEO mixing, four different reaction chemistries wereimulated, representing a range of reaction types. The reactionypes are as follows: (1) bound saporin with anti-saporin IgG inolution [24], (2) generic immunoassay values [20], (3) immobi-ized C reactive protein (CRP) and anti-CRP IgG in solution [13] and

4) immobilized anti-TNT antibodies with trinitrobenzene (340 Da

olecular weight) in solution [10]. Values for the relevant parame-ers were obtained from the literature and are displayed in Table 1.or each reaction, a range of chamber heights was used with appro-

ig. 6. Left: equilibrium binding time as a function of frequency for a 100 �m chamber00 kV/m. Highest velocities are near the 100 Hz range with the phenomena dying outssay enhancement for four different sets of reaction constants. Improvement factors armaximum electric field of 100 kV/m. For each chamber height, ideal electrode geomet

function of electrode and gap widths for a 100 �m chamber height. Right: optimumtry was determined by finding the shortest time to reach equilibrium. Voltage wass of the gap width.

priate scaling of electrode and gap size. An electric field of 100 kV/mwas used with a frequency of 100 Hz. Results of this comparison areshown in Fig. 6 (right).

As the chamber heights get larger, the surface reaction becomesmore transport-limited which makes mixing more beneficial. Themixing enhancement generally increases with chamber heightuntil about 100 �m, where the enhancement plateaus. This plateauis most likely a limitation of ACEO, which, because it is a surfaceeffect, can only extend into the fluid through viscous forces that dieout with distance. For the most transport rate-limited case (Georgeet al. [24]) equilibrium time is reduced by a factor of about 6. For theleast transport-limited case (Vijayendran et al. [10]) no improve-ment is gained from mixing at all. This is to be expected due to thehigh diffusion coefficient of the analyte.

3.2. Fluorescent assay results

3.2.1. Frequency and voltage of the excitation signalDirect adsorption of fluorescent antibodies was performed for a

range of frequencies and voltages. Two different designs of elec-

trodes were employed: (1) a 100 �m electrode with a 25 �medge-to-edge gap and (2) a 20 �m electrode with a 20 �m edge-to-edge gap. The results (Fig. 7 (first and second row)) show thatlower frequencies cause more fluorescent antibodies to attach. The

height using a 70 �m electrode width and a 30 �m gap width. The electric field wasin the kHz range. After 10 kHz, binding time is purely determined by diffusion. Right:e calculated by comparing the equilibrium times for mixed and unmixed samples withry (as shown in Fig. 3) was used. An improvement factor of 1 means no improvement.

372 R. Hart et al. / Sensors and Actuators B 147 (2010) 366–375

Fig. 7. First and second row: direct adsorption assay using 0.01 mg/ml FITC-IgG. The applied signal was 2 Vpp and varied in frequency (displayed above). The top row featureselectrodes that were 20 �m wide with a 20 �m edge-to-edge gap. The bottom image row shows electrodes that were 100 �m wide with a 25 �m edge-to-edge gap. Third row:Increasing voltage (displayed in Vpp) for direct adsorption of 0.01 mg/ml IgG. Solution conductivity was 2 �S/cm and frequency was 100 Hz. Note the increasing fluorescence

)a

till 5 Vpp where electrode damage occurred. Bottom: relative fluorescence unit (RFUusing some images not shown above. For both charts, fluorescent intensity was meshown in separate plots.

image analysis is shown in the left chart in Fig. 7 and shows that atthe lowest frequencies, the fluorescent intensity is at least 3 timesthat of the higher frequencies. The frequency dependence shown bythe images and chart agrees with the simulation results as shownin Fig. 6 (left). Empirical results show that the 20 �m device out-performs the 100 �m device and is contrary to the prediction madeby the simulation which shows the 100 �m device to be superior.

It is hypothesized that this is due to DEP, which is indicated by theincreased fluorescent intensity near the edges of the electrode andat the ends of the fingers. The possibility of the presence of DEP willbe explained in more detail below.

for the respective images. The frequency chart (left) was taken from the 20 �m andsured both near the electrode finger ends and towards the center of the device, each

Using the best frequency value (100 Hz), a range of voltageswas tested for the 100 �m electrode design. This design was useddespite the better performance of the 20 �m design in order toreduce the suspected effects of DEP. The resulting images are shownin Fig. 7 (third row) and a chart showing fluorescent intensity as afunction of voltage is shown below. The measurement area for theimage analysis contained a ratio of gold to glass of 100:25 in order to

match the general device ratio. As expected, the intensity increaseswith voltage, achieving an enhancement of 4 times at 3.5 Vpp (5near the electrode end) until about 5 Vpp where electrode damagewas apparent. It should be noted that due to photobleaching and

R. Hart et al. / Sensors and Actuators B 147 (2010) 366–375 373

Fig. 8. Binding of fluorescent secondary antibodies to IDEs. Images are on top and their respective fluorescent intensity is shown in the chart below. Numbers on top of eachcolumn indicates the enhancement factor which is the ratio of sample intensity compared to the unmixed functionalized surface (C). The blocked surface has been preparedwith BSA only. The functionalized surface has been prepared first with primary antibody and then blocked with BSA. The mixed samples have been supplied with a 100 kV/m,100 Hz electric field. Non-mixed samples received no electrical signal. The lower row contains images of the fully functionalized surface with mixing on different geometryelectrodes, as well as an enlarged image of the 20 �m parallel electrode. These IDEs were supplied with the same signal as those of the top row. The 10 �m castellated device

ly acti

c2

wffctaapetometo the presence of the corners which cause there to be high field

was serendipitously broken in many places during fabrication, allowing only sparse20 �m parallel electrode has been enlarged 5 times compared to the other images.

different experimental conditions, relative fluorescence unit (RFU)values from separate assays cannot be directly compared.

3.2.2. ACEO mixing on a functionalized surfaceIn order to perform selective detection of an analyte, a series

of assays was performed with a fully functionalized surface. IDEsurfaces were prepared as described in Section 2.2.4 and a bindingassay was performed using a 100 kV/m maximum voltage at 100 Hz(1 Vpp for 10 �m gap, 2 V for 20 �m gap, etc.). The results for 20 �mparallel electrodes are shown in Fig. 8 (top) with a chart comparingfluorescent intensity of each device and reaction type (bottom).Negative controls consisted of (1) a completely blocked surfaceunder non-mixing conditions, (2) a blocked surface under mixedconditions, and (3) a functionalized surface that was unmixed. (1)Shows the effectiveness of the blocking agent to prevent nonspe-cific binding, (2) shows the same effectiveness under the presence

of ACEO, and (3) shows a full immunoassay performed on the IDEwithout mixing and serves as a baseline against which the othersamples may be compared. Two different castellated electrodedesigns and the 100 �m parallel electrode pattern were tested aswell and shown in the bottom row of the same figure. The negative

stfifi

vated electrodes. Such an activated electrode is shown in the dotted line. Note the

ontrols for these devices (not shown) are similar to those of the0 �m devices.

Importantly, there was little observable nonspecific bindinghen active mixing was applied to a completely blocked sur-

ace. When comparing activated and inactivated electrodes on fullyunctionalized surfaces, a large enhancement of fluorescent signalan be seen around the electrode edges. Enhancement of signal (inhe center of the IDEs) is shown above each column of the figurend was about 1.4–1.9 times improved compared to an unmixedssay. The 10 �m castellated device shows an area of the electrodeattern where only a single active finger is surrounded by brokenlectrodes on either side. There is manifest enhancement aroundhis single finger compared to its neighbors. The best enhancementccurs near the ends of the electrode fingers reaching an enhance-ent factor of about 6.7 for the straight 20 �m electrode. This

nhancement at the electrode ends may not necessarily be due

trength. Castellated electrodes have many corners and still showhis “finger end-effect.” A likely reason is that at the ends of eachnger, the electrode has “access” to antibodies past the end of thenger and is thereby made much more efficient. For these reasons,

374 R. Hart et al. / Sensors and Actu

Fwt

tphbmne

3

dtbleattoasHtomepvb1sDnhtegc[d

ig. 9. Comparison of 20 �m castellated electrode after a direct adsorption assayith a 3D electric field simulation. The simulation side shows ∇E2, which is propor-

ional to DEP force.

he concentration efficiency at the electrode end will likely sur-ass the predictions made by the 2D simulation. This observationas important design ramifications for future devices, especially foriosensors that may incorporate AC electrokinetic assay enhance-ent. Sensors performing bulk detection across a large surface will

ot benefit from this end-effect but smaller transducers could ben-fit from the local 3D concentrating effects.

.2.3. Consideration of DEP presenceThe presence of DEP is suspected for two reasons. First, the high

egree of concentration of antibodies at the electrode edge is aypical signature of positive DEP. This is present in both the ACEO-inding of primary antibody to a bare surface (Fig. 7) as well as the

abeling of uniformly distributed primary antibody by an ACEO-nhanced secondary antibody reaction (Fig. 8). As an illustration,3D electric field simulation of a castellated electrode shows that

he ∇E2, which is proportional to the DEP force, closely resembleshe pattern produced by the fluorescent antibodies (Fig. 9). Sec-nd, the simulation described in Section 2.1, which considers ACEOlone, shows insignificant variation of antibody binding across theurface, even if the surface diffusion term is reduced or removed.owever, two arguments would suggest that ACEO plays an impor-

ant role. First, DEP is not expected to operate deterministicallyn molecules but instead is typically only useful for manipulatinguch larger particles. A dimensional analysis performed by Lifeng

t al. [15] indicates that molecular collection by DEP should not beossible using this electrode geometry without significantly higheroltages (at least two orders of magnitude). (A special case cane made for DNA however [25].) Second, that frequencies around00 Hz show improved binding, and frequencies greater than 1 kHzhow no improvement is a strong indication of ACEO rather thanEP. At the solution conductivities used, ACEO velocity is greatestear 100 Hz and quickly dies out [19,26]. The DEP force, on the otherand, is expected to be constant in this frequency range. [15] It isherefore suggested that the enhanced antibody collection at the

dges is due to a combination of ACEO and DEP. This suggestion isiven weight by early observations by Washizu et al. [27] recentollection of quantum Dots under a combination of ACEO and DEP28]. However, the quantum dots used were at least 2 times theiameter of an antibody which means the DEP force would be at

ators B 147 (2010) 366–375

least 8 times stronger. Finally, DEP will have a greater impact onthe smaller gap sizes of the 20 �m electrode compared to the largergap size of the 100 �m device. This explains the better collectionof antibodies by the 20 �m device, despite the non-optimal tuningfor ACEO as predicted by the simulation.

4. Conclusions

This work demonstrates the feasibility of using ACEO to improvethe detection time and sensitivity of immunoassays. This improve-ment appears to be aided by DEP as well, despite the small sizeof the molecules being used and the relatively large feature sizeof the devices. Furthermore, the greatest enhancement occurs nearthe ends of the electrode fingers due to an “end-effect.” For the bestadvective mixing, which can shorten the time to reach equilibriumup to a factor of 6, the feature size of the interdigitated elec-trode should scale with the channel height—the electrode widthand gap sizes should be about 70% and 30% of the channel heightrespectively. The height of the chamber also dramatically affectsthe binding enhancement. For the voltage and electrode designssimulated, a 100 �m chamber height or above provided the bestmixing. Operational parameters were also tested with fluores-cent assays. Lower frequencies and higher field strengths exhibitedbetter performance but can also lead to electrode damage and elec-trolysis of water. Safe ranges include frequencies above 100 Hzand field strengths less than 100 kV/m. Finally, complete hetero-geneous immunoassay with a functionalized surface was shown tobe enhanced by the use of combined ACEO and DEP with the bestimprovement occurring at the end of each electrode finger wherethree-dimensional collection of molecules could take place. For thebest design-type, fluorescent signal for these immunoassays wasenhanced by a factor of 1.9 in the center of the device and by 6.7 atthe ends of the device.

Acknowledgements

We would like to thank Jonghyun Oh and Doh-Hyoung Leefor their help with the COMSOL simulations and Erten Ergezenand Johann DeSa for sharing their electrical connection expertise.We are also grateful to Dr. Edward Basgall and Drexel’s Central-ized Research Facility for providing microfabrication expertise andequipment as well as Dr. Kenneth Barbee for his help with fluores-cent intensity measurements.

References

[1] T. Beumer, P. Haarbosch, W. Carpay, Convection during incubation ofmicroplate solid phase immunoassay: effects on assay response and variation,Analytical Chemistry 68 (8) (1996) 1375–1380.

[2] A. Bange, H.B. Halsall, W.R. Heineman, Microfluidic immunosensor systems,Biosensors and Bioelectronics 20 (12) (2005) 2488–2503.

[3] M. Zimmermann, E. Delamarche, M. Wolf, P. Hunziker, Modeling andoptimization of high-sensitivity, low-volume microfluidic-based surfaceimmunoassays, Biomedical Microdevices 7 (2) (2005) 99–110.

[4] V.M. Owen, BIosensors—the needs of the healthcare market, Sensor Review 13(4) (1993) 3.

[5] J. Wang, Microchip devices for detecting terrorist weapons, Analytica ChimicaActa 507 (1) (2004) 3–10.

[6] Z. Ali, Lab-on-a-chip for terrorist weapons management, Measurement andControl 38 (3) (2005) 87–91.

[7] R. Bogue, MEMS sensors: past, present and future, Sensor Review 27 (1) (2007)7.

[8] P.W. Wojciechowski, J.L. Brash, Computer simulation for the study of macro-molecular adsorption with special applications to single-component proteinadsorption, Journal of Colloid and Interface Science 140 (1) (1990) 239–252.

[9] A. Lionello, J. Josserand, H. Jensen, H.H. Girault, Protein adsorption in staticmicrosystems: effect of the surface to volume ratio, Lab on a Chip 5 (3) (2005)254–260.

[10] R.A. Vijayendran, F.S. Ligler, D.E. Leckband, A computational reaction-diffusionmodel for the analysis of transport-limited kinetics, Analytical Chemistry 71(23) (1999) 5405–5412.

tu

proteins in microchannels. Currently, Dr. Noh holds six research grants/awards

R. Hart et al. / Sensors and Ac

[11] M. Sigurdson, D. Wamg, C.D. Meinhart, Electrothermal stirring for heteroge-neous immunoassays, Lab on aChip 5 (2005) 1366–1373.

[12] A. Ramos, H. Morgan, N.G. Green, A. Castellanos, Ac electrokinetics: a reviewof forces in microelectrode structures, Journal of Physics D: Applied Physics 31(1998) 2338–2353.

[13] K.-R. Huang, J.-S. Chang, S.D. Chao, K.-C. Wu, C.-K. Yang, C.-Y. Lai, S.-H.Chen, Simulation on binding efficiency of immunoassay for a biosensor withapplying electrothermal effect, Journal of Applied Physics 104 (6) (2008)064702–064711.

[14] A. Castellanos, A. Ramos, A. Gonzalez, N.G.G. Green, H. Morgan, Electrohydrody-namics and dielectrophoresis in microsystems: scaling laws, Journal of PhysicsD: Applied Physics 36 (2003) 2584–2597.

[15] Z. Lifeng, L. Shengdong, P.J. Burke, J.P. Brody, Towards Single Molecule Manipu-lation with Dielectrophoresis Using Nanoelectrodes, IEEE, Piscataway, NJ, USA,2003.

[16] K.F. Hoettges, M.P. Hughes, A. Cotton, N.A.E. Hopkins, M.B. McDonnell, Optimiz-ing particle collection for enhanced surface-based biosensors, IEEE Engineeringin Medicine and Biology Magazine 22 (6) (2003) 68–74.

[17] J. Wu, N. Islam, M. Lian, High sensitivity particle detection by biased AC elec-troosmotic trapping on cantilever, Istanbul, Turkey, Institute of Electrical andElectronics Engineers Inc. (2006).

[18] H. Morgan, N. Green, AC Electrokinetics: Colloids and Nanoparticles, SRP Ltd.,England, 2002.

[19] J. Oh, R. Hart, J. Capurro, H.M. Noh, Comprehensive analysis of particle motionunder non-uniform AC electric fields in a microchannel, Lab on a Chip 9 (2008)62–78.

[20] G. Hu, Y. Gao, D. Li, Modeling micropatterned antigen-antibody binding kineticsin a microfluidic chip, Biosensors and Bioelectronics 22 (7) (2007) 1403–1409.

[21] N.G. Green, A. Rarnos, A. Gonzalez, A. Castellanos, H. Morgan, Electrothermallyinduced fluid flow on microelectrodes, Journal of Electrostatics 53 (2) (2001)71–87.

[22] R. Hart, J. Oh, J. Capurro, H.M. Noh, “AC Electrokinetic Phenomena Gen-erated by Microelectrode Structures”. JoVe 17, 2008, http://www.jove.com/index/details.stp?id=813.

[23] COMSOL MULTIPHYSICS User’s Guide. (2008).[24] A.J. George, R.R. French, M.J. Glennie, Measurement of kinetic binding constants

of a panel of anti-saporin antibodies using a resonant mirror biosensor, Journalof Immunological Methods 183 (1) (1995) 51–63.

[25] M. Ueda, K. Yoshikawa, M. Doi, Stretching of long DNA under alternating cur-

rent electric fields in a concentrated polymer solution, Polymer Journal 29 (12)(1997) 1040–1043.

[26] N.G. Green, A. Ramos, A. Gonzalez, H. Morgan, A. Castellanos, Fluid flow inducedby nonuniform AC electric fields in electrolytes in microelectrodes. I. Experi-mental measurements, Physical Review E 61 (2000) 4011–4018.

ators B 147 (2010) 366–375 375

[27] M. Washizu, S. Suzuki, O. Kurosawa, T. Nishizaka, T. Shinohara, MolecularDielectrophoresis of Bio-polymers, IEEE, New York, NY, USA, 1992.

[28] M.L.Y. Sin, V. Gau, J.C. Liao, D.A. Haake, P.K. Wong, Active manipulation of quan-tum dots using ac electrokinetics, Journal of Physical Chemistry C 113 (16)(2009) 6561–6565.

Biographies

Robert Hart received his BS degree in Biology with a focus in Computer Science fromCarnegie Mellon University in 2000. He is currently a PhD student in BiomedicalEngineering at Drexel University and expects to graduate in February 2010. He hasindustry experience in Genomics and Automation and his current research interestsinclude microfluidics, electrokinetics and biosensors.

Ryszard Lec has been active in the areas of material science and instrumentationfor more than 20 years. His special interest has been in application of acoustic,ultrasonic, microwave and optical technologies for the design and fabrication ofa variety of sensors for biomedical, chemical, and environmental measurands. Inparticular, in the biomedical area his interest has been focused on immunosen-sors, DNA probes, the detection of bacteria, and the influence of the acousticenergy on the proliferation of human aortic muscle cells. Dr. Lec has received overforty research and science education grants from the National Science Founda-tion, National Institute of Health, Department of Energy, Department of Defense,Du Pont, IBM, BIODE Inc., and other government and private institutions. He haspresented many scientific papers both in the US and abroad and has authoredand co-authored over 50 publications and one book and has been awarded fivepatents.

Hongseok “Moses” Noh received his PhD degree in Mechanical Engineering fromthe Georgia Institute of Technology with the major field in MEMS and the minorin Bioengineering. After joining Drexel University as faculty in 2004 he has devel-oped multi-disciplinary research and teaching programs that focus on biomedicalapplications of MEMS and microfludics technologies in close collaboration withbiologists and medical doctors. His current research projects include microfluidicplatform technology for cell biology and drug screening, implantable microdevicefor the treatment of hydrocephalus, and electrokinetic manipulation of cells and

sponsored by NSF, NIH, and State of Pennsylvania. He has also received the NationalScience Foundation Faculty Early Career Development Award (NSF CAREER) in2008.