Microfluidic Chip for Low-Flow Push-Pull Perfusion Sampling in Vivo

Multilayer soft lithography was used to prepare a poly(dimethylsiloxane) microfluidic chip that allows for in vivo sampling of amino acid neurotransmi...
0 downloads 0 Views 360KB Size
Download PDF
Anal. Chem. 2005, 77, 7067-7073

Microfluidic Chip for Low-Flow Push-Pull Perfusion Sampling in Vivo with On-Line Analysis of Amino Acids Nicholas A. Cellar,† Scott T. Burns,† Jens-Christian Meiners,‡ Hao Chen,‡ and Robert T. Kennedy*,†,§

Department of Chemistry, Department of Physics, and Department of Pharmacology, University of Michigan, Ann Arbor, Michigan 48109

Multilayer soft lithography was used to prepare a poly(dimethylsiloxane) microfluidic chip that allows for in vivo sampling of amino acid neurotransmitters by low-flow push-pull perfusion. The chip incorporates a pneumatically actuated peristaltic pump to deliver artificial cerebrospinal fluid to a push-pull perfusion probe, pull sample from the probe, perform on-line derivatization with o-phthaldialdehyde, and push derivatized amino acids into the flow-gated injector of a high-speed capillary electrophoresis-laser-induced fluorescence instrument. Peristalsis was achieved by sequential actuation of six, 200 µm wide by 15 µm high control valves that drove fluid through three fluidic channels of equal dimensions. Electropherograms with 100 000 theoretical plates were acquired at ∼20-s intervals. Relative standard deviations of peak heights were 4% in vitro, and detection limits for the excitatory amino acids were ∼60 nM. For in vivo measurements, push-pull probes were implanted in the striatum of anesthetized rats and amino acid concentrations were monitored while sampling at 50 nL/min. o-Phosphorylethanolamine, glutamate, aspartate, taurine, glutamine, serine, and glycine were all detected with stable peak heights observed for over 4 h with relative standard deviations of 10% in vivo. Basal concentrations of glutamate were 1.9 ( 0.6 µM (n ) 4) in good agreement with similar methods. Monitoring of dynamic changes of neurotransmitters resulting from 10-min applications of 70 mM K+ through the push channel of the pump was demonstrated. The combined system allows temporal resolution for multianalyte monitoring of ∼45 s with spatial resolution 65-fold better than conventional microdialysis probe with 4-mm length. The system demonstrates the feasibility of sampling from a complex microenvironment with transfer to a microfluidic device for on-line analysis. Measurements of neurotransmitter concentrations in the extracellular space of the central nervous system are necessary for * Corresponding author. Telephone: (734)615-4363. Fax: (734) 615-6462 E-mail: [email protected]. † Department of Chemistry. ‡ Department of Physics. § Department of Pharmacology. 10.1021/ac0510033 CCC: $30.25 Published on Web 10/04/2005

© 2005 American Chemical Society

better understanding of neurochemical signaling involved in behavior, pharmacology, and pathophysiology.1 Microdialysis sampling is one of the most popular methods for such measurements because it enables simultaneous monitoring of multiple analytes, determination of basal concentrations, and relatively longterm measurements.2 The application of microdialysis is often limited by its temporal and spatial resolution. Typical temporal resolution, set by the time required to collect a detectable amount of analyte, is 10-30 min, precluding many types of experiments; however, recent advances in coupling microdialysis to capillary electrophoresis-laser-induced fluorescence (CE-LIF) have allowed up to 3-s temporal resolution, making this method much more compatible with behavioral and mechanistic studies.3-8 Even with this advance, however, spatial resolution remains problematic because microdialysis probes with active sampling length of 1-4 mm can only sample relatively large brain regions in small experimental animals such as rats.9 Two new approaches to in vivo sampling, low-flow push-pull perfusion and direct sampling, provide up to a 500-fold improvement in spatial resolution over microdialysis.10,11 Push-pull perfusion utilizes sampling probes constructed from two concentric or side-by-side tubes. Sample is extracted from one tube while artificial cerebrospinal fluid (aCSF) is delivered through the other tube to replace the sampled volume. Push-pull perfusion was originally practiced using 1-10 µL/min flow rates; however, the low-flow method uses 10-50 nL/min flow rates and consequently causes much less tissue damage and generates better spatial resolution.10 In direct sampling, extracellular fluid is pulled at 1-50 (1) Parrot, S.; Bert, L.; Mouly-Badina, L.; Sauvinet, V.; Colussi-Mas, J.; LambasSenas, L.; Robert, F.; Bouilloux, J.; Suaud-Chagny, M.; Denoroy, L.; Renaud, B. Cell. Mol. Neurobiol. 2003, 23, 793-804. (2) Davies, M. I.; Cooper, J. D.; Desmond, S. S.; Lunte, C. E.; Lunte, S. M. Adv. Drug Delivery Rev. 2000, 45, 169-188. (3) Paez, X.; Hernandez, L. Biopharm. Drug Dispos. 2001, 22, 273-289. (4) Zhou, S. Y.; Zuo, H.; Stobaugh, J. F.; Lunte, C. E.; Lunte, S. M. Anal. Chem. 1995, 67, 594-599. (5) Ciriacks, C. M.; Bowser, M. T. Anal. Chem. 2004, 76, 6582-6587. (6) Lada, M. W.; Vickroy, T. W., Kennedy, R. T. Anal. Chem. 1997, 69, 45604565. (7) Bowser, M. T.; Kennedy, R. T. Electrophoresis 2001, 22, 3668-3676. (8) Lada, M. W.; Kennedy, R. T. Anal. Chem. 1996, 68, 2790-2797. (9) Myers, R. D.; Adell, A.; Lankford, M. F. Neurosci. Biobehav. Rev. 1998, 22, 371-387. (10) Kottegoda, S.; Shaik, I.; Shippy, S. A. J. Neurosci. Methods 2002, 121, 93101. (11) Kennedy, R. T.; Thompson, J. E.; Vickroy, T. W. J. Neurosci. Methods 2002, 114, 39-49.

Analytical Chemistry, Vol. 77, No. 21, November 1, 2005 7067

Figure 1. Monolithic structure formed by multilayer soft lithography of PDMS. Control channels molded in the upper hard layer are sealed against a soft layer with fluidic channels molded into them. The fluidic channels are rounded so they close completely upon application of pressure to the control channel effectively forming a valve.

nL/min flow rates into a single fused-silica capillary implanted in the tissue of interest. The capillary is then removed and fluid analyzed for transmitters. While this seminal work illustrated the potential of these approaches for miniaturized sampling, their initial implementation was of limited practicality because fraction collection with off-line analysis was used, which resulted in disadvantages such as low temporal resolution, loss of real-time monitoring, limited sampling times, and necessity of manual manipulation of low-volume samples. One method of achieving sampling with on-line analysis is to use an in-line pump between the sampling probe and the injector of the instrument so that sample can be withdrawn and analyzed without interruption of sampling. A pump for this application must deliver low nanoliter per minute flow rates, have nanoliter internal volume, and be compatible with biological fluids. Several microfluidic pumps, including electrokinetic, diaphragm, and peristaltic, have been designed that have the potential for satisfying the requirements for in vivo sampling.12-14 In this work, we explored the use of a peristaltic pump fabricated by multilayer soft lithography of poly(dimethylsiloxane) (PDMS) because such pumps can provide flow between 10 and 140 nL/min that is independent of the ionic composition of the pumped liquid and is therefore compatible with high ionic strength extracellular fluids.14-17 Furthermore, multilayer soft lithography offers the opportunity to incorporate sample treatments such as derivatization onto the chip. In multilayer soft lithography, channels with micrometer width and depth are formed on separate levels within a structure by sealing together layers of PDMS with different elasticities that have had trenches molded into them. Elasticity of PDMS is tuned by adjusting the ratio of curing agent to base used in polymerization. Figure 1 illustrates a typical design in which a thin layer of soft PDMS is sandwiched between two hard layers to form channels oriented perpendicularly and separated by a thin elastic (12) Studer, V.; Pepin, A.; Chen, Y.; Ajdari, A. Analyst 2004, 129, 944-949. (13) Sin, A.; Reardon, C. F.; Shuler, M. L. Biotechnol. Bioeng. 2003, 85, 359363. (14) Unger, S. R.; Chou, H.; Thorsen, T.; Scherer, A.; Quake, S. R. Science 2000, 288, 113-116. (15) Fu, A. Y.; Chou, H. P.; Spence, C.; Arnold, F. H.; Quake, S. R. Anal. Chem. 2002, 74, 2451-2457. (16) Hong, J. W.; Quake, S. R. Nat. Biotechnol. 2003, 23, 1179-1183. (17) Quake, S. R.; Scherer, A. Science 2000, 290, 1536-1540.

7068

Analytical Chemistry, Vol. 77, No. 21, November 1, 2005

membrane. Application of pneumatic pressure to channels in the hard layer (control channels) causes the membrane to expand, crimping the channels below (fluidic channels) closed and effectively forming a valve. Three or more of these valves actuated sequentially will pump by peristalsis.14 Such pumps have been used to recycle media for cell cultures17 and to lyse cells on-chip.18 In this work, we have explored the possibility of using a microfluidic pump for driving push-pull perfusion sampling from the brain of live rats and coupling the sampling system on-line to CE-LIF for measurement of amino acid neurotransmitters. Sampling flow rates of 40 nL/min allowed for the collection of electropherograms every 20-24 s that resolved several neuroactive amino acids. Push-pull perfusion also provides means for chemical delivery during sampling, which was demonstrated by measuring dynamic changes caused by perfusion of aCSF with high K+ concentration in vivo. The system combines the advantages of high temporal resolution microdialysis with the improved spatial resolution of low-flow push-pull perfusion sampling in an automated format. MATERIALS AND METHODS Chemicals and Materials. Sodium tetraborate, β-mercaptoethanol (βME), hydroxypropyl-β-cyclodextran (HP-β-CD), ophthaldialdehyde (OPA), o-phosphorylethanolamine (o-PEA), glutamate, aspartate, γ-aminobutyric acid (GABA), dopamine, taurine, serine, glycine, and chloral hydrate were purchased from Sigma (St. Louis, MO). AZ 9260 photoresist and AZ 400K developer were purchased from Cariant (Somerville, NJ). Silicon wafers were obtained from International Wafer Service Inc. (Portola Valley, CA), and PDMS (RTV615A and RTV615B) was purchased from G.E. Silicones (Waterford, NY). All aqueous solutions were prepared using water purified by a Millipore Milli-Q filtration system (Milford, MA), and vacuum filtered through 0.2µm nylon membrane filters. All fused-silica tubing was purchased from Polymicro Technologies (Phoenix, AZ). Chip Design. The microfluidic chip was designed with three, 200 µm wide by 15 µm deep fluidic channels actuated by six, 200 µm wide by 15 µm deep control channels (Figure 2A). One fluidic channel drives aCSF from a reservoir to the “push” arm of the push-pull sampling probe. A second channel removes sample from the “pull” arm of the sampling probe. This channel merges with a third channel that mixes OPA derivatization reagent with the sample. The derivatized amino acids are pumped into the flowgated injector of the high-speed CE-LIF instrument. All three fluidic channels are operated by the same set of control channels. The control channels are actuated by sequential pneumatic pulses delivered by solenoid valves. TTL pulse generating electronics were developed to drive the solenoid valves with a 60° pulse sequence between 10 and 15 Hz. Chip Fabrication. Photomasks of the chip design were created with AutoCAD and printed on an acetate sheets using a laser printer.19 AZ 9260 photoresist was spin coated to a uniform thickness of 15 µm on a 3-in. silicon wafer, baked for 20 s at 110 °C, and exposed to UV light for 10 s through the photomask. The photoresist was developed in a 3:1 solution of water/AZ 400K (18) Wu, H.; Wheeler, A.; Zare R. N. Proc. Natl. Acad. Sci. U.S.A. 2004, 101, 12809-12813. (19) Duffy, D. C.; McDonald, J. C.; Schueller, O. J. A.; Whitesides, G. M. Anal. Chem. 1998, 70, 4974-4984.

Figure 2. Microfluidic chip and connections. (A) The microfluidic chip designed for this application uses six control channels (gray) to actuate valves controlling pumping through three fluidic channels (black). Circles indicate access holes for control (solid) and fluidic (empty) channels. White arrows show the direction of fluid flow through the chip. The chips are 1.2 cm wide by 2.5 cm long and 0.8 cm thick before encasing in the final layer of PDMS. (B) Side view of chip illustrating connections to fluidic and control channels. Only one connection of each type is shown for clarity.

developer, air-dried, and baked at 200 °C for 30 min to round the channels.14 Although it is necessary to round only the fluidic channels for complete valve closure, both the fluidic and control channel molds were baked. PDMS components RTV615A and RTV615B were mixed in a 5:1 ratio for the two hard layers and 25:1 for the soft layer. Vacuum was pulled over the PDMS mixtures for 60 min to remove trapped air. The hard-layer mixtures were poured over the control channel mold and a blank silicon wafer to a depth of 3 mm.20 The softlayer mixture was spun onto the fluidic channel mold to a thickness of 30 µm to create the membrane. The resulting wafers were baked at 80 °C for 90 min to partially cure the rubber. After cooling, the PDMS layer with control channels was cut from its mold and holes were punched with an 18-gauge needle to provide external access to the channels. The control layer was aligned and placed over the fluidic channels of the PDMS membrane. The layers were baked at 80 °C for 60 min to form an irreversible bond between them. The bonded layers were removed from the silicon wafer, and access holes were punched with a 20-gauge needle for the fluidic channels. The fluidic channels were sealed with a blank PDMS hard layer and baked overnight at 80 °C. For pneumatic connections to the control channels, 1-cm lengths of 22-gauge stainless steel tubing were inserted 4-5 mm into 6-cm-long, 1/16-in. Tygon tubing. The exposed stainless steel tube was then inserted into the access holes punched for the control channels (Figure 2B). The Tygon tubing was coupled directly to the nitrogen delivery manifold. Connections to fluidic channels were made with 75 µm i.d. by 360 µm o.d. fused-silica (20) Chen H.; Acharya, D.; Gajraj, A.; Meiners, J.-C. Anal. Chem. 2003, 75, 52875291.

Figure 3. Overview of instrument. The microfluidic chip is placed in-line between the push-pull perfusion probe and the flow-gated injector of the CE instrument. Solenoid valves and a N2 pressurevacuum manifold are employed to actuate control channels in the chip. Fluid pulled from the sample is derivatized with OPA on the chip and transferred through a capillary to a flow-gated CE-LIF instrument. The continuous flow of derivatized sample is serially analyzed by CE-LIF.

capillaries inserted in holes punched during fabrication. Capillaries were cut to 16-cm lengths for the aCSF inlet, OPA inlet, and instrument outlet connections. The aCSF outlet and sampling inlet capillaries for the push-pull perfusion probe were cut to 12 cm. (These capillaries were eventually connected to a 4-cm pushpull perfusion probe, so that their overall length was also 16 cm during sampling.) Finally, the entire device was sealed in a casing of 5:1 PDMS and baked for 2 h to form watertight seals around the external connections. Chips were primed with 70% ethanol at flow rates of 2-4 µL/ min for 30 min to eliminate trapped air prior to filling with aCSF. For push-pull perfusion, the chip was actuated pneumatically with pulses of nitrogen at 11.5 psig delivered at a frequency of 12 Hz. An applied vacuum of 350 mmHg was used to evacuate nitrogen from the control channels between pulses to allow for faster valve reopening.14 Solenoid valves (LHDA1211111H, The Lee Co., Westbrook, CT) and a manifold were used to switch between nitrogen pressure and vacuum for each control channel (Figure 3). Derivatization and CE-LIF. For CE measurements, the chip was placed in-line between the sampling probe and flow-gated injector of the CE-LIF instrument (Figure 3). Amino acids were derivatized on-line in the microfluidic chip to form fluorescent isoindole products with 10 mM OPA, 40 mM βME, 36 mM borate, 0.81 mM HP-β-CD, and 10% methanol as the derivatization solution. Derivatized samples were analyzed on-line by CE-LIF using a previously described instrument with a few modifications.7 Sample was pumped to a flow-gated interface that controlled injections onto the separation capillary.7,21 Injections were performed at 3 kV for 800 ms every 20-24 s. Although individual separations were 10-12 s long, 20-24 s was required between injections to allow enough time for derivatized sample to fill the flow gate prior to injection. Separations were performed using 2222 (21) Hooker, T. F.; Jorgenson, J. W. Anal. Chem. 1997, 69, 4134-4142.

Analytical Chemistry, Vol. 77, No. 21, November 1, 2005

7069

V/cm applied over a 9-cm-long, 10 µm i.d. by 150 µm o.d. fusedsilica capillary. Electrophoresis buffer was 40 mM borate, 0.9 mM HP-β-CD adjusted to pH 9.5 using 1 M NaOH. In some preliminary experiments, 10 mM borate was used with similar additives. LIF detection was performed off-column in a sheath-flow cuvette using the 355-nm line of a diode pumped solid-state laser from DPSS Lasers Inc.(San Jose, CA) for excitation. Fluorescence was collected at 450 nm using a 60× objective through a 450 ( 25 nm band-pass filter and detected using a photomultiplier tube. A 40 mM borate solution at pH 9.5 was used for the sheath-flow buffer. Push-Pull Perfusion. Side-by-side push-pull perfusion probes were fabricated with two, 4-cm-long, 50 µm i.d. by 125 µm o.d. fused-silica capillaries cemented inside a 2-cm length of 250 µm i.d. by 360 µm o.d. capillary with cyanoacrylate, so that 1 mm of the smaller capillaries extended beyond the lumen of the larger capillary for sampling. Two, 1-cm, 250 by 360 µm capillaries were glued to the other end of the 50 by 125 capillaries to allow the probe to be connected to other 360-µm-o.d capillaries. Some in vivo experiments utilized a concentric push-pull probe consisting of a fused-silica capillary threaded through a 27-gauge needle as described previously.10 The pump was used to deliver aCSF consisting of 145 mM NaCl, 2.68 mM KCl, 1.01 mM MgSO4, 1.22 mM CaCl2, 1.55 mM Na2HPO4, and 0.45 mM NaH2PO4 at pH 7.4 through the push arm of the probe. K+ stimulations were performed by 10-min perfusions of high-K+ aCSF (73 mM NaCl, 75 mM KCl, 1.01 mM MgSO4, 1.22 mM CaCl2, 1.55 mM Na2HPO4, and 0.45 mM NaH2PO4 pH 7.4) through the push arm of the probe. Surgical Procedures. The 250-350-g male Sprague-Dawley rats were anesthetized with 200 mg/mL chloral hydrate and placed in a stereotaxic frame for probe implants and experiments. The probe was inserted in the striatum at 1.0 mm anterior of bregma, 2.8 mm lateral of midline, and -6.0 mm deep from dura. After probe insertion, the syringe pump delivering aCSF was disconnected, and the probe was connected to the microfluidic sampling chip. Actuation of the chip was started immediately after probe implantation. RESULTS AND DISCUSSION Pump Design. Previous descriptions of PDMS-based peristaltic pumps provided the basis for pumps used here;14 however, satisfying the demands of nanoliter per minute sampling and analysis from the brain of live animals required several design considerations. The fluidic channel dimensions were chosen to generate the flow rates desired. Three fluidic channels were incorporated to allow for push-pull perfusion sampling and online derivatization. Derivatization on the chip eliminated the need to have a separate, off-chip tee and pump for labeling analytes. To minimize the number of external connections to the chip, a single active control area was chosen to drive fluid through all three channels simultaneously. Six control channels were employed because pilot experiments suggested that this arrangement was superior to a three-channel design for generating sufficient pressure to pump through 30-cm lengths of fused-silica capillary tubing necessary for off-chip sampling, separation, and detection. Channel lengths of 12 mm were chosen to minimize the footprint of the chip while still allowing room for external connections to the sample and analytical system. Proper connections to the fluidic channels and coating the entire device in PDMS after making the 7070 Analytical Chemistry, Vol. 77, No. 21, November 1, 2005

Figure 4. Effect of operational parameters on flow rate through microfluidic pump. (A) Typical result for actuation frequency effect on flow rate with control channels actuated by 13 psig nitrogen and 250 mmHg vacuum. (B) Average result for effect of nitrogen pressure with valves actuated at 10 Hz and 250 mmHg of vacuum. (C) Average result when measuring effect of vacuum with valves actuated at 10 Hz with 15 psig nitrogen. All measurements were from a single chip with errors reported (1 standard deviation (n ) 3).

connections were critical to prevent leaks at the flow rates and back pressures necessary for this application (Figure 2B). With the design described in the Experimental Section, continuous pumping could routinely be performed for over 6 h without leaks or air entering the fluidic channels. Pump Characterization. The dependence of flow rates generated by the chips was tested as a function of valve actuation frequency, applied nitrogen pressure, and applied vacuum. For these experiments, the volumetric flow rate from the converging channels was determined by measuring the mass of water collected at the capillary outlet over 30 min. Flow rate had a linear dependence on frequency from 5 to 25 Hz, corresponding to 25115 nL/min per channel, as expected (Figure 4A). Loss of linearity at higher frequencies has been attributed to incomplete valve reopening with reduced time between pulses.14 Using a light microscope, it was observed that nitrogen pressures of greater than 10 psig were necessary for complete closure of the valves. Increasing applied pressure from 11 to 17 psig nitrogen increased flow rates from 78 to 140 nL/min (Figure 4B). These data are

consistent with a larger fluidic displacement caused by an increased area of valve closure at higher pressures. The flow rate reached a maximum at higher pressures most likely because each valve has a finite cross-sectional area (0.04 µm2), which would place a pressure-independent upper limit on displacement. Application of vacuum to control channels between pulses of pressure was not necessary for pumping; however, at moderate vacuum it allowed for generation of higher flow rates as shown in (Figure 4C). This effect is likely due to faster valve reopening by aiding in the evacuation air from the control channels.16 The cause of flow rate decrease with applied vacuum of >150 mmHg was not investigated; however, a possible explanation is that, above 150 mmHg, the vacuum is great enough to pull the PDMS membrane into the control channel. The increased volume associated with the valve opening too far may pull some fluid back causing the lower flow rates. The flow rates from both chip outlets were measured simultaneously to ensure that all three channels operated at the same flow rate. It was found that actuation at 12 Hz with 11.5 psig nitrogen and 350 mmHg of applied vacuum generated a flow rate of 47 ( 5 nL/min from the aCSF delivery channel and 88 ( 11 nL/min from the instrument channel (n ) 4). Because two fluidic channels drive flow out of the instrument channel, one would expect its flow rate to be about twice that of the aCSF delivery channel, as we observed. These actuation conditions were used for all in vivo measurements because it has been shown that flow rates of 10-50 nL/min are optimal for low-flow-rate push-pull perfusion.10 Coupling to CE-LIF. The complete system shown in Figure 3 was tested in vitro by sampling and analyzing amino acid standards. A flow rate of 47 nL/min per channel provides 14 min from initial mixing of derivatization buffer and sample to injection onto the CE-LIF instrument. This reaction time is longer than optimal because only ∼60 s is required for OPA to react with primary amines and the half-life for some of the fluorescent products, such as glycine and GABA, can be as short as 5 min before the isoindole product begins to degrade.6,22,23 (Because the on-line system maintains a constant time from derivatization to injection, any degradation that does occur should not have an adverse effect on quantification.) In principle, the transfer time can be decreased by decreasing the dead volume in the capillaries connecting the chip to the CE instrument; however, shortening the capillaries makes connections more difficult and smaller inner diameters increase the back pressure. A typical electropherogram collected during in vitro sampling of a solution of amino acids is illustrated in Figure 5A. Separation efficiencies of 80 000-150 000 theoretical plates (measured using statistical moments) were achieved when sampling with the chip. The efficiency is lower than previously reported with this instrument,7 but larger injection volumes were used in this case causing band broadening. Calibration curves were linear from 0.6 to 10 µM (R2 values >0.99). Glutamate and aspartate detection limits were 60 ( 4 and 57 ( 6 nM, respectively (n ) 4). To test the system for monitoring concentration changes, electropherograms were collected while performing step changes in concentration of glutamate and aspartate at the sampling probe (22) Allison, L. A.; Mayer, G. S.; Shoup, R. E. Anal. Chem. 1984, 56, 10891096 (23) Lindroth, P.; Mopper, K. Anal. Chem. 1979, 51, 1667-1674.

Figure 5. In vitro sampling and analysis with chip and on-line CELIF. (A) Electropherogram showing resolution of selected amines that were sampled and derivatized on-line at 6 µM using the system illustrated in Figure 3. Electrophoresis was performed using 2222 V/cm over a 9-cm column. Borate concentration in the electrophoresis buffer was 10 mM. Amino acid peaks are indicated by standard threeletter abbreviation. (B) Temporal response of the system was evaluated by continuously recording electropherograms at 24-s intervals while sampling solutions containing glutamate and aspartate in aCSF. The plot shows peak heights from individual electropherograms as the concentrations were changed as indicated. The times for changes in concentrations are corrected for the dead volume of the system.

(Figure 5B). While the concentration was stable, the relative standard deviation for glutamate and aspartate peak height was 4%. When the concentration was changed while sampling at 47 nL/min per channel, a delay of 27 ( 3 min was observed before the electropherograms recorded the change in concentration. The predicted delay through the 650-nL sampling arm and 1500-nL transfer arm of the probe is 28 min assuming 47 and 88 nL/min flow rates in each arm, respectively. The good agreement of measured and predicted times demonstrate that the connections did not generate significant, unexpected leaks or dead volumes. The response time, defined as the time required for signal to change from 10 to 90% of the maximal signal intensity when the sample concentration changes, was 51 ( 5 s (n ) 8) (see Figure 5B). This response time, which sets the temporal resolution for chemical monitoring, is attributed to flow and diffusional broadening associated with mass transport through the chip and is only ∼2-fold worse than that measured using microdialysis coupled with this CE instrument.7 These initial results suggest that the push-pull perfusion sampling chip offers adequate resolution, stability, and sensitivity for in vivo monitoring of biogenic amines in the brain. Analytical Chemistry, Vol. 77, No. 21, November 1, 2005

7071

Figure 6. On-line electropherogram recorded for sample collected by push-pull perfusion at ∼50 nL/min from the striatum of an anesthetized rat. Borate concentration in the electrophoresis buffer was 40 mM. Peaks were identified by matching migration times with standards. Peak labels are o-PEA, taurine (Tau), and standard threeletter amino acid codes.

Recovery of the push-pull perfusion probes was measured to determine the relationship between the measured concentration and the actual concentration in the sampled solution by comparing peak intensities when sampling through the probe to those collected through direct injection of derivatized amino acid standards using a syringe pump. Recoveries were 81 and 76% for glutamate and aspartate, respectively, confirming the 70-80% previously reported for low-flow push-pull perfusion.10 Absorption of molecules into PDMS can be problematic because of its hydrophobicity and permeability; however, it was unclear if measurement of the relatively polar and charged amino acids would be affected by passing through the PDMS chip.24,25 The rapid temporal response shown in Figure 5B suggests that absorption and release of the amino acids from the PDMS was not a serious detriment to the performance of the chip. To further assess the extent of sample absorption, standards of glutamate, dopamine, GABA, taurine, serine, and glycine at concentrations of ∼1 µM each were directly injected into the flow-gated interface of the CE-LIF with a syringe pump at 47 nL/min per channel, employing either a stainless steel reaction tee or the PDMS chip as a reaction tee. In comparing the signals of amino acids for these two conditions, most were higher by 1-6% with the PDMS tee. The only exceptions were glutamate, reduced by 8%, and GABA, reduced by 1%. After this experiment, the PDMS tee was rinsed with buffer and then imaged by fluorescence microscopy. No detectable fluorescence was observed in or near the channels. These data combined suggest that adsorption was minimal for the amino acids under the conditions used. In Vivo Monitoring. The sampling and on-line analysis system was tested for in vivo measurements by collecting electropherograms while sampling from the striatum of anesthetized rats. A typical electropherogram is illustrated in Figure 6. These data are qualitatively similar to those collected with microdialysis7,11 allowing for measurements of o-PEA, glutamate, aspartate, taurine, glutamine, serine, and glycine in vivo every 20-24 s. Peaks were identified by matching migration times from standards analyzed on the same day. Migration time varied by 0.7% within 1 day and by 3.3% interday. (24) Lee, N. J.; Park, C.; Whitesides, G. M. Anal. Chem. 2003, 75, 6544-6554. (25) Merkel, T. C.; Bondar, V. I.; Nagai, K.; Freeman, B. D.; Pinnau, I. J. Polym. Sci. B 2000, 38, 415-434.

7072 Analytical Chemistry, Vol. 77, No. 21, November 1, 2005

Figure 7. Stability of peak heights for glutamate and aspartate during in vivo sampling by push-pull perfusion at 50 nL/min with online CE-LIF. Electropherograms were continuously recorded at 20-s intervals for 170 min beginning immediately after the probe was implanted in the striatum of an anesthetized rat. After the basal levels stabilize, the RSD of the signal is 7% and 10% for glutamate and aspartate, respectively.

Figure 7 illustrates continuous monitoring of glutamate and aspartate immediately after implant. With continual collection of electropherograms, the microfluidic chip allowed for stable measurements of basal amino acid levels for over 4 h. The decay in signal over the initial 20-30 min is due to trauma during probe implantation that causes elevation of transmitter levels before a return to normal basal levels. This decay is also seen with microdialysis if sampling is begun immediately after microdialysis probe implantation.8 After the signal stabilized, the relative standard deviation of glutamate and aspartate were 7 and 10%, respectively, which is typical for in vivo experiments with this CELIF instrument. Basal concentrations measured using this system were 1.2 ( 0.3, 1.9 ( 0.6, 0.8 ( 0.3, 0.7 ( 0.3, 2.6 ( 0.4, 3.3 ( 1.2, and 1.0 ( 0.1 µM for o-PEA, glutamate, aspartate, GABA, taurine, serine, and glycine, respectively (n ) 4). Of these amino acids, glutamate has been studied the most, allowing quantitative comparison of concentrations measured under similar conditions. Glutamate concentration measured in the striatum of anesthetized rats was 2.0 ( 0.7 µM by push-pull perfusion with off-line analysis,10 1.8 ( 0.1 µM by direct sampling with off-line analysis,11 and 1.7 ( 0.1 µM by calibrated microdialysis sampling.11 The good agreement indicates that the on-line method yields comparable accuracy to these other methods. The precision of the on-line method is slightly worse than that achieved by the other methods. More study will be required to determine the sources of variation; however, possible factors may include variations in flow rate during operation of the chip and that data were pooled from two different sampling probe designs. The concentration of o-PEA has previously been suggested to be a good indicator of membrane disruption and therefore tissue damage caused by probe insertion and sampling with higher concentrations corresponding to more damage.26 o-PEA concentrations with push-pull perfusion were higher than with direct sampling but significantly lower than microdialysis (Figure 8). These results are in agreement with expectations of the effects of the different sampling methods. Push-pull perfusion collects (26) Uchiyama-Tsuyuki, Y.; Araki, H.; Yae, T.; Otomo, S. J. Neurochem. 1994, 62, 1074-1078.

Figure 8. Comparison of o-PEA and glutamate concentrations measured by low-flow-rate microdialysis,11 on-line push-pull perfusion (this work), and direct sampling11 in the striatum of anesthetized rats (n ) 4). * indicates that statistical analyses performed with one-way ANOVA at 95% confidence interval revealed that o-PEA concentrations are significantly different from low-flow microdialysis and bracketed bars.

samples from directly below the implant and therefore is presumably less impacted by tissue disruption than microdialysis, which samples along the track of implantation. Likewise, the combined effects of larger probe size and fluid movement associated with push-pull perfusion may be expected to cause slightly more tissue damage than direct sampling resulting in higher o-PEA levels. Tissue damage with push-pull perfusion may be further limited by using smaller probe tips and lower flow rates, both of which are feasible. Push-pull perfusion offers greater spatial resolution over microdialysis because sampling is performed at the tip of the probe instead of over an active length of 1-4 mm. The active surface area of a 4-mm-long microdialysis probe is 65 times greater than that of push-pull perfusion probes used in this investigation. The improved spatial resolution should enable sampling from smaller brain regions or to sample from smaller species, such as mice. The ability to sample from mice would be useful because gene manipulation technology in mice has allowed development of numerous disease models. Spatial resolution by push-pull perfusion could be further improved by etching the capillary tips or using smaller capillaries. An important advantage of push-pull perfusion is that it allows chemicals to be delivered to a local area while observing their effect on species of interest. As a demonstration of this capability, 75 mM K+ was pumped through the push-pull probe for 10 min while the amino acids were monitored by on-line CE-LIF. Elevated K+ concentrations depolarize neurons, stimulating neurotransmitter release and increasing their concentrations in the extracellular space.27 Figure 9A illustrates a typical recording of glutamate during an individual infusion. Glutamate and aspartate increased 330 ( 40 and 210 ( 30% (n ) 4), respectively, over basal concentrations upon the application of 75 mM K+ (Figure 9B). The K+ infusion experiment also allowed an estimation of the flow rates in vivo. Based on the internal volume of the system and the in vitro flow rate (47 nL/min per channel), we expected a delay from initiation of K+ to detection of increased amino acid of 54 min. The observed delay was found to be 45 ( 7 min (n ) (27) Herrera-Marschitz, M.; You, Z. B.; Goiny, M.; Meana, J. J.; Silveira, R.; Godukhin, O. V.; Chen, Y.; Espinoza, S.; Pettersson, E.; Loidl, C. F.; Lubec, G.; Andersson, K.; Nylander, I.; Terenius, L.; Ungerstedt, U. J. Neurochem. 1996, 66, 1726-1735.

Figure 9. Dynamic measurements by low-flow push-pull perfusion with on-line CE-LIF in striatum of anesthetized rats. (A) Glutamate response, measured as peak height in individual electropherograms recorded at 24-s intervals, during 10-min infusions of 75 mM K+ through the push channel. Bars indicate application of the K+ corrected for dead volume of the system. (B) The averaged responses of glutamate and aspartate to 10-min infusions of 75 mM K+. The responses were normalized to peak heights recorded under basal conditions, i.e., before K+ infusion. Error bars show standard error of the mean of the measurements (n ) 4).

3), suggesting that the in vivo flow rate was slightly higher than the measured flow rate but still in an acceptable range for lowflow-rate push-pull perfusion. CONCLUSIONS This work demonstrates novel application and utility of PDMS pumps fabricated using multilayer soft lithography. Previously such devices have been used to sample from reservoirs or transfer fluids within chips. Their use for external sampling will facilitate many potential applications by allowing sophisticated manipulation and analysis of nanoliter samples collected from complex microenvironments. In the present study, the microfluidic pump allows in vivo sampling by low-flow push-pull perfusion and amine derivatization to be coupled on-line with CE-LIF analysis. The on-line method significantly improves the utility of low-flow pushpull sampling by eliminating manual fraction collection and increasing temporal resolution while maintaining the advantage of high spatial resolution and high recovery compared to microdialysis. The use of microfluidic pump technology for sampling may also provide other improvements for in vivo measurements such as more complex sample manipulation and on-line direct sampling. ACKNOWLEDGMENT This work was supported by NIH NS38476 (R.T.K.). Received for review June 6, 2005. Accepted September 2, 2005. AC0510033 Analytical Chemistry, Vol. 77, No. 21, November 1, 2005

7073