Anal. Chem. 1997, 69, 3413-3418
Generating Sensor Diversity through Combinatorial Polymer Synthesis Todd A. Dickinson and David R. Walt*
The Max Tishler Laboratory for Organic Chemistry, Department of Chemistry, Tufts University, Medford, Massachusetts 02155
Joel White and John S. Kauer
The Department of Neuroscience, Tufts School of Medicine, Boston, Massachusetts 02111
A new approach for rapid, simple generation of uniquely responding sensors for use in polymer-based sensor arrays has been developed. Polymerization reactions between different combinations of two starting materials have been found to lead to many new, unique sensors with responses not simply related to the proportion of the starting materials. This approach is demonstrated in two ways: (a) the use of discrete polymer sensing cones each comprised of a specific monomer combination and (b) the fabrication of a gradient sensor, containing all combinations between the starting and ending monomer concentrations. Gradient sensors were fabricated using two different binary monomer systems, with both systems showing regions of broadly diverse fluorescence responses to organic vapor pulses.
Combinatorial methods1-3 are being employed in a growing number of research areas including drug discovery4 and optimization,5 complex receptor molecule generation,6 catalytic antibody production,7 and superconductor synthesis.8 In the chemical sensor field, researchers are increasingly adopting the array format, in which a series of discrete, cross-reactive sensing regions is used in conjunction with pattern recognition schemes for detecting and identifying a broad range of analytes. Most of these array sensors employ different polymers to provide selectivity, exploiting such properties as polarity,9,10 swelling,9-12 conductiv(1) McDevitt, J. P.; Lansbury, P. T. J. Am. Chem. Soc. 1996, 118, 3818. (2) Fathi, R.; Rudolph, J.; Gentles, R. G.; Patel, R.; MacMillan, E. W.; Reitman, M. S.; Pelham, D.; Cook, A. F. J. Org. Chem. 1996, 61, 5600. (3) Zhang, C.; Mjalli, A. M. M. Tetrahedron Lett. 1996, 37, 5457. (4) Gordon, E M.; Gallop, M. A.; Patel, D. V. Acc. Chem. Res. 1996, 29, 144. (5) Hobbs, DeWitt, S.; Czarnik, A. W. Acc. Chem. Res. 1996, 29, 114. (6) Still, W. C. Acc. Chem. Res. 1996, 29, 155. (7) Schultz, P. G.; Lerner, R. A. Science 1995, 269, 1835. (8) Xiang, X.-D.; Sun, X.; Briceno, G.; Lou, Y.; Wang, K.-A.; Chang, H.; WallaceFreedman, W. G.; Chen, S.-W.; Schultz, P. G. Science 1995, 268, 1738. (9) White, J.; Kauer, J. S.; Dickinson, T. A.; Walt, D. R. Anal. Chem. 1996, 68, 2191. (10) Dickinson, T. A.; White, J.; Kauer, J. S.; Walt, D. R. Nature 1996, 382, 697. (11) Lonergan, M. C.; Severin, E. J.; Doleman, B. J.; Beaber, S. A.; Grubbs, R. H.; Lewis, N. S. Chem. Mater. 1996, 8, 2298. (12) Barnard, S. M.; Walt, D. R. Environ. Sci. Technol. 1991, 25, 1301. S0003-2700(97)00501-5 CCC: $14.00
© 1997 American Chemical Society
Figure 1. Fluorescence image of an optical array, showing eight discrete polymer sensing regions (each 40-50 µm in diameter).
ity,13,14 and sorption.15,16 Regardless of the transduction mechanism of an array sensor, the larger the number of unique sensors in the array, the more information is gathered and thus the better equipped a pattern recognition system is to identify and quantify analytes. Combinatorial polymer synthesis presents an attractive approach for rapidly generating large sets of new polymer matrices with a limited number of starting materials. The present work grew out of an earlier discovery9 that a small amount of one monomer added to a different monomer prior to a photopolymerization reaction produced a sensor exhibiting dramatically different behavior. Our goal was to determine whether a small set of monomers could be copolymerized in different proportions to produce a larger set of unique sensors. Fiber-optic sensors are well-suited for this work for many reasons, including facile polymer incorporation, ease of fabrication and testing, and rapid, information-rich responses. The sensors employed here are constructed by entrapping a solvatochromic fluorescent dye (in this case Nile Red) within a polymer matrix immobilized at the end of the fiber. The sensing mechanism has been described (13) Carey, W. P.; Kowalski, B. R. Anal. Chem. 1986, 58, 3077. (14) Unde, S.; Ganu, J.; Radhakrishnan, S. Adv. Mater. Opt. Electr. 1996, 6, 151. (15) McGill, R. A.; Abraham, M. H.; Grate, J. W. CHEMTECH 1994, (Sept), 27. (16) Grate, J. W.; Patrash, S. J.; Abraham, M. H.; Du, C. M. Anal. Chem. 1996, 68, 913.
Analytical Chemistry, Vol. 69, No. 17, September 1, 1997 3413
Figure 2. Schematic of gradient sensor fabrication apparatus and procedure.
previously.9,10,12 Briefly, the dye shows shifts in emission wavelength depending on the polarity of the local environment. A pulse of vapor administered to the sensor tip alters the polymer microenvironment and gives rise to a complex temporal change in the fluorescence signal of the sensor when monitored at a particular wavelength. The phase, intensity, and shape of these temporal outputs depend directly on the physical and chemical nature of the particular polymer in which the dye is entrapped. In this paper, we present two different approaches to generating sensor diversity through combinatorial methods. First, four discrete combinations of two monomers were used to fabricate four unique sensors, and the responses from each sensor were examined. In the second approach, a polymer stripe was photopolymerized, forming a continuous gradient between two monomers. EXPERIMENTAL SECTION Discrete Sensor Array Fabrication. Two monomers were used: PS802 [(80-85%) dimethyl-(15-20%) (acryloxypropyl)methylsiloxane copolymer] from Gelest and methyl methacrylate (MMA) from Aldrich. Solutions were prepared containing monomer combinations (0.0, 6.7, 33.3, and 50.0% MMA in PS802), dye (Nile Red, 1 mg/mL in chloroform), and initiator (benzoin ethyl ether, BEE, 30 mg/mL). The final solutions were individually polymerized onto the end of a coherent imaging fiber using a photodeposition system, described previously,17 to form several discrete sensing regions across the face of the fiber. The image guides used in this work consist of ∼6000 optical fibers, each 2-3 µm in diameter, packed together in a coherent fashion such that spatial position is maintained from one end of the fiber to the other. Using a series of pinholes, lenses, and objectives, a UV
light beam (330 nm) focused at a distinct location on the proximal fiber face will be transmitted down the length of the fiber and exit the array at precisely the same position. In this way, the polymerization of individual photopolymer solutions can be initiated at designated regions across the distal face of the fiber.18 Here, each of the solutions was polymerized for 5 s, resulting in ∼45-µm-diameter polymer cones. Duplicates of each monomer/ dye mixture were polymerized, yielding a total of eight spatially separated sensing regions (Figure 1). Vapor Testing. The sensor array was tested using an odor delivery/imaging system (previously described9) in the presence of several different vapors. In these experiments, a total of 40 time points were collected over 4 s, with an odor pulse duration of 1 s. The CCD (Princeton Instruments) integration time was set to 25 ms, and pixels were binned 5 × 5 to enhance signal and reduce readout times. The sensor array was illuminated with 535nm light, and emission was monitored at 629 nm using a liquid crystal tunable filter (Cambridge Research Instruments, Cambridge, MA). Gradient Sensor Fabrication. A schematic diagram of the procedure for the gradient sensor fabrication is shown in Figure 2. A photodeposition system was adapted to incorporate a micropositioner (Burleigh Instruments), which was brought into contact with a coarse XY positioner and fiber chuck securing the proximal end of an imaging bundle. UV light (1100 mW) was collected, passed through a pinhole, and focused onto the proximal face of the fiber. The distal end was immersed in 1 mL of a stirred solution containing PS901.5 [(acryloxypropyl)methylsiloxane] (Huls) in chloroform (1:1), with 27 mg/mL BEE (the initiator). Over the course of 22 s, the light beam was scanned across the
(17) Healey, B. G.; Walt, D. R. Anal. Chem. 1995, 67, 4471.
(18) Barnard, S. M.; Walt, D. R. Nature 1991, 353, 338.
3414
Analytical Chemistry, Vol. 69, No. 17, September 1, 1997
Figure 3. Fluorescence temporal response of the eight sensors in the array to saturated vapor pulses of (A) benzene, (B) hexane, (C) 2-propanol, and (D) ethyl acetate. The black bar denotes vapor pulse duration (shown in A). The legend classifies sensors in terms of percent MMA added to the base polymer PS802.
face of the fiber (at 25 µm/s) while 1 mL of PS802 monomer solution (2:3 in chloroform, 30 mg/mL BEE) was simultaneously injected into the vial using a syringe pump. The resulting polymer stripe was therefore a gradient between 100% PS901.5, 0% PS802, and 55.6% PS901.5/44.4% PS802. Both solutions were flushed with nitrogen for 20 min prior to photopolymerization. Using the same procedure, but without the addition of a second monomer to the reaction vessel, a single-component control stripe was polymerized onto the face of the same image guide. The final polymer layers were soaked in a Nile Red (1 mg/mL in toluene) solution for 30 min, rinsed with ethanol, and allowed to dry overnight. A gradient sensor was also prepared using the PS802/MMA system, on a separate imaging fiber, according to the same procedure as outlined above. For this sensor, MMA monomer was slowly added to PS802 (both solutions 1:1 in chloroform and containing 30 mg/mL BEE) during the course of the scanned photopolymerization, yielding a gradient between 100% PS802/ 0% MMA and 50% PS802/50% MMA. The same odor delivery system as that described above was used to test these sensors for their response to various vapors. The imaging software used to acquire and process the fluorescence images was IPLab for PowerPC (Signal Analytics).
Figure 4. Results of cluster analysis performed on the benzene pulse-induced temporal response data for each sensing region in the array: 1 and 2, 0.0%; 3 and 4, 6.7%; 5 and 6, 33.3%; and 7 and 8, 50.0% MMA in PS802.
RESULTS AND DISCUSSION Discrete Sensor Array. Figure 3 depicts the fluorescence output of the array upon exposure to pulses of various saturated vapors. Several observations can be made based on these results. There appear to be four basic temporal curve shapes produced
by the array. Each of the four curve shapes uniquely corresponds to one of the four polymer mixtures used. The responses indicate that each of the polymer combinations experienced a different reaction to the same pulse of vapor and thus has unique sensing Analytical Chemistry, Vol. 69, No. 17, September 1, 1997
3415
Figure 5. (a) Image of two polymer stripes photopolymerized on the distal face of an imaging bundle. The left stripe is a gradient of two siloxanes (PS802 and PS901.5), beginning at the base with PS901.5 alone and ending at a final monomer ratio of 5:4 (PS901.5 to PS802). The right stripe contains only the monomer PS901.5 and was used as a control. In (b-e), temporal data are plotted corresponding to the fluorescence changes observed in 12 selected regions of the polymer stripes (marked by white squares); (b) gradient response to benzene, (c) control response to benzene, (d) gradient response to methanol, and (e) control response to methanol. Pulse duration is indicated by black bars at the bottom of each panel.
properties. The sensor array shows reproducibility; the replicates for each polymer produced curves very similar in shape to one anothersin some cases, nearly identical responses are seen (e.g., 0 and 33.3%). The array’s diversity of response is also interesting, yielding positive, negative, and biphasic fluorescence changes in response to a single vapor application. The individual temporal signatures contain many distinguishing features as well, including different rise times, slopes, recovery rates, and sharp peaks associated with pulse termination. Each of the four sensor groups in the array give different responses to different vapors. This 3416
Analytical Chemistry, Vol. 69, No. 17, September 1, 1997
discrimination capacity is seen both between classes of vapors (such as that shown in Figure 3 between an aromatic and an alcohol) and within a class (e.g., methanol and propanol). Finally, there does not appear to be any obvious progression or relationship between the responses of the four polymer combinations. One might expect different combinations of two monomers to yield sensors with responses that are simply related to the proportion of the two monomers used to construct the sensorsin other words, one type of response from one polymer, another response type from the second polymer, with responses
from combinations of the two following in logical order between these two extremes. In actuality, such a progression is not observed: mixture responses do not appear to be linear combinations of the individual polymer responses. This individuality is most likely due to variation in polymerization kinetics and microstructure formation in the final polymer layer caused by the monomer ratio changes. Subsequent experiments with this array using different vapors, as well as with other arrays prepared in the same way and with other monomer systems, support each of these observations: (1) unique responses from each polymer combination, 2) replicate reproducibility, (3) high diversity of response, (4) multivapor recognition, and (5) absence of trends between polymer combination responses. Cluster analysis was used to compare the responses on a quantitative basis by calculating the dissimilarity between each of the temporal responses generated by the eight sensing regions of the array. “Dissimilarity” is determined by representing each response as a point in multidimensional space and then calculating the Euclidean distance between them. The equation
∑(f
dij ) [
2 1/2
i,t
- fj,t) ]
t)1,T
defines the distance between the temporal responses of sensors i and j. Here, the squared difference between the fluorescence signal from the two sensors at each time point (fi,t and fj,t) is summed over all time points (T ) 40) of the response. An 8 × 8, diagonally symmetrical matrix is thus generated for the overall sensor array. These distances were then used to construct a dendrogram (Figure 4) allowing the isolation of groups, or clusters, of similar responses. The analysis revealed four pairs of distinct responses, corresponding directly to the four polymer combinations used. It is interesting that the cluster analysis does not simply group sensors based on increasing monomer ratio [note that regions 7 and 8 are more closely related to the (1,2)(3,4) group than the (5,6) group], again indicating the nonlinear nature of the relationship between the response and the polymer proportions. Gradient Sensor. In the gradient sensor design, the elements of the sensor array are no longer individual polymer cones, but rather user-defined regions of interest (ROIs) outlining specific areas of the polymer to be monitored (Figure 5a). Following the collection of a sequence of frames, one can use standard drawing tools included in the imaging software to manually select areas of the images from which to measure fluorescence. For the PS901.5/PS802 gradient sensor (Figure 5), 12 ROIs were drawn at various locations along the vertical axes of both the gradient and an adjacent, single-component (PS901.5 monomer only) control stripe. The temporal fluorescence changes for each of the 12 regions in response to vapor pulses (e.g., benzene and methanol) were found to cover a wide range of shapes and intensities for the gradient sensor (Figure 5b and d). The corresponding regions of the control sensor, however, yielded 12 temporal responses of roughly the same shape (Figure 5c and e). For comparison, a matrix of pairwise Euclidean distances (described above) between each sensor response was generated for three separate benzene vapor applications. The distance values in each matrix were then totaled, giving an average summed distance of 31.405 (sm ) 0.9417) for the gradient segments and
Figure 6. Multidimensional scaling plot of the data shown in Figure 4b and c. Filled squares, gradient segment responses; open circles, control segment responses.
12.133 (sm ) 0.6321) for the control. This higher level of diversity among gradient sensing regions was seen throughout the testing process, with other vapors as well as other gradient and control sensors. A standard multidimensional scaling plot (Figure 6) is useful for visualizing the increased dissimilarity between gradient sensor elements and those of the control stripe. Multidimensional scaling produces a two-dimensional representation of the pairwise dissimilarities between the responses of the 12 ROIs for each sensor. Figure 6 shows the gradient sensor points to be much more widely scattered than the control points, indicating a higher level of response diversity. Similar behavior was also seen for the PS802/MMA gradient system (Figure 7). While the control stripe (PS802 only) exhibited temporal curves of very similar shape in response to vapor pulses (in this case, hexane, benzene, and methanol), responses from the corresponding regions of the gradient counterpart were markedly more diverse. Reproducibility. Pulse to pulse responses of a single, discrete combination or a single region in a gradient sensor have been found to be highly reproducible (see related work given in ref 9). A high degree of reproducibility can also be achieved from sensor to sensor when fabrication is performed at the same time with the same solutions (as evidenced, for example, by the nearly identical shapes of replicates in Figure 3). The current setup for gradient polymerization, however, only allows for one gradient to be made at a time, making it difficult to conduct a fair evaluation of gradient to gradient reproducibility. It is likely that these difficulties could easily be overcome with a system built to handle multiple pinholes and fibers. CONCLUSION The ability to produce such variation of response from only two monomers represents a potentially powerful new approach for array sensor preparation. With current CCD and imaging software technology, we can address and measure signals from individual pixels in numbers of up to several thousands; given this fact, the resolution of a gradient approach is most likely to be limited only by the degree to which chemical and/or physical individuality can be produced in microdomains of the polymer gradient. The data presented here support the notion that combinatorial polymer synthesis from a limited set of starting Analytical Chemistry, Vol. 69, No. 17, September 1, 1997
3417
Figure 7. PS802/MMA gradient sensor. Fluorescence images of the gradient (a) and PS802-only control (b) polymer stripes are shown. The nine rectangular boxes represent the polymer regions monitored during the vapor pulse experiments. Responses to hexane, methanol, and benzene vapor pulses (2.6 s in length) are shown respectively in (b-d) for the gradient stripe and in (f-h) for the control stripe.
materials can be used to generate a large set of unique and diverse sensors for analyte detection.
Beth Tabacco and Dr. Fred Milanovich for discussions and helpful suggestions.
ACKNOWLEDGMENT
Received for review May 14, 1997. Accepted June 25, 1997.X
This work was supported by grants from the Office of Naval Research to D.R.W. and J.S.K. and from the National Science Foundation to T. D. (Grant CHE-9256871). We thank Dr. Mary3418
Analytical Chemistry, Vol. 69, No. 17, September 1, 1997
AC970501B X
Abstract published in Advance ACS Abstracts, August 1, 1997.
