Covariance Mapping in the Analysis of Ignitable ... - ACS Publications

The covariance matrix computed from the retention time- ion abundance data ... of given m/z ratios to obtain an estimate of the sample composition bas...
1 downloads 0 Views 268KB Size
Anal. Chem. 2006, 78, 1713-1718

Technical Notes

Covariance Mapping in the Analysis of Ignitable Liquids by Gas Chromatography/Mass Spectrometry Michael E. Sigman* and Mary R. Williams

Department of Chemistry and National Center for Forensic Science, University of Central Florida, Orlando, Florida 32816

The covariance matrix computed from the retention timeion abundance data matrix from gas chromatography/ mass spectrometry analysis of ignitable liquids is shown to be a useful tool for automated identification of ignitable liquids in a database. The absolute value of the elementby-element difference between two normalized covariance matrices is shown to quantitatively differentiate between ignitable liquids composed of complex mixtures of hydrocarbons and is amenable to automated searching of ignitable liquid databases. The covariance mapping method is applied to a matrix-contaminated postburn sample, allowing the determination of a high degree of similarity between the ignitable liquid and a heavily evaporated gasoline.

Optimal chromatographic separation with baseline resolution is not always achieved in the analysis of complex mixtures, as in the case of forensic fire debris analysis. The samples under analysis from fire debris are typically complex mixtures composed of ignitable liquid and matrix components. The analyst’s goal is generally to achieve sufficient resolution within a reasonable time frame to allow for classification of the ignitable liquid following ASTM protocols.1 Ignitable liquids used by perpetrators of arson are typically commercial products and may contain a significant number of components. Gasoline is an example of a commonly encountered ignitable liquid in fire debris analysis. Products from partial combustion and pyrolysis of materials used in building construction and furnishings, as well as volatile constituents normally present in these materials, contribute to what is collectively referred to as “matrix components”. The fire debris analyst generally follows a standard analytical protocol, rather than optimizing chromatographic conditions for each sample. Adherence to a standard protocol is intended to facilitate the identification of highly similar ignitable liquids through comparisons with standards compiled in a database. Standard method ASTM E 1618 * To whom correspondence should be addressed. E-mail: msigman@ mail.ucf.edu. Phone: 407-823-3420. (1) American Society for Testing and Materials Method E1618-01. Standard Test Method for Ignitable Liquid Residues in Extracts from Fire Debris Samples by Gas Chromatography-Mass Spectrometry; American Society for Testing and Materials: W. Conshohocken, PA, 2001. 10.1021/ac058040e CCC: $33.50 Published on Web 01/26/2006

© 2006 American Chemical Society

recommends that each fire debris analysis laboratory maintain its own library of common ignitable liquids.1 In addition to inhouse databases, a large database of ignitable liquid gas chromatography/mass spectrometry (GC/MS) data is available on the Internet.2 Interlaboratory comparisons of GC/MS data can be complicated by variations in chromatographic conditions and columns, resulting in retention times that vary from one laboratory to another. Most laboratories track retention time variations with control charts of retention times for a homologous series of normal alkanes (a “hydrocarbon ladder”) as part of a laboratory quality control practice. A single laboratory generally retains fairly tight control over the retention times of hydrocarbon standards; however, interlaboratory variations in retention times for hydrocarbon standards may be somewhat larger than intralaboratory variations. Interlaboratory comparisons could be facilitated by converting chromatographic retention times to one of the many retention indices; although this is rarely done and no one standard chromatographic index has been applied to the analysis of ignitable liquids. A preferred method of comparing GC/MS analyses from multiple laboratories would obviate the need for retention time indexing. ASTM method E 1618 utilizes extracted ion chromatograms of given m/z ratios to obtain an estimate of the sample composition based on the major ions resulting from chemical classes commonly found in ignitable liquids,1,3 along with target compound analysis, for the classification of an ignitable liquid into one of nine categories. A straight chain hydrocarbon ladder is used to estimate the carbon range of the analytes, which allows the analyst to further classify the material as “light” (C4-C9), “medium” (C8C13), or “heavy” (C8-C20+). While the final classification is useful in helping the analysts and investigators identify an ignitable liquid, the classification scheme is somewhat subjective and not easily amenable to automated database searching. The method relies heavily on comparisons of extracted ion chromatograms, which are readily prepared using commercial software. Classifications of ignitable liquids are general, and the ASTM E 1618 method is only the first step in identifying a specific ignitable liquid. Identification of a specific ignitable liquid is complicated (2) The Ignitable Liquids Reference Database can be found at the universal resource locator http://www.ncfs.ucf.edu. (3) Smith, R. M. Anal. Chem. 1982, 54, 1399A-1409A.

Analytical Chemistry, Vol. 78, No. 5, March 1, 2006 1713

by weathering (evaporation) or biological degradation,4 samplinginduced chromatographic distortion,5 continually changing product formulations, regional differences in product composition, and contributions from matrix components. Databases composed of commercial ignitable liquids common in a geographical area can aid in the identification of a specific ignitable liquid. While many laboratories maintain a limited ignitable liquid database, it may not be feasible for each laboratory to maintain a larger and more comprehensive database. The National Center for Forensic Science maintains a large online database (400 entries and growing) of ignitable liquid GC/MS data.2 Easily automated database search criteria that minimized or removed laboratory-to-laboratory variations would facilitate the search of a single database or the development of networks of databases that can be readily searched. Automated comparison of GC/MS data for complex mixtures taken from different laboratories requires advanced data analysis methods that overcome retention time shifts and interlaboratory variations in methodology. Bertsch has given a detailed account of the data interpretation procedures for ignitable liquids analysis and a review of existing automated software systems.6 Although software-automated approaches to data analysis are helpful, current approaches still rely significantly on visual pattern recognition by the analyst.6 Parallel-column gas chromatographic separation, whereby the sample is split between two GC columns that provide complementary separation, followed by time-of-flight mass spectrometric detection has been shown to provide some analytical advantage in complex samples.7 Using the two-column method, it was found that while methyl tert-butyl ether and benzene in synthetic gasoline samples were poorly resolved, they could be quantitatively determined by the generalized rank annihilation method (GRAM) of data analysis. The GRAM analysis method, which is a chemometric method for enhancing quantitation, is also sensitive to retention time shifts and requires the use of a peak-alignment correction.8,9 Frysinger has applied GC × GC methods to the analysis of fire debris and shown that the method is highly useful in identifying the components of the complex mixture,10 although the GC × GC method introduces a second retention time and therefore increases the challenge of interlaboratory comparisons and database searching, even with software aides.11 Covariance mapping12-15 and coincidence measurements16 have been applied to time-of-flight mass spectrometry to resolve correlated events. Covariance mapping has not previously been (4) Mann, D. C.; Gresham, W. R. J. Forensic Sci. 1990, 35, 913-923. (5) Williams, M. R.; Fernandes, D.; Bridge, C.; Dorrien, D.; Elliott, S.; Sigman, M. E. J. Forensic Sci. 2005, 50, 316-326. (6) Bertsch, W. Forensic Sci. Rev. 1997, 9, 1-22. (7) Prazen, B. J.; Bruckner, C. A.; Synovec, R. E.; Kowalski, B. R. Anal. Chem. 1999, 71, 1093-1099. (8) Fraga, C. G.; Prazen, B. J.; Synovec, R. E. Anal. Chem. 2000, 72, 41544162. (9) Buckner, C. A.; Prazen, B. J.; Synovec, R. E. Anal. Chem. 1998, 70, 27962804. (10) Frysinger, G. S.; Gaines, R. B. J. Forensic Sci. 2002, 47, 471-482. (11) Reichenbach, S. E. Kottapalli,V.; Ni, M.; Visvanathan, A. J. Chromatogr., A 2005, 1071, 263-269. (12) Frasinski, L.; Codling, J. K.; Hatherly, P. A. Science 1989, 246, 10291031. (13) Jukes, P.; Buxey, A.; Jones, A. B.; Stace, A. J. Chem. Phys. 1997, 106, 13671372. (14) Foltin, M.; Stueber, G. J.; Bernstein, E. R. J. Chem. Phys. 1998, 109, 43424360.

1714

Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

applied to the analysis of complex GC/MS data sets. A somewhat related multivariate statistical technique for the analysis of timeor perturbation-dependent data, principal components analysis,17 has been applied to the classification and identification of ignitable liquids18-21 but not as a method of enhancing automated database searching. This paper examines the use of covariance maps to facilitate the comparison of GC/MS data from ignitable liquids samples of the kind often encountered in fire debris analysis. Application of covariance mapping techniques is shown to provide a useful way of visualizing complex GC/MS data, to facilitate interlaboratory data comparisons, and to provide a method of automated database searching. MATERIALS AND METHODS Most of the data sets analyzed in this paper were taken from the Ignitable Liquid Database and Reference Collection (ILRC).2 Data for all entries in the database were collected under standard conditions (discussed below) and reviewed by a panel of fire debris analysts from local, state, and federal crime laboratories prior to publication in the database. The database is freely available to the scientific community for forensic, training, and research purposes. In addition, some of the data sets were analyzed specifically for the purpose of this study, employing the analytical conditions discussed here. GC/MS analyses were conducted under conditions designated “GC/MS-A”, “GC/MS-B”, or “GC/MS-C”. The “GC/MS-A” data, including the data sets taken from the ILRC, were collected on a Hewlett-Packard 6890 gas chromatograph interfaced to a 5973 mass spectrometer. All of the samples from neat solutions were prepared by dilution of 20 µL of liquid into 1 mL of carbon disulfide for GC/MS analysis. Samples of 1 µL were introduced by an Agilent ALS autosampler G2614A with a Merlin septumless injection system and split 50:1 in a split/splitless injector maintained at 250 °C. The mass spectrometer transfer line was maintained at 280 °C, and the source temperature was 230 °C. Mass spectra were repetitively scanned from 30 to 350 m/z following a 2-min solvent delay. Unless otherwise specified, samples were chromatographed on 0.20 mm i.d. × 23.4 m long Hewlett-Packard HP-1 column with a 0.5-µm film thickness. Helium carrier gas was maintained at a constant flow of 0.8 mL/ min on the column, corresponding to a linear velocity of 36 cm/ s. The initial oven temperature of 50 °C was held for 3 min, followed by a temperature ramp of 10 °C/min to a final temperature of 280 °C, which was held for 4 min. GC/MS-B conditions utilized the same instrument used for GC/MS-A; however, the column was changed to a 0.20 mm i.d. × 25.0 m long HewlettPackard HP-1 column with a 0.5-µm film thickness. GC/MS-C utilized a different Hewlett-Packard HP-1 instrument equipped with (15) Feldman, A. B.; Antoine, M.; Lin, J. S.; Demirev, P. A. Rapid Commun. Mass Spectrom. 2003, 17, 991-995. (16) Van Stipkonk, M. J.; Schweikert, E. A.; Park, M. A. J. Mass Spectrom. 1997, 32, 1151-1161. (17) Malinowski, E. R.; Howery, D. G. Factor Analysis in Chemistry; Wiley: New York, 1980. (18) Doble, P.; Sandercock, M.; Du Pasquier, E.; Petoxz P.; Roux, C.; Dawson, M. Forensic Sci. Int. 2003, 132, 26-39. (19) Tan, B.; Hardy, J. K.; Snavely, R. E. Anal. Chim. Acta 2000, 422, 37-46. (20) Sandercock, P. M. L.; DuPasquier, E. Forensic Sci. Int. 2003, 134, 1-10. (21) Sandercock, P. M. L.; DuPasquier, E. Forensic Sci. Int. 2004, 140, 43-59.

Table 1. Designations and Descriptions of Ignitable Liquids Reference Collection and Database (ILRC) Samples sample designationa

ILRC No.b

LPD MPD1 MPD2 HPD DAR ISO OXY

224 227 81 226 164 119 252

GAS1a GAS1b GAS1c GAS2 GAS3 GAS4 GAS5 NAP

301 301 301 96 98 303 105 140

a

ASTM classification

description Ace VM&P Naphtha, unweathered Ace paint thinner (100% mineral spirit), unweathered Exxon Varsol 1, unweathered Ace odorless grade 1 kerosene, unweathered BBQ Pro charcoal lighter, unweathered Exxon Isopar H specialty industrial solvent, unweathered DEFT Clearwood Finish (gloss) fabric/furniture protector, unweathered Hess regular unleaded gasoline, unweathered Hess regular unleaded gasoline, unweathered, Hess regular unleaded gasoline, unweathered, British Petroleum regular unleaded, 25% weathered by volume British Petroleum regular unleaded, 75% weathered by volume Hess regular unleaded, 75% weathered by volume Phillips 66 regular unleaded gasoline, unweathered Lamplight Farms Citronella Torch Fuel lamp oil, unweathered

carbon range

GC/MS conditions

petroleum distillate petroleum distillate petroleum distillate petroleum distillate dearomatized distillate isoparaffinic products oxygenated solvent

6-11 8-13 8-13 8-16 8-11 9-12 6-11

A A A A A A A

gasoline gasoline gasoline gasoline gasoline gasoline gasoline naphthenic paraffinic products

6-14 6-14 6-14 6-14 6-16 7-14 6-13 10-14

A B C A A A A A

Designation used in this note. b Ignitable Liquids Reference Collection and Database sample number.

a 0.20 mm i.d. × 23.6 m long Hewlett-Packard HP-1 column with a 0.5-µm film thickness. A set of 15 samples were selected for comparison, which represent 6 of the 9 ASTM classes of ignitable liquids. Samples were also chosen from the three subclasses (“light”, “medium”, “heavy”) of petroleum distillates, along with five different gasoline samples analyzed under different sets of conditions. The samples analyzed in this paper are listed in Table 1. The gasoline sample designated as GAS1 was analyzed under three different sets of conditions (designated GAS1a, GAS1b and GAS1c), corresponding to GC/MS-A, GC/MS-B, and GC/MS-C to simulate analyses coming from three different laboratories. To simulate a postburn sample, a 500-µL sample of gasoline (ILRC 301) was placed on a 4-cm2 piece of carpet and carpet padding, ignited, and allowed to burn for 2 min before the sample was extinguished by suffocation. The headspace above the sample was collected on a 1-cm2 activated carbon strip by heating the sample in an oven for 16 h at 66 °C. The activated carbon was extracted with carbon disulfide and analyzed by GC/MS under GC/MS-A conditions described above. As a matrix pyrolysis control sample, a 4-cm2 piece of carpet and padding was ignited and allowed to burn for 2 min before extinguishing the sample by suffocation. The sample was placed in a 1-qt can, and the headspace was sampled on activated carbon with subsequent extraction and GC/MS analysis under conditions identical to those used for the gasoline-laden sample. Spectral data from 30 to 200 m/z and corresponding to typically the first 2000 consecutive scans were exported into commaseparated values (CSV) format ASCII files using the Agilent Chemstation 3D-Export option. The final scan number exported was selected to be sufficiently long to include all eluting peaks in the GC/MS total ion chromatogram. The CSV files were read into Excel (Microsoft Inc.) and parsed to select m/z versus scan number data sets, which were exported to Mathematica (Wolfram Inc.) for all matrix manipulations. All matrix visualization graphics were produced with Sigma Plot (Systat Software Inc.). RESULTS AND DISCUSSION Each sample matrix, Y, used in this study was composed of i rows, each corresponding to a single mass scan (designated ti),

and j columns, each corresponding to an m/z ratio (designated mj) ranging from 30 to 200 m/z. The matrix values, y(ti, mj), correspond to the ion abundances in the GC/MS data set. The matrix representing covariance about the origin of the data set, Z, was generated by premultiplying Y by its transpose, as in eq 1.17

Z ) Y TY

(1)

When the calculation of Z is carried out without normalization of Y, as done here, the values in the resulting matrix are inherently weighted in proportion to the magnitude of the values in Y.17 Furthermore, the absolute magnitudes of the elements of Z depend directly on the amount of sample analyzed. To remove the concentration dependence and facilitate a simple mathematical comparison of Z calculated for two samples, it is advantageous to normalize Z. Each covariance matrix Z was normalized such that the sum of all normalized matrix elements equal a value of 1.0, and the normalized covariance matrix is denoted as ZN. In Z, each off-diagonal element corresponds to the covariance of two ions of different m/z ratios.14 The time dependence has been removed in the calculation of ZN, making it a potential tool for visually and quantitatively performing a rapid comparison of complex mixtures analyzed by GC/MS, including samples analyzed under different chromatographic conditions. Examples of four ZN matrices are shown in Figure 1. Figure 1a shows ZN for the heavy petroleum distillate designated HPD in Table 1. Figure 1a clearly shows ions that are indicative of alkanes (m/z 43, 57, 71), cycloalkanes and alkenes (m/z 41, 55), alkylbenzenes (m/z 91, 105), and indanes (m/z 131, 132), which occur in the sample. Figures 1b and 1c correspond to samples ISO and GAS1a, respectively, from Table 1. The graphs of the ZN matrices are readily distinguished visually. Figure 1b clearly shows ions that are indicative of alkanes (m/z 43, 57, 71), which occur in the sample, but does not show the corresponding alkenes and cycloalkanes, which are usually seen in a distillate. Figure 1c clearly shows ions that are indicative of alkylbenzenes (m/z 91, 105), which predominate in gasoline samples. The ZN matrices in Figures 1b and c are easily distinguished from one another and also readily distinguished from Figure 1a. However, Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

1715

Figure 1a (sample HPD) and Figure 1d (sample MPD2) are not as readily distinguished, although careful inspection reveals some differences, particularly in the off-diagonal elements. Two ZN matrices may be compared analytically by calculating a “distance”, D, between the two surfaces. The Manhattan distance, D, is calculated on an element-by-element basis as the absolute value of the difference in magnitude between the two matrix elements.22 The absolute differences are then summed over all matrix elements and the sum is divided by 2, as in eq 2. In eq 2, zN1(i,j) and zN2(i,j) are the elements of two ZN matrices.

∑∑|z D)

i

N1(i,j)

- zN2(i,j)|

j

2

(2)

The values of D can range from 0, for two identical ZN matrices, to a maximum of 1, for two totally nonoverlapping matrices. The values of D calculated for all the samples examined in this work are given in Table 2. Alternatively, a similarity index based on D can be defined as in eq 3, where Dmax is the maximum distance between two ZN matrices. Since Dmax has a theoretical value of 1, S reduces to eq 4 and offers no distinct advantage over D itself. An additional option is to calculate a Euclidean distance by taking the square root of the sum of the square of the differences between corresponding elements in the two matrices.22 The range of Euclidean distances was compressed relative to the Manhattan distances given in Table 2 and was not used as a comparative metric in this work.

S ) 1 - D/Dmax

(3)

S)1-D

(4)

Several observations can be made concerning the values in Table 2. All of the values are observed to fall in the range from 0 to 1, with the maximum value of 0.975 for the distance between the samples specified as ISO and GAS4. This value approaches the maximum value for D and reflects the dramatic difference between the ions observed for a sample containing only branched hydrocarbons (ISO) and a gasoline sample that contains primarily aromatic species (GAS4). The minimum distance between two samples in Table 1, a value of 0.052, was observed for samples GAS1a and GAS1b. These two samples were from the same source and analyzed on the same instrument using different columns having the same bonded phase and only slightly differing column lengths. In addition, the two analyses were separated in time by one year, so that instrument tune values had also changed. The total ion chromatograms for the two samples are highly similar; however, there are slight variations in the relative intensities of some chromatographic peaks, which are manifest as variations in ZN for each sample. The small value of DGAS1a,GAS1b demonstrates that the comparison method has some tolerance for variations in chromatographic profiles. Sample GAS1c corresponds to the same gasoline sample as GAS1a and GAS1b; however, GAS1c was analyzed on a different instrument with a column of significantly different length than the column used for the other two GAS1 Figure 1. ZN matrices for samples (a) HPD (b) ISO (c) GAS1a, and (d) MPD2. The samples are described in Table 1. 1716

Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

(22) Staniloae, D.; Petrescu, B. Environ. Forensics 2001, 2, 363-66.

Table 2. “Distances” between Area-Normalized Covariance Matrices for Samples in Table 1

LPD MPD1 MPD2 HPD DAR ISO OXY GAS1a GAS1b GAS1c GAS2 GAS3 GAS4 GAS5 NAP

LPD

MPD1

MPD2

HPD

DAR

ISO

OXY

GAS1a

GAS1b

GAS1c

GAS2

GAS3

GAS4

GAS5

0.000 0.429 0.423 0.456 0.277 0.511 0.568 0.897 0.866 0.869 0.824 0.962 0.971 0.849 0.497

0.000 0.105 0.152 0.257 0.406 0.463 0.763 0.736 0.749 0.688 0.790 0.810 0.709 0.486

0.000 0.155 0.265 0.443 0.478 0.789 0.761 0.774 0.710 0.813 0.834 0.736 0.428

0.000 0.285 0.400 0.494 0.826 0.799 0.807 0.753 0.859 0.874 0.772 0.422

0.000 0.382 0.525 0.897 0.866 0.870 0.823 0.960 0.968 0.848 0.383

0.000 0.566 0.910 0.880 0.888 0.844 0.969 0.975 0.857 0.569

0.000 0.722 0.694 0.698 0.645 0.777 0.789 0.666 0.638

0.000 0.052 0.136 0.137 0.353 0.307 0.084 0.890

0.000 0.119 0.125 0.387 0.343 0.085 0.864

0.000 0.140 0.457 0.425 0.175 0.868

0.000 0.385 0.356 0.113 0.825

0.000 0.103 0.357 0.944

0.000 0.308 0.955

0.000 0.848

Figure 2. Total ion chromatogram for GAS1a displayed as a positive-going chromatogram, and the total ion chromatogram for GAS1c shown as negative-going. See the text for corresponding chromatographic conditions. Five sets of corresponding peaks (a and a′, b and b′, etc.) are designated above the corresponding peak.

samples. The difference in total ion chromatograms for samples GAS1a and GAS1c can be seen in Figure 2. In Figure 2, the total ion chromatogram for GAS1a is displayed as a positive-going chromatogram, while GAS1c is shown as negative-going and five sets of corresponding peaks (a-e) are marked to facilitate comparison of the two chromatograms. The retention times and the relative peak intensities in the two chromatograms are significantly different. The distance DGAS1a,GAS1c was calculated to be 0.136, which is significantly larger than DGAS1a,GAS1b and demonstrates that the distance between the ZN matrices is not entirely insensitive to changes in instrumental and chromatographic conditions. However, the calculated value for D does show a reasonable tolerance to changes in chromatographic conditions when comparing GC/MS data for a complex mixture measured under substantially different conditions. In the analysis of fire debris samples following ASTM E 1618 guidelines, a bonded phase, methylsilicone, phenylmethylsilicone, or equivalent, capillary column is recommended.1 Column length and temperature programming can vary under ASTM E 1618, so long as each component of the test mixture is adequately separated. Similarly, the distances between unweathered gasoline samples GAS5 and GAS1a and GAS1b are small (∼0.085), whereas a comparison GAS5 with sample GAS1c produces a distance of

0.175. The six pairwise comparisons of all unweathered gasoline samples gave an average distance of 0.109 ( 0.044. Gasoline samples identified in fire debris analysis are often partially evaporated, even up to 90% or more in some cases. The sample GAS2 corresponds to a 25% evaporated gasoline, and samples GAS3 and GAS4 correspond to 75% evaporated gasoline. The average distances between the ZN matrices for the 25% weathered gasoline sample, GAS2, and the unweathered samples GAS1a, GAS1b, GAS1c, and GAS5 was 0.129 ( 0.012. Although the average distance is greater than that observed for the distances between unweathered samples, the standard deviation in the distances make it impossible to distinguish the 25% weathered gasoline from unweathered samples by this technique. The average distance between the unweathered gasoline samples and the two 75% weathered gasolines was 0.367 ( 0.053, whereas the distance between the two 75% weathered gasolines was only 0.103. The results for this limited number of gasoline samples suggest that it would be possible to identify highly evaporated gasoline samples and distinguish them from lightly weathered or unweathered gasoline samples using an automated database search involving the calculation of D for each sample. The sensitivity of the method to distinguishing between weathered and unweathered samples also points to the need to either include weathered samples in a database or provide a method to predict chromatographic distortions resulting from weathering.23,24 Most of the remaining comparisons of the distances between the ZN matrices, given in Table 2, demonstrate that representative samples of the major ASTM classifications of ignitable liquids can be differentiated by D and that a database search based on D would identify complex mixtures of similar composition. For example, the distance calculated by a comparison of sample HPD with samples MPD1 (DHPD,MPD1 ) 0.152) and MPD2 (DHPD,MPD2 ) 0.155) exceeds the distance between samples MPD1 and MPD2 (DMPD1,MPD2 ) 0.105). The comparison demonstrates a subclass differentiation, heavy versus medium petroleum distillate, based on the numerical comparison. A visual comparison of the ZN matrices for HPD and MPD2 can be obtained by comparing Figure 1a and Figure 1d. The values in Table 2 suggest that a minimum D value calculated between an unknown and entries in (23) Hirz, R.; Rizzi, A. M. JFSS 1991, 31, 309-319. (24) Hirz, R.; Rizzi, A. M. Chromotographia 1991, 31, 224-232.

Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

1717

to generate ZN matrices for both data sets. Figure 3a shows the ZN matrix for the “matrix” sample. The principal ions present in the “matrix” sample include m/z 104 and 118, which are indicative of alkylstyrenes.1 The formation of styrene in the pyrolysis of carpet and carpet padding has previously been reported.25 The ZN matrix from the ignited “matrix + gas” sample is plotted in Figure 3b. The most prevalent ions in the sample shown in Figure 3b are m/z 91, 105, and 119, which are indicative of alkylbenzene components present in gasoline. Trace amounts of the “matrix” components can be seen in the “matrix + gas” sample; however, the contributions from the “matrix” components are small. The ZN matrix from the “matrix + gas” sample was compared with the samples in Table 1 by calculating the distance (D). The minimal distances were found between the “matrix + gas” sample and GAS3 (D ) 0.130) and GAS4 (D ) 0.134), both of which correspond to 75% evaporated gasoline samples. The distance calculated between the “matrix + gas” sample and the other GAS samples from Table 1 varied from 0.407 to 0.499. The distance between the “matrix + gas” sample and all other samples in Table 1 ranged from 0.779 to 0.960.

Figure 3. (a) ZN matrix for a burned carpet padding sample. (b) ZN matrix for a gasoline-wetted carpet padding sample that was ignited and extinguished after a partial burn.

a database of samples from different ASTM classes could be used to rapidly screen for possible matches. It is important to note that finding an identical match in a library search with a “real” fire debris sample is unlikely due to the large number of ignitable products available coupled with product variations, regional differences in products, matrix components, etc. The goal of the search is to identify the database entry that is the most similar to the “real” sample. As a surrogate for a “real” sample, a postburn sample created in the laboratory was analyzed. Analysis of a Postburn Sample. The GC/MS data from burned carpet and padding “matrix” sample and an ignited “matrix + gas” sample of gasoline-wetted carpet and padding (see Materials and Methods section) were treated as described above

1718

Analytical Chemistry, Vol. 78, No. 5, March 1, 2006

CONCLUSIONS The distance (D) between the covariance matrices calculated for the GC/MS data from two complex ignitable liquid samples can be used as a metric for rapidly scanning a database for ignitable liquids with similar composition. The distance metric can distinguish between unweathered or lightly weathered gasoline and heavily weathered gasoline samples, as well as distinguishing between subclassifications such as light, medium, and heavy petroleum distillates. The method was shown to be applicable to a matrix-contaminated postburn sample to classify the sample as a heavily evaporated gasoline. ACKNOWLEDGMENT This work was supported under State of Florida Type II Center funding to the University of Central Florida and National Center for Forensic Science. The work was done at the National Center for Forensic Science, a National Institute of Justice program hosted by the University of Central Florida, and a member of the Forensic Science Resource Network. Received for review August 11, 2005. Accepted December 23, 2005. AC058040E (25) Almirall J. R.; Furton, K. G. J. Anal. Appl. Pyrolysis 2004, 71, 51-67.