Chemical Imaging of Pharmaceutical Materials - ACS Publications

Present address: Sean Gilliam, Kaiser Optical Systems Inc., 371 Parkland. Plaza, Ann Arbor, MI 48103. § Saint Louis University. (1) Adar, F. Spectros...
0 downloads 0 Views 438KB Size
Anal. Chem. 2008, 80, 5706–5712

Chemical Imaging of Pharmaceutical Materials: Fabrication of Micropatterned Resolution Targets John F. Kauffman,*,† Sean J. Gilliam,‡,† and R. Scott Martin§ Center for Drug Evaluation and Research, Division of Pharmaceutical Analysis, Food and Drug Administration, 1114 Market Street, St. Louis, Missouri 63101, and Department of Chemistry, Saint Louis University, St. Louis, Missouri 63103 Resolution targets composed of thick poly(ethylene glycol) (PEG) lines on silicon substrates have been fabricated using the method of micromolding in capillaries (MiMIC). Patterns of three parallel lines with equal width and spacing have been prepared, with widths between 5 and 25 µm. Raman chemical images of the PEG-on-silicon devices as well as the metal-on-glass masks used to prepare the devices were measured. The Raman images were used to determine the impulse response of the instrument by comparing the measured images to model functions prepared by convolution of a test impulse function with the object functions of the devices. Impulse widths for PEG-on-silicon targets were approximately two times greater than impulse widths for metal-on-glass targets. The results provide a quantitative measure of the influence of light-matter interactions on the spatial resolution achievable with chemical imaging instruments. This work shows that microfluidic channels can be used to produce robust patterns of PEG on silicon, and these patterns are realistic resolution targets for spectroscopic chemical imaging of pharmaceutical materials. Spectroscopic imaging technology is being increasingly applied to important issues in pharmaceutical development.1–3 Chemically specific particle sizing has allowed researchers to evaluate particle sizes of active pharmaceutical ingredients (APIs) in formulated products such as nasal sprays4 and tablets.5–9 Chemical heterogeneity in process intermediates has also been evaluated with chemically specific imaging technologies.5–9 Process induced polymorphism has been evaluated in finished dosage forms using * To whom correspondence should be addressed. John F. Kauffman, FDA, Division of Pharmaceutical Analysis, 1114 Market St., St. Louis, MO 63101. Phone: 314-539-2168. E-mail: [email protected]. † Food and Drug Administration. ‡ Present address: Sean Gilliam, Kaiser Optical Systems Inc., 371 Parkland Plaza, Ann Arbor, MI 48103. § Saint Louis University. (1) Adar, F. Spectroscopy 2007, 22, 24–28. (2) de Juan, A.; Tauler, R.; Dyson, R.; Marcolli, C.; Rault, M.; Maeder, M. TrAC, Trends Anal. Chem. 2004, 23, 70–79. (3) Rios, M. Pharm. Technol. 2008, 32. (4) Doub, W. H.; Adams, W. P.; Spencer, J. A.; Buhse, L. F.; Nelson, M. P.; Treado, P. J. Pharm. Res. 2007, 24, 934–945. (5) Gendrin, C.; Roggo, Y.; Collet, C. Talanta 2007, 73, 733–741. (6) McGeorge, G. Am. Pharm. Rev. 2003, 6, 94–99. (7) Sasic, S. Pharm. Res. 2007, 24, 58–65. (8) Sasic, S. Appl. Spectrosc. 2007, 61, 239–250. (9) Shah, R. B.; Tawakkul, M. A.; Khan, M. A. J. Pharm. Sci. 2007, 96, 1356– 1365.

5706

Analytical Chemistry, Vol. 80, No. 15, August 1, 2008

Raman chemical imaging (RCI).5–9 Near infrared chemical imaging (NCI) and mid infrared chemical imaging have been applied to forensic analysis of pharmaceutical products.10–12 Terahertz imaging is capable of mapping coating thickness of an entire tablet,13,14 and recent work has shown that chemical imaging can also be performed using terahertz radiation.15,16 When quantitative characterization of material heterogeneity is required, the spatial resolution of chemical imaging instruments must be determined accurately, which is typically performed with metal-on-glass bar targets such as the USAF 1951 target. However, pharmaceutical products are often composed of particulate materials that are nonopaque to chemical imaging excitation sources, and measurable analytical signals can be generated deep within the samples under study. Furthermore both excitation and signal radiation can be scattered beneath the surface.17 As a result, the spatial information content of the signal is often distorted. This phenomenon, known as photon diffusion, photon migration, subsurface scattering, or secondary scattering, can influence the practical resolution achievable with chemical imaging instrumentation.18–20 (See Supporting Information Figure S1 for a schematic diagram and Figure S2 for an example of this phenomenon.) Thus, the practical resolution for pharmaceutical samples is often lower than the optical resolution of the instrument as determined with a metalon-glass target. The significance of this problem is difficult to gauge at present, in part because of a lack of suitable resolution targets that mimic the expected behavior of pharmaceutical materials. (10) Dubois, J.; Wolff, J.-C.; Warrack, J. K.; Schoppelrei, J.; Lewis, E. N. Spectroscopy 2007, 22, 36–41. (11) Hamilton, S. J.; Lowell, A. E.; Lodder, R. A. J. Biomed. Opt. 2002, 7, 561– 570. (12) Roggo, Y.; Edmond, A.; Chalus, P.; Ulmschneider, M. Anal. Chim. Acta 2005, 535, 79–87. (13) Spencer, J. A.; Gao, Z.; Moore, T.; Buhse, L. F.; Taday, P. F.; Newnham, D. A.; Shen, Y.; Portieri, A.; Hussain, A. S. J. Pharm. Sci. 2008, 97, 1543– 1550. (14) Fitzgerald, A. J.; Cole, F. E.; Taday, P. F. J. Pharm. Sci. 2005, 94, 177– 183. (15) Cogdill, R. P.; Short, S. M.; Forcht, R.; Shi, Z.; Shen, Y.; Taday, P. F.; Anderson, C. A.; Drennen, J. K., III J. Pharm. Innov. 2006, 1, 63–75. (16) Wu, H.; Heilweil, E. J.; Hussain, A. S.; Khan, M. A. J. Pharm. Sci. 2008, 97, 970–984. (17) Matousek, P.; Clark, I. P.; Draper, E. R. C.; Morris, M. D.; Goodship, A. E.; Everall, N.; Towrie, M.; Finney, W. F.; Parker, A. W. Appl. Spectrosc. 2005, 59, 393–400. (18) Kop, R. H. J.; Vries, P. d.; Sprik, R.; Lagendijk, A. Phys. Rev. Lett. 1997, 79, 4369–4372. (19) Wiersma, D. S.; Muzzi, A.; Colocci, M.; Righini, R. Phys. Rev. E 2000, 62, 6681–6687. (20) Durduran, T.; Yodh, A. G. J. Opt. Soc. Am. A 1997, 14, 3358–3365. 10.1021/ac800864x CCC: $40.75  2008 American Chemical Society Published on Web 06/25/2008

Metal-on-glass resolution targets can achieve a theoretical modulation depth of 100% in transmission21 and are adequate to characterize the influence of the imaging instrument on resolution. Contrast in reflectance measurements of metal-on-glass targets with NIR imaging systems depends on the differential reflectance between the metal and glass surfaces, which can also result in high modulation depths. Measurements with Raman imaging systems often use the weak glass fluorescence as the signal, and the metal pattern modulates the signal by covering the glass. In this case the metal bars provide slit apertures that limit the F-number in one dimension. When metal-on-silicon resolution targets are used for RCI, the silicon phonon band at ∼520 cm-1 is used as the analytical signal, and contrast is achieved by covering the substrate with the metal pattern. In each of these cases, signal arises primarily from the surface of the target, and diffuse scattering can be minimized with good laboratory practices. Under these circumstances, univariate measurements are adequate to measure contrast. However, contrast measurements based on metal-on-glass or metal-on-silicon targets cannot characterize the loss of resolution due to light-matter interactions when thick, nonopaque, light-scattering pharmaceutical materials are examined. Furthermore, because contrast in pharmaceutical samples depends on the ability to distinguish the spectra of two distinct materials, multivariate methods are often required to analyze RCI and NCI images.22–24 The U.S. Food and Drug Administration has identified the development of standards for spectroscopic instruments as an important element on the critical path from drug discovery to market.25,26 The purpose of this paper is to describe the fabrication of realistic resolution targets for spectroscopic chemical imaging of pharmaceutical materials. The soft lithographic method known as MiMIC (micromolding in capillaries) has been used to create thick chemical patterns of a molecular material on organic and inorganic substrates.27 The molecular materials have been chosen to mimic the behavior of materials found in pharmaceutical products, and the substrates have been chosen to provide varying degrees of spectroscopic contrast. In this paper we focus our attention on Raman chemical images of poly(ethylene glycol) (PEG) on silicon devices. Methods of data analysis have been developed with the aim of providing objective, quantitative information concerning the spatial resolution of spectroscopic chemical imaging instrumentation. The experiments described herein have been designed to determine the influence of material properties and data processing procedures on measures of resolution for real-world samples. Measures of Resolution and Material Considerations It is possible to compute the impulse response (also known as the (21) Boreman, G. D. Modulation Transfer Function in Optical and Electro-Optical Systems; SPIE Press: Bellingham, WA, 2001. (22) Sasic, S.; Clark, D. A.; Mitchell, J. C.; Snowden, M. J. Analyst 2004, 129, 1001–1007. (23) Zhang, L.; Henson, M. J. Appl. Spectrosc. 2007, 61, 1015–1020. (24) Zhang, L.; Henson, M. J.; Sekulic, S. S. Anal. Chim. Acta 2005, 545, 262– 278. (25) U.S. Food and Drug Administration, Innovation or Stagnation: Challenge and Opportunity on the Critical Path to New Medical Products, 2004, http:// www.fda.gov/oc/initiatives/criticalpath/whitepaper.html. (26) U.S. Food and Drug Administration, Innovation or Stagnation: Critical Path Opportunities Report and List, 2006, http://www.fda.gov/oc/initiatives/ criticalpath/reports/opp_report.pdf. (27) Kim, E.; Xia, Y.; Whitesides, G. M. J. Am. Chem. Soc. 1996, 118, 5722– 5731.

point spread function) of an imaging instrument using a resolution target if both the object function, f(x,y), and the image function, g(x,y), of the target are known.21 The object function represents the idealized signal intensity distribution of the object being imaged, without noise or diffraction effects.21 For example, a metalon-glass bar target imaged in transmission would be expected to have a normalized intensity of 1 for pixels associated with the glass image and a normalized intensity of 0 for pixels associated with the metal image. The object function can be estimated if the true shape of the object is known. The image function is the measured signal intensity distribution of the object when it is imaged by the instrument under test. The relationship between the object and image functions is g(x, y) ) h(x, y) * f(x, y)

(1)

where h(x,y) is the impulse response of the imaging system and * denotes the convolution operation. The impulse response of an imaging instrument can be estimated in the spatial position domain by fitting the measured image function to a convolution of a test impulse response with a known object function. (Alternatively, the instrument’s modulation transfer function can be determined in the spatial frequency domain. This topic will be discussed in a separate publication.) The object function of a bar target is spatially invariant along the dimension parallel to the bars and has a square wave shape along the dimension perpendicular to the bars. Thus onedimensional object and image functions are generally used to analyze bar targets, and the procedure must be performed along orthogonal axes of the imaging system. For the purpose of this paper, we define the one-dimensional object function, f(x), as the function that describes the true spatial variation of the target along the axis perpendicular to the bars and the one-dimensional image function, g(x), as the measured signal variation along the axis perpendicular to the bars. Equation 1 still applies in one dimension. The considerations described above suggest that novel devices are required to explore the role of light-matter interactions on the resolution of chemical imaging systems. We have chosen to use microfluidic channels to pattern thick parallel lines of PEG on spectroscopically distinct substrates to form bar targets suitable for determining the impulse response of a spectroscopic chemical imaging system. PEG was chosen because it is organic, transparent to the excitation source, and scatters light and therefore mimics the behavior of typical pharmaceutical materials. The devices were used to compare the resolution of a Raman chemical imaging system measured with metal-on-glass versus PEG-onsilicon targets. The use of microfluidic channels to pattern these devices is important, because the reproducibility of fabrication processes ensures that these types of standards can be made reproducibly in different laboratories. EXPERIMENTAL SECTION The bar targets were fabricated using the MiMIC process,27,28 as described below. Figure 1a is an image of the metal-on-glass mask used to prepare the devices examined in this study. The pattern consists of two square reservoirs (2 mm × 2 mm) (28) Duffy, D. C.; McDonald, J. C.; Schueller, O. J.; Whitesides, G. M. Anal. Chem. 1998, 70, 4974–4984.

Analytical Chemistry, Vol. 80, No. 15, August 1, 2008

5707

Figure 2. Cross section of the PDMS stamp and a PEG-on-glass replica of the device examined in this study. (a) PDMS stamp cross section cut perpendicular to the 15 µm line set. (b) Raised PEG lines, 15 µm wide at the base. Cross-sectional images such as this one were used to measure object functions of the PEG-on-silicon devices.

Figure 1. Brightfield reflectance images of patterns for Raman imaging resolution targets. (a) Metal-on-glass mask showing one reservoir and the lines connecting the reservoirs. (b) Magnified image of the mask lines. (c) Images of a PEG-on-silicon device showing five triplet line sets. The center-to-center distance between line sets is 180 µm.

connected by a series of lines of varying width. Figure 1b is a magnified image of the mask’s line pattern comprised of five sets of triplet lines, each composed of three parallel lines of equal width and equal spacing. A PEG-on-silicon device fabricated as described below is shown in Figure 1c. The widths of the lines range from 5 to 25 µm in 5 µm steps, and the inner distance between reservoirs is 0.5 cm. Hereafter, the term “device” will be used to denote a PEG-on-silicon structure consisting of the five line sets shown in Figure 1c, whereas the term “line set” will be used to denote a triplet of lines with a specific width. The spacing between the centers of the line sets has been fixed at 180 µm to simplify automated measurement of all line sets of the device. 5708

Analytical Chemistry, Vol. 80, No. 15, August 1, 2008

The metal-on-glass mask (a chrome mask prepared to our specifications by Advance Reproductions Corp.) was used to prepare a 10 µm thick (nominal) photoresist-on-silicon embossing master using standard photolithographic techniques.29 Fabrication details are provided in the Supporting Information. Briefly, the embossing master was used to fabricate a poly(dimethylsiloxane) (PDMS) replica of the mask with 10 µm deep channels. The PDMS stamp was sealed to a clean device substrate for final device fabrication. PEG powder was delivered to one PDMS reservoir, melted at 80 °C, and drawn into the channels by vacuum assisted capillary action. After cooling, the PDMS stamp was carefully removed to reveal ∼10 µm thick lines of solid PEG on a silicon substrate. (See Figure 1c.) The PDMS stamp was subsequently cleaned with compressed air, and an identical device was prepared on a glass microscope coverslip. After device fabrication, the coverslip was cleaved to reveal a cross section of the PEG lines. The cross section was imaged with a VHX-500 digital microscope (Keyence Corporation, Woodcliff Lake, NJ) in brightfield reflectance mode at 1000× in order to measure the object function of each line set at a known position along the length of the PEG lines. Figure 2 shows cross section images of the 15 µm line set from a PEG-on-glass device and the PDMS stamp used to prepare the device. Raman chemical images of the line sets were measured (29) Li, M. W.; Huynh, B. H.; Hulvey, M. K.; Lunte, S. M.; Martin, R. S. Anal. Chem. 2006, 78, 1042–1051. (30) Meyer, C. D. Matrix Analysis and Applied Linear Algebra; Society for Industrial and Applied Mathematics: Philadelphia, PA, 2000.

Figure 3. Raman images of the 15 µm line set measured with a 10× objective. (a) Raman image of the PEG-on-silicon device. (b) Raman image of the metal-on-glass mask.

at the position along the length device lines that matched the location of the object function measurement. Raman wide-field images were measured with a FALCON II molecular chemical imaging system (ChemImage Corporation, Pittsburgh, PA). The sample was illuminated with ∼280 mW of 532 nm radiation through the microscope optics. The Raman scattered light was collected and directed through a wide-area liquid crystal tunable filter (LCTF) wavelength selector (9 cm-1 bandwidth) before being imaged onto a charge-coupled device (CCD) camera. Each frame of the hyperspectral image was a full field image of the sample at a specified wavelength. The Raman hyperspectral images analyzed in this work covered the 490-540 cm-1 and 2650-2995 cm-1 ranges with a 5 cm-1 step size, the former capturing the silicon phonon band at ∼520 cm-1 and the latter capturing the PEG band at 2885 cm-1. Signal at each wavelength frame was integrated for 2 min, and each image consisted of 279 × 279 pixels for a total of 77 841 spectra per image. ChemXpert software (ChemImage Corp., v2.1.3) was used to control the instrument and analyze the hyperspectral images. Raman wide-field images of the metal-on-glass mask were generated from the weak fluorescence signal of the glass substrate. A single wavelength frame of each line set measured at 4450 cm-1 was used for data analysis since this wavelength provided the highest signal-to-noise ratio, but other wavelengths gave similar results. Signal heterogeneity due to excitation source heterogeneity was eliminated by dividing each image by a single frame image of bare glass, a process known as flat-fielding. Hereafter, the term “mask image” will be use to denote the images of the metal-on-glass mask. All line set images were measured using 5×, 10×, and 20× objectives, with pixel resolutions of 1.18, 0.55, and 0.27 µm per pixel, respectively. The complete set of lines (5-25 µm widths) was imaged with each objective. Five fields of view were used to measure images of all line sets for the 5× and 10× objectives. The 20× objective required seven fields of view to measure all of the line sets, with the 20 and 25 µm line sets each requiring two fields of view to capture the complete image of the lines. Two pure

component reference images (PEG, Si) were also acquired to support multivariate data processing of the images. Each hyperspectral image was filtered using a 3 × 3 pixel standard deviation filter (cosmic filtering) in order to eliminate dead pixel artifacts and artifacts resulting from interaction of highenergy particles with the CCD camera. Flat fielding of the hyperspectral images was accomplished by standard normal variate scaling followed by vector normalization of the spectrum at each pixel. The same processing steps were applied to the reference images used to construct pure component spectra. Sasic et al.22 have shown that multivariate image processing reduces noise relative to single-wavelength univariate images, and our observations are consistent with their conclusions. (See Supporting Information Figures S3 and S4 for a comparison of univariate and multivariate images and image functions.) The hyperspectral images were thus converted to chemical images using spectral mixture resolution (SMR) as implemented in the ChemXpert software. SMR is a multidimensional linear regression technique30 that assumes that the measured spectrum associated with each pixel is a linear combination of the two known pure component spectra. Chemical images were constructed from the coefficients of the linear combination of pure component spectra, computed pixel-by-pixel. (Mathematical details of the procedure are given in the Supporting Information.) The grid search curve fitting algorithm described below was implemented in Microsoft Excel using Visual Basic for Applications. RESULTS AND ANALYSIS Figure 3a displays a Raman image of the 15 µm line set of the PEG-on-silicon device, and Figure 3b displays a Raman image of the 15 µm line set of the metal-on-glass mask. Each image was collected with a 10× objective. The bright regions of the mask image appear to have more uniform intensity in cross section, whereas the device line images are more intense at the center and more diffuse at the edges. Similar images were collected for each line set with each objective, and all images were analyzed as described below. Analytical Chemistry, Vol. 80, No. 15, August 1, 2008

5709

Image functions for the device were determined from the PEG chemical images, whereas image functions for the mask were determined from the flat-fielded single frame mask images. In each case, the image function was computed with an intensity profile function created for this purpose by ChemImage Corp. (See Supporting Information for a description of this software function.) The object functions of the mask line sets were generated from their brightfield reflectance images by preparing square waves from the measured line widths and spacings. The device object functions were determined by measuring the x (position) and y (height) values for each vertex in the cross-sectional image of the line set and performing a linear interpolation of the y-axis values between the vertices. (See Figure 2.) The channels in the PDMS stamp have a trapezoidal profile, and the PEG structures are also trapezoidal in shape. The maximum heights of the 10-25 µm wide lines were identical, and therefore these were normalized to unit height. The 5 µm line heights were about half the value of the heights of the other lines. The root cause of this observation is still under investigation, but it is conceivable that surface tension effects prevent conformal filling of the 5 µm channels under the conditions of this study. In this work we determined the impulse response of the instrument by comparing a model function (the convolution of the object function and an impulse response function) with the image function as suggested by eq 1. The main advantage of the convolution method is that it allows direct comparison of the measured image function with a model and therefore offers a rigorous test of the model. Its chief disadvantage is that the impulse response must be modeled with a user specified analytical function, and this function is seldom known with certainty. We have attempted to overcome this uncertainty by evaluating Gaussian, Lorentzian, and Pearson impulse functions and selecting the function that minimizes the root mean square (rms) deviation between the image function and the model function. The Pearson function interpolates between Lorentzian and Gaussian lineshapes via a parameter (the “M-value”) that varies from M ) 1 (Lorentzian) to M ) ∞ (Gaussian). When M ) 2, the function is similar to a 50:50 mixture of Lorentzian and Gaussian functions, and when M ) 3, the function is approximately 75% Gaussian. (See Supporting Information for a description of the Pearson line shape function.) The object and image functions were first aligned with one another by shifting the object function to maximize the cross correlation between them. A model function was constructed by convolution of the object function with an impulse response function, and the model function was then compared with the image function. Ten different impulse function lineshapes were examined (Gaussian, Lorentzian, and eight Pearson functions with M ) 1.25, 1.5, 1.75, 2, 2.5, 3, 4, and 5). For each line shape, a grid search of the best fit width was performed by (1) constructing the impulse function with a specified width, (2) convoluting the impulse function with the object function of the line set under study to create a model function, (3) scaling the model function to minimize the rms deviation between the model and image functions, (4) logging the rms deviation for the specified function and width, and (5) repeating steps 1-4 for a range of line widths between 1 and 40 µm. The best fit width was thus determined for 5710

Analytical Chemistry, Vol. 80, No. 15, August 1, 2008

Figure 4. (a) Object function (dashed), image function (squares), and model function (solid) for the 15 µm line set measured with a 10× objective. The best fit impulse function is a Pearson line shape (M ) 1.75) with a 9.5 µm full width at half-maximum (fwhm). The object function intensity has been reduced for clarity. (b) rms deviation plot for the 15 µm line set measured with the 10× objective.

each impulse function line shape. Figure 4a displays the object, image, and best fit model functions for the 15 µm line set measured with a 10× objective. Figure 4b displays a typical rms deviation plot for the grid search of a single line set measured with a single objective using a single impulse response function. The shape of the rms deviation plot demonstrates that the procedure has adequate spatial resolution to identify the best fit impulse width. Best fit impulse widths were determined from the Raman images of each line set measured with each objective for all 10 impulse lineshapes. Similar procedures were applied to both mask and device, for a total of 300 impulse width determinations. For each objective, the best fit impulse function was identified by selecting the model function that minimized the combined rms deviation of all five line sets. The best fit impulse function widths (FWHMs) determined by this procedure for the mask and the PEG-on-silicon device are displayed in Table 1 as a function of line set width. (See Supporting Information Figure S5 for a graphical display of this data.) The best fit impulse widths are inversely proportional to the objective magnification, as expected. Within each objective group, the best fit impulse widths for the mask are nearly independent of line set line width, indicating that impulse widths for metal-on-glass targets can be determined with

Table 1. Best Fit Impulse Function Parameters for Raman Images of the Metal-on-Glass Mask and PEG-on-Silicon Devicea metal-on-glass mask impulse widths (µm)

PEG-on-silicon device impulse widths (µm)

line set width (µm)

5× Pearson (M ) 1.25)

10× Pearson (M ) 1.25)

20× Pearson (M ) 1.75)

5× Pearson (M ) 3)

10× Pearson (M ) 1.75)

20× Pearson (M ) 1)

5 10 15 20 25 avg std deviation RSD

9.0 8.0 8.0 8.0 8.0 8.2 0.4 5%

5.5 5.5 4.5 4.5 4.5 4.9 0.5 10%

2.5 2.0 2.0 2.0 2.0 2.1 0.2 10%

23.0 16.0 17.0 18.0 19.0 19 3 16%

7.0 7.5 9.5 9.5 7.0 8 1 13%

2.5 4.5 5.0 5.5 3.5 4 1 25%

a The best fit impulse width (FWHM) is tabulated for each line set measured with each objective (5×, 10×, and 20×) determined with the impulse function (Pearson) that gave the best fit for the objective. The Pearson M ) 1 function is identical to the Lorentzian lineshape.

high precision using the methods described herein from devices of varying spatial frequency. The averages and standard deviations of the impulse FWHMs are displayed in Table 1 for each objective. The standard deviations of the mask impulse widths are less than the wavelength of the excitation, whereas the average best fit impulse widths are more than 5 times greater than the pixel sampling resolution of the instrument. The best fit impulse widths for the PEG-on-silicon device are also inversely proportional to objective magnification, but are approximately a factor of 2 greater than those observed for the metal-on-glass targets. For each objective, the increased impulse width of the PEG-on-silicon device reflects the contribution of light-matter interactions to the instrument’s impulse response when thick, nonopaque samples are measured. Within each objective group, the line width dependent variation of the best fit impulse width of the device is also greater than the mask variation. The impulse width for the 5 µm line set measured with the 5× objective appears to be an outlier, which may result from the fact that the 5 µm lines are not resolved at this magnification. (See Figure S6 in the Supporting Information for images of the 5 µm line set measured with 5×, 10×, and 20× objectives.) If the impulse width value for the 5 µm line set measured with the 5× objective is eliminated from Table 1, then the standard deviation of the device impulse width is 1 µm for all objectives. Figure 5 displays the best fit impulse functions for the mask (Figure 5a) and device (Figure 5b) measured with the 5×, 10×, and 20× objectives. The impulse function widths diminish as the objective magnification increases in both cases, but the device impulse widths are significantly wider. The difference between the impulse function width of the device and the mask reflects the reduction in resolution that occurs when the sample is thick and nonopaque to the excitation source. DISCUSSION During the course of this investigation, we prepared several devices and analyzed them according to the procedures outlined above. We measured image functions from a variety of regions of interest (ROIs) across each field-of-view and found no systematic variation in the resulting analysis with ROI. We also found that the impulse widths determined from several replicate devices were within experimental error of one another. We compared the results of analysis using object functions that were measured from two separate PEG-on-glass replicates, and these also gave similar

Figure 5. Best fit impulse functions for the 5× (dashed), 10× (dotted), and 20× (solid) objectives. (a) Mask results. (b) Device results. See Table 1 for impulse function parameters.

results that were in agreement within the uncertainties of the measurement. These studies indicate that PEG-on-silicon devices prepared using the MiMIC process offer robust, reproducible targets for chemical imaging systems. Though the work presented here is focused on PEG-on-silicon devices, we have also prepared PEG-on-glass and PEG-on-PDMS devices. Because the MiMIC stamps can be conformally and reversibly sealed to these substrates, the procedures described in this paper will allow direct comparison of the influence of background signal on achievable resolution for similar devices prepared on different substrates. Furthermore, multiple devices can be generated from a single PDMS stamp, which reduces the cost of preparing these resolution targets. Analytical Chemistry, Vol. 80, No. 15, August 1, 2008

5711

The observed best fit impulse functional form differs between objectives and between mask and device. Several factors influence the functional form of the best fit impulse function. For example, when evaluating the instrument response of a full-field chemical imaging instrument, measurement of the image function is performed with respect to the imaging detector coordinates. If the object under test is not perfectly aligned with the detector pixel axes, then the signal intensity can be artificially spread from the center of the image function peaks toward the outside of the peaks. This problem can be solved either by digitally rotating the image prior to measuring the image function or by selecting a ROI that is narrow in the dimension parallel to the lines when computing the image function. Our studies indicate that the image rotation may have a slight influence on the Pearson M-value of the best fit impulse function but does not affect the width of the best fit impulse function. The methods outlined in this paper reflect our current view of best practices for extracting impulse functions from images of resolution targets. These methods have been applied to metalon-glass targets in a manner consistent with common use, as well as PEG-on-silicon targets that provide a more realistic assessment of the resolution that is achievable when imaging pharmaceutical materials. The impulse widths have been shown to be relatively insensitive to the functional form of the impulse function used to generate the model functions. The impulse widths have also been shown to be insensitive to the width of the lines in the test object, as long as the lines can be partially resolved by the instrument. Thus impulse widths measured by these procedures provide robust estimates of achievable resolution with spectroscopic chemical imaging systems. The observations reported here suggest that follow-on studies can be performed with similar devices to examine the influence

5712

Analytical Chemistry, Vol. 80, No. 15, August 1, 2008

of sample topography on resolution. For example, the thickness of the PEG lines and the substrate on which the PEG lines are deposited are expected to influence the resolution. In addition, targets fabricated with the PEG lines embedded in a nonopaque, light scattering material could be used to more fully explore the influence of secondary scattering on resolution. Resolution targets overcoated with thin polymeric layers containing particulate material could be used to explore the ability to image buried structures, as well as the influence of subsurface structures on images thought to arise from the sample surface. Finally, the resolution targets described in this paper can be used to develop quantitative metrics to compare full field imaging instruments with point mapping and confocal instruments. ACKNOWLEDGMENT This research was supported by the FDA Center for Drug Evaluation and Research’s Research for Scientific Review program. The authors wish to thank the scientists at ChemImage Corp. for assistance with many aspects of this project, including the development of the image function measurement algorithm. S.J.G. thanks the Oak Ridge Institute for Scientific Education for postdoctoral research support. SUPPORTING INFORMATION AVAILABLE Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org.

Received for review April 28, 2008. Accepted May 26, 2008. AC800864X