Quantitative Measurement of Cardiac Markers in Undiluted Serum

Dec 9, 2006 - ... Boswell Memorial Hospital, Sun City, Arizona 85351, and School of .... Matthew E. Stewart , Jimin Yao , Joana Maria , Stephen K. Gra...
0 downloads 0 Views 254KB Size
Anal. Chem. 2007, 79, 612-619

Quantitative Measurement of Cardiac Markers in Undiluted Serum Jean-Francois Masson,† Tina M. Battaglia,† Philip Khairallah,‡ Stephen Beaudoin,§ and Karl S. Booksh*,†

Department of Chemistry and Biochemistry, Arizona State University, Mail Code 1604, Tempe, Arizona 85287-1604, SunHealth Heart Center, Boswell Memorial Hospital, Sun City, Arizona 85351, and School of Chemical Engineering, Purdue University, West Lafayette, Indiana 47907

Two mycocardial infarction biomarkers, myoglobin (MG) and cardiac troponin I (cTnI), were quantified at biological levels and in undiluted serum without sample pretreatment using surface plasmon resonance (SPR) sensors. To achieve detection of biomarkers in undiluted serum (72 mg/mL total protein concentration), minimization of the nonspecific signal from the serum protein was achieved by immobilizing the antibody for the biomarkers on an N-hydroxysuccinimide activated 16-mercaptohexadecanoic acid self-assembled monolayer. This monolayer reduces the nonspecific signal from serum proteins in such a manner that short exposure of the sensor to serum prior to analysis prevents any further nonspecific adsorption during analysis. Thus, sensing of MG and cTnI was achieved on the basis of the difference between signals from the active sensor and a reference sensor that captured background interference. This resulted in direct measurement of these biomarkers in undiluted serum. Detection limits for both markers were below 1 ng/mL, which is below the threshold needed to detect myocardial infarction. Detecting biomarkers in the low ng/mL range without signal amplification in such a complex matrix as serum corresponds to a selectivity of 108. The root-meansquare-error (RMSE) of calibration was below 2 ng/mL. There is a great need in medicine to investigate real-time expression of biological markers to better understand their roles in disease progression. These analytes are in biological fluids that contain large amounts of proteins of varied nature. To date, no reported measurement technique has enabled real-time measurement of biomarkers at biological levels in undiluted serum samples mainly due to nonspecific adsorption of the serum proteins to surfaces. To achieve stability in biological media, modification of the sensor surface to minimize adsorption of serum proteins is essential. Recently, Battaglia et al. have reported that by using a coating of N-hydroxysuccinimide activated 16-mercaptohexadecanoic acid (NHS-MHA), the stability of surface plasmon resonance (SPR) sensors was improved in cell culture media during * Corresponding author. Tel.: (480) 965-3058. Fax: (480) 965-2747. E-mail: [email protected]. † Arizona State University. ‡ Boswell Memorial Hospital. § Purdue University.

612 Analytical Chemistry, Vol. 79, No. 2, January 15, 2007

direct measurement of biomarkers correlated to wound healing.1 In this Article, the measurement of myocardial infarction (MI) markers in undiluted serum is performed using NHS-MHA as a non-fouling coating at the surface of SPR sensors while also covalently binding the antibodies to the SPR sensor to achieve detection of analytes at sub-ng/mL concentrations. Recent developments in surface coatings used in SPR sensing allow a reduction of the nonspecific binding while improving specific signals by increasing the amount of active antibodies at the gold surface of the SPR sensors.2,3 One approach used selfassembled monolayers (SAMs) combined with a novel activation of the layer. Pre-reacting the NHS and MHA prior to forming the SAM allowed a greater coverage of NHS-activated 16-mercaptohexadecanoic acid SAM on the gold surface, thus yielding a more sensitive sensor. Simultaneously, this NHS-MHA layer showed promising results for measurement of cytokines in complex media.1 A SPR sensor with this NHS-MHA layer showed negligible nonspecific adsorption of cell culture media proteins to the surface, allowing direct measurement of cytokines related to wound healing. The level of nonspecific proteins in cell culture media is 4 mg/mL. The limits of detection of the cytokines achieved were below 1 ng/mL, without sample pretreatment or signal amplification. Therefore, specificity of >106 was achieved. The specificity is defined here by the ratio of the nonspecific protein concentration in the media to the limit of detection of the analyte. Another approach using polyethyleneglycol monolayers on SPR sensors successfully quantified mouse brain proteins from crude cell lysate.4 Cell lysate contains approximately 11 mg/mL of nonspecific proteins. Limits of detection for mouse brain proteins were as low as 30 ng/mL, corresponding to a specific signal of >4 × 105. These two articles are the only ones published to date reporting biomarker detection in the low ng/mL range in the presence of nonspecific serum protein concentration greater than 1 mg/mL. Of these two approaches, NHS-MHA resulted in lower detection limits and greater specificity. Successful determination of trace biomolecules in undiluted serum has not previously been demonstrated in the literature. The (1) Battaglia, T. M.; Masson, J. F.; Sierks, M.; Beaudoin, S.; Rogers, J.; Foster, K. N.; Holloway, G. A.; Booksh, K. Anal. Chem. 2005, 77, 7016-7023. (2) Masson, J. F.; Battaglia, T. M.; Davidson, M. J.; Kim, Y. C.; Prakash, A. M. C.; Beaudoin, S.; Booksh, K. S. Talanta 2005, 67, 918-925. (3) Masson, J. F.; Battaglia, T. M.; Kim, Y. C.; Prakash, A.; Beaudoin, S.; Booksh, K. S. Talanta 2004, 64, 716-725. (4) Kyo, M.; Usui-Aoki, K.; Koga, H. Anal. Chem. 2005, 77, 7115-7121. 10.1021/ac061089f CCC: $37.00

© 2007 American Chemical Society Published on Web 12/09/2006

nonspecific signal from serum proteins inhibits success in most such analyses. Analytical measurements have been made in synovial fluids,5 diluted or pretreated urine,6,7 pretreated blood for analyte extraction,8 and diluted serum with high analyte concentrations.9-13 To date, a few studies have been reported that analyze biomolecules in diluted serum using SPR.12,14 Typically, dilution of the serum by a factor of 1/1000,15 1/100,16-19 1/50,20 1/40,10 or 1/2011,21-23 is performed. At these dilutions, the nonspecific signal from the serum proteins becomes negligible. At dilution factors smaller than 1/20, the nonspecific signal from serum proteins is large enough that its effect on the SPR output must be considered. For example, in a study of serum samples diluted 10 times, CM-dextran was added to the serum to reduce nonspecific signals by adhering to the serum proteins in solution.24 Limits of quantification of 1 µg/mL were reported for serum samples diluted 10 times,25 but limits of quantification at or below the low-ng/mL range are needed for many applications. Previous studies reported antibody activity studies in 5%, 25%, or 50% serum samples.18,26,27 However, antibodies are often much larger molecules than the antigens of interest here; therefore, the antibodies will produce a much larger signal when binding to an SPR probe. Larger signals are easier to distinguish from nonspecific signals. Detection of an antigen must be possible in solutions with concentrations in the µg/mL range as reported for heparin.28 In the case of the proteins studied here for MI, limits of detection below 1 ng/mL are required. Detecting small molecules at low (5) Kim, J. Y.; Lee, M. H.; Jung, K. I.; Na, H. Y.; Cha, H. S.; Ko, E. M.; Kim, T. J. Exp. Mol. Med. 2003, 35, 310-316. (6) Navratilova, I.; Skadal, P. Supramol. Chem. 2003, 15, 109-115. (7) Shelver, W. L.; Smith, D. J. J. Agric. Food Chem. 2003, 51, 3715-3721. (8) Vollenbroeker, B.; Fobker, M.; Specht, B.; Bartetzko, N.; Erren, M.; Spener, F.; Hohage, H. Int. J. Clin. Pharmacol. Ther. 2003, 41, 248-260. (9) Nishimura, S. Bunseki Kagaku 2001, 50, 79-82. (10) Jongerius-Gortemaker, B. G. M.; Goverde, R. L. J.; van Knapen, F.; Bergwerff, A. A. J. Immunol. Methods 2002, 266, 33-44. (11) Gonzales, N. R.; Schuck, P.; Schlom, J.; Kashmiri, S. V. S. J. Immunol. Methods 2002, 268, 197-210. (12) Disley, D. M.; Blyth, J.; Cullen, D. C.; You, H.-X.; Eapen, S.; Lowe, C. R. Biosens. Bioelectron. 1998, 13, 383-396. (13) Ye, J. S.; Wen, Y.; Zhang, W. D.; Cui, H. F.; Xu, G. Q.; Sheu, F. S. Electroanalysis 2005, 17, 89-96. (14) Liljeblad, M.; Lundblad, A.; Ohlson, S.; Pahlsson, P. J. Mol. Recognit. 1998, 11, 191-193. (15) Li, Y. X.; Zhu, L. D.; Zhu, G. Y.; Zhao, C. Chem. Res. Chin. Univ. 2002, 18, 12-15. (16) Gomara, M. J.; Ercilla, G.; Alsina, M. A.; Haro, I. J. Immunol. Methods 2000, 246, 13-24. (17) Koubova, V.; Brynda, E.; Karasova, L.; Skvor, J.; Homola, J.; Dostalek, J.; Tobiska, P.; Rosicky, J. Sens. Actuators, B 2001, 74, 100-105. (18) Pei, J. H.; Li, X. Y. Electroanalysis 1999, 11, 1266-1272. (19) Vikinge, T. P.; Askendal, A.; Liedberg, B.; Lindahl, T.; Tengvall, P. Biosens. Bioelectron. 1998, 13, 1257-1262. (20) Besselink, G. A. J.; Schasfoort, R. B. M.; Bergveld, P. Biosens. Bioelectron. 2003, 18, 1109-1114. (21) Sellborn, A.; Andersson, M.; Fant, C.; Gretzer, C.; Elwing, H. Colloids Surf., B 2003, 27, 295-301. (22) Johnsson, L.; Baxter, G. A.; Crooks, S. R. H.; Brandon, D. L.; Elliott, C. T. Food Agric. Immunol. 2002, 14, 209-216. (23) Severs, A. H.; Schasfoort, R. B. M.; Salden, M. H. L. Biosens. Bioelectron. 1993, 8, 185-189. (24) Alaedini, A.; Latov, N. Neurology 2001, 56, 855-860. (25) Wong, M. S.; Fong, C. C.; Yang, M. Biochim. Biophys. Acta 1999, 1432, 293-301. (26) Chung, J. W.; Kim, S. D.; Bernhardt, R.; Pyun, J. C. Sens. Actuators, B 2005, 111, 416-422. (27) Mason, S.; La, S.; Mytych, D.; Swanson, S. J.; Ferbas, J. Curr. Med. Res. Opin. 2003, 19, 651-659. (28) Gaus, K.; Hall, E. A. H. J. Colloid Interface Sci. 1999, 217, 111-118.

concentrations is difficult, such that to date no reported detection of small proteins at 1 ng/mL in undiluted serum has been published. Many strategies for either reducing nonspecific adsorption or enhancing the signal from specific binding in the face of nonspecific adsorption have been employed to quantify analytes in serum using sensors. An antibody to HIV-1 p-24 was detected by heating the serum sample prior to analysis.29 Exposure of the SPR surface to 10 M NaOH after ferritin binding in undiluted human serum removed the nonspecific bound proteins while leaving ferritin bound to the surface.30 A sandwich assay for detection of ferritin in 50% serum has been employed.13 Sandwich assays use signal enhancement from the binding of a secondary antibody to the (bound) analyte of interest to help detect the analyte. Nanoparticles have been previously used to enhance the signal from the analyte.31 For measurement of biomarkers in cell culture media previously mentioned, the strategy employed to measure analytes relied on a pretreatment of the SPR sensor surface. Specifically, the SPR sensor was exposed to analyte-free cell culture media for a short period of time prior to analysis to block sensor sites prone to nonspecific adsorption.1 To perform marker monitoring in cell lysate as discussed above, nonspecific adsorption sites were blocked with bovine serum albumin prior to exposure to analyte-containing samples. For MIs, quick diagnosis is vital to patient survival and recovery. One option to perform chemical diagnoses of MIs is to employ SPR technology. However, with SPR sensors, the analysis of the biomarkers must be done directly in undiluted serum or blood because dilution of serum by a factor of 10 will decrease the concentration of myoglobin (MG) and cardiac troponin I (cTnI) below the limit of detection of the SPR sensor. Using similar SPR sensors with antibodies of comparable binding affinities to the target analytes, detectable limits for myocardial markers are at about 1-3 ng/mL, while the physiological levels range from a few ng/mL to 30 ng/mL after a myocardial infarction.32 Although using signal enhancement cannot be totally ruled out for detection purposes, a sensor that would not require signal enhancement would greatly simplify the sensing and reduce the analysis time. Furthermore, the ability of a fiber-optic sensor to achieve biologically relevant detection limits likely enables the possibility to develop the sensors for in-situ and in-vivo analyses. EXPERIMENTAL SECTION Sensor Fabrication. The sensors were hand crafted as described previously.33,34 Two 200 µm diameter fibers were fitted into a custom designed adapter to interface the sensor with the light source and spectrometer. One fiber brought light from the white LED source, and the other returned the reflected light to (29) Hifumi, E.; Kubota, N.; Niimi, Y.; Shimizu, K.; Egashira, N.; Uda, T. Anal. Sci. 2002, 18, 863-867. (30) Lefebvre, S.; Burglen, L.; Reboullet, S.; Clermont, O.; Burlet, P.; Viollet, L.; Benichou, B.; Cruaud, C.; Millasseau, P.; Zeviani, M.; Lepaslier, D.; Frezal, J.; Cohen, D.; Weissenbach, J.; Munnich, A.; Melki, J. Cell 1995, 80, 155165. (31) Kubitschko, S.; Spinke, J.; Bruckner, T.; Pohl, S.; Oranth, N. Anal. Biochem. 1997, 253, 112-122. (32) Masson, J. F.; Obando, L.; Beaudoin, S.; Booksh, K. Talanta 2004, 62, 865-870. (33) Masson, J. F.; Obando, L. A.; Beaudoin, S.; Booksh, K. S. Talanta 2004, 62, 865-870. (34) Obando, L. A.; Booksh, K. S. Anal. Chem. 1999, 71, 5116-5122.

Analytical Chemistry, Vol. 79, No. 2, January 15, 2007

613

the spectrometer and CCD detector. A Jobin-Yvon SPEX 270M spectrometer with an 1800 g/mm grating was used to narrow the spectral range to 42.8 nm. The spectra were collected with an Andor CCD camera. A resolution of 0.0421 nm/pixel was obtained. NHS-MHA Preparation. The NHS-MHA synthesis was a modified Lynn synthesis.35 Prior to synthesizing NHS-MHA, we recrystallized MHA in ethanol to remove any impurity interfering with the reaction and vacuum-dried it overnight using a roughing pump. For NHS-MHA synthesis, 1 g of MHA was dissolved in 20 mL of dioxane. Once dissolved, 0.77 g of dicyclohexylcarbodiimide (DCC) and 0.4 g of NHS were added to the MHA solution. The solution was stirred for 4 h. Dicyclohexylurea, a white precipitate, formed and was filtered out of the solution. The remaining solution was evaporated, and NHS-MHA was collected. NHS-MHA was characterized using NMR and FT-IR. The NHS-MHA SAM was characterized using GATR-FT-IR. Antibody Layer Preparation. The preparation and synthesis of CM-dextran polymer layers have been extensively described.36,37 Preparation of the SAM monolayers is described here. The bare gold surface on the SPR probe was exposed to a 0.005 M solution of NHS-MHA in tetrahydrofuran overnight to form a self-assembled monolayer (SAM). Use of a pure NHS-MHA layer was previously found to maximize the sensitivity of the SPR fiber-optic sensors.38 The SAM was washed with acetonitrile and then rinsed with water for 10 min. MG and cTnI sensors were prepared via an amine coupling reaction of this activated surface with a 700 µg/mL solution of anti-MG (ICN Biomedical, polyclonal rabbit antiserum to human MG) or 50 µg/mL solution of anti-cTnI (Spectral Diagnostics). The surface was allowed to react overnight in HBS at pH 7.4 and 4 °C. HBS was composed of 150 mM NaCl, 10 mM HEPES, 3.4 mM EDTA, and 0.005% Tween 20 (surfactant) in 18 MΩ deionized water. The pH of the HBS solution was adjusted to 7.4 using 2 M NaOH. The nonspecifically bound antibodies were washed away and the nonreacted sites on the monolayer were deactivated by rinsing the probe with an aqueous solution of 1 M ethanolamine with a pH of 8.5 for 10 min. The sensors were stored in HBS at 4 °C until use. Measurement of Antigens in Serum. MG (ICN Biomedicals) solutions were prepared in bovine serum. MG was received as a lyophilized powder and was reconstituted to 1 mg/mL using 18 MΩ deionized water. MG was stored at -20 °C for extended periods of time. A stock solution at an MG concentration of 500 ng/mL was prepared in bovine serum (ICN Biomedicals, 72 mg/ mL total protein concentration) followed by dilutions to the desired concentration with bovine serum. Preparing the stock MG samples for analysis in bovine serum eliminates the dilution of future unknown samples by the addition of MG stock solution. The stock solution for MG was stored at 4 °C until use and was prepared daily. cTnI (Spectral Diagnostics) solutions were prepared in bovine serum. cTnI was received as a solution at 1.69 mg/mL and was stored at -20 °C for extended periods of time. A stock (35) Lynn, M. Immobilized Enzymes, Antigens, Antibodies, and Peptides, Preparation and Characterization; Marcel Dekker: New York, 1975. (36) Lofas, S.; Johnsson, B. J. Chem. Soc., Chem. Commun. 1990, 21, 15261528. (37) Johnsson, B.; Lofas, S.; Lindquist, G. Anal. Biochem. 1991, 198, 268-277. (38) Masson, J. F.; Battaglia, T. M.; Cramer, J.; Beaudoin, S.; Sierks, M. R.; Booksh, K. Langmuir, in preparation. Masson, J. F.; Battaglia, T. M.; Cramer, J.; Beaudoin, S.; Sierks, M. R.; Booksh, K. Anal. Bioanal. Chem. 2006, 386 (7-8), 1951-1959.

614 Analytical Chemistry, Vol. 79, No. 2, January 15, 2007

Figure 1. Nonspecific signal from serum proteins for NHS-MHA (A) and CM-dextran (B). The initial jump when the sensors are placed in serum (at time of 300 s) is caused by a different refractive index of the serum as compared to HBS. The rate of nonspecific binding can be seen in the slopes from 600 to 1500 s. The degree of nonspecific adsorption can be seen in the difference in the probe signals prior to (0-300 s) and following (1500-1800 s) exposure to the serum.

solution at 500 ng/mL was prepared in bovine serum followed by dilutions in HBS pH 7.4 to the desired concentration. The stock solution for cTnI was stored at 4 °C until use and was prepared daily. MG and cTnI solutions were thermally equilibrated in a water bath at 25 °C for 30 min before analysis. The sensor was equilibrated for 15 min in bovine serum before use. The SPR signal was monitored for 5 min in static bovine serum and then transferred to the analyte solution. The analyte measurement was performed in static solutions. The sensor was exposed to bovine serum after analyte measurement for regeneration. Up to three consecutive measurements were obtained for each sensor before antibody degradation reduced the probe sensitivity. Data were acquired at a rate of 1 data point every 3 s. Each data point was an accumulation of three measurements. RESULTS AND DISCUSSION Antigen Detection in Serum for Myocardial Infarction. (1) Stability of CM-Dextran and NHS-MHA in Undiluted Serum. A major breakthrough in biomarker sensing will be realized by measuring analytes directly in serum at biological concentrations. One of the limiting factors is the stability of surface sensors in serum due to the nonspecific binding of serum proteins to surfaces. This process is unavoidable. However, if it can be minimized it could become negligible to the point of allowing direct measurement of analytes in serum. Thus, one of the major characteristics of a novel sensing layer that must be evaluated is its resistance toward fouling by serum. To measure the fouling of the sensor in the presence of serum proteins, SPR sensors containing the antibody support (a monolayer) with covalently bound antibodies are exposed to serum. Previous studies that reported measurement of analytes in serum or diluted serum used CM-dextran as an antibody support. Therefore, the stability of NHS-MHA is hereafter compared to CM-dextran. SPR probes are highly susceptible to fouling from serum proteins, and as a result their output signals are affected significantly by the presence of such proteins in solution,2 hence the need for different antibody support layers. SPR data were acquired for short-term exposure of CM-dextran- and NHS-MHAcoated SPR sensors to undiluted bovine serum at room temperature. The sensors were exposed to HBS for 5 min, then bovine serum for 20 min, and finally HBS for 5 min. The sensor output for this process is shown in Figure 1. Both surface coatings produce sensor signals that reflect the change in the refractive index of the medium from HBS to serum, as well as nonspecific

Figure 2. (A) Long-term exposure to bovine serum of NHS-MHA (black) and CM-dextran (gray)-coated sensors. The size of the data points also represents the error on the measurement. This measurement was performed at 0 °C, hence the slow binding kinetics. (B) Sensor exposure to serum and analyte SPR signal for CM-dextran (gray) and NHS-MHA (black) after pretreatment of the surface to serum to block nonspecific adsorption sites of the sensors. Both panels show significantly more nonspecific binding for CM-dextran sensors than for NHS-MHA-coated sensors.

Figure 3. The observed NSB protein mass on the surface of the sensors (b) can be modeled by the Vroman cascade (solid line) as the sum of three key serum proteins. This model assumed the adsorption of each species individually would be explained by a simple Langmuir-type adsorption. At steady-state such a model predicts the fractional coverage of the probe surface by each protein by eq 4.

adsorption of protein to available sites on the sensors. While this adsorption seems to be unavoidable for these coatings, it may be possible to pretreat a sensor surface with serum prior to monitoring solutions containing antigen. In this manner, the surface would be completely fouled prior to immersion in the antigen-bearing solution, and additional fouling may not occur. In the absence of a pretreatment step, the signal from the nonspecific adsorption would overwhelm the signal of the analyte-antibody binding. Nonspecific adsorption to CM-dextran is less reversible than that to NHS-MHA as is indicated by the greater signal when the sensor is placed in HBS after serum protein adsorption. In Figure 2A, the sensors with NHS-MHA or CM-dextran coatings at 0 °C are exposed to serum, and it can be seen that the system reaches steady-state more quickly in the presence of NHS-MHA than CMdextran. This indicates that the NHS-MHA is more resistant to (39) Green, R. J.; Davies, J.; Davies, M. C.; Roberts, C. J.; Tendler, S. J. B. Biomaterials 1997, 18, 405-413. (40) Green, R. J.; Davies, M. C.; Roberts, C. J.; Tendler, S. J. B. Biomaterials 1999, 20, 385-391. (41) Nordin, H.; Jungnelius, M.; Karlsson, R.; Karlsson, O. P. Anal. Biochem. 2005, 340, 359-368.

protein adsorption than CM-dextran, or, equivalently, that the rate of release of serum proteins from NHS-MHA was larger than the rate of release of these proteins from CM-dextran. The Vroman protein cascade is a concept developed to explain the behavior of surfaces contacted with serum. The cascade dictates that the proteins will adsorb on the surface sequentially, such that the short time adsorption/desorption is dominated by albumin, the most abundant and smallest protein of the cascade, after which IgG dominates, before finally fibrinogen adsorption/ desorption dominates.39,40 Within this framework, the IgG displaces the albumin, after which the fibrinogen displaces the IgG. Figure 2A shows that NHS-MHA either stops the cascade prior to its completion, causes the system to rapidly pass through the entire cascade to reach equilibrium with the most strongly bound species, or offers a minimum number of sites that can participate in the cascade. This suggests that if one pretreats a sensor with serum prior to analyzing a serum solution containing an analyte, the nonspecific binding contribution to the overall sensor output (binding + fouling) will not change between the pretreatment and analyte sensing steps. This, in turn, allows pretreated SPR sensors to be applied in real time in unmodified serum. Figure 2B shows that the NHS-MHA-coated sensor has significantly less fouling in serum than the CM-dextran-coated sensors. To study the effects of the surface coatings on the NSB, five bare gold sensors, four NHS-MHA-coated sensors, four CMdextran-coated sensors, and four MHA-coated sensors were exposed to bovine calf serum, and the SPR shift (from surface fouling) was monitored during the first 20 min of fouling. The data are shown in Figure 3 for one NHS-MHA-coated sensor in terms of the mass per unit area of the probe as a function of time. The surface coverage was approximated on the basis of the calculations from Jung et al.41 for the thickness of the nonspecifically bound (NSB) layer. The SPR signal is reported as the wavelength, in nanometers, of the minimum in the returned light spectrum from the probe. The minimum in this wavelength shifts when the refractive index of the medium in contact with the probe changes, or when binding to the probe occurs. The thickness (d) of an NSB layer can be calculated using

(

)

ld NSBshift d ) - ln 1 2 NSBshiftmax

(1)

Analytical Chemistry, Vol. 79, No. 2, January 15, 2007

615

in which NSB shift is the shift in the minimum in the returned light (nm) at any time, and NSB shiftmax can be calculated from

NSBshiftmax ) m(ηa - ηs)

Γ ) dF

(3)

Protein density is typically 1.3 g/mL.41 Figure 3 shows the calculated value of Γ as a function of time for the case of an SPR probe with an NHS-MHA surface coating immersed into serum. Also shown in Figure 3 is a first generation model for the fouling based on the Vroman cascade. This model assumed the adsorption of each species individually would be explained by a simple Langmuir-type adsorption. At steady-state, when consideration is restricted only to three possible binding species, albumin, IgG, and fibronectin, such a model predicts that the fractional coverage of the probe surface by each protein is given by

N1 NT

)

K1C1

, Θ2 )

3

1+

∑K C

N2

K2C2

)

NT

i i

,

3

1+

i)1

∑K C

i i

i)1

Θ3 )

N3 NT

ΘiNT(mwi) probe area

(6)

(2)

where ηa is the refractive index of the adsorbed layer, and ηs is the refractive index of the solution (1.37 RIU for saline solution with 84 mg/mL serum). If one plots NSBshiftmax as a function of ηa - ηs, one obtains a straight line with a slope m. The refractive index for adsorbed protein is usually 1.57.41 For the SPR spectrometer used here, the sensitivity is 2253 nm/RIU. Thus, the theoretical NSB shiftmax, assuming an infinite layer of adsorbed proteins, is 450 nm. In eq 1, the penetration depth (ld) of the surface plasmon wave is approximately 230 nm for the wavelength range used in these experiments. The NSB surface coverage (Γ) is calculated from the protein density (F) and thickness, as described by

Θ1 )

Γi )

)

K3C3

∂Θi ) ki,aCiΘVNT - kd,aΘiNT ∂t

(7)

where Θv is the fraction of sites that are unoccupied. The fractional coverages Θi are related to the vacant fraction Θv by

1 ) Θ1 + Θ 2 + Θ 3 + Θ v

(8)

Equation 6 can be used to convert the results of eq 7 from Θi to Γi. This model employed six parameters: the ki,a and ki,d for the albumin, IgG, and fibrinogen. Sensitivity analysis of the best-fit parameters demonstrated that modifying any parameter by more than 10% resulted in a significant degradation of the fit of the model to the adsorption data. Also shown in Figure 3 are the calculated values of the coverages due to each of the species in the Vroman cascade, based on the fitting scheme described above. All four coatings led to nonspecific fouling for serum adsorption onto the SPR probe similar to what is shown in Figure 3. Figure 4 shows the adsorption equilibrium constants fit using the Langmuir model to the NSB data, assuming only the albumin, IgG, and fibrinogen adsorb, for each probe coating studied. As can be seen, the NHS-MHA is the most effective coating at minimizing adsorption (lowest equilibrium adsorption constant toward all species). A lesser degree of fouling on the sensor surface offers the possibility of more accurate compensation of the nonspecific signal from serum proteins when a reference sensor is employed. It is

(4)

3

1+

where mwi is the molecular weight of adsorbate i. This approach assumes that all species occupy the same number of sites (1 site) upon adsorption. For each species i, the time-dependent behavior was described using equations of the form

∑K C

i i

i)1

where 1, 2, and 3 represent albumin, IgG, and fibronectin; Θi is the fractional coverage of the probe surface by species i; Ni is the number of surface sites covered by species i; NT is the total number of surface sites available for adsorption; Ki is the equilibrium constant for the adsorption of species i; and Ci is the concentration of each species i in the bulk liquid. Each Ki is defined as

Ki )

ki,a ki,d

(5)

where ki,a is the rate constant for the adsorption of species i, and ki,d is the rate constant for the desorption of species i. To convert the Θi to Γi, one must use 616

Analytical Chemistry, Vol. 79, No. 2, January 15, 2007

Figure 4. Equilibrium adsorption coefficients (molecules/s) for each protein in the Vroman cascade on each probe coating, based on competitive Langmuir adsorption.

Figure 5. SPR signal for the detection of 50 ng/mL myoglobin in serum. (A) represents the raw signal from the analyte sensor, (B) is the signal from the reference sensor, and (C) is the analyte signal when the reference sensor (B) is subtracted from the analyte sensor (A).

well known that biological samples will generally have a replicate error of at least 10%. When probes with a CM-dextran coating that had been subject to a 15 min pretreatment in serum were reinserted into serum, the shift in the subsequent SPR spectrum from nonspecific protein adsorption was about 1 nm, as shown in Figure 2B (gray curve). A 10% error means that the measurement will be accurate to (0.1 nm (one standard deviation). The instrumental noise on the system was typically 0.008 nm SPR shift, 12 times smaller than the error of 0.1 nm. Defining the limit of detection as three standard deviations of the error yields a smallest detectable signal of 0.3 nm SPR shift for CM-dextran-coated sensors in serum. This corresponds to approximately a 25 ng/ mL analyte signal. Because of this variability in the nonspecific binding signal, the SPR signal resulting from analyte binding would therefore be impossible to distinguish from the error on the total signal. Another drawback of the CM-dextran coatings is that the observed signal from nonspecific binding continuously rises for an extended period of time as seen in Figure 2B (gray line). However, if a probe with a surface coating of NHS-MHA is pretreated for 15 min by immersion in serum, is removed from the serum, and then re-immersed, an additional shift in signal due to nonspecific protein binding is only ∼0.1 nm, as shown in Figure 2B (black line). Therefore, a repeatability error of approximately 0.01 nm is expected on a measurement in serum. The signal from nonspecific protein adsorption is very stable after a short pretreatment of the surface. This experiment also suggests that NHSMHA indeed limits the extent to which the Vroman cascade can proceed on the probe surface sufficiently to allow biomarker measurements directly in serum. Using NHS-MHA, the error in the signal reproducibility is approximately equivalent to the system noise (0.008 nm for one standard deviation). Therefore, similar limits of detection for biomarker detection in serum are expected when compared to measurement in saline solution. (2) Signal Measurement in Undiluted Serum Using NHSMHA. NHS-MHA coatings greatly improve SPR sensor stability in undiluted serum as compared to CM-dextran if proper probe pretreatment is applied. Taking advantage of this probe stability, direct measurement of biomarkers in undiluted serum was performed. Figure 5 shows the results of direct SPR probe measurements of MG in undiluted serum. Figure 5A shows the signal from a probe with an NHS-MHA coating and active MG antibodies. This signal contains effects of the RI of the serum, the probe fouling, and the antibody-antigen binding. Figure 5B shows the signal from a reference probe containing an NHS-MHA coating but no active antibodies. The signal from this probe is

influenced only by the probe fouling and the RI of the serum. When the signal in Figure 5B is subtracted from that in Figure 5A, the remaining signal is the result of antibody-antigen binding only. This is shown in Figure 5C. Nonspecific adsorption is still present, but at a level that it can be compensated by the use of a reference sensor. The reference sensor is mainly needed in cases when the sensor is transferred from the blank serum solution to the analyte solution in serum. At that point, there is a spike and a decrease of the SPR coupling wavelength for the reference sensor (Figure 5B). This effect is also present in the active sensor, but convoluted with the binding signal from MG. Increased desorption rate of proteins on surfaces at higher surface concentration is well known as part of the Vroman effect. This effect can be compensated using a reference sensor placed in an analyte-free serum, or using a sensor with denatured antibodies on the surface. For the work presented here, the reference sensor was placed in a blank serum solution. Subtracting the reference sensor signal from that of the sensor detecting the analyte resulted in the binding isotherm of MG that is shown in Figure 5C. The ordinate axis in Figure 5C is negative due to variations in the manufacturing of the analytical and reference probe; note that the two probes differ by ∼7 nm SPRshift when placed in the same serum solution for pretreatment (Figure 5A and B, first 3 min). The measurements displayed in Figure 5 do not have any signal enhancement, and the solution studied was prepared in undiluted bovine serum containing 72 mg/mL of nonspecific proteins. It is notable that the sensor showed selectivity for MG, in the presence of a greater than 1 000 000:1 ratio of nonspecific proteins to analyte, and that the analyte of interest was present in ng/mL concentrations. The successful measurement of such low concentrations of analyte in undiluted serum allows efficient measurement of biomarkers with very minimal sample preparation. Figure 5C shows that the SPR signal equilibrates very quickly, within 5 min. Calibration of the SPR Sensor for MG and cTnI in Undiluted Serum. The sensors were calibrated assuming that a Langmuir isotherm could describe the binding between the analytes of interest and the antibodies.1 Figure 6A illustrates the concentration dependence on the equilibrium binding signal of cTnI, and Figure 6B shows that the binding is consistent with a Langmuir model as the data can be linearized by plotting 1/shift versus 1/concentration of analyte. To prepare Figure 6A and B, a reference sensor with active cTnI antibodies was employed in cTnI-free serum. Figure 6C and D demonstrates that the sensor response to MG also was also consistent with Langmuir antibodyAnalytical Chemistry, Vol. 79, No. 2, January 15, 2007

617

Figure 6. Langmuir isotherm (A) and calibration curve (B) for cTnI and Langmuir isotherm (C) and calibration curve (D) for MG. The dotted lines represent two standard deviations about the regression line.

antigen binding. The Langmuir isotherm assumptions are that only one molecule can be adsorbed per site, only one type of site is present, the adsorption of one molecule does not affect the adsorption of the other molecules, only one adsorbing species is present, the solution is dilute, and the adsorption is reversible.42 The deviation at large concentrations is explained by the fact that not all of the assumptions of the Langmuir isotherm were met. At large concentration, a second molecule could have bound to the antibodies or the bound antigens could have begun to interact with other bound or unbound species, changing their adsorption characteristics. It is significant that the best agreement between the Langmuir model and the observation occurs at low analyte concentrations, which are in the biologically relevant range for both MG and cTnI. MG levels after a myocardial infarction are approximately 15-30 ng/mL, and cTnI levels are approximately 1-3 ng/mL. One of the very important factors to measure for each sensor in undiluted serum are the analytical parameters: limit of detection, RMSE of calibration, and coefficient of variation. These are shown in Table 1 for the sensing of cTnI and MG. The limit of detection was calculated to be three standard deviations from the mean baseline noise. The error on a measurement in serum is 0.01 nm. This is larger than if the SPR signal was measured in saline (0.008 nm). This was caused by the subtraction of the reference signal, which increased the noise in the difference measurement in serum by approximately the square root of 2 relative to the noise in a single measurement in saline. The calculated limits of detection for cTnI and MG were 0.7 and 0.9 (42) Andrade, J. Surface and Interfacial Aspects of Biomedical Polymers; Plenum Press: New York and London, 1985.

618 Analytical Chemistry, Vol. 79, No. 2, January 15, 2007

Table 1. Analytical Performance of the Sensors To Detect MG and cTnI in Undiluted Serum

MW (kg/mol) linearity (ng/mL) LOD (ng/mL) RMSEa (ng/mL) CV (%) K (M-1) biological level (ng/mL) a

cTnI

MG

30.0 up to 50 0.7 1.7 13 2.78 × 109 1-3

17.3 up to 50 0.9 1.5 11 2.80 × 109 15-30

The RMSE was calculated in the biological range of 1-25 ng/mL.

ng/mL. These limits of detection were below the biologically relevant range; therefore, the levels encountered in patients after a myocardial infarction can be directly measured in undiluted serum. Other factors to analyze are the accuracy of the calibration curve and calibration model. The RMSE of the calibration curve in the range of 1-25 ng/mL was 1.7 ng/mL for cTnI and 1.5 ng/ mL for MG, as shown in Table 1. The RMSE was concentrationdependent, such that lower errors were observed at lower concentrations. The data points at 50 ng/mL were not included because of the deviation of the calibration model at these high concentrations. From a practical perspective, sensor errors at concentrations ∼50 ng/mL are not considered important because if such levels were encountered in a patient, immediate care would be required and precise knowledge of the analyte concentration would not be needed. The maximum variation that would be expected in these systems on the measurement was estimated by analyzing two different solutions of the same concentration with two different sensors. For cTnI, the replicate coefficient of

variation (CV) was 13%, while for MG the CV was 11%. This is higher than with the saline solutions (around 3-7%), but excellent for biological samples. Generally, errors of about 20% are widely accepted for biological measurements. SPR sensors were proven capable of detecting biomarkers at low ng/mL levels in a highly complex bovine serum matrix containing 72 mg/mL of nonspecific proteins. Bovine serum was employed as a model matrix because there was no cross-reactivity between the bovine myoglobin or troponin proteins with the human proteins, and because of the similar protein composition and concentration of bovine serum as compared to human serum. Detection limits at 0.7 ng/mL for cTnI in the presence of 72 mg/ mL of other proteins correspond to a selectivity of the SPR sensor of 108. The SPR signal at the detection limit was 0.033 nm, which can be converted into a surface concentration of 1.0 ng/cm2 or 10 pg/mm2 for the SPR sensors based on eqs 1-3. The sensor used in this present study had a 6 mm2 sensing area, which corresponds to a 60 pg detection limit for cTnI (∼2 fmol).

NHS-MHA was applied to the sensors, and it minimized the nonspecific adsorption of serum proteins to a level that allowed MG and cTnI detection and calibration in undiluted serum following pretreatment of the sensors by exposure to serum to block NSB sites. The limits of detection for cTnI and MG were below the biologically relevant ranges encountered in patients after myocardial infarctions. A limit of detection of 0.7 ng/mL for cTnI with a RMSE of calibration of 1.7 ng/mL was observed for cTnI, while the biologically relevant levels are at 1-3 ng/mL. For MG, the limit of detection was 0.9 ng/mL with a RMSE of calibration of 1.5 ng/mL. Biologically relevant levels of MG are above 15 ng/ mL. These biologically relevant levels are encountered in patients in the hours following a myocardial infarction. Using SPR sensors and the NHS-MHA layer, analysis of a patient’s serum is now possible to determine if the patient suffered a myocardial infarction. To perform this sensing, the sensors must be pretreated with serum for 15 min before use to block the nonspecific sites remaining on the sensor.

CONCLUSIONS One of the major obstacles to direct measurement of biomarkers directly in undiluted serum is the signal produced when serum proteins adsorb to sensor surfaces in a nonspecific manner. In the case of the SPR sensor, the signals from nonspecific adsorption and from the biomarker of interest are indistinguishable. Therefore, successful measurement of biomarkers requires minimization of the nonspecific part of the signal measured in serum, and a method to extract the part of the signal due to nonspecific binding using a reference sensor. A monolayer of

ACKNOWLEDGMENT We thank the National Institutes of Health, National Institute of Biomedical Imaging and Bioengineering (R01EB004761), and Margaret Barnhart and Ronald Nieman of the ASU NMR facility for interpreting NMR data.

Received for review June 15, 2006. Accepted October 29, 2006. AC061089F

Analytical Chemistry, Vol. 79, No. 2, January 15, 2007

619