Continuous measurements of a binding reaction using a capacitive biosensor

Biosensors and Bioelectronics 21 (2005) 41–48

Continuous measurements of a binding reaction using a capacitive biosensor Martin Hedstr¨om, Igor Yu. Galaev, Bo Mattiasson∗ Department of Biotechnology, Center for Chemistry and Chemical Engineering, Lund University, P.O. Box 124, SE-221 00 Lund, Sweden Received 30 August 2004; received in revised form 11 October 2004; accepted 14 October 2004 Available online 7 December 2004

Abstract A capacitive biosensor with polyclonal antibodies raised against human serum albumin (HSA) immobilized on a gold transducer has been developed for continuous measurement of HSA in the ␮M-range. A mathematical model has been refined to describe integral HSAbinding curves assuming that (i) binding is essentially irreversible under the conditions used, (ii) the signal is scaled as the number of non-occupied binding sites and (iii) the rate of disappearance of available binding sites is scaled as the number of available binding sites and analyte concentration in solution. Deconvolution of the curves using the mathematical model indicates clearly that it is possible to retrieve concentration profiles (isocratic, linearly or exponentially increasing gradients) of the analyte in the continuous sample flow from the normalized integral binding (NIB) curves. The data presented constitutes the theoretical background and the first step towards the development of an analytical system allowing on-line detection of the concentration profile of the analyte from NIB-curves. Since the system can be used for extended time periods between regeneration steps, a low frequency of regeneration steps can be expected. © 2004 Published by Elsevier B.V. Keywords: Capacitive biosensor; Integral protein-binding curves; Mathematical model; Deconvolution

1. Introduction The interest concerning fast, sensitive and precise techniques for monitoring of target substances in bio-production processes has drastically increased as a consequence of the rapid developments within the area of biotechnology. In the biotech industry the major share of manufacturing cost often lies in the purification steps preceding the formulation of the final product. Monitoring of the downstream steps will become increasingly important both with regard to optimizing the product yield and for documenting the levels of impurities. Furthermore, documentation of process steps will be more important in the future. In addition, rigid regulations concerning the amount of impurities present in the final product together with the high purification costs then imply that analytical tools, capable of on-line monitoring, ∗

Corresponding author. Tel.: +46 46 2228264; fax: +46 46 2224713. E-mail address: [email protected] (B. Mattiasson).

0956-5663/$ – see front matter © 2004 Published by Elsevier B.V. doi:10.1016/j.bios.2004.10.014

are extremely useful when optimizing the fermentation process and down stream purification protocol. According to a recent survey by the Pharmaceutical Research and Manufactures of America (Labrou, 2003), 122 biologics are either in phase III trials or awaiting US Food and Drug Administration (FDA) approval. Some of them, such as therapeutic antibodies, will require large quantities (>1 g per patient per year) of protein to satisfy the market demand. Technical improvements that will increase the production either by increasing expression level in bioreactors or by improving the purification efficiency are thus highly required. One of the ways to improve productivity is to terminate the fermentation stage at the moment when maximum concentration of the target product is produced, avoiding non-productive continuation of the fermentation process often resulting in accumulation of impurities and partial degradation of the target product. To determine this moment, rapid responding on-line assay methods specific for the target product are a necessity.

42

M. Hedstr¨om et al. / Biosensors and Bioelectronics 21 (2005) 41–48

Few methods reported in literature are suitable for continuous measurements of protein-based substances. The most prominent of them is the surface plasmon resonance (SPR) technology (F¨agerstam et al., 1992). SPR-based instruments have become widely used tools for the characterization of proteins and DNA (Malmqvist, 1993). Other potentially suitable techniques for continuous measurements of protein analytes are the method of streaming potential (Glad et al., 1986; Mattiasson and Miyabayashi, 1988; Miyabayashi and Mattiasson, 1990) and flow-injection-based methods capable of performing on-line ELISA assays and quantification of plasmid material during bacterial fermentation (Mattiasson et al., 1990; Nandakumar et al., 2001). The quartz crystal microbalance (Storri et al., 1998; Sadik and Cheung, 2001; Sakti et al., 2001) is another technique with the potential for continuous measurements of proteins. The capacitive biosensor has a gold surface transducer with immobilized specific antibodies. The interactions between the antibodies and the target analyte result in a capacitance change at the solid–liquid interface due to alterations of the electric double layer created in close proximity to the transducer during the applied potential pulses. Several studies, utilizing capacitance measurements, have been reported in literature (Swietlow, 1994; Mirsky et al., 1997; Berggren et al., 1998), all based on discrete pulse injections of the analyte. Mostly antibody–antigen interactions have been exploited as a recognition system. However, other affinity partners have been reported, e.g. immobilized lac-repressor protein (Bontidean et al., 2001) as a recognition element for the corresponding inducers. Traditionally, most biological assays depend on the calibration of the sensor achieved by applying a series of samples with known concentration of the target molecule. The signal produced by the sample is compared with the calibration curve allowing the determination of the concentration of the target analyte. If continuous monitoring is to be set up, some different strategies may be applied. By tradition, immunochemical binding assays were set up based on strongly binding antibodies. For continuous measurements in the true sense, a binding element for the sensing device allowing binding to take place over extended periods of time would be an alternative. This was demonstrated using the streaming potential method where the derivative of the signal registered (equal to the slope of the signal versus time curve) was registered continuously, a value that was directly related to the concentration of antigen in the sample (Miyabayashi and Mattiasson, 1990). Another alternative would be to use weakly binding antibodies where association and dissociation may take place simultaneously such that the amount of bound antigen reflects the concentration of antigen in the sample at each moment. The present study is carried out using a strongly binding antibody, under conditions where the amount of bound antigen is accumulated over time on the sensing element. The signal registered is likewise changing over time. The integral binding curve can, at least in principle, be deconvoluted into

the time profile for the accumulation of the target product during the fermentation. In this work, a platform is presented for the monitoring of a continuously infused protein analyte using the capacitive biosensor.

2. Materials and methods 2.1. Chemicals Polyclonal antibodies raised against human serum albumin (HSA) were purchased from Dako (Glostrup, Denmark). As antigen, pure HSA from ICN Biomedicals (Ohio, USA) was used. Di-(N-succinimidyl)-3,3 dithiodipropionate (DSP) was purchased from Fluka (Buchs, Switzerland). Gold rods (Ø3 mm, 99.99% purity) were bought from Aldrich. All buffers were prepared with water obtained from a milli-Q system, preceded by a reverse osmosis step. All other chemicals were of analytical grade. 2.2. Preparation of electrodes The gold electrodes were polished using alumina slurries with successively decreasing particle size (from 0.1 to 0.05 ␮m) (Struers, Denmark) and ultrasonicated for 10 min. Thereafter the electrodes were placed in Teflon-holders and were plasma cleaned (model PDC-3XG, Harrich, New York) for 15 min. Immediately after this treatment the electrodes were placed in 20 mM DSP in dried dimethyl sulfoxide (DMSO) for 2.5 h and thereafter washed with the same solvent. The transducer with the created self-assembled monolayer (SAM) of DSP molecules (Doblhofer and Fruboese, 1994; Fruboese and Doblhofer, 1995; Esplandiu et al., 2001; Gaspar et al., 2001) was then dried in vacuum for 10 min and placed in an antibody solution (0.05 mg/ml in 100 mM potassium-phosphate buffer (KPB), pH 7.6) over night at 4 ◦ C yielding a covalent immobilization of antibodies to the SAM (Dong and Li, 1997; Ostuni et al., 1999). As a last step in the preparation, the electrodes were placed in a 10 mM 1dodecanethiol ethanolic solution for 20 min in order to cover bare parts of the gold surface. Before placed in the biosensor flow-cell, the electrodes were washed with 10 mM KPB, pH 7.6, containing 0.05% Tween 20, which also was used as running buffer in the system. 2.3. Measuring system The capacitance was measured using perturbation with a potentiostatic step (Fig. 1). The four-electrode system (Berggren et al., 1999) placed in a FIA flow cell (10 ␮l cell volume) consisted of a working electrode made from a Ø3 mm gold rod, a platinum wire as reference electrode, an auxiliary electrode made of platinum foil and finally an Ag/AgCl reference electrode in the outlet stream. During a 50 mV potential pulse created by a potentiostat, the resulting current response was collected and processed by a

M. Hedstr¨om et al. / Biosensors and Bioelectronics 21 (2005) 41–48

43

Fig. 1. Measuring system with a four-electrode system integrated in a 10 ␮l flow-through cell. The results retrieved during the potentiostatic step are processed using a Keithley 575 measurement and control system powered by a computer.

Keithley 575 measurement and control system equipped with an AMM2 master analogue to digital converter module. An acquisition frequency of 50 kHz allowed sampling of 1000 values during the 20 ms potential pulse applied. The potential was stepped back to rest (0 V versus Ag/AgCl) between each pulse. This procedure was repeated and on average 10 pulses were sent to a computer with compatible software and visualized as a plot of capacitance versus time. Simultaneously, data were stored in the computer for manual processing. The absolute value of the capacitance measured on the electrode surface is built up by the contribution from the different layers according to Eq. (1): 1 1 1 1 = + + Ctot Crec. layer Canalyte Cohelm. layer

(1)

where Crec. layer constitutes the capacitive properties derived from the recognition element immobilized at or in the SAM together with the blocking thiol. Crec. layer also includes stored charges as bound solvent molecules between the outer Helmholz layer, Cohelm. layer and the gold surface. The magnitude of the capacitance related to the outer Helmholz layer is strictly dependent on the distance between the metal surface and the electrolyte (Schmickler, 1996). Interactions between analytes and recognition element will result in an increase of this distance, which in turn will lead to a decreased influence of the outer Helmholz layer. According to Eq. (1), the analyte interaction then will result in a registered decrease in total capacitance. In order for the analyte interaction to give maximal contribution to the total measured capacitance, Crec. layer should be as large as possible (Swietlow, 1994).

2.4. Measuring techniques Discrete pulse injections were done using a sample injector with a 250 ␮l loop. Samples, 0.025–2.5 nmol of HSA, were injected and the flow was maintained at 230 ␮l/min. A regeneration step with 0.1 M glycine/HCl buffer, pH 2.2 was performed between each analyte injection. Continuous analyte infusion to the measuring cell was performed in three modes: isocratic infusion together with linearly and exponentially increasing gradients. The linear gradient was formed using a Pharmacia Gradient Mixer GM1 (Pharmacia Biotechnology, 1989) with a variable volume in the mixing chamber. The exponentially increasing gradient was created using a multi channel pump (U4-8R Midi, Alithea, Sweden). In this case the mixing chamber volume (Vm ) was held constant yielding the exponential concentration profile mathematically described in Eqs. (2)–(7): dct F = (c0 − ct ) dt Vm

(2)

F 1 = Vm θ

(3)

dct dt = c0 − c t θ

c 1 1 t dct = dt θ 0 0 c0 − c t [− ln(c0 − ct )]c0t =

t θ

c0 = et/θ → ct = c0 (1 − e−t/θ ) c0 − c t

(4) (5) (6) (7)

where ct is the concentration in the flow cell (␮M), c0 the concentration in feed solution (␮M), F the flow rate (ml/min),

44

M. Hedstr¨om et al. / Biosensors and Bioelectronics 21 (2005) 41–48




t

Vm the volume in mixing chamber (ml), θ the residence time for a fluid element in the mixing chamber (min) and t the time (min).

ln I ∝ −

2.5. Mathematical model

Deconvoluted concentration profile

The mathematical model used to describe the integral binding curves attained during the continuous analyte infusion is based on the assumption that the signal is scaled as the number of non-occupied binding sites: 

I ∝ (1 − [A ]) ,   

t = 0, [A ] = 0, I = 1,

(11)

ln I ∝ −ct

(12)

ln I t For a linear gradient [A] ∝ −

(13)

[A] ∝ t

(14)


t ln I ∝ − t dt,



non-occupied binding sites

ln I ∝ −t 2 ,

(15)

0

t = ∞, [A ] = 1, I = 0

(8)

where I is the normalized signal. At starting time (t = 0) the signal equals 1 whereas at t = ∞, I reaches 0. The signal time profile could then be described by Eq. (9) assuming that the rate of the disappearance of available binding sites is scaled as the number of available binding sites and analyte concentration in solution: d(1 − [A ]) ∝ (1 − [A ])[A] dt

(9)

Under the conditions used, the dissociation of the bound analyte could be neglected. Essentially, the analyte binding is described as irreversible. However, when considering weaker conjugate systems, taking into account dissociation of the analyte, the model becomes more complicated (Sadana and Zhanchi, 1996; Ramakrishnan and Sadana, 2002). The hypothesis also requires convective mass-transport in the measuring cell yielding a straight concentration profile between the liquid bulk phase and the outer Helmholz layer at the electrode interface (i.e. surface reaction limited process) (Bard and Faulkner, 1989). Then, for an isocratic concentration profile of the analyte with the border conditions taken into account [A] ∝ c

c dt

0

(10)

Deconvoluted concentration profile ln I t For an exponential gradient [A] ∝ −

(16)

[A] ∝ k(1 − e−t/θ )

(17)

where k and θ are constants describing the exponential gradient.
t ln I ∝ −k (1 − e−t/θ ) dt, 

0

t θ Deconvoluted concentation profile ln I ∝ kθ 1 − e−t/θ −

[A] ∝ −

kt ln I + θ θ

(18)

(19)

3. Results and discussion The operational sensitivity needed for the monitoring of bio-production processes is dependent on the type of target analyte or process studied. For a target protein produced at levels of a few mg/ml, sensitivities in the ␮M area are

Fig. 2. Signal vs. injected HSA amount dependence for pulse injections of HSA (250 ␮l loop volume, 0.025–2.5 nmol of injected HSA) with intermediate regeneration steps using 0.1 M glycine/HCl buffer, pH 2.2.

M. Hedstr¨om et al. / Biosensors and Bioelectronics 21 (2005) 41–48

45

Fig. 3. Signal vs. injected HSA amount dependence for pulse injections of HSA (250 ␮l loop volume, 0.25–2.5 nmol of injected HSA) without intermediate regeneration steps.

typically demanded, while for toxins or other impurities the sensitivity requirements are 100- to 1000-fold higher. In the present work, a system is described for monitoring in the ␮Mrange. A well-known protein, HSA, has been used as analyte and polyclonal antibodies against HSA have been used as recognition element. The capacitive biosensor was initially investigated with respect to linear range. The measured capacitance change on the electrode surface was proven to be proportional to the concentration of injected analyte where pulse injections of HSA ranging from 0.025 to 2.5 nmol yielded a linear relationship (Fig. 2). By applying pulses of glycine/HCl (0.1 M, pH 2.2) between each analyte injection, 90–96% regeneration was reached. In contrast, consecutive injections of analyte without intermediate regeneration steps resulted in deviation from linearity due to an uneven binding capacity of the immunosurface throughout the dynamic interval (Fig. 3). By saturating the transducer surface through repeated injections of analyte the dynamic binding conditions could be studied. At approximately 50% surface saturation the dynamic binding capacity was diminished by 33%, which ultimately proved the

need of surface regeneration for discrete injections (Fig. 4). Alternatively, in order to avoid regeneration one could try to extract the information from the integral binding curve obtained when the analyte solution continuously flows through the measuring cell until the sensor is saturated with the bound analyte. Intuitively, one could expect that different time profiles of analyte concentration in the sample passing continuously through the measuring cell should result in different integral binding curves. Certainly, without calibration it is impossible to deconvolute the integral binding curve into an exact concentration versus time profile in the sample. Moreover, the inhomogeniety of binding sites will make the situation even more complicated. The normalization of the registered signal having I = 1 on a new electrode at t = 0 and I = 0, when the electrode is saturated with the bound analyte at t = ∞, allows to obtain the integral binding curves, which are electrode-independent, i.e. having the same shape when registered using different electrodes (Fig. 5). The shape of the normalized integral binding (NIB) curves depended indeed on whether isocratic, linearly increasing gradient or exponential gradient of the analyte

Fig. 4. Capacitance changes after pulse injections of HSA followed by electrode regeneration steps with 0.1 M glycine/HCl buffer, pH 2.2.

46

M. Hedstr¨om et al. / Biosensors and Bioelectronics 21 (2005) 41–48

Fig. 5. Normalized integral binding (NIB) curves for continuous HSA assay when using isocratic, linearly and exponentially increasing HSA concentration profiles.

Fig. 6. Continuous infusion of HSA using exponential increased gradient mode. Intermediate exchange of analyte feed with running buffer is reflected as a plateau in mid-interval of the signal curve. The capacitive signal (䊉) and the created exponential HSA concentration profile (––) are presented.

Fig. 7. Continuous isocratic infusion of HSA. The actual capacitive signal (- - -), the theoretically modeled capacitive curve ( ), the created HSA concentration profile (—) and the HSA concentration profile (䊉) deconvoluted from normalized capacitance signal (I) are presented.

M. Hedstr¨om et al. / Biosensors and Bioelectronics 21 (2005) 41–48

47

Fig. 8. Continuous infusion of HSA using linear gradient mode. The actual capacitive signal (- - -), the theoretically modeled capacitive curve ( ), the created HSA concentration profile (—) and the HSA concentration profile (䊉) deconvoluted from the normalized capacitance signal (I) are presented.

Fig. 9. Continuous infusion of HSA using exponential gradient mode. The actual capacitive signal (- - -), the theoretically modeled capacitive curve ( ), the created HSA concentration profile (—) and the HSA concentration profile (䊉) deconvoluted from the normalized capacitance signal (I) are presented.

concentration was analyzed. Fig. 6 shows that the interruption of the analyte flow and the exchange to a pure buffer flow followed by resuming the sample flow resulted in a flat region on the NIB-curves. This supports the assumption of the essentially irreversible analyte binding as no dissociation (i.e. no increase in the signal, I) was observed during the period when only buffer was passed through the system. The commencement of the flat region on the NIB-curve came about 3 min after the flow interruption. The same time period was observed after resuming the sample flow before the decrease in the signal, I. This time delay reflects the dead volume of the system, which could be calculated as about 700 ␮l. The mathematical model describing the analyte binding has been developed assuming that (i) binding is essentially irreversible under the conditions used, (ii) the signal is scaled as the number of non-occupied binding sites and (iii) the rate of the disappearance of available binding sites is scaled as the number of available binding sites and analyte concen-

tration in solution. Deconvolution of the NIB-curves using the mathematical model developed (Figs. 7–9) clearly indicates that it is possible to retrieve the applied concentration profile of the analyte in the sample flow. In fact, the best fitting between the deconvoluted profile and the original HSA concentration profile has been obtained for the exponentially increasing gradient, i.e. the type of profile one could expect in a fermentation process for the accumulation of the target product. One of the most essential prerequisites of the model used is an essentially irreversible binding of the analyte to the recognition element under the conditions used. This assumption was supported by the experimental data (Fig. 6) and allowed us to simplify the mathematical model by neglecting the dissociation of the bound analyte. To meet this condition, a recognition element with a relatively strong analyte binding should be used. Nevertheless, one could extend the model to include the systems with weak interactions by taking into

48

M. Hedstr¨om et al. / Biosensors and Bioelectronics 21 (2005) 41–48

account the dissociation step. Apparently, a numerical differentiation of the NIB-curves will be required in that case. However, in the case of weak interactions with a fast dissociation of the analyte from the recognition element, no integral binding curve could essentially be obtained, as the signal at any time reflects the equilibrium-binding situation at the given time rather than the accumulation of the analyte at the electrode (Sadana and Zhanchi, 1996). Without calibration, it is impossible to obtain the absolute concentrations of the target product, whereas using this approach one could follow the dynamics of the accumulation of the target molecule. In the case of assays directed to the target fermentation product, the sensitivity requirements for the sensor are similar to what is studied in this report, as the product is accumulated during the fermentation at rather high concentrations, usually at levels at or above a few mg/ml. However, an increasing interest to monitor impurities such as endotoxins, nucleic acids, truncated target proteins, etc. will raise the demand on sensitivity of the assay. To conclude, the data presented constitute the theoretical background and the first step towards the development of analytical system allowing on-line detection of the concentration profile of the analyte from normalized integral binding curves.

Acknowledgement The work was supported by the Swedish Competence Center for Bioseparation (CBioSep).

References Bard, A.J., Faulkner, L.R., 1989. Electrochemical Methods. Wiley, USA. Berggren, C., Johansson, G., Bjarnason, B., 1998. An immunological interleukin-6 capacitive biosensor using perturbation with a potentiostatic step. Biosens. Bioelectron. 13 (10), 1061–1068. Berggren, C., Bjarnason, B., Johansson, G., 1999. Instrumentation for direct capacitive biosensors. Instrum. Sci. Technol. 27 (2), 131–140. Bontidean, I., Kumar, A., Cs¨oregi, E., Galaev, I.Yu., Mattiasson, B., 2001. Highly sensitive novel biosensor based on an immobilized lac repressor. Angew. Chem. 113 (14), 2748–2750. Doblhofer, K., Fruboese, C., 1994. Mechanism of the formation of aliphatic thiol-monolayers on gold electrodes. Trans. SAEST 29 (5), 181. Dong, S., Li, J., 1997. Self-assembled monolayers of thiols on gold electrodes for bioelectrochemistry and biosensors. Bioelectrochem. Bioenergetics 42 (1), 7–13. Esplandiu, M.J., Hagenstr¨om, H., Kolb, D.M., 2001. Functionalized selfassembled alkanethiol monolayers on Au(1 1 1) electrodes; surface structure and electrochemistry. Langmuir 17 (3), 828–838.

F¨agerstam, L.G., Frostell-Karlsson, A., Karlsson, R., Persson, B., R¨onnberg, I., 1992. Biospecific interaction analysis using surface plasmon resonance detection applied to kinetic, binding site and concentration analysis. J. Chromatogr. 597 (1–2), 397–410. Fruboese, C., Doblhofer, K., 1995. In situ quartz-microbalance study of the self-assembly and stability of aliphatic thiols on polarized gold electrodes. J. Chem. Soc., Faraday Trans. 91 (13), 1949–1953. Gaspar, S., Zimmermann, H., Gazaryan, I., Cs¨oregi, E., Schumann, W., 2001. Hydrogen peroxide biosensor based on direct electron transfer from plant peroxidase immobilised on self-assembled thiol monolayer modified electrode. Electroanalysis 13 (4), 284–288. Glad, C., Sj¨odin, K., Mattiasson, B., 1986. Streaming potential – a general affinity sensor. Biosensors 2 (2), 89–100. Labrou, N.E., 2003. Design and selection of ligands for affinity chromatography. J. Chromatogr. B: Anal. Technol. Biomed. Life Sci. 790 (1), 67–78. Malmqvist, M., 1993. Biospecific interaction analysis using biosensor technology. Nature 361 (6408), 186–187. Mattiasson, B., Miyabayashi, A., 1988. Recording of the streaming potential over an affinity column in a continuous flow system as a means of quantifying proteins biospecifically. Anal. Chim. Acta 213 (1–2), 79–89. Mattiasson, B., Nilsson, M., Berden, P., Haakanson, H., 1990. FlowELISA: binding assays for process control. Trends Anal. Chem. 9 (10), 317–321. Mirsky, V., Reipl, M., Wolfbeis, O., 1997. Capacitive monitoring of protein immobilization and antigen–antibody interactions on alkylthiol films on gold electrodes. Biosens. Bioelectron. 12 (9–10), 977–989. Miyabayashi, A., Mattiasson, B., 1990. A dual streaming potential device used as an affinity sensor for monitoring hybridoma cell cultivations. Anal. Biochem. 184 (1), 165–171. Nandakumar, M.P., Nordberg-Karlsson, E., Mattiasson, B., 2001. Integrated flow-injection processing for on-line quantification of plasmid DNA during cultivation of E. coli. Biotechnol. Bioeng. 73 (5), 406–411. Ostuni, E., Yan, L., Whitesides, G.M., 1999. The interaction of proteins and cells with self-assembled monolayers of alkanethiolates on gold and silver. Colloids Surf. B 15 (1), 3–30. Pharmacia Biotechnology, 1989. Ion Exchange Chromatography – Principles and Methods. Pharmacia, Sweden. Ramakrishnan, A., Sadana, A., 2002. A mathematical analysis using fractals for binding interactions of nuclear estrogen receptors occurring on biosensor surfaces. Anal. Biochem. 303 (1), 78–92. Sadana, A., Zhanchi, C.A., 1996. Fractal analysis of the influence of nonspecific binding on antigen–antibody binding kinetics for biosensor applications. Biosens. Bioelectron. 11 (8), 769–782. Sadik, A., Cheung, M.-C., 2001. Monitoring the specific adsorption of proteins using the electrochemical quartz crystal microbalance electrodes. Talanta 55 (5), 929–941. Sakti, S.P., Hauptmann, P., Zimmermann, B., Buhling, F., Ansorge, S., 2001. Disposable HSA QCM-immunosensor for practical measurement in liquid. Sens. Actuators 78 (1–3), 257–262. Schmickler, W., 1996. Electronic effects in the electric double layer. Chem. Rev. 96 (8), 3177–3200. Storri, S., Santoni, T., Minunni, M., Mascini, M., 1998. Surface modifications for the development of piezoimmunosensors. Biosens. Bioelectron. 13 (3–4), 347–357. Swietlow, A., 1994. Capacitive biosensor. Doctoral Thesis, Lund University, Lund.