Anal. Chem. 2003, 75, 5767-5774
An Integrated Microfluidic System for Reaction, High-Sensitivity Detection, and Sorting of Fluorescent Cells and Particles Petra S. Dittrich and Petra Schwille*
Experimental Biophysics Group, Max-Planck-Institute for Biophysical Chemistry, Am Fassberg 11, D-37077 Go¨ttingen, Germany
Presented is a novel approach for an integrated micro total analysis system (µTAS) based on a microfluidic on-chip device that supports ultrasensitive confocal detection of fluorescent cells and particles and subsequently allows for their precise sorting in the fluid phase with respect to spectroscopic properties, such as brightness and color. The hybrid silicone elastomer/glass chip first comprises a branched channel system to initiate fluid mixing and to hydrodynamically focus the sample solution down to a thin flow layer, matching the size of the confocal detection volume placed at that position and, thus, providing a high detection efficiency. In the subsequent on-chip module, the dispersed cells or particles can be sorted into two different output channels. The sorting process is realized by a perpendicular deflection stream that can be switched electrokinetically. The performance of the automated sorting routine is demonstrated by precise partition of a mixture of differently colored fluorescent beads. Moreover, the specifically branched channel geometry allows for direct implementation of reaction steps prior to detection and sorting, which is demonstrated by inducing a selective recognition reaction between the fluorescent protein R-phycoerythrin and a mixture of live bacterial cells exhibiting or lacking the respective surface antigens. Due to the advancement of microfabrication techniques, predominantly triggered by the microelectronics industry, the availability of integrated chips with complex structures in micrometer and submicrometer dimensions has greatly improved in recent years. In the field of applied and analytical chemistry, this development was accompanied by the design, fabrication, and use of microfluidic systems, including capillary networks, miniaturized pumps, valves and mixers.1-3 The availability of more complex liquid handling devices also accounts for their increasing relevance in biological applications, including modules for separation and analysis of fluid samples, for example, for detection, * Corresponding author’s current address: Institute of Biophysics/Biotec, Dresden University of Technology, c/o MPI-cbg, Pfotenhauerstr. 108, D-01307 Dresden, Germany. Phone: +49-351-210-1444. Fax: +49-351-210-1409. E-mail:
[email protected]. (1) Reyes, D. R.; Iossifidis, D.; Auroux, P. A.; Manz, A. Anal. Chem. 2002, 74, 2623-2636. (2) Auroux, P. A.; Iossifidis, D.; Reyes, D. R.; Manz, A. Anal. Chem. 2002, 74, 2637-2652. (3) Mitchell, P. Nat. Biotechnol. 2001, 19, 717-721. 10.1021/ac034568c CCC: $25.00 Published on Web 09/16/2003
© 2003 American Chemical Society
sequencing, and sizing of DNA.4-6 Various kinds of biochemical reactions have so far been successfully carried out within microcapillary systems, including enzymatic and immunoassays7,8 and polymerase chain reactions,9 with low sample consumption, short reaction times due to, for example, efficient heat transfer, and with low (re)production and operating costs of the respective microchips. One of the key challenges in microfluidics certainly consists of the integration of multiple reaction and separation units on a single chip in combination with the total analysis of the whole sample. Downscaling of reaction systems to dimensions comparable to biological structures is an important step toward the precise handling and manipulation of small particles, cells, cell organelles, and even macromolecules, opening up fascinating prospects for fluid-phase screening of large populations of microorganisms and molecules, and thereby providing a platform for efficient isolation of samples with desired properties. High-throughput screening is widely used in biology and biotechnology in the search for novel and improved molecular or cellular function.10-12 In cellular screens, which are often based on brightness patterns, size and shape of the cells, protein expression, or enzyme activity can be determined, for identifying specific structures within the cells and for investigation of cell functions. However, since commercially available cell sorters are rather expensive and limited with respect to their implemented sorting parameters, many efforts have been undertaken to find low-cost alternatives on small microchips, providing more elaborate separation steps. Most often, microdevices designed for particle separation and sorting are based on physical properties of the sample, such as size and charge in capillary electrophoresis.13 Separation of particles with different sizes can also be achieved by chip-based filtering structures, (4) Van Orden, A.; Cai, H.; Goodwin, P. M.; Keller, R. A. Anal. Chem. 1999, 71, 2108-2116. (5) Chou, H. P.; Spence, C.; Scherer, A.; Quake, S. Proc. Natl. Acad. Sci. U.S.A. 1999, 96, 11-13. (6) Foquet, M.; Korlach, J.; Zipfel, W.; Webb, W. W.; Craighead, H. G. Anal. Chem. 2002, 74, 1415-1422. (7) Wang, J. Electrophoresis 2002, 23, 713-718. (8) Hatch, A.; Kamhols, A. E.; Hawkins, K. R.; Munson, M. S.; Schilling, E. A.; Weigl, B. H.; Yager, P. Nat. Biotechnol. 2001, 19, 461-465. (9) Kopp, M. U.; de Mello, A. J.; Manz, A. Science 1998, 280, 1046-1048. (10) Katsuragi, T.; Yoshiki, T. J. Biosci. Bioeng. 2000, 89, 217-222. (11) Georgiou, G. Adv. Protein Chem. 2001, 55, 293-315. (12) Nolan, J. P.; Sklar, L. A. Nat. Biotechnol. 1998, 16, 633-638. (13) Jacobson, S. C.; Hergenro¨der, R.; Koutny, L. B.; Ramsey, J. M. Anal. Chem. 1994, 66, 1114-1118.
Analytical Chemistry, Vol. 75, No. 21, November 1, 2003 5767
employing sieves and sieves-like obstacles,14,15 and by use of the fast diffusion of small molecules compared to larger ones, enabling a passive filtering perpendicular to the laminar flow direction.16 Furthermore, single cells and particles have been successfully fixated and manipulated by optical traps17 or dielectrophoresis.18 Another strategy is based on the indirect manipulation of dissolved particles in a transporting liquid stream by precisely steering the liquid flow. Employing manipulation of small “containerlike” liquid quantities rather than particles, cells, and molecules, the sorting routine can be uncoupled from molecular properties and made amenable to various kinds of secondary control parameters. However, this larger flexibility with respect to molecular or cellular properties to be screened for requires highly dynamic systems for manipulation of small liquid quantities. For precisely controlling the flow of the sample solution, that is, the velocity as well as the direction into different output channels, the need for suitable valves and switches, being fast and reliable on one hand and easy to handle on the other, are indispensable. The two most prominent strategies to induce controllable movement of the fluid in a microchannel are pressure-driven flow and electroosmotic flow. In the former case, pumps and valves are necessary to regulate the flow properties. Although previous approaches integrated peristaltic pumps directly in a multilayer microchip,19 we tried in the first step to implement magnetic valves outside of the chip into the connected tubing system. In that way, the fabrication of the less sophisticated microchip can be much facilitated, and it can be more easily replaced by other designs. The various interfaces of microfluidic channels, tubing systems, and valves are difficult to handle, though. In the case of electroosmotic flow20 induced by a high applied voltage (electrical fields strength up to a few hundred V/cm) between miniature electrodes inserted into the channel input and output reservoirs, the direction of the flow can simply be regulated by electrically switching the respective electrodes in different output channels. Since the response of electroosmotic flow to the direction of the applied field is instantaneous, electrical manipulation usually provides much shorter switching times and thus, theoretically yields much higher sorting speeds. However, because of the strong dependence on local channel structures and material as well as fluid composition, electroosmotic pumping is frequently accompanied by instabilities or by an overlaid electrophoretic movement of the particles, but also by a rather low tolerance of living material, such as cells,19 toward the high required electrical fields to be applied for large throughput rates. Consequently, structures that expose the cells to strong electric fields over a long range of on-chip processing usually result in low cell survival rates. (14) Carlson, R. H.; Gabel, C. V.; Chan, S. S.; Austin, R. H.; Brody, J. P.; Winkelmann, J. W. Phys. Rev. Lett. 1997, 79, 2149-2152. (15) Chou, C. F.; Bakajin, O.; Turner, S. W. P.; Duke, T. A. J.; Chan, S. S.; Cox, E. C.; Craighead, H. G.; Austin, R. H. Proc. Natl. Acad. Sci. U.S.A. 1999, 96, 13762-13765. (16) Weigl, B. H.; Yager, P. Science 1999, 283, 346-347. (17) Grover, S. C.; Skirtach, A. G.; Gauthier, R. C.; Grover, C. P. J. Biomed. Opt. 2001, 6, 14-22. (18) Fiedler, S.; Shirley, S. G.; Schnelle, T.; Fuhr, G. Anal. Chem. 1998, 70, 1909-1915. (19) Fu, A. Y.; Chou, H. P.; Spence, C.; Arnold, F. H.; Quake, S. R. Anal. Chem. 2002, 74, 2451-2457. (20) Fu, A. Y.; Spence, C.; Scherer, A.; Arnold, F. H.; Quake, S. R. Nat. Biotechnol. 1999, 17, 1109-1111.
5768
Analytical Chemistry, Vol. 75, No. 21, November 1, 2003
Exploiting the benefits of both pressure and voltage-induced fluid manipulation, the miniaturized sorting device presented here combines the two standard modes of fluid transport in microchannels: The main transport of the sample solution through the channel system is pressure-driven, providing a fast and stable flow velocity and, consequently, high througphut rates, while the switching is realized by a perpendicular deflection channel being regulated by electrokinetic flow, directly followed by a branching of the main channel, as introduced by Duffy et al.21 Since the electrokinetic flow is easy to control and initiated quasi-instantly when the electrodes are activated, the switching routine in the deflection channel can be operated very rapidly. However, for a shallow branching of the main channel, only small deflecting stimuli are required to direct the sample fluid in either of the branches. Thus, the applied voltage in the deflection channel can be held very low, providing a high cell viability of this sorting device. For fabrication of the microchip, we used standard replica molding techniques established by so-called soft lithography.22,23 A silicon wafer fabricated by photolithography processes is used as a mold. The casted PDMS chip structure which comprises the channel system can be covered with a glass plate for sealing and optical adaptation. The reproduction of the chip is easy and fast and can be carried out under standard lab atmosphere without the need for a clean room. Besides the manipulation, another important task in microfluidics is the precise and quantitative analysis of the manipulated fluid samples or particles. Often, the detection is based on optical methods, such as fluorescence microscopy and spectroscopy, allowing extremely fast analysis of the sample solution with high sensitivity and specifity, with the additional benefit of being contact-free and, thus, substantially noninvasive. Using confocal fluorescence setups with measurement volumes in the subfemtoliter regime, it has been shown that high signal-to-noise ratios up to 103 can be achieved.24 Because of the tiny detection volumes, the scattering background of the solution can be largely suppressed, and the detection of single fluorescent dye molecules is supported, even in small structured channels where conventional optical techniques usually suffer from stray light contributions from the channel walls.25 Confocal detection can be easily performed with two or more spectrally distinguishable fluorescence species,26 which adds more specificity and flexibility for analyzing more complex reaction schemes. In addition to that, by employing a hardware correlator unit for data processing, the flow velocity of the sample inside the channel can be simultaneously analyzed and controlled by fluorescence correlation spectroscopy (FCS) within the same setup.27,28 In the following, the proper functioning of our newly designed microfluidic device with respect to ultrasensitive detection, (21) Duffy, D. C.; Schueller, O. J. A.; Brittain, S. T.; Whitesides, G. M. J. Micromech. Microeng. 1999, 9, 211-217. (22) Xia, Y.; Whitesides, G. M. Annu. Rev. Mater. Sci. 1998, 28, 153-184. (23) McDonald, J. C.; Duffy, D. C.; Anderson, J. R.; Chiu, D. T.; Wu, H.; Schueller, O. J. A.; Whitesides, G. M. Electrophoresis 2000, 21, 27-40. (24) Mets, U ¨ .; Rigler, R. J. Fluoresc. 1994, 4, 259-264. (25) Do¨rre, K.; Brakmann, S.; Brinkmeier, M.; Han, K. T.; Riebeseel, K.; Schwille, P.; Stephan, J.; Wetzel, T.; Lapczyna, M.; Stuke, M.; Bader, R.; Hinz, M.; Seliger, H.; Holm, J.; Eigen, M.; Rigler, R. Bioimaging 1997, 5, 139-152. (26) Schwille, P.; Bieschke, J.; Oehlenschla¨ger, F. Biophys. Chem. 1997, 66, 211-228. (27) Magde, D.; Webb, W. W.; Elson, E. L. Biopolymers 1978, 17, 361-376.
Figure 1. The microstructure with the integrated channel system is positioned on a confocal microscope. The sample solution was excited by the 488-nm line of an Ar ion laser, and the green and red spectral parts of the emission light were collected by two detectors. The data were correlated and analyzed online by hardware correlation and multichannel analysis. For sorting procedures, the electrodes connected to the computer could be activated via a feedback loop from the detection module.
analysis, and quantitation of fluorescent cells and particles, as well as the efficient sorting of differently colored species, is demonstrated. In contrast to standard flow cytometry applications, in which several external preparation routines (i.e., inducing specific reactions) are required prior to cell counting and sorting, we show the successful on-chip integration of such reactions in the same microstructure where the sorting is performed, providing the first approach to a truly multifunctional chip for cell handling. Thus, reaction and preparation times can be reduced, and the cell handling can be simplified significantly. As a proof of principle for biological applications to be performed in our fluidic devices, the specific recognition of surface proteins in the outer membranes of bacterial cells by a fluorescent ligand is demonstrated. The respective assay consists of Escherichia coli bacterial cells expressing a transmembrane protein with a short peptid sequence displayed on the cell surface which exhibits a high binding affinity to R-phycoerythrin. R-Phycoerythrin is a fluorescent protein extracted from red algae that is commonly used in standard cytometry experiments. Its fluorescence is due to a monodisperse population of prosthetic groups (phycobilines) absorbing over a broad spectral range of blue light and emitting orange-red light. In the context of biotechnological cell sorting, the approach used in our test assay is better known as epitope mapping for screening of large peptide libraries.29,30 Hereby, bacterial cells transformed with a so-called surface display vector express a transmembrane protein (carboxyterminally truncated intimin), that serves as a carrier protein to present foreign peptides on the surface of bacterial cells. Epitopepresenting cells can then easily be identified and isolated by immunofluorescence staining with monoclonal antibodies. MATERIALS AND METHODS Optical Setup. All sorting experiments were carried out in a confocal setup (Figure 1) realized in an inverted microscope (Olympus IX70). For excitation of the fluorescent sample, the blue 488-nm line of an Ar ion laser (Lasos, Jena, Germany) was coupled (28) Dittrich, P. S.; Schwille, P. Anal. Chem. 2002, 74, 4472-4479.
into the microscope by parallel epi-illumination (with underfilled back aperture) via a dichroic mirror (DCLP 495; all optical filters purchased from AHF, Tu¨bingen, Germany) and focused by the objective (Olympus PlanApo 40×, NA 1.15). Fluorescent light was collected by the same objective and, after passing the first dichroic mirror, was split into its green and red spectral parts by a second dichroic mirror (DCLP 520). Both detection beam paths were projected to the respective entrance slits (diameter 100 µm) of two optical fibers in the image plane, representing the confocal pinholes. For efficient suppression of background light (Rayleigh stray light and Raman scattering) and to further enhance detection selectivity, additional band-pass filters were placed in front of the optical fiber apertures in the green (516DF10) and red (630DF60) detection channels. The fluorescence signal was detected by avalanche photodiodes (SPCM-200, EG&G, Optoelectronics, Vaudreuil, Canada) and split by a pulse divider. By splitting the signal, simultaneous data processing with a PC-correlator card (ALV, Langen, Germany) for fluorescence correlation analysis, and automated control of the sorting device by another hardware processor ADWin (ADWin-LD, Ja¨ger Messtechnik, Lorsch, Germany) could be facilitated. The ADWin card counts the recorded TTL pulses from the photodiodes in a predefined integration time (i.e., photon counts per unit time, measured in bins of typically 1-100 ms), and activates the electrodes in a feedback loop. Calibration measurements of the optical setup were carried out with Rhodamine Green at a laser intensity of 50 kW/cm2. The dimensions of the detection volume determined by fluorescence correlation spectroscopy (FCS)24,26 were usually ∼0.8 µm in axial and ∼4.0 µm in the lateral direction, resulting in a detection volume of roughly a few femtoliters (∼2.0 fL). To test the correct functioning of the sorting device and to count fluorescent beads after the sorting procedure, a Hg arc lamp (Olympus U-RFL-T) was used for wide field illumination, and the (29) Christmann, A.; Wentzel, A.; Meyer, C.; Meyers, G.; Kolmar, H. J. Immunol. Methods 2001, 257, 163-173. (30) Christmann, A.; Walter, K.; Wentzel, A.; Kratzner, R.; Kolmar, H. Protein Eng. 1999, 12, 797-806.
Analytical Chemistry, Vol. 75, No. 21, November 1, 2003
5769
Figure 2. SEM image of the silicon wafer used as the mold. The main channel of the structure shown here was 25 µm wide and 7.6 µm deep.
images were recorded by a CCD camera (CCD 260 sw/w, Spindler & Hoyer). Fabrication of the Microchannels. The microstructures were casted from a silicone mold comprising the channel geometry and topology in the negative (upstanding) form. The silicon wafers were partly custom-made (GeSim, Grosserkmannsdorf, Germany), coated with a thin 200-nm Teflon layer, and partly self-made at the Cornell Nanofabrication Facility, Ithaca, NY, employing standard photolithography processes. Briefly, a silicon wafer was spin-coated first with hexamethyldisilazane (HMDS) and, second, with positive photoresist (Shipley 1813) at 3500 rpm (ramping time 3 s, spin time 30 s). After a soft bake (1 min on a 115 °C hot plate), it was exposed to UV light through a photolithography chrome mask with the specific pattern design. Afterward, the wafer was baked in NH3 vapor to reverse the tone of the photoresist (image reversal). Its entire area was exposed to UV light and developed in tetramethylammonium hydroxide (TMAH, Shipley MF 321). Using deep reactive ion etching (DRIE), rectangular channel walls could be obtained (Figure 2). For sorting of large particles, the channel width was 25 µm in the main channels and 10 µm in the perpendicular deflection channel, with a channel depth of 14.7 µm. For sorting of smaller particles and bacterial cells, the channel widths were 25 µm and 15 µm, respectively, with a depth of 7.6 µm. The lengths of all microchannels were ∼1.5 cm. Large numbers of cheap and disposable structures could then easily be obtained by imprinting the complementary mold topology into silicon elastomer. In this routine, 2-3 mL of viscous elastomer kit (Sylgard 184 (Dow Corning)) with a mixing ratio of 10:1 of component A/B (A, silicone prepolymer; B, curing agents) was poured onto the silicon wafers and hardened on a heating plate for 30 min at 110 °C. The hardened elastomer could then easily be peeled off the wafer, cut to the desired sizes of ∼2 × 3 cm, and punched at the end of the microchannels for introducing the liquid. Finally, it could be sealed with a thin glass plate for excellent adaption to the optical detection system, which was realized simply by manually pressing a coverslip of ∼170 µm thickness to the channel top. The adhesion forces between the silicon elastomer and the glass coverslip rendered the channels sufficiently tight for the here-applied pressures on the fluidic 5770
Analytical Chemistry, Vol. 75, No. 21, November 1, 2003
sample. Without further treatment, the channels could then be filled with the buffer solution. This technique allows one to reproduce feature sizes in (sub)micrometer dimensions22,23 in PDMS in a standard lab. The soft elastomer material even tolerates small contaminations, such as dust particles between the chip and cover glass without seriously interfering with the functioning of the device. Additionally, we could not observe a degradation of the mold, even after ∼100 casting procedures. Sorting Routine. The silicon elastomer structure was positioned on a revolving x, y, z-stage, while the exact positioning could be controlled by a CCD-camera. In the punched fluid reservoirs at the two ends of the deflection channel, the electrode wires were inserted and connected to a home-built electrical controller for the automated switching of the polarity. The control unit was hooked up to a voltage generator (model Highpac A2K5 20HR, Oltronix; applied voltage, 40 V, corresponding to electrical field strength in the microchannel of 20-30 V/cm) and driven by the PC ADWin card. The software interface that translates the input parameters such as particle speed, brightness, and color, into the desired regulation parameters for efficient sorting was self-written in a card-specific language, AdBasic (Ja¨ger Messtechnik, Lorsch, Germany), in combination with Test Point (Capital Equipment Corp., Billerica, MA) for real-time presentation of the data, which enabled the control of the sorting process on the monitor. Whenever the fluorescence emission rate from the detectors exceeded a predefined threshold, the electrodes were activated, either instantaneously or after a certain delay time reflecting the duration for a particle to travel from the detection to the selection sites in the channel. The interval of electrode activation could also be varied accordingly. For calibration and sorting of fluorescent beads, we used polysterene beads (diameters 2.5 and 1 µm; emitting blue (450 nm), green (515 nm), and yellow (560 nm); all beads purchased from Molecular Probes) diluted in water (concentration, ∼106 beads/mL). Prior to every sorting process of individual beads, a microchip was freshly prepared, and the entire channel system was filled with filtered and degassed water solution. The flow parameters in the microchannels were carefully calibrated by adjusting the pressures on the filling holes of each channel to guarantee proper hydrodynamic focusing and to obtain equal selection probability in both output branches of the main channel in the absence of the deflection flow in the perpendicular channel. This calibration was routinely performed by imaging the distribution of a concentrated bead solution that was nonfluorescent in the wavelength region used for the sorting. Finally, ∼2 µL of a premixed suspension of green and yellow fluorescent beads was added to the water reservoir of the main channel, and the automated sorting routine was started, running usually for 10-30 min. After the sorting, the processed sample solution could be recollected from the output reservoirs. During the time of the sorting, the performance of the microsorter was stable, and in principle, the microstructure could be used for several successive sorting experiments. However, to exclude mistakes in the sorting process resulting from residual beads adsorbed in the microchannels or left in the reservoirs, the microchip was replaced for each new experiment.
Figure 3. Principle of the sorting device, exemplified with a highly concentrated suspension of fluorescent beads. After hydrodynamic focusing, the beads entered both output channels with equal probability (left image). By activation of a perpendicular electroosmotic flow, the suspension could be directed into one desired output channel, depending on the polarity of the electrodes (middle and right image).
Experimental Protocol for Bacterial Recognition Assay. Bacterial cell cultures Escherichia coli BMH 71-1830 transformed with the bacterial surface display vector pASKInt100phycoA were grown overnight at 37 °C (25 µg/mL Chloramenphenicol). In a subculture being also grown at 37 °C until it reached OD550 > 0.5, the expression of the transmembrane protein was induced by addition of anhydrotetracycline (200 µg/mL). After 3 h of incubation time at room temperature, the cells were pelleted by centrifugation and stored at 4 °C. Prior to the experiment, they were freshly resuspended in PBS buffer solution at pH 7.2 (140 mM NaCl, 10 mM KCl, 6.4 mM Na2HPO4, 2 mM KH2PO4) in small fractions (∼1 µL pellet in 200 µL of buffer). R-Phycoerythrin (4 mg/mL) was diluted in PBS buffer solution 1:400 for incubation in a reaction tube and 1:4000 for the on-chip reaction. The mircochannels were filled first with PBS buffer solution enriched with 0.3 µg/mL BSA to prevent cells and R-phycoerythrin from adhesion to the channel walls. After calibration of the flow parameters, the R-phycoerythrin solution was introduced into the side channels. Finally, the cell suspension was added into the main channel. RESULTS AND DISCUSSION Principle of the Sorting Device. The proper functioning of the sorting device, involving the different units for (a) fluorescence detection and analysis and (b) the automated sorting process for cells and particles, is shown in Figure 3. To illustrate the performance of the sorting module, the flow of a highly concentrated fluorescent bead solution is directly vizualized with the CCD camera by illuminating the field of view with a Hg arc lamp. At the first intersection, the sample in the main channel is focused hydrodynamically to a narrow stream, as introduced by Knight et al.31 for mixing reactions. This design guarantees high detection efficiency in the sorting device by laterally confining the sample solution to a cross section that can be largely covered by the tiny detection volume. By this procedure, the cells reach the excitation/detection point successively in line, like beads on a chain, and a complete particle counting and analysis is guaranteed. Thus, hydrodynamic focusing eliminates the need for extremely narrow channels that are difficult to handle and suffer from a frequent clogging or, on the other hand, for widening the detection volume, which would result in decreased detection (31) Knight, J. B.; Vishwanath, A.; Brody, J. P.; Austin, R. H. Phys. Rev. Lett. 1998, 80, 3863-3866.
sensitivity.32 Additionally, by adding a reactive substance in the side channels instead of just buffer solution, an integrated reaction/sorting system can be easily performed. Since the particles to be sorted have large hydrodynamic radii and, consequently, show only slow diffusion perpendicular to the flow direction, they are well retained in the center of the channel33 until they reach the sorting branch, where both exits are taken with equal probability if the hydrodynamic pressure of the output channels is equalized (and lower than the pressure at the input channels). To direct the particle stream into one desired output reservoir, the deflection stream in the perpendicular channel has to be activated. The flow in this narrow deflection channel, likewise with equalized pressures at both ends in the resting state, is controlled electrokinetically. By applying a voltage, a weak electrokinetic flow toward the cathode is induced, pushing the particles only a few micrometers away from the center of the channel, such that their probability to enter the right channel is significantly enhanced. This deflection flow can be switched in both directions. For particle sorting, the electroosmotic flow is activated discontinuously, that is, whenever the transit of a cell through the detection volume is indicated by the fluorescence recording unit. To obtain good sorting results, two main challenges have to be met, as will be explained in more detail in the following. The first lies in the complete high-sensitivity detection of all sample particles being successively lined up by hydrodynamical focusing, the second is the accurate sorting step according to the particles’ properties, as determined in the detection process. High-Sensitivity Detection. As described above, detection is performed in a confocal geometry with the focal spot being placed in the center of the main channel. Fluorescent particles dispersed in the transported fluid are efficiently excited by the laser focus. Because of the small dimensions of the detection volume of only ∼0.8 µm in the flow direction and ∼4.0 µm perpendicular to it (see Experimental Section), and because of the large detection efficiency of the optical setup (up to 4%), scattering background from the fluid is greatly reduced. Hydrodynamic focusing on the other hand ensures that all particles indeed pass the detection volume. The flow velocity in the flanking channels has to be (32) Do ¨rre, K.; Stephan, J.; Lapczyna, M.; Stuke, M.; Dunkel, H.; Eigen, M. J. Biotechnol. 2001, 86, 225-236. (33) Demas, J. N.; Wu, M.; Goodwin, P. M.; Affleck, R. L.; Keller, R. A. Appl. Spectrosc. 1998, 52, 755-762.
Analytical Chemistry, Vol. 75, No. 21, November 1, 2003
5771
Figure 4. The fluorescence bursts in the green and red emission channels signal the throughput of a green or red fluorescent bead (left). Simultaneously, the flow velocity of the beads can be determined by fluorescence correlation spectroscopy (right). Each curve was generated within a measurement interval of 10 s and fitted with the theory model in eq 1. From the so-determined flow time, τf (corresponding to the time that the particles need in average to traverse the detection volume), the flow velocity can be derived. The hydrodynamic flow employed for net transport of the sample solution exhibits a parabolic velocity profile with a rather broad distribution of transit times, typically ranging from 2 to 6 ms. The corresponding flow velocities thus range between 0.55 and 1.65 mm/s.
optimized with respect to tight focusing but also to maximum throughput. For detection of the beads with a diameter of ∼2.5 µm, the sample solution was focused down to a small stream of 1-2 µm width (Figure 3). By comparison of the number of beads registered by the detector and their total number as being counted in the output wells, an increase of the particle detection efficiency from 50% without hydrodynamic focusing to 99% could be determined for a distance of 75 µm between the focusing branch and sorting branch. The small difference with respect to complete detection was most probably due to occasional sticking of beads to each other, making them indistinguishable for the detectors. Another rare error source is the “crawling” of beads on the bottom of the channel, thereby surpassing the detection volume. Similar results were obtained for 1-µm beads in a 25-µm-wide and 7.6µm-high channel. In the setup shown here, fluorescence correlation spectroscopy can be easily adapted as an ideal tool to precisely control local flow velocity. The determination of actual velocity distributions within the channel is a crucial requirement for adjusting the delay times between particle detection and deflection channel activation in an automated sorting routine. An inherent consequence of the hydrodynamic transport applied to the sample solution in the main channel is the parabolic flow profile, leading to a relatively broad distribution of measured bead velocity perpendicular to the flow direction, with a maximum in the center of the channel and very slow mobility near the channel bottom (and top). This inhomogeneity has to be taken into consideration when presetting the ideal switching time. Typical autocorrelation curves for a flowing bead suspension are plotted in Figure 4. The flow velocity can be calculated based on the average time that the particles need to transit the laser focus. If diffusion of the particles can be neglected, this so-called flow time, τf, is determined using the following expression,27 in which G(τ) is the autocorrelation function within the time, τ, and N is the average particle number inside the detection volume.
G(τ) )
[ ( )]
1 τ exp N τf
2
(1)
In standard experiments, the characteristic τf is between 2 and 6 5772
Analytical Chemistry, Vol. 75, No. 21, November 1, 2003
ms. Considering the size of the particles (diameter dparticle, 2.5 µm in Figure 4) and the dimensions of the detection volume (dfocus ) 0.8 µm), the flow times correspond to average velocities of v ) (dfocus + dparticle)/τf ) 0.55-1.65 mm/s. If the detection spot in the channel is situated, for example , 15 µm before the sorting branch, the particle travels another 9-27 ms to reach this branch after being detected. During this time interval, all steps required for data analysis have to be completed, and the deflection flow has to be activated (the exact switching parameters being predefined in the controlling computer program). The temporal resolution of the detection unit is determined by the optimum integration time for photon counting, that is, the transit time of particles through the detection volume. For integration times significantly shorter than the transit time, the fluorescence bursts induced by the particles get fragmented into a respective number of successive smaller events, and the proper definition of a certain photon yield threshold to indicate a transit event gets difficult. On the other hand, integration times that are too long reduce the signal-to-noise levels (photons per burst) by collecting only background for a certain amount of time. The optimal binning frequency for counting and analyzing of cells with typical transit times of up to 6 ms, consequently, is around 160 Hz. One of the most prominent sorting parameters, due to its simplicity of quantitation, is certainly the absolute brightness of cells and particles under fluorescence excitation. However, because of the limited size of the detection volume and its inhomogeneous illumination profile (assumed to show Gaussian intensity distribution in all three dimensions), there is a certain heterogeneity of fluorescence burst sizes, depending on where exactly the particles enter and leave the focal spot. This effect is particularly strong perpendicular to the flow direction, where because of the channel geometry, no hydrodynamic focusing is obtained. By introducing a rigorous burst-size threshold, selection of the brightest species out of a mixed sample is relatively simple, but some losses of bright species as a result of their nonideal transit pathways through the focal spot are inevitable. Therefore, relative rather than absolute, measurement parameters are by far more reliable sorting criteria. Such a parameter is, for example, the ratio of fluorescence count rates in two
Table 1. Sorting Results and Sorting Speed for a Mixture of Green and Red Fluorescent Beads with Different Initial Fractionsa initial
fraction reservoir 1
reservoir 2
sorting speed
0.58:0.42 0.21:0.79 0.09:0.91
94.7:5.3 94.8:5.2 84.6:15.4
3.2:96.8 14.6:85.4 0.9:99.1
0.38/s 0.79/s 0.68/s
a
The values always describe the fraction of green vs red beads.
different detection channels corresponding to spectrally different detection ranges. Hereby, differently colored species, for example, green and red, can be easily differentiated: each time when a fluorescent burst is detected, indicating the throughput of a particle, the ratio of signal intensity in both detection channels is calculated. Independent of the total number of photons detected in the burst, the so-determined ratio reliably determines the spectral properties, that is, the color of the particles which can then be used for activation of the sorting process (Figure 4). The use of a two-color detection setup thus extends the range of possible sorting parameters to various interesting assay systems: inter- or intramolecular dynamics accompanied by spectroscopic changes due to FRET (fluorescence resonance energy transfer) or spectral shifts can easily be resolved, as will be demonstrated below. Sorting Process. To test the fidelity of the sorting process, a 1:1 mixture of green and red fluorescent beads was introduced into the main channel, both bead types being excitable by the 488-nm line of the Ar ion laser. Because of their brightness, the laser intensity in this experiment was reduced to only 100 W/cm2 to prevent saturation of the detectors. The sorting routine was run in an automated mode, as explained in the Materials and Methods Section, and was stopped after 200-500 successive events (i.e., detected beads), corresponding to a typical running time of 10 min. After the sorting experiments, the total numbers of beads in the output reservoirs were determined by recording a CCD image of the whole reservoir under broad fluorescence wide-field illumination by an arc lamp and counting the bright spots by simple image analysis. The sorting accuracy calculated from the fractions of green and red fluorescent beads found in both detection wells after the sorting process is listed in Table 1 for different initial fractions. It could be shown that the sorter is equally well suitable for differentiation of evenly distributed species as well as for “fishing” of rare events, that is, the enrichment of a rare species in one of the collection wells. Although the efficient fractionation of two evenly distributed species requires careful pressure calibration of both output channels, as explained above, enrichment of rare species can also be reliably obtained when the channel for the majority species is slightly preferred. Typical processing rates that still yield accurate sorting are between 0.3 and 1 event/s, although the theoretical limit of the sorting frequency in this specific setup is 10 Hz. This limit is principally imposed by the binning time for the detection and the activation time of the electrodes for inducing the electroosmotic deflection flow. However, since the use of highly concentrated sample suspensions decreases the sorting accuracy, the flow rate needs to be increased for high-speed sorting.
Specific Recognition and Sorting of Live E. coli Bacterial Cells. To ensure biological significance of the microfluidic cell sorter, a test run was first performed with the goal of quantifying the cellular survival rate after being subjected to pumping, hydrodynamic focusing, and electrokinetic sorting on the chip. This quantitation was done by counting the cells when passing the detection volume, due to their autofluorescence at the excitation wavelength 488 nm and intensity I ) 500 W/cm2. All cells were directed to one output well by a continuously activated electroosmotic flow in the deflection channel (electrical field strength 30 V/cm). After collecting the cells and reculturing them on a medium plate, the number of colonies could be counted. The recovery rate, given by the fraction of colonies on the plate with respect to the number of cells detected in the confocal volume, is typically between 80 and 95%. This rate not only describes the fraction of cells surviving the sorting process inside the channel, but also takes into account all kinds of other losses, for example, due to adsorption at the channel walls or incomplete recollection from the output well. We observed decreasing recovery rates for high electrical field strengths (only 60% for 100 V/cm). To demonstrate the successful combination of a reaction and a sorting procedure on the same chip, the specific recognition of transmembrane proteins on the surface of E. coli bacterial cells by the fluorescent label R-phycoerythrin was observed. For the on-chip reaction, a freshly prepared E. coli cell suspension was introduced into the middle input reservoir of the main channel. Each side channel was filled with a solution of the fluorescent protein R-phycoerythrin. At the intersection between the main and side channels, both the hydrodynamic focusing of the cell suspension and the binding reaction were initiated. Figure 5a,b shows typical fluorescence burst traces for high-speed analysis of bacterial cells before and after the binding reaction with R-phycoerythrin. A statistical overview over 500 single events is given by the histograms in Figure 5c,d. The signal bursts induced by the native, nonlabeled bacterial cells in buffer solution were mainly induced by scattering and autofluorescence, with a typical ratio of 0.64 between green and red fluorescence. This ratio increases up to 1.22 after addition of R-phycoerythrin from the side channels. Since the R-phycoerythrin solution increases the background in the red detection channel as a result of its autofluorescence in the nonbound state, the traces are offsetcorrected, that is, the additional steady-state fluorescence from free Phycoerythrin in the red detection channel is simply subtracted. The results are compared to an in vitro reaction of highly concentrated R-phycoerythrin solution to a cell suspension in a test tube for 15 min. In a following cell analysis inside the microchannels, the signal ratio of red to green fluorescence for these cells was shifted to 3-4.5, indicating a high reaction rate under these conditions. However, the sensitivity of the detection was sufficient to prove whether a specific reaction had occurred, even after the short time interval between the addition of R-phycoerythrin at the first channel branch and the detection point which was positioned ∼50 µm downstream. The available reaction time for typical flow velocities of 0.5-1 mm/s is only 50-100 ms. In this way, the on-chip reaction reduces time-consuming preparation steps. Analytical Chemistry, Vol. 75, No. 21, November 1, 2003
5773
Figure 5. Fluorescence traces recorded in the red and green detection channels before (a) and after (b) the surface staining of bacterial cells by the fluorescent protein R-phycoerythrin in the sorting device. For each cell passage, the ratio of the two signals is calculated (squares), serving as the main criterion for the downstream sorting process; all cells above/below a preset threshold (here, ratio )1) are sorted to the left/right. A statistical overview is provided by the histograms (c, d) created for 500 cell passages: the main ratio increases after addition of R-phycoerythrin from 0.64 to 1.22, when the detection point is positioned ∼50 µm downstream. For comparison, the results for an extern incubation in a test tube for 15 min are shown (main ratio, 3.68). The signal ratio remains stable with a maximum of ∼0.65 for bacterial cells that do not express the transmembrane protein (d) and, consequently, do not bind to R-phycoerythrin for the same reaction conditions.
CONCLUSIONS As a fluid-phase alternative to conventional fluorescenceactivated cell sorters (FACS), the inexpensive microdevice presented here, which can be mounted on a confocal setup, allows for extremely sensitive detection of fluorescent species down to the single molecule detection level and implies all benefits of microfluidic systems, including the possibility of integration of reaction units before the detection unit or after the sorting process. The combination of hydrodynamic liquid pumping in a main channel with a fast electrokinetic switching module in a deflection channel enables the precise detection, counting, and sorting of living cells and particles in an automated routine. The sorting unit makes use of the inherent probability distribution that directs 50% of the particles to either side already. To enrich particles with specific properties in one desired output channel, only small modifications of the probability distribution have to be made, which is achieved by activating the deflection channel. Thus, the required energy for one single selection process is minimized compared to, for example, alternative geometries in which particles have to be actively “picked” into a selection channel perpendicular to the main transport channel. This strategy particularly pays off if more than one selection step are assembled in line, because the probability for selecting incorrectly then decays exponentially. In this case, the processivity or speed of the sorting module could be substantially increased if less care had to be spent to optimize the yield in a single selection step. For future applications of cell sorting modules in the fluid phase, we thus propose that the assembly of many single sorting units in line is the way to proceed.
Recognition reactions analogous to those presented here are widely employed in flow cytometry; thus, a wide field of applications in cell biology, biotechnology and medicine is opened up. The future potential certainly lies in the highly specific, ultrasensitive, and fast screening of large libraries of cells, cell organelles, aggregates, and small fluorescent particles, yielding information about the distribution of subpopulations. On the other hand, screening for rare events with a following enrichment can also be achieved when the sensitive detection is ensured. As outlined above, the optical setup allows for detection of weak fluorescent signals. To attain the sorting of single (macro)molecules tagged with a fluorescent dye,12,34 we propose to cage them in appropriate compartments, for example, in vesicles or emulsions, or to bind them to beads. Hereby, investigations of single-molecule reactions and isolation of such “containers” with specific characteristics may soon be feasible.
(34) Eigen, M.; Rigler, R. Proc. Natl. Acad. Sci. U.S.A. 1994, 91, 5740-5747.
AC034568C
5774
Analytical Chemistry, Vol. 75, No. 21, November 1, 2003
ACKNOWLEDGMENT We thank A. Christmann and H. Kolmar (Department of Molecular Genetics and Preparative Molecular Biology, University of Go¨ttingen) for kindly providing us with the bacterial clone transformed with the surface display vector and Tobias Kohl (Experimental Biophysics Group) for discussion and practical support. Financial Support was provided by the German Ministry for Education and Research (Biofuture Grants nos. 0311845 and 16SV1257) and the German Exchange Organization DAAD (Travel Grant to CNF, Ithaca, NY). Received for review May 28, 2003. Accepted August 13, 2003.