Anal. Chem. 1997, 69, 507-513
All-Solid-State Miniaturized Fluorescence Sensor Array for the Determination of Critical Gases and Electrolytes in Blood Alfredo E. Bruno,*,† Steven Barnard,‡,§ Marisel Rouilly,† Adrian Waldner,‡ Joseph Berger,‡ and Markus Ehrat†
Bioanalytical Research and Central Research Laboratories, Ciba-Geigy Ltd., 4002 Basel, Switzerland
We describe a six-channel, all-solid-state, miniaturized fluorescence sensor array for the precise determination of blood analytes for medical diagnostic purposes. The device features superblue LEDs as light sources, GRIN optics, and photodiodes, assembled according to pigtailing procedures (Bruno, A. E.; et al. Trends Anal. Chem. 1994, 13, 190-198). The numerical aperture of the fluorescence optics is 0.46, rendering a collection efficiency of 2.4%. The performance of this instrument has been evaluated in terms of dynamic response, linearity, channel reproducibility, reversibility, long-term drifts, photobleaching of indicator, cross-talk, ionic strength, and selectivity in pH measurements. The responses of the pH sensing membranes were optimized in the physiological range. Responses are linear with typical values of ∼1.5 V/pH unit, with limits of decision of 24 mV, which corresponds to pH resolutions of 0.03 pH unit. Under continuous illumination, using calibration buffers, the sensors display nonstatistical differences within 2 standard deviations over a period of 6 h, and it is shown that, under discontinuous illumination, the membranes can be used in more than 2000 measurements without need of calibration, in contrast to electrochemical sensors which require periodic calibration. After selecting the appropriate combination of LEDs, excitation and emission filters, and sensing membranes, the instrument was used to determine the concentrations of various critical blood analytes in buffer solutions in the various channels. Similar measurements in untreated blood reproduce the reported results. Patient care often requires the routine determination of various critical blood analytes (CBAs) in untreated blood. The basic menu of CBAs to be measured, with the concentrations of clinical significance and tolerances, is summarized2,4 in Table 1. In hospitals, CBAs are traditionally measured using electrochemical sensors in centralized laboratories. Although electrochemical methods are rather complicated to use and expensive (primarily because, to remedy the inherent instability of the electrodes, *Reprint requests: alfredo
[email protected]. † Bioanalytical Research Laboratory. ‡ Central Research Laboratory. § Present address: Chiron Diagnostics, 63 North St., Medfield, MA 02052. (1) Technical Specifications of the Ciba Corning series 400 Blood Gas System. Chiron Diagnostics, 63 North St., Medfield, MA. (2) Meyerhoff, M. E. Trends Anal. Chem. 1993, 12, 257-266. (3) Wolfbeis, O. S., Ed. Fiber Optics Chemical Sensors and Biosensors, Volumes I and II; CRC Press: Boston, 1991. S0003-2700(96)00855-4 CCC: $14.00
© 1997 American Chemical Society
Table 1. Most Required Critical Blood Analytes,2,3 Their Clinical Ranges, and Their Concentration Tolerancesa CBAs
clinical range
tolerance
pH p(CO2) p(O2) [K+] [Na+] [Ca2+] [Cl-] lactate glucose creatinine
6.6-7.8 pH units 10-40 mmHg 30-100 mmHg 2.5-6.5 mM 120-160 mM 0.80-1.5 mM 95-110 mM 1-110 mM 20-65 mg/dL 0.7-3.5 mg/dL
(0.02 pH unit (3 mmHg (5 mmHg (0.2 mM (3 mM (5 µM (2 mM 5 mg/dL 5 mg/dL 10%
a [Cl-], lactate, glucose, and creatinine are target substances that have not been demonstrated with the present device.
continuous calibrations of the electrodes are imperative), and despite the enormous progress that has been made in the development of optical sensors in the research front during the last decade,3,4 this technology endures by being constantly adapted to the changing market. However, the limitations of electrochemical technology may be reached as the market demands growth in directions unsuitable for electrochemistry, i.e., very low cost diagnostics, single-use measurement, or in vivo analysis. The ability of electrochemistry to fulfill these future needs will probably be limited by the inherent problems of this technology: instability of signal, expensive and multistep fabrication techniques, limitation of menu expansion, and the need for frequent calibration. As an alternative to the electrochemical methods, optical methods have shown enormous progress on the research front and possess many advantages;3,4 still, they have not lived up to their expectations commercially. Optrodes have not delivered on their promise to bring new advantages to users and manufacturers because their performance has not met the specifications of their electrochemical equivalents (Table 1). This unfulfilled promise can be understood in terms of the unsolved technical hurdles still facing optical technology with regard to chemical recognition and signal transduction. Optical technology is driven by new developments in both chemistry and optoelectronic hardware. However, both types of developments are not mutually exclusive; e.g., if the input/output of light is highly efficient, photobleaching would be reduced, translating in a reduction in sensor calibration frequency. Another overriding factor that precludes instrument (4) Arnold, M. A. Anal. Chem. 1992, 64, 1015A-1024A. (5) Thompsom, R. B.; Lakowicz, J. R. Anal. Chem. 1993, 65, 853-856.
Analytical Chemistry, Vol. 69, No. 3, February 1, 1997 507
manufacturers from switching from electrochemistry to optics is cost. Research prototypes reported in the literature often rely on expensive components, traditional optics, and sophisticated electronics. The use of some indicators which are incompatible with existing bright light emitting diodes (LEDs) or laser diodes and membrane deposition methods which are not suitable for replication is also on the list of arguments from manufactures. Among the various optical sensing modes, fluorescence is the preferred one because of the higher selectivity and lower sampling volume required. The presence of an analyte correlates with the changes in fluorescence intensity3,4 or with the fluorescence lifetime5 of indicators embedded in a polymeric matrix. These polymers are immobilized by, e.g., spin-coating or thick-film technology6 on a glass or plastic substrate, forming a thin planar membrane permeable to the analyte of interest. Some sensors use a coating at the distal end of a fiber7 or a coating of a singlemode or a bifurcated fiber.7,8 Optical devices rarely operate at theoretical limits. The inefficient use of the available light, propagation losses, noise due to mechanical vibrations and schlieren effects, drifts due to thermal expansions, etc. are some of the factors that preclude ideal performance. To overcome these problems, and taking advantage of the great number of miniature optical devices that have been made available to optical designers, Bruno et al.9 recently revisited an old approach, pigtailing, to construct small optical detectors for chemical analysis. In this article we describe a six channel, all-solid-state, miniaturized fluorescence sensor instrument, featuring superblue10 LEDs as excitation sources, GRIN optics to input/output light, and photodiodes as light detectors assembled according to pigtailing1 procedures. The six fluorescence channels can be used to determine simultaneously the concentrations of various critical blood analytes by selecting the appropriate combination of excitation light (i.e., LED and excitation filter), emission filter, and sensing membrane. The performance of this instrument has been evaluated in terms of response, reproducibility, reversibility, photobleaching of indicator, selectivity, etc. in the determination of pH, using new chemical formulations for the transducing membranes. The instrument was also used to determine several of the analytes listed in Table 1. INSTRUMENTAL SECTION Pigtail Approach. The aim behind the pigtail approach9 is to minimize, or completely avoid, the number of glasslike media/ air optical interfaces in the optical paths by gluing the components with index-matching adhesives. Several problems, known to degrade instrumental performance, are in this way addressed simultaneously: (i) reflections and strong refractions are avoided, (ii) mechanical problems are reduced, and (iii) thermal equilibrium is reached fast. If total pigtailing is achieved, the final device is an all-solid-state optical device subject to a minimum of light losses, which lacks noise and drifts due to vibrations, schlieren effects, and thermal instabilities. Because their focusing properties would disappear upon gluing, conventional lenses are not suitable for pigtailing purposes, and we employ gradient-index (6) Gala´n-Vidal, C. A.; Mun ˜oz, J.; Domı´nguez, C.; Alegret, S. Trends Anal. Chem. 1995, 14, 225-231. (7) Hauser, P. C.; Tan, S. S. Analyst 1993, 118, 991-995. (8) Hale, Z. M.; Payne, F. P. Sens. Actuators B 1994, 17, 233-240. (9) Bruno, A. E.; Maystre, F.; Krattiger, B.; Nussbaum, P.; Gassmann, E. Trends Anal. Chem. 1994, 13, 190-198. (10) Matsuoka, T. Adv. Mater. 1996, 8, 469-471.
508
Analytical Chemistry, Vol. 69, No. 3, February 1, 1997
(GRIN) lenses. The other solid-state components which complete the present device are LEDs and photodiodes (PDs). The pigtailing of LEDs to GRIN lenses involves the grinding or drilling of the plastic body of single LEDs down to almost the emitting crystal (as shown in Figure 3 of ref 9) and then gluing it with index-matching glues. The roughness in the ground surface is not crucial because it disappears upon gluing with indexmatching glue. Higher reproducibility was obtained by grinding many LEDs at the same time using specially made tools. Two types of glues are employed here: UV curing (e.g., NOA 81 from Norland Products, New Brunswick, NJ) and two-component epoxy composites (e.g., Araldit 2020, Ciba-Geigy, Basel, Switzerland). Filters. A crucial step in the construction of a fluorescence instrument is the choice of filters for the excitation and emission light. Small interference filters, obtained by cutting large ones, are still bulky when compared to the dimensions of the other optical components employed here (i.e., thickness of a few millimeters in the substrate would already result in a degradation of the quality in the GRIN optics employed). To overcome this difficulty, we took advantage of a technology developed by Schmidt et al.11 to transfer interference filter coatings, of micrometer thickness, from a large glass substrate onto the surfaces of GRIN lenses. As this technology also allows the deposition of interference coatings in materials having low melting points, such as plastics, we also used it to pigtail interference filter12 onto LEDs and PDs. The bandwidth of the thus manufactured filters (O.I.B. GmbH, Jena, Germany), after pigtailing, is 15 or 20 nm, with a transmittance of more than 75%, a precision in the transmission maximum of (4 nm, and extinction ratios of better than 1000 at the blue and red sides of the transmission peak. Harmonics transmission peaks are suppressed upon adding subsequent filtering layers. The best filtering is obtained when the filter is sandwiched between two 1/4 pitch lenses, because rays strike the filters in a collimated fashion (see Figure 2 of ref 12). However, for practical reasons, we mostly used filters pigtailed at the end of 1/2 pitch lenses or at the flat window of LEDs or PDs, as shown in Figure 1. LEDs and PDs. LEDs are solid-state light sources requiring low driving power (e.g., could be driven even by batteries), displaying excellent intensity stabilities of around ∆I/I < 10-5 (i.e., as good as the current power supply), making them ideal sources for sensors. Most of the benefits of using LEDs in chemical sensors have been reported.13 To be added is the ∼100-fold increase in the brightness of recently available GaN blue LEDs as compared to LEDs based on SiC used in previously reported sensors.3,13 Progress made last year10 by Nakamura’s team at Nichia Chemical Industries Ltd. (Japan) has resulted in a 10100 times increase in the luminosity of GaN- and InGaN-based superblue LEDs (we measured luminosities of 3.6 µW (5% when driven by 20 mA). The emission spectra of these LEDs have a cutoff at ∼400 nm, with a maximum at 470 nm, which makes them ideal excitation sources for fluorescein-labeled indicators. For the experiments requiring higher illumination, we used the multiquantum well Nichia NLPB-500 superblue LED. However, as the efficiency of the collection optics and the fluorescence yields of the membranes increased, (11) Schmidt, E.; Peupemann, J.; Do¨ring, H. Photonics Spectra 1995, May, 126128. (12) Schmidt, E. Laser Optoelektron. 1996, 28, 50-51. (13) Holobar, A.; Benes, R.; Weigl, B. H.; O’Leary, P.; Rasport, P.; Wolfweis, O. S. Anal. Methods Instrum. 1995, 2, 92-100.
Figure 1. Cross sections (x-z and y-z) of the optical system of one of the six identical channels comprising an optical head. PD, photodiode; IMG, refractive index-matching glue; GRIN-1 and GRIN-2, emission and excitation lenses, respectively; NA1 and NA2, numerical apertures of the excitation and emission systems, respectively; 1-5, Propagation paths of five significant fluorescence rays.
good S/N ratios were also obtained with standard blue LEDs (Siemens LB5410). Inexpensive, flat front, plastic-covered, blue-enhanced PDs (BPW34B from Siemens) were used as photodectectors. Interference filters were pigtailed (not only to LEDs but alternatively) to12 PDs to select the emission spectra of interest. GRIN Optics. Multimode optical fibers have been the preferred choice in the quest for miniaturization of previous optical sensors,3 but not ours. Our approach is based on GRIN optics.14 In contrast to the case with GRIN lenses, light emerges from fibers in a divergent fashion, requiring the use of additional optical elements (e.g., ball or GRIN lenses) at their ends to be able to focus the light and augment the light collection field. Gradedindex rod lenses are short segments (∼5-10 mm for 1/4 and 1/2 pitch lenses, respectively) of thick optical fibers (1-3 mm) with refractive index n, having a radial distribution approximately described by16
(
n(r) ) n0 1 -
)
r2A 2
(1)
where no is the refractive index at r ) 0, r is the radial coordinate, and A is the refractive index gradient constant provided by the manufacturer in mm-2 units. According to eq 1, light rays entering (14) SelFoc Product Guide (SPGE0695), Nippon Sheet Glass Co., Tokyo, Japan, 1995. (15) Haugland, R. P. Handbook of Fluorescent Probes and Research Chemicals; Molecular Probes Inc.: Eugene, OR 1992. (16) Saari, L. A.; Seitz, W. R. M. Anal. Chem. 1982, 54, 821-823.
the lens would be attracted by higher refractive indexes and thus bent toward the axis. The focal length of a GRIN rod is related to the gradient constant15 A by
f)
n0A
1/2
1 sin(A1/2Z)
(2)
where Z is the length of the rod in millimeters. The pitch P of a lens is defined as
P)
ZA1/2 2π
(3)
where one pitch describes the longitudinal distance in a GRIN lens traveled by a ray after propagation of an entire sinusoidal period. The ray-tracing of rays entering a GRIN lens with a radial coordinate r1 at angle θ1 will emerge with coordinates r2 and θ2 according to the following transformation matrix:15
[
1 1/2 sin(A1/2Z) r2 ) cos(A Z) n0A1/2 θ2 -n0A1/2 sin A1/2Z cos A1/2Z
[ ]
]
[ ] r1 θ1
(4)
Two cross sections (z-x and z-y) of the optical system of one of the six identical sensors of the optical head is shown in Figure 1. It comprises17 two identical 2 mm diameter, ∼1 cm long, P ) 1/2 (17) Munkholm, C.; Walt, D. R.; Milanovich, F. P.; Klainer, S. M.; Klainer, S. M. Anal. Chem. 1986, 58, 1427.
Analytical Chemistry, Vol. 69, No. 3, February 1, 1997
509
Figure 3. Chemical structure of 4-acrylamidofluorescein.
Figure 2. Schema with the architecture of a six-channel optical sensor array to analyze critical blood analytes. The coordinates are indicated in Figure 1.
GRIN lenses (i.e., half sine wave) having a numerical aperture NA ) 0.46 (NA ) n sin R, thus 2R ) 36°) and A1/2 ) 0.3039 mm-1 (SelFoc SLW-type lenses from Nippon Sheet Glass, Tokyo, Japan). A 1 mm pinhole is placed right in front of the PDs and LEDs to spatially define the viewing zone of the fluorescence membrane. The lenses are glued into a machined, anodized aluminum chassis, which allows the lenses to be positioned with a precision better than 50 µm. GRIN-1 images the LED crystal into the 45° polished facet (i.e., excitation) where the fluorescence membrane is located, whereas GRIN-2 collects part of the induced fluorescence and brings it to the PD (i.e., emission). The flat facet of the emission lens (GRIN-2) is in contact with the curved side of the excitation lens (GRIN-1), defining a volume which is filled with indexmatching glue (see cross section z-y in Figure 1). The collection efficiency, φ, of a GRIN lens, glued to the fluorescence media with index-matching materials (having n ≈ 1.5), can be calculated with
[ (
φ (%) ) 100 × sin2
)]
arcsin(NA/n) 2
(5)
Six-Channels Optical Head. A diagram showing the architecture of the six-sensor fluorescence head is shown in Figure 2. The metallic chassis has six sets of identical pairs of holes to accept the six microfluorescence systems (Figure 1). The distance between the microfluorescence systems is 7 mm along the y dimension. The fluorescence head is used to determine simultaneously the concentration of various CBAs by selecting (i) the sensing membrane, (ii) the appropriate combination of excitation light (i.e., LED and excitation filter), and (iii) the emission filter. The substrate of the fluorescence membrane separates the optical head from the fluid cell. This cell is made of a black plastic material which brings the flowing liquid from one membrane to the other. In most cases, this cell is simply a channel 0.5 mm deep, 1 mm wide, extending from channel 1 to 6. The bottom of the cell is tilted to prevent reflections from reaching the PDs. The whole head is surrounded by an electronically thermostated metallic body (not shown in Figure 1) at T ) 37.5 °C, acting also as a Faraday cage. The LEDs and PDs are wired to a 25-pin connector which is plugged into an analog electronic box. This customized box allows the operator to select the driving current of each LED from 510 Analytical Chemistry, Vol. 69, No. 3, February 1, 1997
5 to 50 mA and the gain factors for each PD from 1.0 × 108 to 5.0 × 1011. The output is read by an A/D converter from a PC, where the data are displayed on-line and subsequently analyzed. The PC also controls the fluid-cycler to select, e.g., the pH sequence of the fluids entering the cell. SENSOR TECHNOLOGY It is well documented that fluorescein and its derivatives, having absorption spectra matching the emission maxima of blue LEDs, show a strong dependence of their fluorescence behavior on the pH of the environment.15 Weak fluorescence is observed in acidic media, whereas strong fluorescence accompanied by a bathochromic shift occurs in basic solutions. The pKa of the phenolic group in fluorescein derivatives varies in the range of 6.5-8.0, thus covering the clinical range (Table 1). A key issue to achieve a good sensor is the proper immobilization of the fluorescent pH indicator dye on a substrate to realize the transducer. Different approaches were described using immobilization on glass,16-18 or cellulose16 or by copolymerization of an acryloyl derivative of aminofluorescein with acrylamide derivatives.17 Despite this progress, several drawbacks still exist for a practical use of the sensors, and the reproducibilities of the immobilization procedures on glass and on cellulose are not as good as needed for commercial applications. The loading of the support is critical with respect to a high background signal affecting the sensitivity of the sensor, and the background signal depends on the substrate used.16 On the other hand, short response times in the range of seconds were observed with the polyacrylamide support.17 The immobilization approach outlined in this work is similar to a reported one,17 consisting of the preparation of a copolymer from two acrylamide derivatives and 4-acrylamidofluorescein (shown in Figure 3). Synthesis of 4-Acrylamidofluorescein. Five grams (14.4 mmol) of 4-aminofluorescein (Fluka AG, Buchs, Switzerland) are first suspended in 200 mL of acetone and cooled to 0 °C. Then, 1.26 mL (15.5 mmol) of acryloyl chloride, dissolved in 2 mL of acetone, is added dropwise to the suspension over 10 min and stirred for 3 h at ambient temperature. The crystals are then filtered off over a frit and washed with acetone and ether. Drying of the brown powder affords ∼5.5 g (94%) of the desired compound. The physical properties of 4-acrylamidofluorescein are as follows: melting point > 200 °C; mass spectrometry (field desorption) renders 402 (100%), 348 (60%), whereas that calculated for C23H15NO6 is 401; the absorption spectrum of a c ) 7.54 × 10-5 M solution shows λmax ) 442 nm in EtOH with ) 9970 and a λmax ) 500 nm in EtOH + 2% 0.01 N NaOH, with ) 89 100. Synthesis of Copolymer. In a reaction flask connected to vacuum and nitrogen, 2.19 g of N,N-dimethylacrylamide, 2.81 g (18) Offenbacher, H.; Wolfbeis, O. S.; Fu ¨ rlinger, E., Sens. Actuators 1986, 9, 73.
of N-tert-butylacrylamide, 200 mg of 4-acrylamidofluorescein, and 25 mg of azobis(isobutyronitrile) were dissolved in 15 mL of dimethyl sulfoxide. The reaction flask was then purged with nitrogen, sealed, and heated at 60 °C for 48 h in a water bath to achieve polymerization. Subsequently, the contents of the reaction flask were diluted with 100 mL of methanol, and the solids precipitated in 2 L of water and then were dried under vacuum for 24 h. The dried precipitate was further purified twice using this latter procedure. Finally, the polymer was vacuum-dried at 60 °C for 2 days. This protocol leads to 3.6 g (69%) of a polymer with Tg ) 156 °C. From IR spectroscopy, the content of N-tertbutylacrylamide was determined to be 45.7 wt %. Membrane Deposition. The pH-sensitive copolymer is dissolved in methanol at a concentration of 50 mg/mL. One milliter of the polymer solution was deposited on a 150 µm thick, 70 × 70 mm2 glass substrate (Balzers, Liechtenstein) and spincoated for 30 s at 3000 rpm (using a Convac 1001 spin-coater), resulting in a transparent film of ∼2 µm thickness. The coated substrates were cut in 7 × 7 mm2 pieces, producing ∼100 individual sensors per glass wafer using an automated diamond cutter instrument (Karl Su¨ss Ritzautomat RA120, Mu¨nchen, Germany). The sensors were stored in the dark at room temperature for months without observable degradation. RESULTS AND DISCUSSION Optical Configuration. The efficiency of the optical system shown in Figure 1 was optimized in terms of the angles R and β, and the lateral shift of GRIN-2 with respect to GRIN-1 was optimized by ray-tracing, using eq 4. The theoretical predictions were subsequently corroborated experimentally. The dominant criteria in the optimization process were the optical efficiency (i.e., I/O of light to/from the membrane), followed by aspects aiming to facilitate construction. A compromise between these two factors resulted in R ) β ) 45° and, the z axis of GRIN-2 is shifted in the -x direction with respect to the optical axis of GRIN-1 in such way that the polished edges of both rods coincide, as shown in the x-z cross section of Figure 1. Because the 45° polishing results in shortening in the middle of the GRIN lenses of 1 mm (i.e., their diameters are 2 mm), the lenses were purchased longer to stop the polishing at the right length. The numerical aperture of the GRIN lenses used in both excitation and emission is NA ) 0.46, and, as the LEDs used display a high directivity, the illumination spot contains about 30% of the total LED power in the absence of a filter. The fluorescence collection efficiency of the optical system, calculated with eq 5, is φ ) 2.4%. Although this configuration is not perfect for imaging purposes (i.e., it is subject to astigmatism along the x dimensions, mainly due to the 45° facet of both GRINs), it is very efficient to input/ output light to a fluorescence membrane. As a result of the 45° facet in GRIN-1, the image of the LED crystals (of 250 × 250 µm2) appears slightly enlarged at the fluorescence membrane (∼280 × 300 µm2). When using superblue LEDs driven by 20 mA pigtailed13 to an interference filter and a GRIN lens, this spot is illuminated by ∼1 µW. The LED-induced fluorescence from the membrane, which propagates toward the PD within the ∼55° distorted cone defined by the NA of the emission system, such as rays 3 and 4 in Figure 1, is collected by the excitation rod GRIN-2 after a short propagation path through GRIN-1, and the index-matching glue reaches the PD after crossing a 1 mm pinhole. On the other hand, fluorescence rays outside this NA, such as rays 1, 2 and 5, walk off the collection system.
Figure 4. Fluorescence signal response with pH varying from 6 to 8 in steps of 0.15 pH unit over a period of 30 min. Each measurement lasted 1 min.
Early optical configurations were based on 1/4 pitch (instead of 2 pitch) GRINs. The 1/4 pitch lenses illuminate a larger surface with a lower photon density and thus minimize photobleaching of the membrane, and the positioning of the lenses is less critical. However, photobleaching was not a dominant issue, and 1/4 pitch lenses resulted in higher scattering light. Another optical configuration explored was involved switching the positions of the PD and LED (i.e., perpendicular excitation and tilted emission), but the results were not better than those obtained with the configuration shown in Figure 1. The LEDs employed are multiquantum well having a wide emission spectrum. Single-quantum well LEDs, having a narrow emission bandwidth (i.e., delivering more power per nanometer), would have been more suitable. However, they are more expensive than the filter/LED combination used to select the excitation wavelength for the construction of prototypes that sufficed to obtain good fluorescence S/N. However, for largescale production, the price of customized LEDs will drop, and its use would eliminate the step of pigtailing filters to LEDs. Linearity, Reversibility, and Reproducibility. Figure 4 displays the fluorescence signal changes induced by buffers having a pH varying from 6 to 8 in discrete steps of 0.15 pH unit over a period of 30 min, each measurement lasting 1 min. The sensors display a high degree of reversibility, as shown in Figure 5 and are sufficiently fast (the response time at signal ) 90% is less than 0.5 min). The calibration curves for each sensor as a function of time are shown in Figure 6. They are constructed by plotting data as in Figure 4, normalizing them for intensity and offsetting the first pH measurement to zero (i.e., pH ) 6.6 correspond to intensity ) 0 in Figure 6). All signal responses are linear with pH over the entire physiological range, but they display different slopes, indicating that the dynamic response varies from channel to channel, in spite of the fact that they have fluorescence membranes from the same batch. This is, in part, due to the fact that channels 1-3 are slightly different from channels 4-6. The former set was built with the excitation and 1/
Analytical Chemistry, Vol. 69, No. 3, February 1, 1997
511
Figure 5. Repetitive titration curves (four) for three pH values covering the physiological range.
Figure 7. Sensor response with pH. The linear physiological range is indicated, and the dotted lines indicate the confidence limits for the calibration function for 95% confidence. For this sensor, the dynamic response is 1.41 V/pH unit, and the limit of detection is 24 mV, corresponding to a resolution of 0.033 pH unit.
Figure 6. pH vs fluorescence intensity calibration curves of six sensors from the same optical head. The time for each measurement, correlated to the pH, is also given in the lower horizontal scale. The raw data for each channel are identical to those shown in Figure 4 and have been normalized for intensity and offset to zero for the first pH measurement (i.e., at time ) 0 min or pH ) 6.6). Sensor 1, ) O; 2, 9; 3, #; 4, +; 5, *; and 6, 2.
emission filters glued between two 0.25 pitch lenses (as shown in Figure 2 of ref 12), whereas in the second set of channels, the interference filters are glued directly on the flat window12 of the LEDs and PDs. The average dynamic responses associated with channels 1-3 and 4-6 are 3.8 ( 0.7 and 1.76 ( 0.70 V/pH unit, respectively. The uncertainty of (0.7 V/pH unit in the dynamic response among the channels constructed with identical components is mainly due to differences in their optical I/O efficiencies. Although care is taken in the assembling of the optical components, the gluing steps are difficult to replicate. Variations among the characteristics of LEDs, PDs, electronics, and membrane quality also contribute to the observed variations in the channel responses. The spin-coating process, as the deposited polymer flows from the center to the outer edges, produced sensors having typical thickness variations of about 5% within the same batch and within 10% from different batches. Considering that the six channels are intended for sensing different CBAs, the lack of reproducibility in channel replication is not a dominant issue. Sensor Behavior. The response of a typical sensor in the physiological range is 1.41 V/pH unit as shown in Figure 7. The dotted lines indicate the confidence limits for the calibration function for 95% confidence. The statistical limit of decision associated with these measurements is 24 mV, which corresponds to a resolution of 0.033 pH unit. This noise has contributions from temperature and pressure variations caused by the fluids flowing 512 Analytical Chemistry, Vol. 69, No. 3, February 1, 1997
Figure 8. Long-term stability of the sensor membranes over a period of more than 9 h for pH ) 6.7 (*, 6.9 (#), 7.1 (O), 7.2 (4), 7.4 (3), 7.7 (+), 8.0 (×), and to 8.5 (b).
through the cell at ∼1 mL/min and, to a lesser extent, due to electronic noise. Although this resolution is not as good as the state-of-the-art obtained by commercial instruments based on electrochemical detection of 0.02 pH unit, it provides sufficient precision for a clinical decision. Titrations curves, such as those shown in Figure 4, were generated continuously for periods of ∼9 h for various pH values within the physiological range to determine the long-term drift of the total system (i.e., including chemistry, instrumentation, and fluidics). The results are summarized in Figure 8. Over 6 h of continuous operation, there is no statistical significant difference within 2 standard deviations, whereas during 9 h the difference is within 3 standard deviations in the titration curves. Such a sensor would need calibration only once every 6 h to comply with the suggested clinical tolerances, or once every 9 h if the sensor is precalibrated. However, as in these measurements the LEDs continuously illuminate the membrane during the entire test, and the degradation in the performance of the sensor is believed to be caused primarily by photobleaching, it is expected that the lifetime of the membranes would be much longer if they were illuminated only during the measurements (i.e., a linear extrapolation from the present data renders a lifetime of ∼2 weeks). Alternatively, if the LEDs were powered only during the actual
measurement of ∼10 s, a 6 h illumination period would correspond roughly to 2160 measurements. Such an estimate shows the advantage of optical systems over electrochemical systems that require 24 h operation with calibrations every hour. In this regard, the ability of reducing calibration frequency to run the instrument would add significant value to clinical diagnostics by reducing the amount of biologically hazardous wastes and calibration solutions, thereby lowering costs per test. In other words, an optical instrument increases the usable instrument time into sample measurements, rather than calibrations. CONCLUDING REMARKS The present optical device realizes a balance between the optoelectronic and the chemical nature of the sensing system to obtain the described technical benefits. For long-term stability, it was found that it is important to work at the highest amplification level possible (10 pA/V) to require less excitation power to reduce photobleaching, while the sensor chemistry was optimized to provide a robust signal without sacrificing dynamic response. The weakest aspect in the pigtailing procedures is channelto-channel replication, translating in the observed lack of reproducibility (Figure 6). However, as in a commercial instrument, each sensor will be used to determine the concentration of different CBAs, thus the best parameter to judge the sensors is linearity. In this regard, signal responses are linear with pH over the entire physiological range (Figures 4 and 6).
The present instrument has also being demonstrated with membranes to be suitable for measurement of p(CO2), p(O2), [Ca2+], [Na+], and [K+], and the instrument performance was comparable to that reported here for pH. Furthermore, first pH measurements in untreated blood reproduce the herein-reported results in buffers. Current measurements aim to characterize degradation mechanisms of membranes by blood over a long periods (e.g., as shown in Figure 8). As medical diagnostics moves into a more cost-sensitive phase, it is vital to offer alternatives that could bring added benefits. In this regard, the most important advantage of optical sensors, as compared to electrochemical ones, appears to be the possibility of both a significant reduction in calibration protocols for comparable performances and a simplification in the manufacturing of the disposable part of the sensors. ACKNOWLEDGMENT We thank T. Allison (Chiron Diagnostics, Sudbury, England) for the design and construction of the electronic boxes and G. Kraus for the statistical analysis. Received for review August 21, 1996. Accepted November 11, 1996.X AC960855N X
Abstract published in Advance ACS Abstracts, January 1, 1997.
Analytical Chemistry, Vol. 69, No. 3, February 1, 1997
513
