Anal. Chem. 2004, 76, 7391-7395
Internal Standard in Surface-Enhanced Raman Spectroscopy A. Lore´n, J. Engelbrektsson, C. Eliasson,§ M. Josefson,† J. Abrahamsson,‡ M. Johansson, and K. Abrahamsson*
Analytical and Marine Chemistry, Department of Chemistry and Bioscience, Chalmers University of Technology, SE-412 96 Go¨teborg, Sweden
A method is presented for the use of SAM layers as internal standards for calibration in surface-enhanced Raman spectroscopy. Three cyano-containing compounds were attached to gold colloids via a metal-sulfur bond and evaluated for spectral stability and normalization capacity. The results show that the analyte, rhodamine 6G, and the internal standard signal enhancement covaried, and it was possible to quantify the analyte with PLS. The fact that the enhancing substrate was chaotic assemblies with large variation in signal enhancement shows the versatility of this method. Surface-enhanced Raman spectroscopy (SERS) has during recent years proved to be a useful assay for a number of biologically relevant molecules.1-4 The enhancement, caused by the interaction of the analyte with a metal surface, makes it possible to reach detection limits in the lower nanomolar range. Solid bulk substrates, such as electrodes, metal films, and surface structures, prepared by lithography provide reproducible sensors for use in solution and gas phase.5 The strong distance dependence of the enhancement effect decreases the applicability of such fixed surfaces, since the object under study must be brought very close to the surface. Colloidal substrates lack the uniformity of those prepared on solid support but can instead be dispersed throughout the object of study, and even inside macrostructures, such as cells and tissue samples. The preparation of colloidal substrates has been studied since ancient times, but size and shape of the colloids are still difficult to control completely. Colloidal substrates can be prepared in many different shapes, where spheres and rod-shaped colloids are the most common, but others, such as cubes, triangles, shells, and composites, can * To whom correspondence should be addressed. E-mail:
[email protected]. † Present address: Pharmaceutical Analytical R&D, AstraZeneca, SE-431 83 Mo ¨lndal, Sweden. ‡ Present address: Department of Paediatrics, Sahlgrenska University Hospital, Go ¨teborg University, SE-416 85 Go ¨teborg, Sweden. § Present address: Department of Pure and Applied Chemistry, University of Strathclyde, 295 Cathedral Street, Glasgow, G4 0PL, Scotland. (1) Vo-Dinh, T.; Houk, K.; Stokes, L. Anal. Chem. 1994, 66, 3379-3383. (2) Riboh, J. C.; Haes, A. J.; McFarland, A. D.; Yonzon, C. R.; Van Duyne, R. P. J. Phys. Chem. B 2003, 107, 1772-1780. (3) McLaughlin, C.; MacMillan, D.; McCardle, C.; Smith, W. E. Anal. Chem. 2002, 74, 3160-3167. (4) Sulk, R.; Chan, C.; Guicheteau, J.; Gomez, C.; Heyns, J. B. B.; Corcoran, R.; Carron, K. J. Raman Spectrosc. 1999, 30, 853-859. (5) Kennedy, B. J.; Milofsky, R.; Carron, K. T. Anal. Chem. 1997, 69, 47084715. 10.1021/ac0491298 CCC: $27.50 Published on Web 11/11/2004
© 2004 American Chemical Society
also be manufactured.6,7 However, there is always a distribution of shapes in each batch, which will influence the surfaceenhancement properties. Other difficulties arise from the interaction between the colloid particles, and between the colloids and the sample matrix, in the form of aggregation and chemisorption of compounds to the surface. The difference in surface enhancement between a chemisorbed and a physisorbed analyte is another source of variation, especially with measurement areas in the nanoscale. In a controlled macroscale environment, the variation in surface enhancement is easily corrected for by addition of an internal standard to the samples; however, this does not solve the problem of variable chemisorbed versus physisorbed signals, which can differ as much as 3 orders of magnitude.8 In less controlled environments, such as biological systems, the actual concentration of the internal standard in each location cannot be guaranteed and another solution must be found. One approach to counter many of these difficulties at once is to coat the metal substrate with the standard to create a self-assembled monolayer (SAM). SAMs form stable layers with a high coverage, and chemisorption of an analyte to the surface, as well as interaction between the substrate and the matrix, is minimized. Properly selected, the SAM will become an internal standard firmly attached to the substrate, while retaining the possibility of nanoscale work, since SAMs readily form on colloidal as well as bulk substrates. The criteria for an optimal internal standard SAM could be summarized as follows: the coating should be homogeneous and stable, both spectrally and chemically. The signal intensities from the internal standard should be comparable to those of the analyte and one or more bands should be in spectrally silent regions. Internal standards with only a few spectral bands are preferable for easier data evaluation. In order for the internal standard to be useful for measurements in biological systems, the coated colloid should be water soluble. Measurements have been performed with SERS on SAMcoated metal surfaces. Alkanethiolate monolayers (C3-C18) have been shown to form homogeneous layers, and they do not present available metal sites for chemisorption of an analyte9,10 Properly (6) Sun, Y. G.; Xia, Y. N. Analyst 2003, 128, 686-691. (7) Tao, A.; Kim, F.; Hess, C.; Goldberger, J.; He, R. R.; Sun, Y. G.; Xia, Y. N.; Yang, P. D. Nano Lett. 2003, 3, 1229-1233. (8) Otto, A., Ed. Light Scattering In Solids IV; Springer-Verlag: New York, 1984. (9) Kennedy, B. J.; Spaeth, S.; Dickey, M.; Carron, K. T. J. Phys. Chem. B 1999, 103, 3640-3646. (10) Olson, L. G.; Lo, Y. S.; Beebe, T. P.; Harris, J. M. Anal. Chem. 2001, 73, 4268-4276.
Analytical Chemistry, Vol. 76, No. 24, December 15, 2004 7391
chosen, they promote adsorption of analytes to the active surface.5 It has also been proposed that they can be used for normalization purposes.11 However, they lack selective bands in spectrally silent regions. Another consideration is that the analyte signal intensities decrease rapidly with increasing chain length as the distance of the analyte from the metal surface increases. In this work, the usefulness of an internal standard SAM substrate with specific bands in an otherwise relatively silent spectral region is shown. Normalization and evaluation is performed with multivariate statistics capable of discrimination even when spectral overlap exists between internal standard and analyte. Several SAMs with different spectral properties are tested with regard to chemical and spectral stability. EXPERIMENTAL SECTION Materials. Standard solutions (0.1-13 µM) of Rhodamine 6G (R6G) were prepared in 0.1 M phosphate buffer (pH 7.2). Salt stability test were performed in 0-100 mM NaCl solutions. The three internal standards used were 4-cyano-N-(2-mercaptoethyl) benzamide (CMEB), 3-mercaptopropionitrile (MPN) (SigmaAldrich), and 4-mercaptobenzonitrile (MBN) (Sigma-Aldrich), and they were dissolved in ethanol (1 mM). The SAM-coated colloids were prepared in two different ways. Procedure 1: Gold colloids (Ted Pella Inc.) (60 nm in diameter) were coated with SAMs by adding 1 mM concentrations of the three internal standard solutions to colloidal solutions.. The coated colloids were dried onto cover glass slides and repeatedly rinsed with ethanol before measurement. Procedure 2: The gold colloids were coated after thorough rinsing with 11.6 M hydrogen peroxide overnight. The solution was subsequently applied to Biacore chips (SIA kit). After the chips were dry, another rinsing cycle was performed including ammonia and hydrogen peroxide. The chips were washed with large amounts of MQ water and ethanol. The chips were then soaked in NaCl solution (1 M), and as the last step, 1 mM MBN solution (in ethanol) was added. After 24 h, the colloids on the chip were coated with SAM layers and ready for use. The second procedure was only used for the final calibration. CMEB was synthesized as follows: Cysteamine (0.9 g, 11.7 mmol) was dissolved in a solution of dichloromethane (DCM) (5 mL)/pyridine (2 mL). Triethylamine (6.5 mL, 46.8 mmol) was added in one portion to the resulting mixture at room temperature. After stirring for 10 min under argon atmosphere, the reaction mixture was cooled to 0 °C. 4-Cyanobenzoyl chloride (2.062 g, 12.5 mmol) dissolved in DCM (5 mL) was added dropwise. The reaction was stirred at room temperature for 18 h under argon atmosphere. Thereafter, the reaction mixture was filtered and concentrated under reduced pressure to give a colorless oil. Purification by flash chromatography on silica gel using a gradient of 5-40% ethyl acetate in petroleum ether (40-60 °C) gave pure CMEB at moderate yield (512 mg, 19,3%), as a colorless oil: H NMR (CDCl3, 400 MHz) δ 7.9 (d, 2 H), 7.5 (d, 2H), 6.6 (b, 1 H). 3.6 (q, 2 H), 2.8 (m, 2 H), 1.4 (t, 1 H); HRMS m/z calculated for C10H10N2OS: 206.0514. Found: 206.0505 All chemicals, if not specifically stated, were purchased from commercial sources. Surface-Enhanced Raman Spectroscopy. SER spectra were acquired by excitation with a 633-nm He/Ne laser via a 60× (NA (11) Deschaines, T. O. C., Keith T Appl. Spectrosc. 1997, 15, 1355-1359.
7392
Analytical Chemistry, Vol. 76, No. 24, December 15, 2004
1.3 water immersion) objective, and the backscatter light was filtered, dispersed, and detected using a Labram INV microspectrometer (Jobin-Yvon) with a cooled NIR enhanced CCD. XY maps were collected over metal clusters with a step size of 1 µm. At least four different clusters, containing 100 spectra each, were mapped for each concentration level and internal standard-coated colloid. For the MBN-coated colloid, a total of 2500 spectra between 500 and 2500 cm-1, with a resolution of 4 cm-1, were collected with an integration time of 1 s and a laser power of 60 µW. For the CMEB- and the MPN-coated colloids, the integration time was 10 s. Multivariate Analysis. Principal component analysis (PCA) and partial least squares (PLS) were performed in Simca 10.0.4 (Umetrics AB). Simulation of the theoretical behavior of a system with analyte and internal standard is shown as reference. A polynomial background subtraction of spectra was performed, and intensity spikes due to cosmic rays were removed prior to data analysis. Polynomial background subtraction was used since, in our experience, it yields better and more consistent results for these types of measurements. No scaling was performed on the data in the PCAs, and the data were centered in the PLS model. The models were diagnosed by R2 and Q2, which represent their ability to describe and predict the variation in the data, respectively.12,13 RESULTS AND DISCUSSION Stability. The chemical stability of the monolayers is an important aspect since a homogeneous layer with complete coverage is necessary to prevent chemisorption of analyte and matrix compounds to the metal surface, which could negate the effect of an internal standard. Three different SAM internal standard layers were tested: MBN, MPN, and CMEB (Figure 1). The internal standards were chosen to contain a spectrally selective band in a silent region, CN at 2300 cm-1. These compounds are sufficiently small to allow large surface enhancement of the analyte and are readily preparable either with thiol functionality or in a form where thiol functionality can easily be added. The internal standard monolayers are firmly attached to the metal surface by their thiol bond and, once attached, yield a signal in the same range as those normally obtained from highly scattering dyes such as R6G. Earlier experiments with strong scatterers attached to the surface yielded signal strengths of such a magnitude that the analyte signal was difficult to detect even when the analyte was R6G, or similar dyes, at high concentrations.14 Spectra of the three internal standards are shown in Figure 1, and MBN and CMEB show sharp spectral bands in contrast to MPN, which exhibits broad unresolved features in the region between 1100 and 1700 cm-1. Evaluation of such broad features presents some difficulties even when multivariate techniques are employed due to the similarity to the SERS background. Spectra obtained from the MPN-coated colloids showed an unpredictable variation, which could be due to incomplete cover(12) Hawkins, D. M.; Basak, S. C.; Mills, D. J. Chem. Inf. Comput. Sci. 2003, 43, 579-586. (13) Wold, S.; Trygg, J.; Berglund, A.; Antti, H. Chemom. Intell. Lab. Syst. 2001, 58, 131-150. (14) Loren, A.; Engelbrektsson, J.; Eliasson, C.; Josefson, M.; Abrahamsson, J.; Abrahamsson, K. Nano Lett. 2004, 4, 309-312.
Figure 1. Surface-enhanced Raman spectra of the internal standards CMEB (a), MPN (b), and MBN (c).
age of the colloids. This was a somewhat surprising result since C3-thiols have previously been shown to produce stable monolayers as well as high and reproducible surface enhancement.9 Experiments with addition of pyridine in order to confirm whether free sites were present for chemisorption did unfortunately give inconclusive results. The other two internal standards did not show any large spectral variations due to illumination or to addition of salt. Calibration. The multivariate analysis of internal standard containing SERS spectra was started with a simulation. Two pure spectra were used, one internal standard and one R6G spectrum. The R6G analyte spectrum was multiplied by the factors 0.1, 0.5, 1.0, 2.5, and 5 in relation to the internal standard. The analyte and the internal standard spectra were added and the total magnitude of the spectral sum was multiplied by a random
number. In total, 100 spectra were created at each concentration level. Further, normally distributed additive and multiplicative noise was introduced for each value in the spectra.. This yielded a PCA score plot as in Figure 2. Each line in the score plot represents one concentration. The spectra with a high SERS amplification will be most distant from the origin irrespective of the actual concentration. It was found that the spread around each line is related to the amount of multiplicative noise introduced. It is, however, not clear why the angular spread increases for higher analyte concentrations. In Figure 2, the additive noise was set to a standard deviation of 400 counts and the multiplicative noise was set to a relative standard deviation of 30%. These levels were exaggerated to get clearly visible effects. The additive noise in the spectra is much lower for the MBN-coated colloids (Figure 4) with a mean standard deviation of 20 counts at the baseline segment at 1750-2200 cm-1. Principal component analysis of spectra collected at four different concentrations of R6G and the internal standards CMEB and MBN yielded the score and loading plots shown in Figure 3. The loading plots show that the first principal component describes the variation in the total signal intensities, i.e., the enhanced signal of both R6G and internal standard. The second principal component describes the relative difference between internal standard and analyte, compared to the first loading. The different concentrations are found as lines in the score plot reflecting the relative proportions of analyte and internal standard, in the same way as in the simulation, but the spread due to local variations in the signal are much larger. The large local signal variations are probably caused by defects in the SAM layers, leaving sites available for adsorption of the analyte. It was speculated that the incomplete coverage could be due to bad preparation of the naked gold colloid. Therefore, the cleaning protocol of the colloids was dramatically changed. To clean the colloidal solution in several steps, the colloids were immobilized on gold-coated chips. Principal component analysis performed on five different concentration levels of analyte on immobilized, MBN-coated colloids (Figure 4) was similar to the simulation. To reduce the problem of local variation, the spectra of each concentration were separated into groups of ∼20 spectra and averages were calculated. The loadings for the PCA model in Figure 4 show separation between the internal standard and the analyte, with little or no interference from the background. In contrast to commonly used calibration data sets, where the intensities are directly related to concentration, we get data sets where the intensities depend to a large extent on the enhancement and to a minor extent on concentration. In our case, it is the proportion between the analyte spectrum and the internal standard spectrum that is used by the PLS calibration. Since there are two variations present, one the total magnitude and the other the proportion between the standard and the analyte, at least two PLS components will be needed to make a calibration model. When this method was applied directly to the measured data, the Q2 level was ∼0.7, which is too low for a quantitative PLS calibration. When ∼20 individually acquired spectra were averaged to yield one entry in the calibration data set, the Q2 value increased to 0.92 for the resulting model. This indicates that it is possible to make a quantitative calibration using the internal standard, but Analytical Chemistry, Vol. 76, No. 24, December 15, 2004
7393
Figure 2. Plot of the PCA components 1 and 2 derived from simulation. Average spectra were not subtracted, and the variable weights were set to 1. Each line represents a concentration level, and the spread around the line is a simulated maximal (5% variation in proportion between analyte and internal standard.
Figure 3. PCA of spectra measured at different concentrations of R6G with CMEB (a, b) and MBN (c, d). The left-hand graphs are the score plots for the two first principal components, and the right-hand graphs show the corresponding loading plots.
the single spectrum spot size in the present measurement is too small to yield a good average that can be directly connected to the Y-reference value. The reason for this is that Y-values for the calibration are only defined in the macroscale of the sample, i.e., 7394
Analytical Chemistry, Vol. 76, No. 24, December 15, 2004
when a sufficiently large area of the sample is included to get a representative average concentration. The final calibration was performed with MBN as internal standard, and the model contained four PLS components. The
Figure 4. PCA of original data used for the PLS model in Figure 5. The left graph is a score plot with classes according to concentration. The right graph shows the corresponding loadings for the first two principal components.
Figure 5. Actual versus predicted plot of R6G concentration in the test set used for validating the PLS model. Root-mean-square error of prediction is 0.5 mM (n ) 35).
first two are the variations described above. The next two components correct for variations in the spectral baseline remaining after the polynomial baseline subtraction. Five chips were used, one for each concentration. Several clusters were mapped from each chip, and the spectra were averaged in groups of 20. The calibration set contained 70 averages from the 5 concentrations, and the test set contained 35 averages. As can be seen in the actual versus predicted plot in Figure 5, despite the large intensity variation between the spectra used in the model, a linear correspondence can be found after four components. The larger variation in the prediction of 2.5 µM may be due to the larger variation in intensities (Figure 4). CONCLUSION We have shown that it is possible to do quantitative measurements using an internal standard SAM on gold colloids. To achieve
the required stability of the SAM layers, proper cleaning of the surfaces is essential. Remaining local variations can be handled by averaging over a larger area in combination with multivariate evaluation. ACKNOWLEDGMENT Queen Silvia Children Hospital Research Foundation and The Swedish Research Council are greatly acknowledged for financial support.
Received for review June 14, 2004. Accepted September 27, 2004. AC0491298
Analytical Chemistry, Vol. 76, No. 24, December 15, 2004
7395
