Forensic Discrimination of Copper Wire Using Trace Element

Jul 9, 2014 - A solution-based inductively coupled plasma mass spectrometry method for measuring trace element concentrations in high-purity copper wa...
2 downloads 11 Views 4MB Size
Download PDF
Article pubs.acs.org/ac

Forensic Discrimination of Copper Wire Using Trace Element Concentrations Joshua R. Dettman,† Alyssa A. Cassabaum,† Christopher P. Saunders,†,‡ Deanna L. Snyder,† and JoAnn Buscaglia*,§ †

Oak Ridge Institute for Science and Education, Federal Bureau of Investigation, Laboratory Division, 2501 Investigation Parkway, Quantico, Virginia 22135, United States ‡ Department of Mathematics and Statistics, South Dakota State University, Harding Hall 213, Box 2220, Brookings, South Dakota 57007, United States § Counterterrorism and Forensic Science Research Unit, Federal Bureau of Investigation, Laboratory Division, 2501 Investigation Parkway, Quantico, Virginia 22135, United States S Supporting Information *

ABSTRACT: Copper may be recovered as evidence in highprofile cases such as thefts and improvised explosive device incidents; comparison of copper samples from the crime scene and those associated with the subject of an investigation can provide probative associative evidence and investigative support. A solution-based inductively coupled plasma mass spectrometry method for measuring trace element concentrations in high-purity copper was developed using standard reference materials. The method was evaluated for its ability to use trace element profiles to statistically discriminate between copper samples considering the precision of the measurement and manufacturing processes. The discriminating power was estimated by comparing samples chosen on the basis of the copper refining and production process to represent the within-source (samples expected to be similar) and between-source (samples expected to be different) variability using multivariate parametric- and empirical-based data simulation models with bootstrap resampling. If the false exclusion rate is set to 5%, >90% of the copper samples can be correctly determined to originate from different sources using a parametric-based model and >87% with an empirical-based approach. These results demonstrate the potential utility of the developed method for the comparison of copper samples encountered as forensic evidence.

T

such as a power source, timer, or detonator. Linking the copper wire used in an IED to a roll of wire associated with a suspect or to wires used in unsolved crimes could provide probative circumstantial evidence for associating a suspect to the IED or generate investigative leads. Trace element profiles hold promise for providing a link or discriminating between copper samples associated with a suspect and a crime scene to supplement visual comparison, fingerprints, DNA, and toolmark examinations. Trace element profiles have been used to compare a diverse range of materials such as polymers,5 solder,6 radiological materials,7,8 glass,9−11 lead,12 aluminum foil,13 paint,14 drugs,15,16 tape,17 fibers,18 geological materials,19 food,20 beer,21 ink/toner,22 diamonds,23 and oil.24 A link between a suspect and a crime is suggested if the crime sceneassociated copper sample and the suspect-associated copper sample are determined to be analytically indistinguishable (i.e., the two samples cannot be statistically distinguished considering

he potential for the presence of copper as forensic evidence has increased recently because of an increase in the frequency of the theft of copper wiring and piping associated with an increase in the market value of recycled copper coupled with the recent economic downturn. Copper sources such as air conditioners, construction sites, copper production facilities, and wiring and piping inside the walls of buildings are targeted for theft. Property damage often exceeds the value of the stolen copper by several times, in addition to other economic and social consequences such as injury or death of the thief, loss of electrical power, defacing of public art or memorials, and loss of crops.1,2 Local, state, and federal agencies have responded in some cases by forming Copper Theft Task Forces3 and introducing legislation to curb metal theft. For example, a bill was recently introduced in the U.S. Senate to make metal theft (of which copper theft comprises 96% in the United States) a federal crime, impose harsher penalties for theft, and discourage payment for stolen metal.4 Another application of copper wire of forensic interest is the use of copper wires in homemade electronic devices such as improvised explosive devices (IEDs). Copper wires are used in IEDs to provide an electrical connection between components © 2014 American Chemical Society

Received: April 14, 2014 Accepted: July 9, 2014 Published: July 9, 2014 8176

dx.doi.org/10.1021/ac5013514 | Anal. Chem. 2014, 86, 8176−8182

Analytical Chemistry

Article

effects.26−28 A blank and a SRM were also prepared and analyzed for every 12 samples digested to account for day-to-day drift. Analyte Selection and ICPMS Measurement. Twentyfive potential analyte elements were selected if they were comined with copper, commonly found in high-purity copper, or potentially alloyed with copper to modify its properties.32,33 Eight elements (Ag, As, Bi, Co, Ni, Pb, Sb, and Se) were found at concentrations above the quantitation limit and consistently measured with good precision. Only these elements were used in the data analysis. An Element2 ICPMS instrument (ThermoScientific, Bremen, Germany) was used to measure trace element concentrations via external calibration with internal standardization. Isotopes and measurement conditions are listed in Table 1. All intensities were the average of five instrument replicates.

the precision of the measurement and manufacturing processes). In a forensic scenario, this would involve the measurement and statistical comparison of the trace element profiles of the small section of copper left behind when thieves cut a copper pipe/wire in a theft case or a sample of wire from an IED with a sample of copper associated with a suspect. These examinations require an analytical and data analysis method with a high discriminating power, that is, a method that correctly determines that two samples randomly selected from a population are from different sources a large percentage of time (i.e., minimizes the false exclusion error rate) while controlling at an acceptable level the rate of incorrect assertion of analytically indistinguishable crime scene- and suspect-associated samples (i.e., the false positive error rate).25 Here, a method for measuring trace element profiles in copper using inductively coupled plasma mass spectrometry (ICPMS) was developed using standard reference materials (SRMs). The discriminating power of the method was estimated using a parametric-based and an empirical-based model to simulate additional data based on samples selected to represent withinand between-source covariance. The forensic significance of a finding that two compared samples are analytically indistinguishable is not addressed herein and will be the topic of a separate paper.

■

Table 1. Analyte Measurement Details analyte 107 121

Ag Sb

206,207,208 209

Bi 59 Co 60 Ni 75 As 77 Se

EXPERIMENTAL SECTION

Sample Preparation. All cleaning, digesting, and diluting operations that required exposing samples or solutions to open air were performed in a laminar flow, ultraclean workstation (Microzone Corp., Ottawa, ON). A wafering saw (IsoMet 1000 model 112180, Buehler, Lake Bluff, IL) was used to cut copper rod samples, and wire clippers were used to cut wire samples. National Institute of Standards and Technology (NIST, Gaithersburg, MD) SRMs 394, 395, 398, and 399 were used as received. Three SRMs (39X 17867 batch AA, 39X 17869 batch AB, and 39X 17866 batch AC) from MBH Analytical Ltd. (Barnet, England) and one SRM (075) from the Institute for Reference Materials and Measurement (Geel, Belgium) were analyzed for troubleshooting perceived inaccurate measurements of Co concentrations in the NIST SRMs. Copper samples were cleaned in 30% (w/w) nitric acid (Certified ACS Plus grade, Fisher Scientific, Pittsburgh, PA) and rinsed twice with 18.8 MΩ cm deionized water to remove surface contamination. After being washed and dried, 0.1−0.6 g samples were weighed in perfluoroalkoxy (PFA) microwave digestion vessels and dissolved in 2.5 mL of concentrated nitric acid (70% Certified ACS Plus grade). For some samples, an additional 0.5 mL of concentrated nitric acid was required for complete dissolution of the copper solid. After the evolution of NO2 gas had reached completion, deionized water was added such that the ratio of nitric acid to deionized water added (by volume) was 1. Solutions were microwave digested with the following program: 5 min ramp to 180 °C and then a 5 min hold at 180 °C using a MARS5 microwave system (CEM Corp., Matthews, NC). Each final solution for measurement of trace element concentrations contained approximately 1000 ppm of copper and 10 ppb of Li, In, Tl, Sc, and Y internal standard in 5% nitric acid (doubly distilled Plasma Pure Plus grade, SCP Science, Champlain, NY). Elements for internal standards (Li, In, Tl, Sc, and Y) were chosen if they were absent from high-purity copper solutions, such that all analytes were within 30 amu of one internal standard to account for drift and mass-dependent matrix

a

Pba

resolution

dwell time (ms)

low low low low medium medium high high

10 10 10 10 50 50 50 50

The three isotopes of Pb were summed prior to further data analysis.

Where possible, the mass spectrometer was operated with lowresolution ion slits to maximize sensitivity unless medium or high resolution was required to resolve a spectral interference. Method quantitation limits (QLs) were estimated by carrying 11 blank solutions through the sample preparation process and measuring the standard deviation in trace element concentrations (QL = 10 × the standard deviation of blank concentrations).29−31 Continuing calibration verification standards were analyzed every 12 samples. Concentrations are the average concentration of three subsamples of each copper sample carried independently through the sample preparation process, unless noted, and are expressed as parts per million in solid copper. Data Analysis. R version 3.01 with mclust, mvtnorm, and MASS packages was used for data analysis on an early 2013 Macbook Pro Retina with an Intel i7 processor and 16 gigabits of DDR3 RAM. Further details of the statistical analysis are provided throughout the Results and Discussion.

■

RESULTS AND DISCUSSION Method Accuracy and Precision. The sample preparation procedure detailed in the Experimental Section was developed using SRMs. The results of six replicate analyses of NIST SRMs 394 and 398 are shown in Figure 1 to demonstrate the accuracy and precision of the developed procedure. All analytes except Se in SRM 394 and Co in both SRMs were within 2.4−23% (median = 10%) of the certified concentrations with replicate precisions of 1.1−6.5% (median = 2.0%) relative standard deviation (RSD). The inaccuracy is likely due to the imperfect matching of the m/z and ionization energy of the four internal standards with those of all of the analytes. Alternative calibration strategies including standard additions, isotope dilution, and matrix-matched calibration curves were also

8177

dx.doi.org/10.1021/ac5013514 | Anal. Chem. 2014, 86, 8176−8182

Analytical Chemistry

Article

inaccuracy. Four additional SRMs from other vendors were also analyzed with Co concentrations within 10, 16, 6.0, and 1.6% of the certified concentrations; therefore, Co was retained in the data set. To verify the reproducibility of the method over different days, SRM 398 was measured three additional times each on four days (for a total of 18 replicates) over the course of the study. The pooled average accuracy was 1.0−18% (median = 10%) with precisions of 4.3−10% (median = 5.3%) RSD over the course of the study. Discriminating Power: Comparison of Between- and Within-Source Covariances. To estimate the discriminating power of the developed ICPMS method, we selected samples to estimate the between-source covariance (wanted covariance) and within-source covariance (unwanted covariance). The between-source samples were intended to elucidate the population distribution of mean trace element concentrations from different sources. The within-source samples will assess the covariance expected from a single source. In other words, the between-source samples address the spacing in concentration means between sources, whereas the within-source samples provide the spread in concentration of each of the sources. Estimation of Covariance between Sources. To select samples to estimate between-source covariance, the time interval required to obtain samples that can be considered to originate from different sources (i.e., have distinguishable trace element profiles) had to be determined. To do this, samples of each of 30 consecutively produced rod coils (production for 90 min) were analyzed. Note that copper rod was sampled for this portion of the study because of difficulties in obtaining the number of copper wire samples required without disruption of the manufacturing process.33 As demonstrated in Figure 2, the trace element profile changed by a statistically significant amount at least once over the course

Figure 1. Average measured analyte concentrations divided by certified concentrations. Uncertainty is expressed as the standard deviation (n = 6).

explored but were determined to be impractical or impossible with the currently available materials. In the envisioned forensic comparison scenario in which crime scene- and suspectassociated samples are compared in a single analysis, precision is the primary metric of importance, whereas accuracy is important for interlaboratory comparisons. The poorer precision for Se in SRM 394 was due to a single replicate of SRM 394 having a concentration significantly higher than those of the other replicates, perhaps because of contamination. The measurement of the Co concentration had poor accuracy but excellent precision (2.3% and 1.2% RSD for NIST SRMs 394 and 398, respectively). A Co spike added prior to dissolution was 100% recovered, eliminating the loss of Co in the sample preparation procedure, matrix effects, and Co calibration standard contamination as possible causes of the

Figure 2. Trace element concentrations (parts per million in the solid) of samples of 30 consecutively produced rod coils. Replicate analyses of three subsamples of each copper sample are connected by a line for the sake of visualization. 8178

dx.doi.org/10.1021/ac5013514 | Anal. Chem. 2014, 86, 8176−8182

Analytical Chemistry

Article

Figure 3. Trace element concentrations (parts per million in the solid) measured from samples of every 30th copper rod coil over 10 days of refinery production. Replicate analyses of three subsamples of each copper sample are connected by a line for the sake of visualization.

and each group has the same covariance matrix (EEE covariance structure). The small number of groups indicates strong clustering of the trace element concentration vectors for the 93 copper rod samples. This should lead to well-estimated parameters for the mixture model and reliable simulation of additional data to estimate the discriminating power. The 93 copper rod samples were divided into one of two groups on the basis of the maximum likelihood estimate of cluster membership35 with a summary shown as a pairwise biplot in Figure 4. The covariance for some elements (Ag and Sb, for example) can also be seen by comparing the shape of the trace element concentrations over time in Figure 3. The latent variable(s) leads to high- and low-concentration groups, suggesting that, for example, an upset in the efficiency of the refining procedure that impacts all elements may be responsible. The covariance matrix for the eight trace element concentrations from the 93 rod samples (Σb) is the between-source covariance needed to estimate the discriminating power (Table S1 of the Supporting Information). Estimation of Covariance within a Source. The covariance matrix for within-source samples must be determined to provide an estimate of the spread about the mean concentration of individual sources (within-source covariance). To do this, the trace element concentrations were measured from each of three randomly selected No. 12 AWG copper wires (manufactured by the same copper refinery that produced the 93 copper rods) every 5 ft along the length of a 70 ft wire. Thus, the estimated within-source covariance will include measurement variability and the heterogeneity along 70 ft of copper wire (arising, in part, from the potential cross-sectional heterogeneity in trace element concentrations in copper rod, which may be stretched longitudinally during the wire production process).33 The covariance matrices of the three wires were used to determine the overall wire covariance matrix, i.e., the within-

of the production of the 30 rod coils (compare, e.g., Ag concentrations from samples of rod coil 1 and rod coil 16). On this basis, the samples from different sources of copper needed to estimate the between-source covariance were taken from every 30th rod coil produced over a 10-day production period (12 calendar days). The trace element concentrations for these 93 rod samples (plotted in Figure 3) were then used to estimate the between-source covariance in the trace element profiles of copper wire. The multivariate trace element concentration covariance matrix for these 93 rod samples (between-source covariance) was then estimated. A Shapiro−Wilk test for univariate normality on the trace element concentrations individually indicated that the data do not follow the Gaussian distribution. One potential explanation is that because of the large number of variables in the refining process, one or more latent variables may be present, leading to two or more overlapping Gaussian trace element concentration distributions. Examples of causes for latent variables include differences in the percentage of copper produced from sulfide or oxide ore (and corresponding differences in the refining procedure)33 and changes in the efficiency of any of the refining processes. To account for latent variables, multivariate Gaussian mixture model distributions34 with varied numbers of groups and group covariance structures were fit to the eight-dimensional distribution of trace element concentration vectors using the R mclust package as detailed in the Supporting Information.35 The quality of the mixture model distributions with respect to the experimental data distribution was compared using a Bayesian Information Criteria (BIC) metric.36 The two-group model with an EEE covariance structure35 has a maximal BIC of 1448 (see Figure S1 of the Supporting Information), indicating this was the best fit model; this signifies that there is at least one latent variable leading to two separate trace element concentration distributions (two groups), 8179

dx.doi.org/10.1021/ac5013514 | Anal. Chem. 2014, 86, 8176−8182

Analytical Chemistry

Article

Figure 4. Bivariate plots of trace element concentrations (parts per million in the solid) showing classification of copper rod samples into two groups (teal and red points). For example, the plot in the top left corner displays the concentration of Ag on the x-axis and Sb on the y-axis.

Figure 5. Trace element concentrations (parts per million in the solid) sampled along the length of three 70 ft wires. Replicate analyses of three subsamples of each copper sample are connected by a line for the sake of visualization.

covariance matrices (VVV covariance structure) using mclust.35 The quality of the fits was then compared with a BIC metric, which indicated that it is much more reasonable that the three wires have a common covariance matrix (BIC = 1847) than the covariance matrix for one or more wires are different (BIC = 1757). The common within-source covariance matrix (Σw) is given as Table S2 of the Supporting Information.

source covariance. There are two options for the covariance matrices of the three wires: either there is a common wire covariance matrix, or one or more wire covariance matrices are different. The more likely hypothesis was determined by fitting the distribution of trace element concentration vectors of the three copper wires with three distributions with common covariance matrices (EEE covariance structure) and different 8180

dx.doi.org/10.1021/ac5013514 | Anal. Chem. 2014, 86, 8176−8182

Analytical Chemistry

Article

Discriminating Power Based on a Parametric Bootstrap Model. The experimental data were used as a basis for simulating additional data. Simulated pairs of samples were repeatedly generated (bootstrap resampling37) and statistically compared to provide a reliable estimate of the percentage of copper samples that can correctly be discerned (discriminating power). This analysis was conducted as follows. (1) Two trace element concentration mean vectors were simulated using the parameters for the mixture model equation (listed in the Supporting Information) determined using the 10-day, 93-rod sample data set (samples known to originate from different sources). (2) Five replicate samples were simulated about each of the mean vectors with a covariance defined by Σw. (3) Hotelling’s T2 test over a range of p-values (i.e., over a range of false inclusion rates) was used to compare the two mean vectors. (4) Steps 1−3 were repeated 10000 times. The fraction of the 10000 simulated copper sample pairs correctly determined to originate from different sources is the discriminating power at each false inclusion rate. The discriminating power for this parametric bootstrap approach is plotted in Figure 6.

compared to all other mean vectors (a total of 4278 comparisons). The fraction of sample pairs correctly determined to originate from different sources is the discriminating power at each false inclusion rate. Selected discriminating powers for this empirical approach are listed in Table 2. Results similar to those of Table 2. Discriminating Powers for Selected Hotelling’s T2 Test p-Values from the Empirical, Two-Level BootstrapBased Approach p-value of Hotelling’s T2 test

discriminating power

0.001 0.01 0.05 0.10 0.15

0.0383 0.3537 0.8720 0.9697 0.9911

parametric bootstrap resampling are obtained with >87% of samples correctly concluded to originate from different sources at a 5% false inclusion rate.

■

CONCLUSIONS A small variability in the trace element profile along 70 ft of No. 12 AWG wire compared to the variability in the trace element profile over the course of 10 days of copper production leads to a high discriminating power for the developed ICPMS method. High and similar discriminating powers were estimated using parametric-based (>90%) and empirical-based (>87%) data simulation approaches and support the use of copper wire evidence as a valuable technique for the comparison of evidence in criminal and terrorism cases. A source of variability in the trace element profiles for which these experiments do not account is the variability between manufacturers. While this between-source variability is limited by the efficiency of the refining process and ASTM standards for high-purity copper,38,39 the discriminating power determined here is likely an underestimate. Future work will focus on trace element variations between manufacturers, extending the conclusions to copper pipe (which is manufactured by a process similar to that used for copper wire), the potential impact of commercial wire distribution patterns on discriminating power, and suggested steps for determining local discriminating power for use in a criminal case.

Figure 6. Discriminating power as a function of the p-value of Hotelling’s T2 test when comparing between-source samples using parametric bootstrapping. Discriminating powers at specific p-values are given in the inset.

At a false inclusion rate of 5% (i.e., a 5% chance that two samples are identified as originating from the same source when they actually originate from different sources), >90% of the samples were able to be correctly identified as originating from different sources. If a smaller false inclusion rate is required, such as 1%, >36% of the samples could still be correctly discriminated. If a higher false inclusion rate of 10% is determined to be acceptable, >98% of the samples could be correctly discriminated. Discriminating Power Based on an Empirical, Two-Level Bootstrap Model. The discriminating power was also estimated using the 93 experimentally measured between-source mean trace element concentration vectors (rather than simulating them from the mixture model as described above) as follows. (1) Two experimentally measured mean vectors were selected from the between-source 93-rod sample data set. (2) Five replicate samples were simulated about each of the two mean vectors with a covariance defined by Σw. (3) Hotelling’s T2 test over a range of p-values (i.e., over a range of false inclusion rates) was used to compare the two mean vectors. (4) Steps 1−3 were repeated until each pair of experimentally measured mean vectors was

■

ASSOCIATED CONTENT

S Supporting Information *

Details of the multivariate Gaussian mixture probability density distribution used to estimate the between-source covariance and the between- and within-source covariance matrices. This material is available free of charge via the Internet at http:// pubs.acs.org.

■

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Telephone: (703) 6324553. Notes

The authors declare no competing financial interest. 8181

dx.doi.org/10.1021/ac5013514 | Anal. Chem. 2014, 86, 8176−8182

Analytical Chemistry

■

Article

(22) Trejos, T.; Flores, A.; Almirall, J. Spectrochim. Acta, Part B 2010, 65, 884−895. (23) Dalpe, C.; Hudon, P.; Ballantyne, D.; Williams, D.; Marcotte, D. J. Forensic Sci. 2010, 55, 1443−1456. (24) Kim, Y.; Kim, N.; Park, S.; Lee, D.; Lee, J. H. Forensic Sci. Int. 2013, 230, 58−67. (25) Aitken, C. G. G.; Taroni, F. Statistics and the Evaluation of Evidence for Forensic Scientists, 2nd ed.; Wiley: Chichester, England, 2004. (26) Olivares, J. A.; Houk, R. S. Anal. Chem. 1986, 58, 20−25. (27) Tan, S. H.; Horlick, G. J. J. Anal. At. Spectrom. 1987, 2, 745−763. (28) Beauchemin, D.; McLaren, J. W.; Berman, S. S. Spectrochim. Acta, Part B 1987, 42, 467−490. (29) Harris, D. C. Quantitative Chemical Analysis; W. H. Freeman and Co.: New York, 2003. (30) Taylor, J. K. Quality Assurance of Chemical Measurements; Lewis Publishers, Inc.: Chelsea, U.K., 1987. (31) Ingle, J. D.; Crouch, S. R. Spectorchemical Analysis; Prentice Hall: Upper Saddle River, NJ, 1988. (32) Sanchez, J. Freeport McMoRan Copper and Gold, personal communication, 2012. (33) Joseph, G., Ed. Copper: Its Trade, Manufacture, Use, and Environmental Status; ASM International: Materials Park, OH, 1999. (34) Frühwirth-Schnatter, S. Finite Mixture and Markov Switching Models; Springer: New York, 2006. (35) Fraley, C.; Raferty, A. E.; Murphy, T. B.; Scrucca, L. University of Washington Technical Report 597; University of Washington: Seattle, 2012. (36) Wasserman, L. All of Statistics: A Concise Course in Statistical Inference; Springer: New York, 2004. (37) Efron, B.; Tibshirani, R. J. An Introduction to the Bootstrap; Chapman & Hall/CRC: Boca Raton, FL, 1993. (38) Standard Specification for Electrolytic Copper Cathode. ASTM Standard B115, 2010; ASTM International: West Conshohocken, PA, 2000. (39) Standard Specification for Copper Rod Drawing Stock for Electrical Purposes. ASTM Standard B49, 2010; ASTM International: West Conshohocken, PA, 1998.

ACKNOWLEDGMENTS We thank Freeport McMoRan Copper and Gold Inc. for helpful discussions about the copper mining and refining process and for providing copper samples and a site tour. We acknowledge Robert Koons, who performed initial proof-of-concept work on trace elements in copper, which inspired this research project. This research was supported in part by an appointment to the Visiting Scientist Program at the Federal Bureau of Investigation (FBI) Laboratory Division, administered by the Oak Ridge Institute of Science and Education, through an interagency agreement between the U.S. Department of Energy and the FBI. This is publication 14-06 of the Laboratory Division of the FBI. Names of commercial manufacturers are provided for information only, and inclusion does not imply endorsement by the FBI or the U.S. Government. The views expressed are those of the authors and do not necessarily reflect the official policy or position of the FBI or the U.S. Government.

■

REFERENCES

(1) Steinhauer, J. Unusual Culprits Cripple Farms in California. The New York Times; The New York Times Co.: New York, July 31, 2007. (2) Mitchell, K. Salem man badly burned in attempted metal theft, police say. The Oregonian; Advance Publications: Portland, OR, August 31, 2009. (3) Garner, M. K. What’s behind recent rash of copper thefts? The Atlanta Journal-Constitution; Cox Enterprises: Atlanta, August 3, 2008. (4) Dinan, S. Senate moving to make copper theft a federal crime. The Washington Times; June 11, 2013. (5) Buscaglia, J. The Forensic Characterization of Polyethylene Films by Elemental Analysis Using Total-reflection X-ray Fluorescence (TXRF) Spectrometry. Ph.D. Dissertation, City University of New York, New York, 1999. (6) Koons, R.; Buscaglia, J.; Havekost, D. 51st Annual Meeting of the American Academy of Forensic Sciences, Orlando, FL, 1999. (7) Goldberg, M.; Finck, M.; Chamberlain, D.; Carney, K. Annual Meeting Proceedings of the Institute of Nuclear Materials Management, 2011; a128/121−a128/125. (8) Varga, Z.; Katona, R.; Stefanka, Z.; Wallenius, M.; Mayer, K.; Nicholl, A. Talanta 2010, 80, 1744−1749. (9) Koons, R.; Buscaglia, J.; Bottrell, M.; Miller, E. In Forensic Science Handbook, 2nd ed.; Saferstein, R., Ed.; Prentice-Hall: Upper Saddle River, NJ, 2001. (10) Weis, P.; Dücking, M.; Watzke, P.; Menges, S.; Becker, S. J. Anal. At. Spectrom. 2011, 26, 1273−1284. (11) Trejos, T.; Koons, R.; Weis, P.; Becker, S.; Berman, T.; Dalpe, C.; Dücking, M.; Buscaglia, J.; Eckert-Lumsdon, T.; Ernst, T.; Hanlon, C.; Heydon, A.; Mooney, K.; Nelson, R.; Olsson, K.; Schenk, E.; Palenik, C.; Pollock, E.; Rudell, D.; Ryland, S.; Tarifa, A.; Valadez, M.; van Es, A.; Zdanowiczq, V.; Almirall, J. J. Anal. At. Spectrom. 2013, 28, 1270−1282. (12) Koons, R.; Buscaglia, J. J. Forensic Sci. 2005, 50, 341−351. (13) Koons, R.; Peters, C.; Merrill, R. J. Forensic Sci. 1993, 38, 302− 315. (14) Nishiwaki, Y.; Watanabe, S.; Shimoda, O.; Saito, Y.; Nakanishi, T.; Terada, Y.; Ninomiya, T.; Nakai, I. J. Forensic Sci. 2009, 54, 564−570. (15) Chan, K.; Tan, G.; Wong, R. Anal. Lett. 2012, 45, 1122−1132. (16) Koper, C.; van den Boom, C.; Wiarda, W.; Schrader, M.; de Joode, P.; van der Peijl, G.; Bolck, A. Forensic Sci. Int. 2007, 171, 171−179. (17) Dobney, A.; Wiarda, W.; de Joode, P.; van der Peijl, G. J. Anal. At. Spectrom. 2002, 17, 478−484. (18) Koons, R. J. Forensic Sci. 1996, 41, 199−205. (19) Bong, W.; Nakai, I.; Furuya, S.; Suzuki, H.; Abe, Y.; Osaka, K.; Matsumoto, T.; Itou, M.; Imai, N.; Ninomiya, T. Forensic Sci. Int. 2012, 220, 33−49. (20) Taylor, A.; Branch, S.; Day, M.; Patriarca, M.; White, M. J. Anal. At. Spectrom. 2006, 21, 439−491. (21) Mahmood, N.; Petraco, N.; He, Y. Anal. Bioanal. Chem. 2012, 402, 861−869. 8182

dx.doi.org/10.1021/ac5013514 | Anal. Chem. 2014, 86, 8176−8182