On-chip microdialysis system with flow-through sensing components

Biosensors and Bioelectronics 22 (2007) 2422–2428

On-chip microdialysis system with flow-through sensing components Yi-Cheng Hsieh a,1 , Jeffrey D. Zahn b,∗ a b

Bioengineering Department, Pennsylvania State University, B18 Hallowell Building, University Park, PA 16802, United States Bioengineering Department, Pennsylvania State University, 224 Hallowell Building, University Park, PA 16802, United States Received 9 May 2006; received in revised form 16 August 2006; accepted 25 August 2006 Available online 17 October 2006

Abstract Microdialysis probes have been used for diabetes treatment as continuous monitoring system coupled to a glucose sensor. An on-chip microdialysis system with in-line sensing electrodes is demonstrated. As a first step towards greater biosensor integration with this miniaturized microdialysis system, a stacked system with in-line sensing electrodes was developed. Impedance electrodes sputtered within the microchannels were used to determine fluid electrical resistance from a dialyzed phosphate buffered saline (PBS) solution, which characterizes solution conductivity as a function of PBS concentration. The permeability of the membrane to the salt ions was obtained as 0.246 ± 0.028 ␮m/s (15 nm pores). Subsequently, experiments measuring PBS dialysis in the time-domain at 64.4% recovery were conducted. The PBS concentration of the reservoir was changed in both a step response and sinusoidally with an 800 s period. The subsequently measured impedance indicates that the system is able to continuously track concentration changes in the reservoir with a 210 s system response delay. Most of this delay is due to the dead volume within the tubing between the syringe pumps and the microsystem. In addition, the predicted response was modeled using linear systems theory and matches the experimental measurements (r = 0.98). This system is expected to have the proper sensitivity to track physiologically relevant concentration changes of biomolecules such as glucose (which has a physiological maximum change rate of ∼4 mg/dl min with a periodicity of 1 h or greater) with minimal lag time and amplitude reduction. © 2006 Elsevier B.V. All rights reserved. Keywords: Microdialysis; Glucose; Diabetes; Impedance; Microfluidics

1. Introduction Diabetes mellitus is a clinically heterogeneous group of disorders characterized by elevated blood glucose levels resulting from deficiency in insulin secretion or insulin resistance, or a combination of both. There are two major types of diabetes. Type 1 diabetes comprises about 5–10% of cases in the diabetes mellitus syndrome. The onset is usually abrupt and occurs at a young age. Type 1 diabetes is caused by autoimmune ␤-cell destruction in the pancreas leading to loss of insulin secretion and absolute insulin deficiency. Patients with type 1 diabetes rely on exogenous insulin to sustain life and are prone to ketosis, which is characterized by large amounts of ketones in the body, secondary to excessive breakdown of fat caused by insuf∗

Corresponding author. Tel.: +1 814 865 8090; fax: +1 814 863 0490. E-mail addresses: [email protected] (Y.-C. Hsieh), [email protected] (J.D. Zahn). URL: http://biomems.bioe.psu.edu/, http://biomems.bioe.psu.edu/. 1 Tel.: +1 814 865 6744; fax: +1 814 863 0490. 0956-5663/$ – see front matter © 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.bios.2006.08.044

ficient insulin in a person with diabetes mellitus. Type 2 diabetes comprises about 90–95% of cases in the diabetes mellitus syndrome. It results from insulin resistance in muscle, liver and adipose tissue. Patients may require insulin for correction of fasting hyperglycemia and are usually not prone to ketosis. Although type 2 diabetes mellitus is strongly suggested to be genetically associated, some non-genetic factors including increasing age, high caloric intake, obesity, central adiposity and sedentary lifestyle are also identified (LeRoith et al., 2000). According to American Diabetes Association, there are 18.2 million people in the United States, or 6.3% of the population, who have diabetes. The total estimated cost of diabetes to the USA in 2002 was $132 billion and the number of diabetics is expected to rise 44% in the next 20 years (American Diabetes Association, 2003). In diabetics, the consistently high glucose levels results in long-term complications including retinopathy, nephropathy and neuropathy which often leads to amputation of extremities. The 10 years Diabetes Control and Complications Trial (DCCT) was designed to determine whether intensive glycemic control affects the complications in type 1 diabetes mellitus (DCCT,

Y.-C. Hsieh, J.D. Zahn / Biosensors and Bioelectronics 22 (2007) 2422–2428

1993; LeRoith et al., 2000). More than 1400 patients with type 1 diabetes participated in this project. The results showed that intensive therapy reduced the development of diabetic long-term complications. This study reinforced the role of intensive insulin therapy for glycemic control. Results from the United Kingdom Prospective Diabetes Study (UKPDS) have also demonstrated the benefits of intensive insulin therapy for type 2 diabetics (UKPDS, 1998; LeRoith et al., 2000). However, one of the disadvantages of tight glycemic control was a tripling of the risk of hypoglycemic incidences highlighting the need for a feedback controlled insulin infusion system or ‘artificial pancreas’. One of the critical components of such a system is a continuous microdialysis based glucose sensing system (Heinemann, 2003; Lodwig and Heinemann, 2003; Kubiak et al., 2004). Microdialysis is a continuous sampling technique based on controlling the mass transfer rate of small molecules across a semipermeable membrane while excluding the larger ones. Microdialysis systems are usually placed (inserted or implanted) inside the tissue of interest with an isotonic perfusion fluid flowing through the system and diffusional exchange occurring between the perfusate and the surrounding interstitial fluid (ISF) (Rosdahl et al., 1998; Heinemann, 2003; Petrou et al., 2003; Poscia et al., 2003; Schoonen and Wientjes, 2003; Ekberg et al., 2005). Since the dialysis process does not change or affect the surrounding fluid, it is viewed as a tool for continuous monitoring. This paper highlights the work towards a miniaturized microdialysis monitoring system with continuous glucose sensing capability. Since glucose sensors can be built based on thin film fabrication, it is easy to integrate the sensing part with an on-chip microdialysis system (Lambrechts and Sansen, 1992; Gooding et al., 1998; Yang et al., 2002; Fang et al., 2003). As a first step towards this goal, continuous sensing is demonstrated by measuring changes in solution conductivity in a dialyzed solution. A microdialysis system with gold sensing electrodes for ion impedance sensing is presented here (Ayliffe et al., 1999; Collins and Lee, 2004). 2. Theory By controlling the mass transfer rate of molecules of different sizes across a semipermeable membrane, microdialysis is considered to be a continuous sampling technique. Microdialysis systems can be roughly categorized into microdialysis probes and microdialysis chips. The commercially available probes are inserted into the tissue of interest directly. In diabetes treatment, these probes are usually inserted subcutaneously into either the abdomen or forearm, sampling glucose from the interstitial space (Steinkuhl et al., 1996; Rosdahl et al., 1998; Perdomo et al., 2000; Beyer et al., 2003; Petrou et al., 2003; Schoonen and Wientjes, 2003; Ekberg et al., 2005). For the second design, a commercially available dialysis membrane is usually used and sandwiched between two microfabricated chips (Kuo et al., 2003; Pan et al., 2003). Other microchip-based systems used a photopatterning technique to have a built-in nanoporous dialysis membrane within the device (Song et al., 2004). The concentration gradient between the perfusate and the component of interest in the ISF (e.g. glucose) is the driving

2423

force to transport molecules to the lumen of the probe. The concentration of the molecules of interest in the perfusion fluid at the output of the microdialysis probe can be correlated to the concentration within the ISF. The term, recovery, is the ratio of concentration of the component of interest, Cout , the outlet concentration, to C∞ , the bulk concentration in the surrounding fluid: recovery =

Cout C∞

(1)

Microdialysis probes have been used for diabetes treatment as continuous monitoring system coupled to a glucose sensor but can struggle to obtain high recoveries of analytes while the sampling probes and glucose sensors are fabricated as separate pieces and then assembled resulting in a large dead volume (Perdomo et al., 2000; Petrou et al., 2003; Poscia et al., 2003; Pijanowska et al., 2004; Ekberg et al., 2005). Even though the microdialysis chips have the advantage of easy integration, they are not designed to have direct contact with the sampled environment and still rely on extracting fluid which is then infused into the microsystem. In this study, we present a novel on-chip microdialysis system which solves the problems mentioned above. The device is based on thin-film fabrication, which is easily integrated with in situ biosensors, and direct polymer bonding onto the microfluidic channels are designed for direct contact between the dialysis membrane and tissue of interest. The integration of a biosensor directly with the microdialysis system is designed to allow high recovery of analytes with a smaller diffusional surface area and lower perfusion flow rates resulting in a less invasive, more precise microdialysis probe. In addition, the large diffusional surface area to microchannel volume ratio and short diffusional path will allows higher recoveries and faster equilibration times for higher frequency sampling rates. Even though the device was designed for direct contact with the tissue of interest, in order to create an environment for testing the device in the time domain more efficiently, a second fluidic channel was needed. Thus, the system is modeled as a two compartment systems with a reservoir channel bonded on top of the semipermeable membrane, as shown in Fig. 1a. By balancing the diffusional flux of material across the microdialysis membrane and the convective flux of the perfusion fluid, a relationship between the outlet concentration and perfusion flow rate is obtained as     Cr,out − Cd,out 1 1 ln = −K0 A (2) + Cr,in − Cd,in Qr Qd where Cr,in and Cr,out are the inlet and the outlet concentration of the reservoir channel. Cd,in and Cd,out are the dialysate inlet and outlet concentration of the perfusion flow. Qr is the reservoir channel flow rate and the Qd is the perfusion flow rate. Ko is the overall molecular permeability and A is the diffusional surface area. In this work, Cd,in is zero. Since the reservoir flow rate is much higher than the perfusion flow rate, Cr,in and Cr,out are assumed to be equal values, represented by C∞ . Cd,out is the measured Cout . In addition, when Qr is much larger than Qd

2424

Y.-C. Hsieh, J.D. Zahn / Biosensors and Bioelectronics 22 (2007) 2422–2428

Fig. 1. (a) Model of the microdialysis system. Schematic of the mass transfer. (b) The schematic of the stacked microdialysis system. The depth and width of PDMS channels are 30 and 1500 ␮m. The depth and width of SU-8 channels are 15 and 400 ␮m. (c) Electrodes, SU-8 and PDMS channels under brightfield microscopy. (d) Photo of a complete device.

(1/Qr  1/Qd ), then 1/Qr can be ignored on the right side of the equation. Therefore, after making these assumptions Eq. (2) can be simplified as     Cout 1 = −K0 A (3) ln 1 − C∞ Qd Eq. (3) can be rearranged to represent recovery as a function of perfusion flow rate, Qd : Cout = 1 − e−K0 A/Qd C∞

(4)

Also by plotting the left-hand side of Eq. (3), as a function of perfusion flow rate (1/Qd ), the slope of the line is K0 A, so the permeability of the membrane to the molecules of interest may be determined to characterize the system functionality. 3. Fabrication There are three main parts of the described device: the microdialysis monitoring chip, a polydimethylsiloxane (PDMS) reservoir channel piece and a PDMS mixer. Previously, a complete fabrication protocol for the microdialysis chip without sensing components has been developed (Hsieh and Zahn, 2005a,b). The current system is fabricated using a polycarbonate track-etch membrane (100 or 15 nm pore sizes) directly bonded onto SU-8

microfluidic channels with the gold resistance sensing electrodes patterned and sputtered in the bottom of the microchannels. A complete fabrication protocol for this device was developed with a non-leaking and reproducible bond between the SU-8 layer and the membrane. The gold sensing electrodes were first fabricated using liftoff techniques on clean glass slides. Shipley 1818 photoresist was used to define the electrode areas, separated by 40 ␮m, and ˚ thick Chrome/Gold layer. The followed by sputtering a 3000 A glass slides were then placed in an acetone solution until all the unwanted Cr/Au peeled off. Access holes of the device were drilled by estimation before constructing the fluidic channels. SU-8 2010 negative photoresist is the material for the fluidic channels. The glass slides were again cleaned before applying SU-8. The SU-8 microfluidic channel was determined to be 15 ␮m thick after the standard SU-8 fabrication procedures. Polycarbonate track-etch membrane (pore size 15 nm, 6 ␮m thick, Whatman Inc.) was chosen as the microdialysis membrane. The membrane and the device were treated with oxygen plasma for 90 s (300 mTorr, 50 W). Before applying the membrane on the 15 ␮m deep SU-8 fluidic channels, the device was wetted with a clean cotton wiper followed by a lamination bond. The device was next placed on a hot plate of 120 ◦ C for 30 min to enhance the bonding, which subsequently results in a strong non-leaking bond. Using this approach, the bonding between

Y.-C. Hsieh, J.D. Zahn / Biosensors and Bioelectronics 22 (2007) 2422–2428

2425

trodes is V, and the distance apart of the electrodes is l, then the current flowing between the electrodes is given by Ohm’s law: I = GV where G is the solution conductance, and is given by   A e0 (n+ z+ u+ + n− z− u− ) G= l

Fig. 2. Experimental setup for achieving time varying reservoir concentrations. The reservoir concentration is controlled by varying the relative flow rate from the two input syringe pumps. Top-right corner shows the PDMS micromixer piece. The width of the mixing channel is 200 ␮m and split into five sub-streams to enhance the mixing process by 25 times. The third syringe pump is used to create the perfusion flow. Both the reservoir concentration profile and solution impedance are computer controlled.

SU-8 and polycarbonate membrane is highly reproducible. A schematic of the completed device is shown in Fig. 1b and the fabricated device is shown in Fig. 1c and d. In a previous study using glucose microdialysis, a 0.8 mm diameter ring was placed on top of the dialysis channel as a reservoir (Hsieh and Zahn, 2005a,b). The concentration within the reservoir could be changed manually but in a very limited manner. Instead of placing a ring as a reservoir, a stacked system was developed (Fig. 1b and c) here to allow time varying reservoir concentrations in a more efficient and precise way requiring a second layer of fluidic channels. The PDMS reservoir layer was added onto the device to allow continuous reservoir fluid concentration changes and determine how the microsystem responds. Standard soft lithography procedures were employed by using SU-8 2010 for the mold of 30 ␮m deep reservoir fluidic channels (Duffy et al., 1998). PDMS was poured onto the mold and placed in an oven at 65 ◦ C for 30 min for curing. The membrane surface of the device and the PDMS layer were both treated with oxygen plasma (300 mTorr, 100 sccm O2 , 50 W) for 1.5 min and placed in an oven of 65 ◦ C for at least 4 h to cure. The PDMS piece was reinforced by glue along the edge. No leakage was observed in the device during the data collection. The same soft lithography technique was utilized to build a supportive PDMS micromixer for controlling the input concentration into the reservoir, which was then bonded with a clean glass to close the open surface. Fig. 2 shows the geometry of the mixer. The width of the mixing channel is 200 ␮m and split into five sub-streams resulting in a 40 ␮m diffusional path length. 4. Experimental set-up Phosphate buffered saline (PBS) solution of varying concentration (1× PBS solution: 10 mM phosphate buffer, pH 7.4, 140 mM NaCl, 3 mM KCl) was used to characterize the microdialysis system. If the potential difference between the elec-

(5)

(6)

where A is the electrode area, e0 the unit charge, n the number of the ions in solution, z the valance of the ion, and u is the mobility. It can be seen that the solution conductance depends on the nature of the ions dissolved and the concentration of these ions (through n+ and n− ). The experimental setup is shown in Fig. 2. PE10 tubing (INTRAMEDIC Polyethylene Tubing, i.d.: 0.28 mm, o.d.: 0.61 mm) was used to connect the device to the syringes. Three syringe pumps were employed for time domain microdialysis. Two syringe pumps were required for changing the solution concentration within the reservoir PDMS channel by alternating the relative flow rates from two syringes, where one is filled with 1× PBS solution, the other is filled with DI water. The micromixer was used to generate a well-mixed flow without concentration gradients introduced into the reservoir channel. The flow stream is split into n alternating sub-streams to yield a shorter diffusion pathway between the PBS solution and DI water. The total mixing time is decreased a factor of n2 (based on 1.3 × 10−9 m2 /s free diffusion coefficient of Na+ , and a 10 ␮L/min flow rate). The total mixing channel width is 200 ␮m and split into five substreams of alternating fluid composition (PBS and DI water). Therefore, the diffusion distance is reduced to 40 ␮m and the fluid emerging from the micromixer which feeds the reservoir channel is considered well mixed. The third syringe pump was used for producing a steady perfusion flow of DI water within the SU-8 lower channel. An impedance meter (ESI impedance meter 252) was connected to the contact pads of the microdialysis device and a computer was used to collect the impedance data continuously through a National Instruments digital to analog (D/A) (SCB68 National Instruments) card. A LabView program was written to control the flow rates of the two alternating syringe pumps continuously and to receive the impedance signal from the D/A card simultaneously. 5. Results and discussion Initially, the relationship between the solution conductance and PBS concentration was obtained over the experimental operating range. The measured solution conductance as a function of PBS concentration is shown in Fig. 3. At higher PBS concentrations the solution conductance is found to be linearly proportional to PBS concentrations. However, at lower PBS concentrations the conductance dependence on solution concentration has a larger slope. This is because the molar conductance of a dilute solution is higher than a concentrated solution (Harned and Owen, 1950). Therefore, the calibration curve was split

2426

Y.-C. Hsieh, J.D. Zahn / Biosensors and Bioelectronics 22 (2007) 2422–2428

Fig. 3. Left: calibration of PBS conductance as a function of PBS concentration. Right: PBS microdialysis recovery as a function of perfusion flow rate. The permeability of the membrane at this condition was obtained as 0.246 ± 0.028 ␮m/s.

into a low concentration and high concentration regime with each region fitted to a linear response. Next, the polycarbonate membrane with 15 nm pore size was tested for PBS microdialysis. PBS recovery as a function of perfusion flow rates was obtained with a device of 700 ␮m × 10 mm microdialysis area. The PDMS upper channel was infused with 1× PBS solution at constant flow rate of 10 ␮L/min and the SU-8 lower channel was infused with DI water at flow rates between 0.1 and 1.3 ␮L/min (Fig. 3). Outlet solution conductance from the perfusion fluidic channel at different flow rates were obtained and converted to concentration based on the calibration curve. Using Eq. (3), the permeability of the polycarbonate membrane to ions was determined to be 0.246 ± 0.028 ␮m/s based on three sets of experimental data (Table 1). The recovery at the lowest perfusion flow rate, 0.1 ␮L/min, is 64.4%. Experiments with the time varying concentrations within the PDMS reservoir channel were subsequently conducted using both a step change and sinusoidally changing concentration, to evaluate how the device responds to concentration fluctuations within the reservoir fluid. The system response was first demonstrated with step concentration changes in the reservoir at a constant total flow rate of 10 ␮L/min. The fluid concentration of PDMS channel was initially 0.2× PBS, which was changed stepwise to 0.5× PBS. The perfusion flow rate remained constant at 0.1 ␮L/min, which results in a 64.4% recovery. The step response of the system was recorded and the system was able to reach a new steady state value after the step concentration change within about 230 s (Fig. 4a). The system response can be modeled using linear systems theory by convolving the input waveform with the impulse response. For a given step response, g(t), the impulse response, Table 1 Membrane permeability data Experiment

K0 Am (␮m3 /s)

K0 (␮m/s)

Set 1 Set 2 Set 3

1,901,988 1,757,818 1,510,008

0.272 0.251 0.216

Average STD

0.246 0.028

h(t), could be obtained by taking first derivative of the step response: h(t) = g˙ (t)

(7)

With the impulse response, the output signal can be predicted by convolution: signalout = signalin ∗ h(t)

(8)

Next, the concentration of the reservoir channel was varied in a sinusoidal fashion from 0.2× PBS to 0.8× PBS by varying the relative flow rate of the two input syringe pumps while the total PDMS flow rate remained at 10 ␮L/min. The period of the input signal was 800 s with a concentration change every 40 s. This period was chosen because it is a sufficient time scale for tracking physiological concentration changes such as glucose fluctuations within a diabetic patient. Computational simulation of the device has also been conducted using FEMLAB (Fig. 4b) for the step change and the sinusoidal input. Fig. 4c shows the experimental, linear systems prediction and computational simulation sinusoidal response of the system. Experimentally, the system was able to reach the predicted recovery with a phase lag of 210 s. Both input and output signal were converted into the frequency domain by Fourier transforms and had an obvious frequency peak at 0.00125 Hz, which represents the 800 s sinusoidal input period. A Pearson’s correlation coefficient has been determined between the experimental data and the system response predicted by linear systems theory as well as between the FEMLAB simulation and experimental data and are r = 0.98 and 0.97, respectively. The simulation results showed that the predicted phase lags were about 80 s for a step change and 60 s for sinusoidal input at the 0.1 ␮L/min flow rate. The delay due to the dead volume of the mixing device and the tubing connection is estimated to be 150 s, which matches the experimental estimation based on the flow rate and the tubing dead volume. In order to validate the analytical model, the initial assumption that Cr,in and Cr,out are the same value can be verified by substituting the experimentally determined recovery values and permeability into Eq. (2). This showed that Cr,out and Cr,in will differ by less than 1%. Furthermore, the permeability K0 was estimated to be only 3% less when the previously ignored 1/Qr

Y.-C. Hsieh, J.D. Zahn / Biosensors and Bioelectronics 22 (2007) 2422–2428

2427

The pressure drop along the microchannel was also calculated and can be neglected when compared with the atmospheric pressure. Also, the osmotic pressure-induced flux across the membrane due was found to be negligible when compared with the diffusional flux due to the concentration gradient. The permeability was obtained as 0.246 ␮m/s (STD = 0.028 ␮m/s, Dm = 1.476 ␮m2 /s) and it is in good agreement with the theory. Accordingly, the system should reach 99% recovery at a perfusion flow rate of 0.023 ␮L/min, which is attainable by a normal syringe pump. Since the major mass transfer limitation is the diffusion through the polycarbonate membrane, the time for the system to come into equilibrium with the surrounding fluid tss can be estimated as tss ≈

d2 Dm

(9)

The tss calculated for the ions is 24.4 s (d is 6 ␮m for the polycarbonate membrane). This means that a rapid channel equilibrium is possible for rapid continuous sensing. Real time ion monitoring was possible due to the in-line sensing electrodes within the microfluidic channel. The predicted outputs from the step change and sinusoidal continuous monitoring experiments followed the functionality of the input signal with an appropriate phase lag and amplitude reduction. These results not only show good agreement with analytical predictions but also the stability of the device over long periods of time. The phase lag seen in the step and sinusoidal monitoring was about 230 and 210 s. The major contribution to this lag was mostly due to the dead volume within the tubing between the syringe pumps and the microsystem, which was measured to be approximately 150 s, and not due to mass transfer limitations. As a result, future studies should include a built-in mixer within the PDMS layer to minimize the tubing dead volume and hence this lag time. Future work will include packaging the microdialysis system as an integrated probe with electrochemical glucose sensors to be placed subcutaneously for in vivo testing. Also, different semipermeable membranes will be utilized and tested in order to achieve more efficient dialysis. Fig. 4. Time domain of PBS dialysis at a perfusion flow rate of 0.1 ␮L/min with a recovery of 64.4%. The flow rate in the upper PDMS channel was 10 ␮L/min. (a) System response to a step reservoir concentration change The concentration was initially 0.2× PBS, went up a step of concentration change to 0.5× PBS. The input signal is normalized by the recovery of 64.4%. Within about 230 s the outlet concentration reaches the steady state value. (b) Representative results from the FEMLAB simulation. The simulation consists of two subdomains, the semipermeable polycarbonate membrane and the perfusion flow channel. (c) The experimental, predicted and simulated response of the microdialysis systems to a sinusoidal reservoir input. The period of the input signal is 800 s. The input signal is normalized by the recovery of 64.4%. The correlation between the experimental data and the predicted and simulation response are r = 0.98 and 0.97, respectively.

was taken into consideration, which is within the experimental variance measured. Therefore, these assumptions are consistent with experimental data and the mathematical model presented in Eqs. (3) and (4) can represent the microdialysis system well.

6. Conclusions A complete fabrication protocol has developed for this onchip microdialysis system. As the first step toward the integration of glucose sensor and the system, in-line sensing electrodes were used to determine fluid resistance from a dialyzed phosphate buffered saline (PBS) solution. This integration was used to collect data successfully. The result shows that the system is able to continuously track concentration changes in the reservoir. In the future, a mixer will be combined within the PDMS layer and the sensing electrodes will be evolved into a glucose sensor. This system is expected to have the proper sensitivity to track physiologically relevant concentration changes of biomolecules such as glucose (maximum change rate ∼4 mg/dl min with periodicity of 1 h or greater) with minimal lag time and amplitude reduction.

2428

Y.-C. Hsieh, J.D. Zahn / Biosensors and Bioelectronics 22 (2007) 2422–2428

Acknowledgement This work is supported by the American Diabetes Association Amaranth Diabetes Fund Grant no.1-05-JF-58. References American Diabetes Association, 2003. Diabetes Care 26, 917–932. Ayliffe, H.E., Frazier, A.B., Rabbitt, R.D., 1999. J. Microelectromech. Syst. 8 (1), 50–57. Beyer, U., Fleischer, A., Kage, A., Haueter, U., Ehwald, R., 2003. Biosens. Bioelectronics 18 (11), 1391–1397. Collins, J., Lee, A.P., 2004. Lab Chip 4 (1), 7–10. Diabetes Control Complications Trial Group, 1993. N. Engl. J. Med. 329 (14), 977–986. Duffy, D., McDonald, J., Schueller, O.J.A., Whitesides, G.M., 1998. Anal. Chem. 70 (23), 4974–4984. Ekberg, N.R., Wisniewski, N., Brismar, K., Ungerstedt, U., 2005. Clin. Chim. Acta 359 (1–2), 53–64. Fang, A., Ng, H.T., Li, S.F.Y., 2003. Biosens. Bioelectronics 19 (1), 43–49. Gooding, J.J., Praig, V.G., Hall, E.A.H., 1998. Anal. Chem. 70 (11), 2396–2402. Harned, H.S., Owen, B.B., 1950. The Physical Chemistry of Electrolytic Solutions, 2nd ed. Reinhold, New York. Heinemann, L., 2003. Diabetes Technol. Therap. 5 (4), 545–561. Hsieh, Y.C., Zahn, J.D., 2005a. Sens. Actuators B: Chem. 107 (2), 649–656. Hsieh, Y.C., Zahn, J.D., 2005b. In: Bright, V.M., Fedder, G.K. (Eds.), Proceedings of the 18th IEEE International Conference on MEMS. IEEE Inc., Piscataway, NJ, pp. 802–805. Kubiak, T., Hermanns, N., Schreckling, H.J., Kulzer, B., Haak, T., 2004. Diabetic Med. 21, 487–490.

Kuo, T.C., Cannon, D.M., Chen, Y., Tulock, J.J., Shannon, M.A., Sweedler, J.V., Bohn, P.W., 2003. Anal. Chem. 75 (8), 1861–1867. Lambrechts, M., Sansen, W., 1992. Biosensors: Microelectrochemical Devices. Institute of Physics, London, UK. LeRoith, D., Okefsky, J.M., Taylor, S.I. (Eds.), 2000. Diabetes Mellitus—A Fundamental and Clinical Text. Lippincott Williams & Wilkins, Philadelphia, USA. Lodwig, V., Heinemann, L., 2003. Diabetes Technol. Therap. 5 (4), 572–586. Pan, M., Guo, X., Cai, Q., Li, G., Chen, Y., 2003. Sens. Actuators A: Phys. 108 (1), 258–262. Perdomo, J., Hinkers, H., Sundermeier, C., Seifert, W., Martinez Morell, O., Knoll, M., 2000. Biosens. Bioelectronics 15 (9), 515–522. Petrou, P.S., Moser, I., Jobst, G., 2003. Biosens. Bioelectronics 18 (5), 613– 619. Pijanowska, D.G., Sprenkels, A.J., van der Linden, H., Olthuis, W., Bergveld, P., van den Berg, A., 2004. Sens. Actuators B: Chem. 103 (1–2), 350– 355. Poscia, A., Mascini, M., Moscone, D., Luzzana, M., Caramenti, G., Cremonesi, P., Valgimigli, F., Bongiovanni, C., Varalli, M., 2003. Biosens. Bioelectronics 18 (7), 891–898. Rosdahl, H., Hamrin, K., Ungerstedt, U., Henriksson, J., 1998. Am. J. Physiol.: Endocrinol. Metabol. 274 (5), 936–945. Schoonen, A.J.M., Wientjes, K.J.C., 2003. Diabetes Technol. Therap. 5 (4), 589–598. Song, S., Singh, A.K., Shepodd, T.J., Kirby, B.J., 2004. Anal. Chem. 76 (8), 2367–2373. Steinkuhl, R., Sundermeier, C., Hinkers, H., Dumschat, C., Cammann, K., Knoll, M., 1996. Sens. Actuators B: Chem. 33 (1), 19–24. UK Perspective Diabetes Study Group, 1998. The Lancet 352 (9131), 837–853. Yang, H., Chung, T.D., Kim, Y.T., Kim, H.C., 2002. Biosens. Bioelectronics 17 (3), 251–259.