1. Introduction

Research into chemical and biochemical sensors has progressed dramatically in the past two decades. At present, much research work is focused on the development of systems capable of multi-analyte detection in a single sample, for environmental, clinical or security applications. Optical sensors have great potential in this field because of their ability to probe surfaces and films using a range of optical phenomena with low noise and high sensitivity. In addition, they have advantages in speed and immunity from interference over electronic approaches, and permit in-situ sensing and real-time measurements [1]. Optical sensors are also suitable for miniaturization and for remote and multi-analyte sensing. Fluorescence-based array biosensors hold particular promise, due to their high specificity and tolerance to temperature changes and to non-specific binding, compared to refractive index based sensors. Duveneck et al have demonstrated a grating-coupled thin-film planar waveguide array biosensor exhibiting a sensitivity improvement of about two orders of magnitude compared to conventional fluorescence excitation and detection systems [2], and Ligler et al. have employed an end-fire-coupled thick planar waveguide array biosensor to detect six toxins simultaneously at concentrations down to 0.5 µg/L [3], for example. Recently, Burke et al have demonstrated an alternative approach to multianalyte waveguide fluorescence sensing, in which the fluorescent molecules are excited externally and the fluorescence is collected in multiple ridge waveguides upon which the sensing films had been deposited [4].

An optical biosensor employing fluorescence-based detection of the binding of tagged biomolecules to the surface of an optical waveguide chip is described here. The fibre-pigtailed chip consists of a channel waveguide circuit which distributes evanescent excitation light to 32 separate sensing patches on the chip surface. Bio/immuno-chemistry may be used to sensitise each of the 32 patches to a specific analyte and a microfluidic system is used to automatically handle the sample injection over the sensor surface, enabling rapid, simultaneous and high-sensitivity fluorescence detection of up to 32 pollutants. A fibre-coupled detection array monitors the 32 separate fluorescence signals, and software controls the laser, fluidics, data acquisition and processing of the fluorescence signals and records the laser power and ambient and chip temperature. In this paper we describe sensor chip fabrication and operation and give results for detection of a key pollutant, estrone, in water. Estrone pollution, which arrives in waterways as a byproduct of pharmaceuticals, is potentially deleterious to human wellbeing due to its hormonal activity. A characteristic immunoassay calibration curve has been generated that yields a detection limit of 1 ng/L with a range (without dilution) up to about 1 µg/L. Water samples can be analysed directly without time-consuming concentration techniques.

Monitoring water quality and identifying pollution sources are emerging as important tasks in the future management of rivers, which are major sources of water for human consumption. The sensor chip described here is a central part of a cost-effective on-line water-monitoring instrument designed to rapidly and simultaneously measure a plurality of low molecular weight organic pollutants, with the potential for remote control and surveillance.

2. Sensor device fabrication

The multisensor chip was fabricated by potassium ion-exchange in BK7 glass [5]. An aluminium masking film was deposited on the substrate, and the waveguide circuit was defined by photolithographically defining and etching tracks in this film, following the schematic layout shown in Fig 1. Ion exchange was carried out by immersing the masked substrates in KNO₃ at 400°C for 2 hours to produce channel waveguides. This approach allows excellent control of the spatial distribution of optical power at the surface and offers the potential for more dense multisensor integration. The tracks were defined to be 2.5 µm wide up to the end of the Y-junction splitter, with parabolic tapers widening to 30µm introduced into each waveguide branch after the Y-junction splitters to reduce the optical power density at the waveguide surface, and hence the rate of fluorophore photobleaching [6]. Reducing the laser power would also reduce the rate of photobleaching but would diminish the fluorescent signal, degrading the signal to noise ratio (S/N). Waveguide broadening reduces the incident intensity upon each fluorophore while proportionally increasing the number of fluorophores illuminated and is therefore expected to maintain the peak signal power. The 32 sensor windows of 1.5 mm length and 0.3mm width were photolithographically defined in photoresist on the chip as shown in Fig. 1, and a 1µm thick silica isolation layer was sputtered on to the surface. The 32 windows were then opened in this film by lift-off in acetone, allowing access of the liquid analyte to the surface of the waveguide (and the evanescent field) at these positions. The ends of the chips were polished to allow fibre butt-coupling resulting in an overall length of 67mm and width of 15mm.

 

Fig. 1. Schematic diagram of multisensor chip

Download Full Size | PDF

The isolation layer prevents excess losses due to contact with the flow cell and environment outside the sensing regions. In order to improve the sensitivity of the sensor, by enhancing the intensity of the excitation light at the waveguide surface in the windows, a 35 nm thick high index film of Ta₂O₅ was deposited over the entire surfaces of the chip by RF sputtering in an atmosphere of 1:9 oxygen:argon at a total pressure of 11mTorr [6]. A polarisation-maintaining (PM) fibre pigtail with PM connector was permanently bonded to the input end of the sensor chip with UV-curing epoxy. Fig. 2 shows a photograph of a sensor chip before pigtailing.

 

Fig. 2. Photograph of a sensor chip

Download Full Size | PDF

3. System operation

The apparatus for immunoassay measurements is shown in Fig. 3. Light from a semiconductor laser emitting approximately 5mW at 637±2nm is coupled into the input waveguide of the sensor chip using the polarization maintaining single mode fibre pigtail. This input power is divided equally into four parallel waveguides using three Y-junction splitters and, in the 32 exposed sensing regions, the evanescent field is able to interact with the analyte, forming spatially separated sensing spots. If a fluorophore is brought within a few 100nm of the sensing spot the evanescent field of the guided light will excite the fluorophore, resulting in fluorescence. This fluorescence is collected without the use of lenses by an array of 1mm core diameter high numerical aperture polymer optical fibres located directly under the sensor chip, so that the fibre ends are ~1mm from the fluorescent molecules at the waveguide surface. Crosstalk between patches is minimized by ensuring that the emissive regions of adjacent patches are not within the angular acceptance cone (NA) of the fibres. The fluorescence is filtered to remove stray excitation light with interference filters held above the photodetector windows, and detected by a silicon photodiode with integral amplifier having a noise equivalent power (NEP) specified as 100fW.Hz-^½. The signal is amplified to give a responsivity of ~50mV/pW and low-pass filtered at 1.4Hz. A micro-flow-cell is affixed to the chip over the 32 patches to supply sequences of solutions to the sensor surface. The cell, with one inlet and one outlet, was molded from acrylic polymer to provide a single chamber of width 5 mm, length 26 mm and depth 35 µm which covered all 32 patches so that water samples can ultimately be analysed for 32 different analytes simultaneously. The pumps and valves which supply the solutions, the laser, and the data acquisition system are controlled by a computer integrated in the housing. The fluorescence power from each sensing site, the laser power, the ambient temperature and the temperature at the sensor are recorded at a rate of 8 samples/s after analog to digital conversion using a commercial card, and the data are logged and analysed on the PC. The temperature was measured using LM35 semiconductor temperature sensors in the photodetector housing and at the bottom surface of the sensor chip, to allow study of any temperature-dependent effects and potentially to allow temperature compensation. The optical detection limit of the full system, defined as 3×NEP, has been found experimentally to be equivalent to ~500fW of fluorescence power. The fluorophore chosen for this work was Cy5.5 (Amersham Pharmacia). The wavelength of peak absorption of Cy5.5 is approximately 675nm and there are other dyes, such as Cy5 and Alexa 633, with peak absorptions closer to the 637nm excitation wavelength used here. However, using Cy5.5, the wavelength shift between the excitation wavelength and the emission wavelength is ~50nm, compared with ~25nm for Cy5, so that scattered excitation light is easier to filter from the signal. Further, Cy5.5 has a flat shoulder in its absorption spectrum near 637nm, reducing the effects of any wavelength variation in the excitation source.

 

Fig. 3. Experimental apparatus

Download Full Size | PDF

4. Surface modification and immunoassays

This multisensor platform may be applied to a wide range of analytes according to the surface attachment protocol. In this case, characterisation of chip and instrument performance for immunofluorescence sensing was carried out with the single analyte, estrone, to allow direct patch-to-patch comparisons. Initial characterisation using multiple analytes would not have allowed separation of optical performance issues from biochemical variations. The entire sensor surface was chemically modified in order to render the chip specific to estrone. In the case of multiple analytes, each patch would be modified separately by microdispensing specific chemicals only onto the patches, reducing the usage of reagents. The procedure adopted was designed to reduce non-specific binding and to enable repeated use [7]. The chip was cleaned and trimethoxysilane was applied to the dried surface for 1 hour. The silanised surface was rinsed and blown dry at room temperature and aminodextran was coupled to the silanised surface. The sensor surface was then immersed in the analyte derivative estrone carbonate acid dissolved in N-Dimethylformamide (DMF) together with N, N’-Dicyclohexylcarbodiimide (DCC). This procedure leads to a high surface density of binding sites specific to estrone antibodies. Non-specific binding of molecules other than estrone antibodies is limited due to the shielding of the glass surface by the aminodextran [7].

The performance of the biosensor was demonstrated by measuring the response to eight known concentrations of the analyte estrone in deionised water, ranging from 0 ng/L (blank) to 10 µg/L. 100 µL of antibody solution containing 60 ng/mL of labelled affinity purified polyclonal anti-estrone antibody in 10-fold phosphate buffered saline (PBS) were added to 900 µL of each standard solution and mixed thoroughly to allow the analyte to bind to the fluorescent-labelled antibody molecules, according to its concentration. The incubated solution was then pumped over the sensor surface as described below. Those antibodies which have not been bound to analyte molecules are free to bind to the sensor surface. The output of one sensing channel during a sensor test cycle using Cy5.5 labelled anti-estrone and a blank sample is shown in Fig. 4, and progressed as follows: A constant flow of PBS was established through the micro-flow-cell, and the background signal, due to laser breakthrough and background fluorescence, was measured. While the incubated sample was being loaded and injected into the flow-cell, the laser was turned off to prevent the onset of photobleaching. After binding of the incubated sample at the sensor surface for ~10 minutes, the cell was flushed with PBS again and the laser turned on, allowing fluorescence from the Cy5.5 dye to be detected. Fig. 4 shows that, once the background has been subtracted, a peak signal of ~155pW was obtained and that this was bleached in a few seconds. Nonetheless, there is no difficulty in acquiring this signal, and the signal to noise ratio is of order 1000. The value of this peak signal for all the patches is shown in Fig. 6 for an antibody concentration of 10ng/mL and is discussed below. After bleaching, the surface was regenerated by 0.5% sodium dodecyl sulphate (adjusted with HCl to pH 1.8) to remove the antibodies bound to the surface, so that another measurement could be carried out, with the entire cycle time taking less than 18 minutes. Surface regeneration may be carried out up to 400 times without significant degradation of the surface chemistry.

 

Fig. 4. Sensor test cycle for zero estrone in water (blank) using 6ng/mL anti-estrone

Download Full Size | PDF

A calibration curve for normalised fluorescence power against estrone concentration was obtained by repeating this procedure using the 3ng/mL dye-labelled antibodies and water samples containing estrone in the concentration range 0–10ppb (µg/L), prepared as described above. The blanks were repeated 9 times to yield a good statistical sample and all other concentrations were repeated thrice. The averaged data were fitted to a logistic function, representing a close approximation to the shape of a typical immunoassay curve, and an example is given in Fig. 5, which shows the estrone calibration curve for one sensing patch normalised to the mean of the signals for the blanks. In the case of multianalyte sensing, all 32 patches can be calibrated simultaneously using a mixture of appropriate antibodies and target analytes, so that the calibration procedure does not have to be repeated 32 times.

 

Fig. 5. Averaged calibration curve for estrone, using 3ng/mL anti-estrone

Download Full Size | PDF

The detection limit (LoD) for estrone for this sensing patch was found to be just below 1 ng/L, and the test mid point of the response curve occurred at a concentration of 40 ng/L. The LoD was taken to be the concentration in Fig. 5 at which the signal has fallen below the blank value by thrice the standard deviation of the blanks. Concentrations up to 1µg/L are measurable and higher concentrations could be measured by diluting the sample.

The excitation power is expected to decay along the waveguides due to propagation losses, so that a comparison of the signal from each patch for the same analyte is expected to yield information on the sensor uniformity and the waveguide losses. Figure 6 shows the peak power recorded, with the background subtracted, for all 32 sensing patches for zero estrone concentration (blank) using an antibody concentration of 10ng/mL. Patches 1–4 are nearest the waveguide input and patches 29–32 are nearest the waveguide outputs. The triangles show the signal from each patch with co-directional propagation of light and of liquid flow through the cell. In this case, several patches at the input end have saturated the detectors, due to the high antibody concentration, but beyond patch 16 the signal tends to fall. (Saturation can be avoided by further optimisation of the amplifier gain or the antibody concentration.) From these data it appears that the waveguide losses may be high. However, reversing the direction of liquid flow results in a signal which is higher at the output end of the waveguides, as shown by the circles in Fig 6. This is believed to be due to the antibody concentration along the flow-cell reducing as antibodies bind to the surface.

 

Fig. 6. Fluorescent signal magnitude along the chip for co- and contra-directional fluid flow (wrt light propagation).

Download Full Size | PDF

In these experiments, the entire surface of the chip exposed to the analyte (~100mm²) can bind anti-estrone, while in future multisensing experiments only one individual patch of ~0.5mm² area, will be modified for the same analyte by microdispensing specific chemicals [8], so that in real use antibodies will not be expended except at the specific sensing patch. There is nonetheless a significant contribution to patch-to-patch nonuniformity from optical losses along the channels, as 12 channels in the flow path were saturated in the former case and only three channels were saturated when the flow direction was reversed.. These losses may be reduced by process developments such as improvements in silica and tantalum pentoxide film deposition to reduce absorption and scattering, and by improving fibre to waveguide mode-matching. Automated calibration of the sensor chips can be used to compensate for optical nonuniformities in operation.

The first multi-analyte test of this sensor, successfully carried out with a mixture of six analytes, atrazine, bisphenol A, estrone, isoproturon, sulphamethizole, and propanil, will be reported elsewhere. The detection limit of ~1ng/L is two orders of magnitude better than that required by EU legislation for organic pollutants, and it is expected that improvements in sample handling and signal processing will reduce this further. The ability to regenerate the sensor automatically after an assay, as part of an automated protocol, means that the sensor and system may be left unattended for months, in normal operation, before a chip needs replacing. There is potential for further miniaturisation, increased integration and reduced use of reagents as the fluorescent signal is obtained from an irradiated area of ~0.05mm2 and volume below 20pL.

5. Conclusions

An integrated optical fluorescence-based multisensor chip has been realised and integrated with fluidics, detection system, surface chemistry, immunochemistry, and a computer for control and signal processing. The performance of this biosensor system has been demonstrated with the key pollutant estrone. A detection limit of 1 ng/L was achieved, with a range up to 1µg/L. The use of microspotting to modify individual patches on the sensor surface will allow future measurement of up to 32 multiple pollutants simultaneously in one sample. The instrument is being produced in portable form where it can be taken to a site and connected to the internet for remote control and data handling, and field trials are underway.