Hydrodynamic Flow Profiling in Microchannel Structures by Single

Anal. Chem. 2000, 72, 3260-3265

Hydrodynamic Flow Profiling in Microchannel Structures by Single Molecule Fluorescence Correlation Spectroscopy Michael Go 1 sch, Hans Blom,† Johan Holm,‡,§ Toni Heino,‡ and Rudolf Rigler*

Department of Medical Biophysics, Karolinska Institute, S-171 77 Stockholm, Sweden

In this paper we demonstrate high spatial resolution hydrodynamic flow profiling in silicon wafer based microchannels using single molecule fluorescence correlation spectroscopy (FCS). We have used confocal fluorescence microscopy to detect single tetramethylrhodamine (TMR-4-dUTP) biomolecules traversing a ∼l fL volume element defined by an argon laser beam focus. By elevating a (∼10-10 M) reservoir of diluted analyte, a continuous hydrodynamic flow through the microstructure could be accomplished. The microchannel was then scanned with a diffraction-limited focus in ∼1-µm steps in both the vertical and the horizontal directions to determine the flow profile across a 50 × 50 µm2 channel. The flow profile measured was parabolic in both dimensions, thereby showing a Poiseuille laminar flow profile. Future microstructures can hereby be nondestructively investigated with the use of high spatial resolution confocal correlation microscopy. The first entire micromachined fluidic system was demonstrated two decades ago.1 During the past decade, micromachined fluidic systems have found increased application in such areas as chemical analysis, biological and chemical sensing, molecular separation, and environmental monitoring. Potential benefits of introducing micro systems include improved accuracy, lower power and sample consumption, disposability, and lower cost. However, none of these benefits is guaranteed simply by using micromachined structures, because the effect of scaling down a fluidic system can introduce other problems. One such problem is the design of a proper macro-to-micro interface in order to ensure proper sample delivery without leakage. Another problem is the probing of functions in these microstructures. In particular, probing of microfluidics, i.e., flow velocities and flow profiles, with high spatial resolution has not been solved yet. Currently, different sized beads are introduced in microstructures to probe the microfluidics.2,3 These beads can clog the microsystem, making further use of the system impossible. Their size (∼1 µm) also * Corresponding author. Tel.: +46-8-728 6801. Fax: +46-8-326505. E-mail: [email protected]. † Department of Electronics, KTH Electrum 229, SE-16440 Kista, Sweden. ‡ ACREO AB, Electrum 236, SE-164 40 Kista, Sweden. § Present address: IBSEN Microstructures A/S, Gammelgaardsvej 65, DK3520 Farum, Denmark. (1) Terry, S. C.; Jerman, J. H.; Angell, J. B. IEEE Trans. Electron Devices 1979, ED-26, 1880-1886. (2) Chen, Z.; Milner T. E.; Dave, D.; Nelson, J. S. Opt. Lett. 1997, 22, 64-66.

3260 Analytical Chemistry, Vol. 72, No. 14, July 15, 2000

contributes to a poor resolution of the measured flow functions. Probing and handling of microfluidics in microstructures is a key area of the technological challenge of transferring micromachined structures from science to, for example, user-friendly medical diagnostic tools.4 To this end one needs to be able to monitor the flow in the microsystem to be able to control it. Measurements of microfluidics in microstructures have previously been performed using a variety of techniques: Laser Induced Fluorescence (LIF) for flow visualization of markers,5 weight hydraulic measurements,6 optical Doppler tomography,7 and NMR imaging.8 A problem with the first and the third methods is the introduction of small beads in the microstructure, which can disturb the flow properties and clog the microchannel. The second method uses no beads, but depends on experimentally derived formulas with no possible spatial resolution of the flow in a microsystem. In the last method a resolution of only 10 µm is obtained. Fluorescence correlation spectroscopy (FCS), however, is a method with high spatial resolution and very short measurement times (approximately milliseconds). No beads9 are needed in FCS. Instead, the flow properties can be measured by simply adding a highly diluted concentration of fluorescent molecules into the microchannel. Since the early 1970s FCS has progressively developed as a method for investigation of chemical kinetics,10-12 diffusion dynamics,10,12 rotational motion,13 excited-state kinetics,13,14 and laminar flow.15 Through the use of a confocal epi-illuminated microscope, the detection and characterization of single fluorescent molecules in solution at room temperature also (3) Santiago, J. G.; Wereley, S. T.; Meinhart, C. D.; Beebe, D. J.; Adrian, R. J. Exp. Fluids 1998, 25, 316-319. (4) Zhou, Z. Y.; Ye, X. Y.; Li, Y.; You, Z.; Cui, T. H.; Yang, Y.; Jiang, X. N.; Hu, M.; Xiong, S. S.; Wang, W. D.; Cai, H. N.; Zou, J.; Zhang, L. J. Inst. Eng. 1998, 38, 7-14. (5) Boer, G.; Dodge, A.; Fluri, K.; van der Schoot, B. H.; Verpoorte, E.; de Rooij, N. F. Micro Total Analysis Systems; Harrison, J., Vanden Berg, A., Eds.; Kluwer Academic Publisher: Dordrecht, 1998. (6) Li, H.; Gale, R. J. Langmuir 1993, 9, 1150-1155. (7) Wang, W.; Lui, Y.; Sonek, G. J.; Berns, M. W.; Keller, R. A. Appl. Phys. Lett. 1995, 67, 1057-1059. (8) Manz, B.; Stilbs, P.; Jo¨nsson, B.; Soderman, O.; Callaghan, P. T. J. Phys. Chem. 1995, 99, 11297-11301. (9) Brinkmeier, M.; Do ¨rre, K.; Stephan, J.; Eigen, M. Anal. Chem. 1999, 71, 609-616. (10) Elson, E. L.; Magde, D. Biopolymers 1974, 13, 1-27. (11) Widengren, J.; Rigler, R. J. Fluoresc. 1996, 7, 211-213. (12) Magde, D.; Elson, E. L.; Webb, W. W. Biopolymers 1974, 13, 29-61. (13) Ehrenberg, M.; Rigler, R. Chem. Phys. 1974, 4, 390-401. (14) Ehrenberg, M.; Rigler, R. Q. Rev. Biophys 1976, 9, 69-81. (15) Magde, D.; Webb, W. W.; Elson, E. L. Biopolymers 1978, 17, 361-376. 10.1021/ac991448p CCC: $19.00

© 2000 American Chemical Society Published on Web 06/10/2000

became possible.16-18 The use of single molecules avoids the problem of flow disturbances and clogging, even in channels as narrow as 5 µm. In FCS, the fluorescent light emitted from these individual molecules is measured during their passage through a very small, laser-induced, diffraction-limited volume element. The emitted light is further analyzed by the intensity autocorrelation function, providing information about chemical rate coefficients, diffusion coefficients, and flow velocities. The high spatial resolution of a flow profile, which has not previously been achieved, is an additional benefit of the FCS method. By scanning the small volume element, defined by the diffraction-limited, high numerical aperture focus of the microscope lens across the microchannel, flow velocities in a microsystem can be measured. The small dimension of the illuminated volume element gives the method its desired high spatial resolution and also ensures that the flow is translationally uniform in the volume probe.15 The uniformity of the translation used to map out the flow profile in a microstructure is the key quality in this FCS method. The determination and control of flow velocities is, as mentioned before,4 of crucial importance for applications such as single-molecule DNA sequencing.19 All types of generated flows, including electroosmotic or hydrodynamic, can be determined and controlled with the use of this FCS-scanning principle. It is the purpose of this paper to show that we can measure flow velocities up to 50 mm/s in a femtoliter volume element with FCS and also determine flow profiles in small microstructures using fluorescent molecules. In addition, we show that FCS is sensitive enough to see single dye molecules in a microstructure at flow velocities as high as 25 mm/s. THEORY Fluorescence Correlation Spectroscopy. The idea of this analysis is based on observing the fluctuations in intensity of individual molecules in a small open volume element defined by a laser beam with Gaussian intensity profile. The collected fluorescence intensity emitted from a fluorophore molecule is given by

r b,t)dV r ∫ I(b)C(

i(t) ) qQ

V

(1)

Here, q stands for the detection quantum efficiency of the detectors as well as geometrical and optical filtering losses inherent in the experimental arrangement. Q is the fluorescence quantum yield, and C(rb,t) is the concentration of fluorescent molecules at position br and time t. I(rb) is the combination of collection efficiency function and the excitation intensity and is approximated to be Gaussian distributed (16) Rigler, R.; Widengren, J. BioScience 1990, 3, 180-183. (17) Rigler, R.; Widengren, J.; Mets, U ¨ . Fluorescence Spectroscopy; Wolfbeis, O. S., Ed.; Springer-Verlag: Berlin, Germany, 1993; Chapter 2. (18) Nie, S.; Chiu, D. T.; Zare, R. N. Science (Washington, D.C.) 1994, 266, 1018-1021. (19) Do¨rre, K.; Brakmann, S.; Brinkmeier, M.; Kyung-Tae, H.; Riebeseel, K.; Schwille, P.; Stephan, J.; Wetzel, T.; Labczyna, M.; Stuke, M.; Bader, R.; Hinz, M.; Seliger, H.; Holm, J.; Eigen, M.; Rigler, R. Bioimaging 1997, 5, 139-152.

I(b) r ) I0 e-2(x +y )/(w0 )e-2(z )/(z0 ) 2

2

2

2

2

(2)

where ω0 and z0 are the radius and the height of the volume element at which I(rb) has decreased by a factor of e-2. The number of fluorescent molecules in the volume element at a given time is determined by the intensity fluctuation of the fluorescent signal. This fluctuating signal is dependent on concentration variation in the volume element. The concentration variation with time, t, about equilibrium mean concentration, C h, can be expressed as δ C(rb,t) ) C(rb,t) - C h . Then, the concentration correlation function can be defined as

g(b, r b′,τ) r ) 〈δC(b,t)δC( r b,t r + τ)〉

(3)

assuming a stationary random process (ergodicity). The number of molecules traversing the small open volume element can change by diffusion10 or by applying a flow.15 Combining flow with diffusion gives the hydrodynamic equation

dδC(b,t) r r -B V(b)‚∇δC( r b,t) r ) D∇2δC(b,t) dt

(4)

Here, ∇ ) d2/d2x, d2/d2y, d2/d2z, D is the diffusion coefficient of the examined molecules, and B V(rb) is the vectorial flow velocity. In our case, the vectorial flow is considered as uniform translation flow in the channel and can be expressed by

B V(b) r ) Vy(x,z)

(5)

where Vy is in the direction of the flow and is perpendicular to the height, z0, of the volume element. The spatial size of the volume element is sufficiently small to assume the velocity to be uniform in its x-y plane and constant in time; however, the velocity may vary along zˆ. The substitution of eqs 4 and 5 in eq 3 yields the concentration correlation function:15

g(b, r b′,τ) r )

{

C h

exp -

8πDz0

4Dτ

[ ( )] nπ

∑exp - D 2z

2

n

}{ }

(x - x′)2 + (y - y′ - Vτ)2

0

nπz

2

τ cos

2z0

cos

1+

nπz′ 2z0

(6)

Here, 2z0 is the entire length of the illuminated region imaged on the fluorescence detector. The normalized autocorrelation function is calculated from the intensity fluctuation, δi ) i - jı, with the fluorescence intensity of eq 1:

G(τ) )

〈δi(0)δi(τ)〉 〈i〉2

+1) r b′)g( r b, r b′,τ) r dV dV′ ∫ ∫ I(b)I( +1 (qQ∫ ∫ I(b)C r h dV dV′)

q2Q2

V V′

2

(7)

V V′

Analytical Chemistry, Vol. 72, No. 14, July 15, 2000

3261

Inserting eq 6 in eq 7 and executing the integration gives

G(τ) )

1 A exp N

{( ) } 2

τ

τflow

A +1

where

(

A) 1+

τ

)( () ) -1

1+

τdiff

w0 z0

2

τ

-1/2

τdiff

(9)

τdiff is the average diffusion time and τflow the average flow time of the fluorescent molecules through the open volume element. From this autocorrelation function, the flow time, τflow, is deduced by analysis of the experimental data. The flow velocity is further calculated by13

V)

w0 τflow

(9) Figure 1. Layout of the microchip (33 × 7 mm) and SEM picture of the etched microchannel.

Hydrodynamic Flow. The simple hydrodynamic law, ∆p ) Fg∆h, determines the relation between the pressure difference and the height of a liquid column. An applied pressure difference, ∆p, over a liquid-filled microchannel results in a stationary flow with velocity V. The velocity of the liquid is fastest in the center of the channel and zero at the lateral faces. Between the center and the lateral faces a velocity gradient dV/dx is established due to friction forces. This gradient results in a parabolic velocity profile inside the microchannel, given by the Hagen-Poiseuille equation20

V(x) )

∆p 2 (d - x2) 2ηl

(10)

where η denotes the viscosity, l is the length, and d the half width of the microstructure. To ensure laminar flow, the Reynolds number, R ) VlF/η, has to be kept below R ≈ 2000. For squared microchannels with a cross section of 50 × 50 µm2, as used in our experiments, this sets the maximum velocity in the microstructure to approximately V ) 70 mm/s. EXPERIMENTAL SECTION Microstructures. All microstructures (see Figure 1) used in the flow profiling experiments were fabricated by ACREO AB, Kista, Sweden21 using standard micromachining technologies, which include thin film deposition, photolithography, etching, and wafer bonding. In the first step, the microchannels were patterned onto an oxidized silicon wafer using photo resist and optical lithography. The patterned oxide mask was subsequently dryetched with reactive ion etching (RIE). Deep reactive ion etching (DRIE) was thereafter used to etch 50-µm-deep microchannels into the silicon wafer. With this anisotropic etching method it is possible to obtain vertical side walls and an essentially planar bottom surface independent of crystallographic orientation. As a second step in the microfabrication, a 500-µm-thick HOYA SD-2 (20) Streeter, V. L.; Wylier, E. B.; Bedford, K. W. Fluid Mechanics, 9th ed., McGraw-Hill Int.: New York, 1998; Chapter 6. (21) ACREO AB is the product of the merge of IMC (Industrial Microelectronics Center) and IOF (Swedish Institute of Optical Research).

3262 Analytical Chemistry, Vol. 72, No. 14, July 15, 2000

glass wafer had thin layers of chromium and gold deposited onto it by evaporation. These chromium-gold layers were thereafter patterned, using photolithography and wet etching. The exposed glass was then isotropically wet etched in diluted hydrofluoric acid to define the connections to the microchannels. The glass wafer was then cleaned and aligned before anodic bonding22 to the silicon wafer. After this bonding procedure, the glass lid was thinned down to a thickness of 170 µm. Finally, the remaining thin glass membrane was mechanically punctured at the connections to give access to the microchannels. Apparatus. The experimental hydrodynamic flow setup is displayed in Figure 2. A multiline (457-, 488-, and 514.5-nm, 7 mW) output of an Ar+ laser (Lasos, model LGK 7812 ML/1) was expanded six times by a beam expander (f ) 600 mm and f ) 100 mm plano-convex glass lenses). The 514.5-nm wavelength was selected by an excitation filter (Omega Optical 515N-B3) and prefocused by an f ) 150 mm lens in front of a confocal epi-illuminated microscope. The light was reflected by a dichroic mirror (Omega Optical 565DRLP02) into a 63× NA 1.2 water immersion objective, corrected to the thickness of a 170-µm cover slip (Zeiss Plan-Neofluar). The intensity under the objective was measured to be 2 mW by a silicon photodiode (Graseby Optronics 371, silicon head model 262). Fluorescence emission from the detection volume element was collected by the same objective and passed through a band-pass filter (Omega Optical 565DF50) that discriminate against Rayleigh and Raman scattered light. Finally, the emission was passed through a circular selfmanufactured 30-µm pinhole to reject out-of-focus light and subsequently focused by a lens onto a single photon avalanche photodiode (EG&G model SPCM-100). The photo-induced TTL pulses from the detector were passed to a PC-based correlator card (ALV-5000E) which calculates the autocorrelation function. Procedure. In our experiments the microstructure was mounted in a self-manufactured Plexiglass holder. The microstructure was laid into a recess in a thin Plexiglass plate and covered with another Plexiglass plate with polyethylene tubing (22) Wallis, G.; Pomerantz, D. J. Appl. Phys. 1969, 40, 3946.

Figure 2. Schematic diagram of the hydrodynamic flow setup.

connections. The complex was pressed together with M4 screws with O-rings inserted between the microstructure and the Plexiglass cover to get the system watertight. The microstructure was filled with highly diluted (10-10 M) TMR-4-dUTP (FluoroRed Amersham Pharmacia Biotech, Piscataway, NJ) in HPLC grade water. The sample reservoir was held in a 10-mL plastic syringe and connected to the inlet of the microstructure via a polyethylene tube (L 0.58 mm, Intramedic, Clay Adams). Different flow velocities in the microstructure were obtained by changing the height of this reservoir. We scanned the microchannel with the laser focus and recorded the flow velocities for every position. Reversed flow could also be achieved when the syringe reservoir was held below the microstructure. This permitted us to perform repeated measurements without any need for reloading the microchannel after each scan. RESULTS AND DISCUSSION Calibrations. Free hanging droplet experiments of Rhodamine6G molecules have previously been performed23 in order to determine the lateral size of the volume element, i.e., the laser beam diameter, 2w0. With the diffusion constant of Rhodamine6G as a reference value, we have determined the lateral focus to be 0.4 µm in diameter. This small laser focus was later used to probe the flow profile inside the microchannel. In a hangingdroplet experiment, performed with a 0.3 nM TMR-4-dUTP solution, the time for free diffusion through this small laser focus was measured. By this, the autocorrelation function of the observed fluctuation was deduced to τdiff ) 0.046 ms. This diffusion time was fixed in the flow model (eq 9) to determine the different flow velocities in the flow-profiling experiments. To establish the dependence of flow velocity on the reservoir height above the (23) Rigler, R.; Metz, U ¨ .; Widengren, J.; Kask, P. Eur. Biophys J. 1993, 22, 169175.

Figure 3. Autocorrelation curves for different flow velocities in the 50-µm microchannel.

inlet of the microstructure, a calibration curve as shown in Figure 4 was determined. In this calibration the laser focus was centered in the middle of the 50-µm microchannel, where the signal-tonoise ratio reaches an optimal value, because of the minimal scattering interference from the microchannel surfaces. The measured autocorrelation data (see Figure 3) for the flow times, τflow, were fitted with eq 8 and thereafter converted to flow velocities, V, with the use of eq 9. The time to acquire sufficient datapoints for an autocorrelation curve was less than 3 s. Data collection time could even be reduced to a fraction of a second for velocities above 10 mm/s. Flow velocities >0.5 mm/s were measured with an error rate of around 10%, due to uncertainties in the diffusion constant, D, of TMR-4-dUTP. As can be seen from Figure 4, the dependence of flow velocities on different reservoir heights is linear as expected. This velocity dependence satisfies the Hagen-Poiseuille law, which states that Analytical Chemistry, Vol. 72, No. 14, July 15, 2000

3263

Figure 4. Calibration curve for the flow velocity as a function of height of the reservoir.

the pressure difference (which is proportional to our height difference ∆h) is proportional to the flow velocity, V20 (eq 10). Flow Properties. High flow velocities, up to 25 mm/s, are easily obtained by slightly elevating the reservoir (see Figure 4). Brinkmeier et al.9 showed that strong electric fields with voltages up to 4000 V can induce high flow velocities in microstructures as well. However, the application of high voltage requires specific precautions as well as the implementation of wires into the microstructures. In addition, the generation of high flow velocities in similar microstructures (50 × 50 µm2 × 3000 µm) with electroosmotic pumping is difficult to obtain. For reverse flow, the reservoir has to be moved physically. This step is time-consuming and leads to irregularities in the laminar flow. In this case, electroosmotic flow would be of advantage, because the flow direction can be changed almost instantaneously.24 A flow velocity of several millimeters per second is desirable for high-throughput screening.25 The hydrodynamic flow is, for such experiments, a simple way to induce high flow velocity and avoid high voltages at the same time. Flow Profile in the Microstructure. For flow profiling in the microstructure, a continuous hydrodynamic flow of 15 mm/s was applied, by setting the reservoir with 0.3 nM TMR-4-dUTP at a level of 20 cm. This value was taken from the calibration curve in Figure 4. By scanning the 50-µm microchannel with the small 0.4µm laser focus from top to bottom and from left to right in micrometer steps, we measured the flow times of the molecules. The determination of a hydrodynamic flow profile in microstructures with a spatial resolution of 0.4 µm has, to our knowledge, not been presented yet. It opens a new and simple way of flowvelocity determination in microsystems. Figure 5 illustrates these measurements after the flow times have been transformed into flow velocities using the autocorrelation curves. The figure shows a parabolic flow profile in the vertical and horizontal directions, respectively. These measurements show a clear hyperbolic profile caused by the flow gradient directed toward the center of the microchannel, as predicted by eq 10. However, other flow studies in microchannels found deviations from laminar theory.26 (24) Holm, J.; Elderstig, H.; Kristensen, O.; Rigler, R Nucleosides Nucleotides 1997, 16, 557-562. (25) Brinkmeier, M.; Do ¨rre, K.; Riebeseel, K.; Rigler, R. Biophys. Chem. 1997, 66, 229-239.

3264 Analytical Chemistry, Vol. 72, No. 14, July 15, 2000

Figure 5. Flow profiles in the microchannel scanned in (a) vertical and (b) horizontal directions at an applied reservoir height of 20 cm (V ) 15 mm/s).

Figure 6. Single molecules burst from inside a 50-µm microchannel with an applied flow of (a) 10 mm/s and (b) for pure diffusion in a hanging droplet.

Single Molecule Detection at High Flow Velocity. For the detection of single molecules, fluorescence confocal spectroscopy is the right tool.27,28 The flow time of the TMR-4-dUTP molecules (26) Mala, G. M.; Li, D. Int. J. Heat Fluid Flow 1999, 20, 142-148. (27) Rigler, R.; Mets, U ¨ . Proc. SPIE-Int. Soc. Opt. Eng. 1993, 1921, 239-48.

in the microchannel can easily be increased well above the free diffusion time, τdiff, just by elevating the reservoir. An applied flow of 10 mm/s gives a sample throughput per second, which is more than 20 000 times higher than observed by Brownian motion. This means that 20 000 times more fluorescent molecules will pass the volume element. If single molecule measurements of such a high flow velocity are to be studied by photon burst analysis, the concentration of the TMR-4-dUTP molecules has to be reduced by at least a factor of 10 000 to ensure single-molecule detection. Figure 6 shows a photon burst analysis of a 0.01 pM solution of TMR-4-dUTP molecules at a flow of 10 mm/s inside the microchannel. In the single-molecule regime, such a high flow has, to our knowledge, never been studied before. Even though the molecule transverses the volume element at a very high velocity, enough photons can still be collected and the single molecule bursts can clearly be distinguished. The photon burst analysis was evaluated with a fast multichannel scaler PC-card (MCS) and compared with a previously made hanging droplet experiment. The prefocus lens in the setup was therefore changed to f ) 400 mm to make this comparison possible. The hanging-droplet burst analysis is shown in the lower part of Figure 6. The single-molecule measurements show that in our setup a signal-to-noise ratio of 10:1 could be obtained for TMR-4-dUTP molecules in silicon microstructures. For flow velocities of 10 mm/s a signal-to-noise ratio of 4:1 could still be obtained. The detection of single dye molecules coupled to a nucleotide is the

prerequisite for DNA sequencing in microstructures.19 The result is encouraging for further research in the field of single-molecule DNA sequencing.

(28) Mets, U ¨ .; Rigler, R. J. Fluoresc.1994, 4, 259-64.

AC991448P

CONCLUSION We have used FCS to perform high spatial resolution flow profiling of a 50-µm microstructured channel in a simple and precise manner, by scanning across the channel with a diffraction limited laser focus. This scanning FCS method can deliver a full two-dimensional flow profile with micrometer resolution in the microchannel quickly and easily. Properties which are important in high-throughput screening, such as high hydrodynamic flow velocities, can be monitored with this method.25 Even single molecules at high flow velocities (∼10 mm/s) can be detected, which is important in single-molecule DNA sequencing.19 ACKNOWLEDGMENT We thank the Swedish National Board of Industrial and Technical Development (NUTEK, KOFUMA), the Swedish Research Council for Engineering Sciences for financial support and the Swedish Institute for supporting M.G. Furthermore, we thank Dr. Zeno Fo¨ldes-Papp for all biochemical preparations, Lennart Wallerman for expert workshop assistance, and finally, Professor Gunnar Bjo¨rk for stimulating discussions and for carefully reading through the manuscript. Received for review December 17, 1999. Accepted April 11, 2000.

Analytical Chemistry, Vol. 72, No. 14, July 15, 2000

3265