Single Fiber Identification with Nondestructive Excitation–Emission

Technical Note pubs.acs.org/ac

Single Fiber Identification with Nondestructive Excitation−Emission Spectral Cluster Analysis Krishnaveni Appalaneni,† Emily C. Heider,† Anthony F. T. Moore,† and Andres D. Campiglia*,†,‡ †

Department of Chemistry, University of Central Florida, P.O. Box 25000, Orlando, Florida 32816-2366, United States National Center for Forensic Science, University of Central Florida, 12354 Research Parkway, Suite 225, Orlando, Florida 32826, United States

‡

S Supporting Information *

ABSTRACT: Identification methods for single textile fibers are in demand for forensic applications, and nondestructive methods with minimal pretreatment have the greatest potential for utility. Excitation−emission luminescence data provide a threedimensional matrix for comparison of single-fiber dyes, and these data are enhanced by principal component analysis and comparison of fibers using a statistical figure of merit. No dye extraction methods are required to sample the spectra from a single fiber. This approach has been applied to the analysis of single fibers to compare closely matched dye pairs, acid blue (AB) 25 and 41 and direct blue (DB) 1 and 53. In all cases, the accuracy of fiber identification was high and no false positive identifications were made.

T

are indistinguishable by eye and with similar absorbance spectra (Figure 1) were used for the purpose of identification. Four dyes from two different classes, acid blue (AB) 25 and 41 and direct blue (DB) 1 and 53, were selected to demonstrate the applicability of this technique. These dyes were selected primarily due to the difficulty in distinguishing their spectra from one another (Figure 1) and because they represent two of the eight major classes of dyes (acid dyes, basic dyes, azoic dyes, direct dyes, disperse dyes, sulfur dyes, reactive dyes, and vat dyes).4 Even with higher dimensionality data provided by the excitation−emission matrices (EEMs), comparison of spectra for purposes of fiber identification requires a statistical figure of merit. A variety of approaches to classification of fibers has been published previously. Palmer and co-workers12 developed a system of dividing fibers into classes based on perceived color of the source garment, followed by visual evaluation of absorbance spectra using terms such as “trough,” “peak,” “shoulder,” and “noise.” This approach may be useful for rapid and preliminary comparison but could be enhanced by statistical rigor. More recently, statistical analysis methods have been employed in forensic evaluations with success in a variety of areas. Barret and co-workers investigated the application of hierarchical cluster analysis, principal component analysis (PCA), and discriminant analysis (DA) in classifying dyed hair specimens.13 Yu et al.14 reported the evaluation of Entellan New fiber mount using PCA. Recent innovative application of statistical analysis to forensic samples has been reported by Sikirzhytski et al.15 and Bueno et al.;16 in those

he development of analysis methods for the identification of dyes in textile fibers has become increasingly important for samples of historical,1 archeological,2,3 or forensic4 investigations. For each case, analysis is optimally conducted from a single fiber, either because the sample is precious or a bulk sample is unavailable to investigators. Currently, many methods for the analysis of dyes in fibers have been developed for cases when the material composition of the fiber does not provide exclusive identifying information (e.g., cotton, wool, and other natural fibers are ubiquitous). Highly sensitive and selective methods of single fiber-dye analysis include electrospray ionization mass spectrometry,5,6 surface enhanced Raman spectroscopy,7−9 and capillary electrophoresis.10 In the above cases, single fiber analysis is possible, but pretreatment7 or extraction techniques5,10 are required that can potentially damage the precious fiber sample. Single fiber analysis without pretreatment has been reported by Markstrom and co-workers11 using a liquid crystal tunable filter microspectrophotometer, acquiring visible absorbance spectra with only the fixture of a fiber on a glass slide. Fiber identification using this method may be achieved by comparison of spectra, provided that the two-dimensional absorbance data are sufficiently distinct. As discussed by Huang et al., such distinction may not always be attained using only absorbance spectra.6 Single fiber identification can be substantially enhanced by acquiring data of higher dimensionality than that obtained by absorbance spectra, as well as the inclusion of statistical methods of classification. This paper describes the nondestructive acquisition of 3-dimensional excitation−emission luminescence using a combination of two commercially available instruments: a fluorimeter fiber-optically coupled with an epi-illumination Olympus microscope. Fibers treated with a variety of dyes that © 2014 American Chemical Society

Received: September 4, 2013 Accepted: January 16, 2014 Published: January 16, 2014 6774

dx.doi.org/10.1021/ac500021h | Anal. Chem. 2014, 86, 6774−6780

Analytical Chemistry

Technical Note

Figure 1. Normalized absorbance spectra (A and C) and emission spectra (B, 304 nm excitation; D, 342 nm excitation) of fiber dyes used for identification of single fibers. Bright field images of single fibers with each dye type are inset in each plot.

Acrylic 864, Cotton 400, and Nylon 361 from the same source. For solution measurements, 50% v/v methanol/water dye solutions were prepared using HPLC grade methanol purchased from Fischer Scientific and nanopure water was obtained from a Branstead Nanopure Infinity water purifier. Tweezers, blades, and scissors used for isolating the fibers from fabric were cleaned with methanol and were visually examined under UV-light (254 nm) to check the presence of fluorescence impurities. Absorption Spectroscopy. Absorbance measurements were made with quartz microcuvettes (1 cm path length and 2 mm width) that can hold a maximum of 700 μL. A single beam spectrophotometer (model Cary 50, Varian) equipped with a 75 W xenon pulsed lamp, 20 nm fixed band-pass, and a maximum scan rate of 24 000 nm·min−1 was used to record the absorbance. Fluorescence Microscopy. A commercially available spectrofluorometer (FluoroMax-P from Horiba Jobin Yvon) fiber-optically coupled with an epifluorescence microscope (BX-51 from Olympus) was used to acquire excitation− emission matrices. The spectrofluorometer is equipped with a continuous 100 W pulsed xenon lamp with broadband illumination from 200 to 2000 nm. Excitation and fluorescence spectra were recorded with two spectrometers holding the same reciprocal linear dispersion (4.25 nm·mm−1) and accuracy (±0.5 nm with 0.3 nm resolution). Both diffraction gratings had the same number of grooves per unit length (1200 grooves· mm−1) and were blazed at 330 nm (excitation) and 500 nm (emission). A photomultiplier tube (Hamamatsu, model R928) with spectral response from 185 to 850 nm was used for fluorescence detection operating at room temperature in the photon-counting mode. Commercial software (DataMax) was used to computer-control the instrument.

cases, a variety of multivariate statistical methods was applied to the analysis of Raman data for the forensic identification body fluid traces15 and gunshot residue.16 Although EEMs were previously reported for the analysis of historical textiles,1 the purpose of the work was completely different to the one reported here. In the early work reported by Nakamura et al.,1 dyes of different colors (yellow, red, purple, and blue) were identified by visual comparison of EEMs recorded from textile cloths and natural dye extracts of known chemical composition. EEMs were acquired by placing textile cloths at the distal end of a reflectance fiber-optic probe connected to a commercial spectrofluorimeter. Our work presents a rigorous statistical comparison of visually indistinguishable fibers based on EEMs recorded from single fibers. To the extent of our literature search, this is the first report on the use of principal component cluster analysis for the identification of visually indistinguishable single dyed fibers. Multiple (10) spots along the fiber were sampled to provide a training set for comparison to other fibers, threads, and regions of fabric. Statistical figures of merit for correct identification of fiber dyes are described so that identification of single evidential fibers to other single fibers, threads, or bulk materials may be accomplished with 99% confidence.

■

EXPERIMENTAL SECTION Reagents and Materials. Fabric dyes including AB 25, DB 1, and DB 53 were supplied by Sigma-Aldrich (www.sigmaaldrich.com). AB 41 was acquired from Acros Organics (www. acros.com). The dyes were applied to different types of fabrics. AB 25 and AB 41 were applied to spun nylon 351. DB 1 and DB 53 were used to dye cotton 400. Fabrics were acquired from Testfabrics, Inc. (West Pittston, PA) where the fabrics were dyed. Undyed fibers were obtained from samples of 6775

dx.doi.org/10.1021/ac500021h | Anal. Chem. 2014, 86, 6774−6780

Analytical Chemistry

Technical Note

spectra shown in the Supporting Information. Fluorescence spectra (λexc = 530 nm) from fibers from five different regions of a cloth dyed with DB 1 were measured and, following background subtraction, the average peak intensity of the five fibers on diverse regions of the same cloth was 10 000 ± 2000 au, a 20% variation in intensity. This variation demonstrates the need for normalization in the sample pretreatment and demonstrates the applicability of identification of fibers from different parts of the cloth that can possess highly variable dye concentrations. A data matrix containing the training set spectra was created with the intensity of a given spot on the training fiber contained on a row and each wavelength in a different column. Each spot on the training set was represented as a row in the original data matrix. A square covariance (Z) matrix was calculated:

The microscope and spectrofluorometer are connected via commercially available fiber-optic bundles. The sample compartment of the spectrofluorometer was equipped with a fiber-optic platform (Horiba Jobin-Yvon) that optimizes collection efficiency via two concave mirrors. The microscope is equipped with two 50/50 beam splitters, one for the ultraviolet and the other for the visible spectral region. A 40×Visible (Olympus UPlanSApo 40×) objective lens was used for light collection. A rotating pinhole wheel, with various diameters of 0−1000 μm, is located between the 50/50 beam splitter and the mirror that directs fluorescence emission to the CCD camera (iDS UI-1450SE-C-HQ USB-camera) of the microscope for imaging or the emission fiber bundle of the spectrofluorimeter. An in-house fabricated rectangular sample holder with a hole of approximately 3 mm was made to hold the single fibers on the microscope stage. Collection of Excitation−Emission Matrices. Excitation−emission matrices (EEMs) were collected with excitation wavelength ranging between 350 and 675 nm at 5 nm increments and emission wavelength ranging between 430 and 800 at 1 nm increments using cutoff filters. Two pairs of fabric (8 in. × 10 in.) were chosen for the experiments: nylon dyed with AB 25 and AB 41 and cotton dyed with DB 1 and DB 53. Fibers from each pair are found to be indistinguishable when compared under white light using a 40×-Visible objective (Figure 1). Three fabric elements were sampled with EEM data: single fibers, single threads, and single fibers from different bulk regions of fabric. To acquire EEM data within a fiber, a single fiber pulled from a thread was fixed against the hole of the sample holder using tape so that approximately 3 mm of the length of the fiber can be used for measurements. A quartz slide and coverslip were used with the sample holder. EEMs were collected on five randomly chosen spots on the fiber; then, the sample holder was inverted so the reverse side of the fiber could be sampled, and EEMs were collected from five spots on the other side of the fiber. Similarly, individual threads (composed of 10 fibers) were also sampled by pulling a thread from each fabric. Ten EEMs were collected from each of the 10 fibers isolated from the thread. Finally, individual spectra from different fibers collected from separate bulk regions of fabric were collected to evaluate variation in spectra from different regions. Comparison of Excitation−Emission Matrices. Training set EEM data (10 spectra from different spots on an individual fiber) were collected for each dyed fiber. To identify the excitation wavelength at which the emission spectra showed the maximum deviation, the training spectra were averaged and, for a given dye pair (e.g., AB 25 and AB 41), the averaged training spectra were subtracted and the square of the residuals calculated for each wavelength point in the matrix. The excitation wavelength that generated the greatest difference in the emission spectra was identified from the maxima in the squared residuals plots. Data Analysis. Emission spectra showing maximum differences between dye pairs were identified by scrutinizing the maxima in the residuals plots, and the excitation wavelength corresponding to the highest points in the residuals plots was used for cluster analysis. Data were pretreated by baseline correction and normalization prior to calculation of principal component eigenvectors. This treatment eliminated complexities that arise from fiber identification of fabrics that have nonuniform dye concentrations. This variation in concentrations on different regions of the same cloth is shown for the

Z = DTD

(1)

Eigenvectors (Q0), a set of orthonormal vectors describing the wavelength variation in the D matrix, and the diagonal matrix of eigenvalues (λ0) were calculated by diagonalizing the Z matrix:

Q 0TZQ 0 = λ0

(2)

The majority of vectors in the eigenvector matrix describe noise contained in the data matrix, while only a small number of eigenvectors with the largest eigenvalues contain the majority of the variance in the spectra. Malinowski17 has developed a rigorous test to determine the number of principle components and exclude eigenvectors that describe only noise, using a Fischer’s F-ratio of reduced eigenvalues. To calculate reduced eigenvalues, the eigenvalues are normalized and weighted according to their distribution so that the reduced eigenvalue is proportional to the variance in the data. The equation for the Fratio calculation is: s

F(1, s − n) =

∑ j = n + 1 (r − j + 1)(c − j + 1)

λn

(r − n + 1)(c − n + 1)

∑ j = n + 1 λj 0

s

(s − n)

(3)

This equation calculates the F-ratio for a given eigenvalue to the next smallest eigenvalue to test the null hypothesis. If there are r rows and c columns in the data matrix resulting in s eigenvalues (λ), the nth eigenvalue is being tested for significance. The F-ratio is calculated with s − n degrees of freedom. For the direct blue data set, two eigenvectors were required to describe 99.9% of the variance in the data set, and according to the reduced eigenvalue F-test, the null hypothesis is rejected for two eigenvectors using an F-ratio with 99% confidence. Similarly, acid blue principle components required only two eigenvectors according to the F-test with two eigenvectors accounting for 99.4% of the variance in the training set. Accordingly, the eigenvector matrix was truncated to exclude noise vectors. Analysis of Dye Clusters. Principle component scores were calculated for each spectrum in the training set by multiplying the data matrix by the truncated eigenvector matrix. The scores were mean centered, normalized, and plotted to evaluate the clusters formed by the individual dyes in the training set. The shapes of the clusters were elliptical, so equations describing the shapes of the ellipses were determined. First, the center of mass for each cluster was calculated by averaging the x- and y-coordinates for each point in a training cluster. The angle of the skew of the ellipses was determined by 6776

dx.doi.org/10.1021/ac500021h | Anal. Chem. 2014, 86, 6774−6780

Analytical Chemistry

Technical Note

Figure 2. EEMs for single fibers with dyes AB 25, AB 41, DB 1, and DB 52. Visual characterization of fibers based on EEMs alone would be challenging due to similarity in 3-dimensional matrices.

Figure 3. Plots showing the square of the residuals of excitation−emission matrices for each dye pair. (A) Squared residuals for AB 25 and AB 41. (B) Squared residuals for DB 1 and DB 53.

⎛ cos 2 α ⎛1 sin 2 α ⎞ 1⎞ (x − h)2 + 2cos α sin α⎜ 2 − 2 ⎟ ⎜ 2 + 2 ⎟ ⎝ ⎝ a b ⎠ a b ⎠ ⎛ sin 2 α cos2 α ⎞ 2 (x − h)(y − k) + ⎜ 2 + ⎟(y − k ) = 1 ⎝ a b2 ⎠

fitting the training data to a linear least-squares best fit line, the slope of which is equal to the tangent of the skew angle. The radii of the ellipses were calculated from the training set clusters by calculating the distance of each point from the center of each ellipse along the axis of the rotation angle and combined to give the standard deviation along the major axis of the ellipse. The standard deviation of the training cluster along the axis perpendicular to the rotation angle was also calculated. The standard deviations of the training set points parallel (major axis) and perpendicular (minor axis) to the angle of rotation were used to determine the boundaries of an ellipse, centered at a point (h,k) representing the center of mass of the training cluster and rotated by angle, α. The equation for a rotated ellipse with center (h,k) is given by:

(4)

In this equation, a is the standard deviation and the major axis of the ellipse along angle α, and b is the standard deviation and minor axis perpendicular to the rotation angle. Boundaries of ellipses within three standard deviations of the center of the training cluster are shown in Figure 4. Unambiguous assignment of some fibers, threads, and regions within the cluster within three standard deviations of the center 6777

dx.doi.org/10.1021/ac500021h | Anal. Chem. 2014, 86, 6774−6780

Analytical Chemistry

Technical Note

Figure 4. Principal component scores plots for fibers, threads, and regions of each dye pair (A) AB 25 and AB 41 and (B) DB 1 and DB 53. Ellipses represent a three-standard deviation boundary from the center of the training cluster generated from a single fiber with each dye.

Table 1. Classification of Fibers, Threads, and Regions Dyed with AB 25, AB 41, DB 1, and DB 53 Based on Proximity (within Three Times the Radii of Cluster Ellipse) to the Center of the Training Clusters with 99% Confidence Level accuracy (%) AB 25 AB 41 DB 1 DB 53

false positive

threads

regions

fibers

threads

regions

fibers

threads

regions

100 100 80 90

100 90 100 100

100 100 100 80

0 0 0 0

0 0 0 0

0 0 0 0

0 0 20 10

0 10 0 0

0 0 0 20

scores were calculated from emission spectra at the excitation wavelengths showing the maximum residual sum of squares between emission spectra (see Figure 3). In the case of the AB dye pair, the greatest residual in emission occurred with an excitation wavelength of 560 nm. In the case of DB, two maxima in the residuals plots were observed at 580 and 420 nm excitation. For principal component analysis, the emission at 580 nm was selected because the noise in this region of the spectrum was diminished relative to that at 420 nm. Principal component scores cluster plots for similar dye pairs are shown in Figure 4, in which the radii of the ellipses drawn around the cluster correspond to three times the standard deviation, of each training cluster. In many cases, the line demarcating three standard deviations indicates that the fibers, threads, and regions belong to the cluster with the correct dye identity, and in all cases, fibers, threads, and regions are excluded from classification with the incorrect cluster. In some cases, scores from a given fiber fall outside the three standard deviation boundary, yielding false negative results. Hence, an additional statistical test using F-ratios for the variances of the training clusters was employed. The accuracy and rate of false positive and false negative identification of a given fiber based on the F-test are shown in Table 1, while a complete listing of F-values for all fibers, threads, and regions is provided as Supporting Information. Principal component scores from AB 25 fibers, threads, and regions were tested for correct classification using the F-test (eq 5, above). With 99% confidence, all were classified correctly and none were misidentified as AB 41. Similarly, for the AB 41 validation set containing fibers, threads, and regions, all except a single thread was classified correctly at the 99% confidence level. The thread excluded from the AB 41 cluster was a false negative but was not classified as AB 25 dye. For the case of DB 1, 80% of the fibers were correctly classified, with two false negatives, but their distance from the DB 53 cluster excluded false positive identification. All the

is, in some cases, quite apparent. However, the scatter outside the boundary of three standard deviations requires additional statistical characterization, as described in the following. An F-test was utilized to compare the variance of the training cluster ellipse to the distance of a given questionable point (representing a fiber, thread, or region) from the ellipse. It was important to consider the direction of the point from the center of the cluster, since the distance from the boundary of the ellipse to its center depends upon the direction in which the boundary is measured. An equation for a line formed by the point in question and the center of the ellipse was solved simultaneously with the equation for the ellipse to determine the distance of the ellipse boundary in the direction of the questionable point to the center of the ellipse. The F-ratio for the variance of the point (di2) and the boundary of the cluster (dellipse2) was given by F=

di 2 dellipse 2

false negative

fibers

(5)

To classify a fiber, thread, or region as belonging to a given training set, the F-ratio must be less than the critical F-value at the desired confidence level. With 99% confidence, the critical F-value is 10.56.

■

RESULTS AND DISCUSSION The excitation−emission spectra acquired from fibers provide a 3-dimensional surface to seek spectral features that can be used for identification. Examples of such EEM data from single fibers are shown in Figure 2. However, identification of single fibers by comparison of EEM data requires statistical figures of merit, particularly in cases where dye characteristics, such as spectra, are similar. Determination of principal component scores from training spectra, acquired from a single fiber, facilitated comparison of other single fibers with the same fabric composition and the same or similar dyes. Principal component 6778

dx.doi.org/10.1021/ac500021h | Anal. Chem. 2014, 86, 6774−6780

Analytical Chemistry

Technical Note

Figure 5. (A) Emission spectra (λex = 395 nm) from undyed fibers from Acrylic 864, Cotton 400, and Nylon 361. (B) Principal component scores from the undyed fibers showing resolution of the fibers’ scores into distinct clusters.

variance of the EEM spectra at different regions along the length of the fiber provides a useful training set that can be used as the basis for principal component cluster analysis. The statistical approach to identification was demonstrated using challenging dyes with similarities both in two-dimensional absorbance spectra and in three-dimensional EEM data. By limiting the criteria for identification to a 99% confidence level, no false positive identification events are observed. This may be of particular significance when fibers possess forensic importance and inaccurate identification has legal repercussions.

threads and regions dyed with DB 1 were correctly identified at 99% confidence. In the case of direct blue 53 fibers, threads, and regions, 90% of fibers, 100% of threads, and 80% of regions were correctly identified as dyed with DB 53. No fibers, threads, or regions were incorrectly classified as DB 1. The false exclusion of fibers, threads, or regions from training clusters reflects the limitations of obtaining a training set from only a single fiber, for which subtle differences in spectra might result in exclusion of the sample from correct identification. These false negative results could be diminished if larger training sets were obtained from threads or multiple threads from different regions of a fabric, rather than a single fiber. For a forensic investigation scenario in which this fiber identification scheme could be employed, false positive identification (i.e., concluding that fibers match when they do not) is more destructive than a false negative result. Given the hazards of incorrectly matching fibers, the results presented here indicate that 99% confidence levels could be employed, which only yielded false negatives, and no false positive identification fibers resulted from the analysis. Although the dyes investigated in this study were selected because they represent a difficult identification case, it is worth investigating the potential applicability of cluster analysis to the general case. For this purpose, endogenous fluorescence emission spectra from single fibers of Acrylic 864, Cotton 400, and Nylon 361 fabrics were also acquired. The spectra collected with 395 nm excitation were used to calculate principal component scores. The resulting spectra and cluster plot from the fibers are shown in Figure 5. Despite the inherently weak endogenous fluorescence of the fibers and the similarity in the spectral features, plotting principal component scores on a three-component coordinate axis revealed discrete separation of the fibers with different composition and without any applied dye. This result shows the promise of the method in the more general case, although thorough databases of dyes and fabric types will be necessary before general utility of this method can be attained.

■

ASSOCIATED CONTENT

S Supporting Information *

Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Fax: +1 407 823 2252. Notes

The authors declare no competing financial interest.

■ ■

ACKNOWLEDGMENTS The authors acknowledge the National Institute of Justice for financial support (Grant # 2011-DN-BX-K553). REFERENCES

(1) Nakamura, R.; Tanaka, Y.; Ogata, A.; Naruse, M. Anal. Chem. 2009, 81, 5691−5698. (2) Proefke, M. L.; Rinhart, K. L.; Raheel, M.; Ambrose, S. H.; Wisseman, S. U. Anal. Chem. 1992, 64, 105A−111A. (3) Wouters, J.; Rosariu-Chirinos, N. J. Am. Inst. Conserv. 1992, 31, 237−255. (4) Goodpaster, J. V.; Liszewski, E. A. Anal. Bioanal. Chem. 2009, 394, 2009−2018. (5) Tuinman, A. A.; Lewis, L. A.; Lewis, S. S. A. Anal. Chem. 2003, 75, 2753−2760. (6) Huang, M.; Russo, R.; Fookes, B. G.; Sigman, M. E. J. Forensic Sci. 2005, 50, 1−9. (7) Leona, M.; Stenger, J.; Ferloni, E. J. Raman Spectrosc. 2006, 37, 981−992. (8) Wustholtz, K. L.; Brosseau, C. L.; Casadio, F.; Van Duyne, R. P. Phys. Chem. Chem. Phys. 2009, 11, 7350. (9) Casadio, F.; Leona, M.; Lombardi, J. R.; Van Duyne, R. P. Acc. Chem. Res. 2010, 43, 782−791.

■

CONCLUSIONS The acquisition of excitation−emission matrices for nondestructive analysis of dye in single textile fibers can provide valuable identification information. The coupling of two commercially available instruments enhances the luminescence capabilities above that of the individual instruments, allowing the acquisition of a complete training set for fiber dye identification from an individual fiber. Accounting for the 6779

dx.doi.org/10.1021/ac500021h | Anal. Chem. 2014, 86, 6774−6780

Analytical Chemistry

Technical Note

(10) Xu, X.; Leijenhorst, H.; Van Den Jove, P.; de Koeijer, J.; Logetnberg, H. Sci. Justice 2000, 41, 93−105. (11) Markstrom, L. J.; Mabbott, G. A. Forensic Sci. Int. 2011, 209, 108−112. (12) Palmer, R.; Hutchinson, W.; Fryer, V. Sci. Justice 2009, 49, 12− 18. (13) Barrett, J. A.; Siegel, J. A.; Goodpaster, J. V. J. Forensic Sci. 2011, 56, 95−101. (14) Yu, M. M. L.; Sandercock, P. M. L. J. Forensic Sci. 2012, 57, 70− 74. (15) Sikirzhytski, V.; Sikirzhytskaya, A.; Lednev, I. K. Forensic Sci. Int. 2012, 222, 259−265. (16) Bueno, J.; Sikirzhytski, V.; Lednev, I. K. Anal. Chem. 2012, 84, 4334−4339. (17) Malinowski, E. R. J. Chemom. 1988, 3, 49−60.

6780

dx.doi.org/10.1021/ac500021h | Anal. Chem. 2014, 86, 6774−6780