Subscriber access provided by UNIVERSITY OF LEEDS
Article
Raman Spectroscopy of Blood for Species Identification Gregory McLaughlin, Kyle C. Doty, and Igor K. Lednev Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/ac5026368 • Publication Date (Web): 28 Oct 2014 Downloaded from http://pubs.acs.org on November 2, 2014
Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.
Page 1 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
Raman Spectroscopy of Blood for Species Identification Gregory McLaughlin†, Kyle C. Doty†, and Igor K. Lednev* Department of Chemistry, University at Albany, 1400 Washington Avenue, Albany, NY 12222, United States †
Both authors contributed equally to this work
*
Corresponding author at: Department of Chemistry, University at Albany, 1400 Washington
Avenue, Albany, NY 12222, United States. Tel: 1-518-591-8863; fax: 1-518-442-3462 E–mail address: [email protected] (I.K. Lednev) Keywords: Raman spectroscopy, Chemometrics, Blood, Serology, Human origin, Species identification Abstract The species identification of a blood stain is an important and immediate challenge for forensic science, veterinary purposes, and wildlife preservation. The current methods used to identify the species of origin of a blood stain are limited in scope and destructive to the sample. We have previously demonstrated that Raman spectroscopy can reliably differentiate blood traces of human, cat and dog (Virkler et al. Anal. Chem. 2009) and, most recently, built a binary model for differentiating human vs. animal blood for eleven species integrated with human existence (McLaughlin et al. Forensic Sci. Int. 2014). Here we report a satisfactory classification of blood obtained from eleven animal classes and human subjects by statistical analysis of Raman spectra. Classification of blood samples was achieved according to each sample’s species of origin which enhanced previously observed discrimination ability. The developed approach
1 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
does not require the knowledge of a specific (bio)chemical marker for each individual class, but rather relies on a spectroscopic statistical differentiation of various components. This approach results in remarkable classification ability even with intrinsically heterogeneous classes and samples. In addition, the obtained spectroscopic characteristics could potentially provide information about specific changes in the (bio)chemical composition of samples, which are responsible for the differentiation. 1. Introduction Interspecies blood analysis is a well-researched topic in analytical chemistry, forensic science and biochemistry. Applying analytical techniques to identify the species of a blood stain is particularly of interest to the fields of forensic casework, veterinary science, and wildlife preservation. Currently, species identification can be accomplished via a wide array of analytical techniques. The analysis of blood by high-performance liquid chromatography (HPLC) and mass spectrometry (MS) is an emerging analytical technique for species identification. Several HPLC methods have been developed which can simultaneously identify a substance as blood and determine its species of origin. One study incorporated this technique using reverse-phase HPLC to analyze fresh blood and bloodstains from humans and 28 animal species1. The sensitivity of this approach was increased using ultraviolet excitation with fluorescence detection2. Espinoza et al. expanded this method further by analyzing and quantitating the concentrations of blood in bloodstains and blood mixtures from over 50 different animal species3. Also, Espinoza et al. demonstrated that species differentiation can be substantiated with mass spectrometry4. This technique detected minor interspecies molecular mass differences in α- and β-chains (α/β-pairs) of hemoglobin from 62 different species. Most recently, a reverse-phase LC-MS method was utilized to differentiate human blood from dog and cow blood5. In that study, Steendam et al. 2 ACS Paragon Plus Environment
Page 2 of 18
Page 3 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
demonstrated the use of a single test to determine the identity and species of a biological sample. For most of these studies, species distinction was accomplished by targeting differences in hemoglobin in a destructive manner. Vibrational spectroscopy stands apart from other analytical techniques due to its nondestructive nature and the limited sample preparation required. A spectroscopic approach is therefore desirable in forensic applications, or where the sample size is in trace amounts6. Raman spectroscopy is a type of vibrational spectroscopy which also has the potential as a nondestructive confirmatory identification method for blood7-9 and other body fluids7,10-16. This technique uses a monochromatic light source to irradiate a sample. Spectra are generated by detecting inelastically scattered light by the sample. A Raman spectrum contains numerous distinctive bands which correspond to specific molecular vibrational modes17. Vibrational spectroscopy has been explored for the purposes of species identification of blood samples, initially with a study conducted by De Wael et al.18 This group reported that blood particles originating from human, cat and dog samples could not be visually differentiated through their Raman or infrared spectra. Using a novel approach, Virkler et al. demonstrated that differentiation of human, cat and dog blood samples is possible using Principal Component Analysis (PCA) modeling of Raman spectroscopic data19. Separation between these classes within a PCA model exceeded a 99% confidence interval. The application of statistical models to Raman spectra enhances the selectivity of Raman spectroscopy and has been used extensively to build similar classification models20,21. Most recently we demonstrated that this approach could be expanded to as many as eleven animals and built a binary model for the discrimination of human and non-human blood samples22. Additionally, for that study, all unknown samples were classified correctly (either human or non-human). 3 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Here, we analyzed blood from a wide survey of species (12 in total). The animal (nonhuman) samples were selected to represent three groups: animals that are bred for domestication (cat, dog, horse and cow), those that are consumed as food (chicken, cow, pig, and rabbit), and those that are integrated with human existence (mouse, rat, opossum and raccoon). To account for the complexity of the dataset, a Partial Least Squares-Discriminant Analysis (PLS-DA) classification model was used. These models have more class discrimination power than PCA and provide automatic soft prediction of classes23. The PLS-DA classification model was used to differentiate a training dataset of human and animal blood spectra. As an internal validation step, predictions were made for a set of human and animal samples. Predictions were also performed on cow spectra and human spectra, all of which were excluded from the training dataset to serve as an external validation step. The constructed model presents a superb ability to correctly identify human and animal blood traces to their specific species of origin. 2. Materials and Methods 2.1 Raman Microscope and Blood Samples A Renishaw inVia confocal Raman spectrometer with attached Leica microscope and a Renishaw PRIOR automatic stage were used for data collection for all blood species experiments. The instrument was calibrated with a silicon standard before all measurements. Spectra were accumulated with a 20x long-range objective with 785 nm excitation in the spectral range of 250-1800 cm-1. Laser power at the sample was approximately 4.0 mW. A Raman map consisting of 35 spectra were collected from each of the samples. WiRE software version 3.2 was used to operate the instrument.
4 ACS Paragon Plus Environment
Page 4 of 18
Page 5 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
Animal (non-human) blood from cow, cat, dog, horse, pig, mouse, opossum, raccoon, rabbit, rat and chicken were obtained from Bioreclamation, LLC. For each animal species, 2 mL of blood were procured from 10 individual donors. Animal blood was collected randomly from both genders to ensure donor diversity. Human blood was obtained from Bioreclamation, LLC without consideration to a particular gender, race or age. Blood samples were kept frozen until sample preparation. For each blood sample, approximately 30 µL was placed on an aluminum covered microscope slide and allowed to dry for at least 60 minutes. All Raman measurements were obtained within 48 hours of initial sample preparation. 2.2 Data Preparation and Statistical Treatment All data preparation and construction of statistical models were performed with the PLS Toolbox 7.0.3 (Eigenvector Research, Inc.) operating in MATLAB version R2010b. The spectra were truncated to the data range 252-1709 cm-1. For each sample, the 35 spectra were baseline corrected with a sixth-order polynomial and normalized by area. These 35 spectra from each donor were averaged to form a single spectrum representing one sample. The averaged spectra were imported into a data matrix. The dataset containing 110 total spectra was finally mean centered before models were calculated. All models were cross-validated using the Venetian blinds method24. All model predictions used the most probable class setting. 2.3 Internal and External Validation To evaluate the performance of the classification model, a set of ten samples were prepared from the available blood samples. Five human and five animal samples were arbitrarily chosen and assigned as unknown #1-10 (internal validation). To account for samples and species classes outside of the model, a set of ten cow spectra and four human spectra were prepared. The 5 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
human blood spectra were obtained from donors outside the training set donors (external validation). All samples for validation were prepared and analyzed in the same way as the blood used in the training dataset. 3. Results and Discussion The main objective of this study was to develop a method to identify the species of origin of blood samples through their Raman spectra. The model selection, laser power and data processing steps were selected based upon preliminary experimentation. A PLS-DA model was chosen to build a classification model using a training dataset containing 110 Raman spectra from human and the aforementioned animal samples (excluding cow). The PLS-DA model was constructed by classifying each spectrum as the specific species class to determine if a correct assignment could be made. The model was tested by attempting to classify ten internal samples (human and animal) and ten external samples (cow) which were not part of the training dataset. 3.1 Spectral Analysis of Training Dataset For each blood sample, a Raman spectral map of 35 points was collected. The 35 spectra within each map were preprocessed by baseline treatment, normalized by total area and averaged. The training dataset consisted of ten mean spectra from every species considered (excluding ten cow samples) for a total of 110 mean spectra. The eleven preprocessed averaged spectra for each species class analyzed are shown in Figure 1. The prominent features in a Raman spectrum of blood correspond to specific vibrational modes related to hemoglobin25,26, therefore extremely large differences would not be expected. From visual inspection, all spectra seem to be identical in terms of the number of spectral features and their location. This is not surprising because the components in the blood of all species vary but not significantly. 6 ACS Paragon Plus Environment
Page 6 of 18
Page 7 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
However, there does appear to be variations in the relative intensity of several overlapping bands, particularly in the region 1265-1200 cm-1 (Figure 1, shaded area), which corresponds to the stretching vibrational modes of hemoglobin25. It is possible that these variances are due to the interspecies differences in the hemoglobin structure. Nonetheless, these spectral variations are not significant enough to distinguish these classes visually. 3.2 Species Classification A PLS-DA model was constructed using the 110 averaged spectra obtained from the training dataset. Each mean spectrum in the training dataset was assigned to its specific class before building the model. This model was built with 14 latent variables (LVs) as opposed to the four LVs used in our previous binary classification study22. This was the minimum number of LVs that delivered satisfactory classification. For reference, loading variables 1-6 and a plot of the Hotelling T2 and Q Residual values from the PLS-DA model constructed are provided in Supplemental Figure 1 and 2, respectively. A two-dimensional (2D) scores plot of this model is displayed in Figure 2, showing the separation of the human class (purple triangles) which is best isolated from all animal classes using LVs #1 and #6. This plot also shows complete separation of the chicken class (grey squares) from the human class, which proved to be difficult in other models22. The PLS-DA model was used to predict the individual species assignment for each sample. The vast majority of spectra were correctly classified into their respective classes. The most distinct species were cat, horse, pig, rabbit, rat, mouse, human, chicken and opossum; there were no false positive or false negative assignments for these classes. The two species that were hardest to classify were dog and raccoon; where each of these classes contained one false positive and one false negative assignment. Interestingly, this model was able to effectively 7 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Page 8 of 18
distinguish chicken and human samples, as these caused errors in our previous study22. The cross-validated classification results for all classes are summarized in a confusion table (Table 1). This table shows the number of positively identified samples, along with a total percentage of true positive classifications for each individual species. Table 1. Internal cross-validated classification results for constructed species specific PLS-DA model. Actual class membership Predicted class Cat
Cat
Dog
Horse
Pig
Rabbit
Rat
Mouse
Opossum
Raccoon
Chicken
Human
10
0
0
0
0
0
0
0
0
0
0
Dog
0
9
0
0
0
0
0
0
1
0
0
Horse
0
0
10
0
0
0
0
0
0
0
0
Pig
0
0
0
10
0
0
0
0
0
0
0
Rabbit
0
0
0
0
10
0
0
0
0
0
0
Rat
0
0
0
0
0
10
0
0
0
0
0
Mouse
0
0
0
0
0
0
10
0
0
0
0
Opossum
0
0
0
0
0
0
0
10
0
0
0
Raccoon
0
1
0
0
0
0
0
0
9
0
0
Chicken
0
0
0
0
0
0
0
0
0
10
0
Human
0
0
0
0
0
0
0
0
0
0
10
Percent true positive
100
90
100
100
100
100
100
100
90
100
100
The performance of this model to identify human samples is most significant to the forensic community. There were no internal false positive or false negative assignments for this class. Likewise, all of the species classes had perfect classification results with the exception of raccoon and dog. These results demonstrate that this model works well and has real-world applicability, which is particularly important in a both a forensic and veterinary setting as well as for wildlife preservation. 3.3 Internal and External Validation Results To confirm the performance of the model, predictions were calculated for twenty-four spectra – ten spectra originating from ten cow blood samples and fourteen spectra originating 8 ACS Paragon Plus Environment
Page 9 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
from internal animal blood samples and four external human samples (Section 2.3). The ten unknown human and animal blood samples were used as an internal validation step to test the model’s repeatability. Spectra accumulated from cow and four additional human samples were excluded from the training dataset and thus serve as an external validation. The cow spectra were excluded to test how the model would classify samples outside of the training dataset. This is an important consideration since it is practically impossible to compile training data for every animal species. The four additional human samples (unknown 11-14) were excluded to test the model’s ability to accurately predict an untrained human sample. The prediction results for these twenty-four test spectra are displayed in Table 2. Table 2. Summary of model predictions from internal and external validation test spectra. Spectrum Unknown 1 Unknown 2 Unknown 3 Unknown 4 Unknown 5 Unknown 6 Unknown 7 Unknown 8 Unknown 9 Unknown 10 Unknown 11 Unknown 12 Unknown 13 Unknown 14 Cow 1 Cow 2 Cow 3 Cow 4 Cow 5 Cow 6 Cow 7 Cow 8 Cow 9
Predicted class Chicken Horse Human Rabbit Human Mouse Raccoon Human Human Human Human Human Human Human Rat Horse Horse Cat Pig Horse Horse Horse Horse
Actual Identity Chicken Horse Human Rabbit Human Mouse Raccoon Human Human Human Human Human Human Human Cow Cow Cow Cow Cow Cow Cow Cow Cow
9 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Cow 10
Pig
Page 10 of 18
Cow
All of the internal validation samples were classified correctly with no false positive assignments. Since the cow spectra were not included as part of the training dataset the “cow” class was not a valid option for the model to choose for predictions. In Figure 2 it can be seen that one cow blood spectrum is separated from all animal spectra and closer to the human spectra. Although most of the cow (external validation) samples were assigned to either the horse or pig class, none were predicted to be human. The internal and external validation spectra predictions for the human class are displayed in Figure 3. This plot shows the model scores for every spectrum in the training dataset and those to be predicted (unknown and cow samples). The unknown samples that were human (#3, #5, #8, #9, #10, #11, #12, #13, #14) and non-human (#1, #2, #4, #6, #7) were very well-resolved. These results demonstrate that it would be very unlikely for a cow spectrum to be misclassified as human, but very well could misclassify as to the specific animal species of origin. However, this result does not preclude the possibility of some unconsidered species misclassifying as human. More importantly, all four of the external human blood samples were correctly predicted to the human class. The characteristics of a PLS-DA model are defined by the shape of each component loading variable. Each loading variable contains numerous peaks throughout the fingerprint region of the Raman spectrum. This implies that there is not one characteristic peak within the Raman spectrum of blood that can be used to distinguish one class from another. It has been reported that blood composition of animals and humans is different; relating to concentrations of individual components (e.g. red blood cell count or total leucocytes) as well component structure and other properties27. We believe a possible explanation of the discriminatory ability of this approach is based on the subtle interspecies differences in the hemoglobin structure. For 10 ACS Paragon Plus Environment
Page 11 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
example, there are 18 amino acid substitutions in both the alpha and beta chains (36 total) of hemoglobin between horse and human28. This could alter the conformation of the protein thereby affecting its Raman spectrum. Following this hypothesis, Raman spectra of human and gorilla blood should be most similar since the total amino acid profile of hemoglobin differs in only three residues. Collectively, these biochemical interspecies differences in human and animal blood could account for the spectroscopic variances observed. The influence of these factors on the Raman spectrum of blood has not been extensively investigated and is beyond the scope of this study. 4. Conclusions The combination of Raman spectroscopy and chemometric models was demonstrated to be a powerful tool toward the identification of the species of a blood sample. A chemometric PLS-DA model was constructed using a training dataset formed from Raman spectral data collected from human blood and ten animal species’ blood. The model demonstrated excellent internal classification ability with all species, most notably the human class which had zero occurrences of false negative and false positive assignments. Furthermore, the model performed well under two performance measures – an internal test of repeatability and external validation where the model tested ten animal (cow) and four human spectra which were excluded from the initial training dataset. This approach was demonstrated to be effective at classifying intrinsically heterogeneous samples even when the discriminating biomarker or feature is unknown. The classification performance and the non-destructive nature of Raman spectroscopy make this approach well-suited for the species identification of an unknown blood sample, especially for forensic analysis. This is particularly important in cases where the sample is limited and destructive techniques of species identification cannot be afforded. Furthermore, the increasing 11 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
capabilities of on-field Raman analysis29 support the idea that rapid species identification is feasible. The method described conforms well to other Raman methods in the field, in that the same spectral data used to identify the sample as blood can then be used for a deeper determination of the species of origin.
Acknowledgements This project was supported by Award No. 2011–DN–BX–K551 awarded by the National Institute of Justice, Office of Justice Programs, U.S. Department of Justice (I.K.L.). The opinions, findings, and conclusions or recommendations expressed in this publication are those of the authors and do not necessarily reflect those of the U.S. Department of Justice.
ASSOCIATED CONTENT Supporting Information The information included here contains two figures to better explain some of the results discussed in section 3.2 (Species Classification). The first figure shows latent variables 1-6 which were used to build the model for species blood differentiation. The second figure shows scores of two metrics (Hotelling T2 and Q Residuals) for all samples used in the training dataset as well as for the external cow samples. This material is available free of charge via the Internet at http://pubs.acs.org.
12 ACS Paragon Plus Environment
Page 12 of 18
Page 13 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
References (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17)
(18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28)
(29)
Inouel, H.; Takabe, F.; Takenaka, O.; Iwasa, M.; Maeno, Y.; Int. J. Legal Med. 1990, 104, 9-12. Andrasko, J.; Rosén B. J. Forensic Sci. 1994, 39, 1018-25. Espinoza E. O.; Kirms M. A.; Filipek M. S. J. Forensic Sci. 1996, 41, 804-11. Espinoza, E. O., Lindley N. C.; Gordon K. M.; Ekhoff J. A.; Kirms M. A. Anal. Biochem. 1999, 268, 252-61. Steendam, K.; De Ceuleneer, M.; Dhaenens, M.; Van Hoofstat, D.; Deforce, D. Int. J. Legal Med. 2013, 127, 287-298. Gebel, E. Anal. Chem. 2009, 81, 7862. Virkler, K.; Lednev, I. K. Forensic Sci. Int. 2008, 181, e1–e5. Virkler, K.; Lednev, I. K. Anal. Bioanal. Chem. 2010, 396, 525-34. Sikirzhytskaya, A.; Sikirzhytski, V.; Lednev, I. K. J. Biophotonics 2014, 7, 59-67. Virkler, K.; Lednev, I. K. Forensic Sci. Int. 2009, 193, 56-62. Virkler, K.; Lednev, I. K. Analyst 2010, 135, 512-517. McLaughlin, G.; Sikirzhytski, V.; Lednev, I. K. Forensic Sci. Int. 2013, 231, 157-166. Sikirzhytski, V.; Sikirzhytskaya, A.; Lednev, I. K. Anal. Chim. Acta 2012, 718, 78-83. Sikirzhytski, V.; Sikirzhytskaya, A.; Lednev, I. K. Appl. Spectrosc. 2011, 65, 1223-1232. Sikirzhytskaya, A.; Sikirzhytski, V.; Lednev, I. K. Forensic Sci. Int. 2012, 216, 44-48. Virkler, K.; Lednev, I. K. Forensic Sci. Int. 2009, 188, 1-17. Nafie, L.A. In Handbook of Raman spectroscopy: From the Research Laboratory to the Process Line; Lewis, I. R.; Edwards, H. G. M., Ed.; Marcel Dekker: New York, 2001; pp 1-10. De Wael, K.; Lepot, L.; Gason, F.; Gilbert, B. Forensic Sci. Int. 2008, 180, 37–42. Virkler, K.; Lednev, I. K. Anal. Chem. 2009, 81, 7773-7777. McLaughlin, G.; Lednev, I. K. Am. J. Anal. Chem. 2012, 3, 161-167. Sikirzhytski, V.; Virkler, K.; Lednev, I. K. Sensors 2010, 10, 2869-2884. McLaughlin, G.; Doty, K. C.; Lednev, I. K. Forensic Sci. Int. 2014, 238, 91-95. Varmuza, K.; Filzmoser, P. Introduction to Multivariate Statistical Analysis in Chemometrics; CRC Press: Boca Raton, 2009; pp 195-207. Wise, B. M.; Gallagher, N. B.; Bro, R.; Shaver, J. M.; Windig, W.; Koch, R. S. PLS Toolbox; Eigenvector Research, Inc.: Wenatchee, 2005. Premasiri, W. R.; Lee, J. C.; Ziegler, L. D. J. Phys. Chem. B 2012, 116, 9376-9386. Lemler, P.; Premasiri, W. R.; DelMonaco, A.; Ziegler, L. D. Anal Bioanal Chem 2014, 406, 193. Altman, P. L. Blood and Other Body Fluids; Federation of American Societies for Experimental Biology: Washington, D.C., 1961; pp 109-118. Zuckerkandl, E.; Pauling, L. In Linus Pauling: Selected Scientific Papers. Biomolecular Sciences, Vol. 2; Kamb, B., Pauling Kamb L., Pauling, P.J., Kamb, A., and Pauling Jr., L., Ed.; World Scientific: River Edge, 2001; pp 1305-1306. Izake, E. L. Forensic Sci. Int. 2010, 202, 1-8.
13 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Figure captions Figure 1. Comparison of the preprocessed averaged Raman spectra from the 12 species analyzed. Each trace is the mean spectrum (after baselining and normalizing by area) for the particular species noted. Figure 2. A 2D scores plot showing separation of the human class from all other animal classes based on LVs #1 and #6. Green dotted line is demonstrating human class separation. Figure 3. Prediction scores for human class using the species specific PLS-DA model with cow (red triangles), internal validation (red stars), and external human (green circles) spectra loaded as predictions. Red dotted line represents the default classification threshold.
14 ACS Paragon Plus Environment
Page 14 of 18
Page 15 of 18
Rat
Raccoon
Rabbit
Pig
Relative Intensity (arbitrary units)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44
Analytical Chemistry
Opossum
Mouse
Horse
Dog
Cow
Chicken
Cat
Human 1600
1400
1200 1000 800 Raman shift (cm-1)
600
400
Figure 1. ACS Paragon Plus Environment
Analytical Chemistry
-3
3 x 10
Scores on LV 6 (2.48%)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
Page 16 of 18
Cat Dog Horse Pig Rabbit Rat Mouse Opossum Raccoon Chicken Human Cow (external) Internal unknown External human
#14 #12 #3 #5
#4
#10 #9
#8
#13 #11 #7
0 #6
#2
#1
-3
-3 x 10 -0.015
0
Scores on LV 1 (50.25%)
Figure 2. ACS Paragon Plus Environment
0.015
Page 17 of 18
1.2
0.8
Predicted Class Human
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
Analytical Chemistry
Cat Dog Horse Pig Rabbit Rat Mouse Opossum Raccoon Chicken Human Cow Internal unknown External human
#10 #9
#5
#8 #3
#14 #13 #12
#11
0.4
#7 #6 #4
0
-0.4
#2 #1
20
40
60
Sample
80
100
Figure 3. ACS Paragon Plus Environment
120
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Page 18 of 18
Pig Dog Raccoon Rat Cow
Rabbit Cat Opossum Human Horse Chicken Mouse
ACS Paragon Plus Environment
Article
Raman Spectroscopy of Blood for Species Identification Gregory McLaughlin, Kyle C. Doty, and Igor K. Lednev Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/ac5026368 • Publication Date (Web): 28 Oct 2014 Downloaded from http://pubs.acs.org on November 2, 2014
Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.
Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.
Page 1 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
Raman Spectroscopy of Blood for Species Identification Gregory McLaughlin†, Kyle C. Doty†, and Igor K. Lednev* Department of Chemistry, University at Albany, 1400 Washington Avenue, Albany, NY 12222, United States †
Both authors contributed equally to this work
*
Corresponding author at: Department of Chemistry, University at Albany, 1400 Washington
Avenue, Albany, NY 12222, United States. Tel: 1-518-591-8863; fax: 1-518-442-3462 E–mail address: [email protected] (I.K. Lednev) Keywords: Raman spectroscopy, Chemometrics, Blood, Serology, Human origin, Species identification Abstract The species identification of a blood stain is an important and immediate challenge for forensic science, veterinary purposes, and wildlife preservation. The current methods used to identify the species of origin of a blood stain are limited in scope and destructive to the sample. We have previously demonstrated that Raman spectroscopy can reliably differentiate blood traces of human, cat and dog (Virkler et al. Anal. Chem. 2009) and, most recently, built a binary model for differentiating human vs. animal blood for eleven species integrated with human existence (McLaughlin et al. Forensic Sci. Int. 2014). Here we report a satisfactory classification of blood obtained from eleven animal classes and human subjects by statistical analysis of Raman spectra. Classification of blood samples was achieved according to each sample’s species of origin which enhanced previously observed discrimination ability. The developed approach
1 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
does not require the knowledge of a specific (bio)chemical marker for each individual class, but rather relies on a spectroscopic statistical differentiation of various components. This approach results in remarkable classification ability even with intrinsically heterogeneous classes and samples. In addition, the obtained spectroscopic characteristics could potentially provide information about specific changes in the (bio)chemical composition of samples, which are responsible for the differentiation. 1. Introduction Interspecies blood analysis is a well-researched topic in analytical chemistry, forensic science and biochemistry. Applying analytical techniques to identify the species of a blood stain is particularly of interest to the fields of forensic casework, veterinary science, and wildlife preservation. Currently, species identification can be accomplished via a wide array of analytical techniques. The analysis of blood by high-performance liquid chromatography (HPLC) and mass spectrometry (MS) is an emerging analytical technique for species identification. Several HPLC methods have been developed which can simultaneously identify a substance as blood and determine its species of origin. One study incorporated this technique using reverse-phase HPLC to analyze fresh blood and bloodstains from humans and 28 animal species1. The sensitivity of this approach was increased using ultraviolet excitation with fluorescence detection2. Espinoza et al. expanded this method further by analyzing and quantitating the concentrations of blood in bloodstains and blood mixtures from over 50 different animal species3. Also, Espinoza et al. demonstrated that species differentiation can be substantiated with mass spectrometry4. This technique detected minor interspecies molecular mass differences in α- and β-chains (α/β-pairs) of hemoglobin from 62 different species. Most recently, a reverse-phase LC-MS method was utilized to differentiate human blood from dog and cow blood5. In that study, Steendam et al. 2 ACS Paragon Plus Environment
Page 2 of 18
Page 3 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
demonstrated the use of a single test to determine the identity and species of a biological sample. For most of these studies, species distinction was accomplished by targeting differences in hemoglobin in a destructive manner. Vibrational spectroscopy stands apart from other analytical techniques due to its nondestructive nature and the limited sample preparation required. A spectroscopic approach is therefore desirable in forensic applications, or where the sample size is in trace amounts6. Raman spectroscopy is a type of vibrational spectroscopy which also has the potential as a nondestructive confirmatory identification method for blood7-9 and other body fluids7,10-16. This technique uses a monochromatic light source to irradiate a sample. Spectra are generated by detecting inelastically scattered light by the sample. A Raman spectrum contains numerous distinctive bands which correspond to specific molecular vibrational modes17. Vibrational spectroscopy has been explored for the purposes of species identification of blood samples, initially with a study conducted by De Wael et al.18 This group reported that blood particles originating from human, cat and dog samples could not be visually differentiated through their Raman or infrared spectra. Using a novel approach, Virkler et al. demonstrated that differentiation of human, cat and dog blood samples is possible using Principal Component Analysis (PCA) modeling of Raman spectroscopic data19. Separation between these classes within a PCA model exceeded a 99% confidence interval. The application of statistical models to Raman spectra enhances the selectivity of Raman spectroscopy and has been used extensively to build similar classification models20,21. Most recently we demonstrated that this approach could be expanded to as many as eleven animals and built a binary model for the discrimination of human and non-human blood samples22. Additionally, for that study, all unknown samples were classified correctly (either human or non-human). 3 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Here, we analyzed blood from a wide survey of species (12 in total). The animal (nonhuman) samples were selected to represent three groups: animals that are bred for domestication (cat, dog, horse and cow), those that are consumed as food (chicken, cow, pig, and rabbit), and those that are integrated with human existence (mouse, rat, opossum and raccoon). To account for the complexity of the dataset, a Partial Least Squares-Discriminant Analysis (PLS-DA) classification model was used. These models have more class discrimination power than PCA and provide automatic soft prediction of classes23. The PLS-DA classification model was used to differentiate a training dataset of human and animal blood spectra. As an internal validation step, predictions were made for a set of human and animal samples. Predictions were also performed on cow spectra and human spectra, all of which were excluded from the training dataset to serve as an external validation step. The constructed model presents a superb ability to correctly identify human and animal blood traces to their specific species of origin. 2. Materials and Methods 2.1 Raman Microscope and Blood Samples A Renishaw inVia confocal Raman spectrometer with attached Leica microscope and a Renishaw PRIOR automatic stage were used for data collection for all blood species experiments. The instrument was calibrated with a silicon standard before all measurements. Spectra were accumulated with a 20x long-range objective with 785 nm excitation in the spectral range of 250-1800 cm-1. Laser power at the sample was approximately 4.0 mW. A Raman map consisting of 35 spectra were collected from each of the samples. WiRE software version 3.2 was used to operate the instrument.
4 ACS Paragon Plus Environment
Page 4 of 18
Page 5 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
Animal (non-human) blood from cow, cat, dog, horse, pig, mouse, opossum, raccoon, rabbit, rat and chicken were obtained from Bioreclamation, LLC. For each animal species, 2 mL of blood were procured from 10 individual donors. Animal blood was collected randomly from both genders to ensure donor diversity. Human blood was obtained from Bioreclamation, LLC without consideration to a particular gender, race or age. Blood samples were kept frozen until sample preparation. For each blood sample, approximately 30 µL was placed on an aluminum covered microscope slide and allowed to dry for at least 60 minutes. All Raman measurements were obtained within 48 hours of initial sample preparation. 2.2 Data Preparation and Statistical Treatment All data preparation and construction of statistical models were performed with the PLS Toolbox 7.0.3 (Eigenvector Research, Inc.) operating in MATLAB version R2010b. The spectra were truncated to the data range 252-1709 cm-1. For each sample, the 35 spectra were baseline corrected with a sixth-order polynomial and normalized by area. These 35 spectra from each donor were averaged to form a single spectrum representing one sample. The averaged spectra were imported into a data matrix. The dataset containing 110 total spectra was finally mean centered before models were calculated. All models were cross-validated using the Venetian blinds method24. All model predictions used the most probable class setting. 2.3 Internal and External Validation To evaluate the performance of the classification model, a set of ten samples were prepared from the available blood samples. Five human and five animal samples were arbitrarily chosen and assigned as unknown #1-10 (internal validation). To account for samples and species classes outside of the model, a set of ten cow spectra and four human spectra were prepared. The 5 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
human blood spectra were obtained from donors outside the training set donors (external validation). All samples for validation were prepared and analyzed in the same way as the blood used in the training dataset. 3. Results and Discussion The main objective of this study was to develop a method to identify the species of origin of blood samples through their Raman spectra. The model selection, laser power and data processing steps were selected based upon preliminary experimentation. A PLS-DA model was chosen to build a classification model using a training dataset containing 110 Raman spectra from human and the aforementioned animal samples (excluding cow). The PLS-DA model was constructed by classifying each spectrum as the specific species class to determine if a correct assignment could be made. The model was tested by attempting to classify ten internal samples (human and animal) and ten external samples (cow) which were not part of the training dataset. 3.1 Spectral Analysis of Training Dataset For each blood sample, a Raman spectral map of 35 points was collected. The 35 spectra within each map were preprocessed by baseline treatment, normalized by total area and averaged. The training dataset consisted of ten mean spectra from every species considered (excluding ten cow samples) for a total of 110 mean spectra. The eleven preprocessed averaged spectra for each species class analyzed are shown in Figure 1. The prominent features in a Raman spectrum of blood correspond to specific vibrational modes related to hemoglobin25,26, therefore extremely large differences would not be expected. From visual inspection, all spectra seem to be identical in terms of the number of spectral features and their location. This is not surprising because the components in the blood of all species vary but not significantly. 6 ACS Paragon Plus Environment
Page 6 of 18
Page 7 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
However, there does appear to be variations in the relative intensity of several overlapping bands, particularly in the region 1265-1200 cm-1 (Figure 1, shaded area), which corresponds to the stretching vibrational modes of hemoglobin25. It is possible that these variances are due to the interspecies differences in the hemoglobin structure. Nonetheless, these spectral variations are not significant enough to distinguish these classes visually. 3.2 Species Classification A PLS-DA model was constructed using the 110 averaged spectra obtained from the training dataset. Each mean spectrum in the training dataset was assigned to its specific class before building the model. This model was built with 14 latent variables (LVs) as opposed to the four LVs used in our previous binary classification study22. This was the minimum number of LVs that delivered satisfactory classification. For reference, loading variables 1-6 and a plot of the Hotelling T2 and Q Residual values from the PLS-DA model constructed are provided in Supplemental Figure 1 and 2, respectively. A two-dimensional (2D) scores plot of this model is displayed in Figure 2, showing the separation of the human class (purple triangles) which is best isolated from all animal classes using LVs #1 and #6. This plot also shows complete separation of the chicken class (grey squares) from the human class, which proved to be difficult in other models22. The PLS-DA model was used to predict the individual species assignment for each sample. The vast majority of spectra were correctly classified into their respective classes. The most distinct species were cat, horse, pig, rabbit, rat, mouse, human, chicken and opossum; there were no false positive or false negative assignments for these classes. The two species that were hardest to classify were dog and raccoon; where each of these classes contained one false positive and one false negative assignment. Interestingly, this model was able to effectively 7 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Page 8 of 18
distinguish chicken and human samples, as these caused errors in our previous study22. The cross-validated classification results for all classes are summarized in a confusion table (Table 1). This table shows the number of positively identified samples, along with a total percentage of true positive classifications for each individual species. Table 1. Internal cross-validated classification results for constructed species specific PLS-DA model. Actual class membership Predicted class Cat
Cat
Dog
Horse
Pig
Rabbit
Rat
Mouse
Opossum
Raccoon
Chicken
Human
10
0
0
0
0
0
0
0
0
0
0
Dog
0
9
0
0
0
0
0
0
1
0
0
Horse
0
0
10
0
0
0
0
0
0
0
0
Pig
0
0
0
10
0
0
0
0
0
0
0
Rabbit
0
0
0
0
10
0
0
0
0
0
0
Rat
0
0
0
0
0
10
0
0
0
0
0
Mouse
0
0
0
0
0
0
10
0
0
0
0
Opossum
0
0
0
0
0
0
0
10
0
0
0
Raccoon
0
1
0
0
0
0
0
0
9
0
0
Chicken
0
0
0
0
0
0
0
0
0
10
0
Human
0
0
0
0
0
0
0
0
0
0
10
Percent true positive
100
90
100
100
100
100
100
100
90
100
100
The performance of this model to identify human samples is most significant to the forensic community. There were no internal false positive or false negative assignments for this class. Likewise, all of the species classes had perfect classification results with the exception of raccoon and dog. These results demonstrate that this model works well and has real-world applicability, which is particularly important in a both a forensic and veterinary setting as well as for wildlife preservation. 3.3 Internal and External Validation Results To confirm the performance of the model, predictions were calculated for twenty-four spectra – ten spectra originating from ten cow blood samples and fourteen spectra originating 8 ACS Paragon Plus Environment
Page 9 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
from internal animal blood samples and four external human samples (Section 2.3). The ten unknown human and animal blood samples were used as an internal validation step to test the model’s repeatability. Spectra accumulated from cow and four additional human samples were excluded from the training dataset and thus serve as an external validation. The cow spectra were excluded to test how the model would classify samples outside of the training dataset. This is an important consideration since it is practically impossible to compile training data for every animal species. The four additional human samples (unknown 11-14) were excluded to test the model’s ability to accurately predict an untrained human sample. The prediction results for these twenty-four test spectra are displayed in Table 2. Table 2. Summary of model predictions from internal and external validation test spectra. Spectrum Unknown 1 Unknown 2 Unknown 3 Unknown 4 Unknown 5 Unknown 6 Unknown 7 Unknown 8 Unknown 9 Unknown 10 Unknown 11 Unknown 12 Unknown 13 Unknown 14 Cow 1 Cow 2 Cow 3 Cow 4 Cow 5 Cow 6 Cow 7 Cow 8 Cow 9
Predicted class Chicken Horse Human Rabbit Human Mouse Raccoon Human Human Human Human Human Human Human Rat Horse Horse Cat Pig Horse Horse Horse Horse
Actual Identity Chicken Horse Human Rabbit Human Mouse Raccoon Human Human Human Human Human Human Human Cow Cow Cow Cow Cow Cow Cow Cow Cow
9 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Cow 10
Pig
Page 10 of 18
Cow
All of the internal validation samples were classified correctly with no false positive assignments. Since the cow spectra were not included as part of the training dataset the “cow” class was not a valid option for the model to choose for predictions. In Figure 2 it can be seen that one cow blood spectrum is separated from all animal spectra and closer to the human spectra. Although most of the cow (external validation) samples were assigned to either the horse or pig class, none were predicted to be human. The internal and external validation spectra predictions for the human class are displayed in Figure 3. This plot shows the model scores for every spectrum in the training dataset and those to be predicted (unknown and cow samples). The unknown samples that were human (#3, #5, #8, #9, #10, #11, #12, #13, #14) and non-human (#1, #2, #4, #6, #7) were very well-resolved. These results demonstrate that it would be very unlikely for a cow spectrum to be misclassified as human, but very well could misclassify as to the specific animal species of origin. However, this result does not preclude the possibility of some unconsidered species misclassifying as human. More importantly, all four of the external human blood samples were correctly predicted to the human class. The characteristics of a PLS-DA model are defined by the shape of each component loading variable. Each loading variable contains numerous peaks throughout the fingerprint region of the Raman spectrum. This implies that there is not one characteristic peak within the Raman spectrum of blood that can be used to distinguish one class from another. It has been reported that blood composition of animals and humans is different; relating to concentrations of individual components (e.g. red blood cell count or total leucocytes) as well component structure and other properties27. We believe a possible explanation of the discriminatory ability of this approach is based on the subtle interspecies differences in the hemoglobin structure. For 10 ACS Paragon Plus Environment
Page 11 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
example, there are 18 amino acid substitutions in both the alpha and beta chains (36 total) of hemoglobin between horse and human28. This could alter the conformation of the protein thereby affecting its Raman spectrum. Following this hypothesis, Raman spectra of human and gorilla blood should be most similar since the total amino acid profile of hemoglobin differs in only three residues. Collectively, these biochemical interspecies differences in human and animal blood could account for the spectroscopic variances observed. The influence of these factors on the Raman spectrum of blood has not been extensively investigated and is beyond the scope of this study. 4. Conclusions The combination of Raman spectroscopy and chemometric models was demonstrated to be a powerful tool toward the identification of the species of a blood sample. A chemometric PLS-DA model was constructed using a training dataset formed from Raman spectral data collected from human blood and ten animal species’ blood. The model demonstrated excellent internal classification ability with all species, most notably the human class which had zero occurrences of false negative and false positive assignments. Furthermore, the model performed well under two performance measures – an internal test of repeatability and external validation where the model tested ten animal (cow) and four human spectra which were excluded from the initial training dataset. This approach was demonstrated to be effective at classifying intrinsically heterogeneous samples even when the discriminating biomarker or feature is unknown. The classification performance and the non-destructive nature of Raman spectroscopy make this approach well-suited for the species identification of an unknown blood sample, especially for forensic analysis. This is particularly important in cases where the sample is limited and destructive techniques of species identification cannot be afforded. Furthermore, the increasing 11 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
capabilities of on-field Raman analysis29 support the idea that rapid species identification is feasible. The method described conforms well to other Raman methods in the field, in that the same spectral data used to identify the sample as blood can then be used for a deeper determination of the species of origin.
Acknowledgements This project was supported by Award No. 2011–DN–BX–K551 awarded by the National Institute of Justice, Office of Justice Programs, U.S. Department of Justice (I.K.L.). The opinions, findings, and conclusions or recommendations expressed in this publication are those of the authors and do not necessarily reflect those of the U.S. Department of Justice.
ASSOCIATED CONTENT Supporting Information The information included here contains two figures to better explain some of the results discussed in section 3.2 (Species Classification). The first figure shows latent variables 1-6 which were used to build the model for species blood differentiation. The second figure shows scores of two metrics (Hotelling T2 and Q Residuals) for all samples used in the training dataset as well as for the external cow samples. This material is available free of charge via the Internet at http://pubs.acs.org.
12 ACS Paragon Plus Environment
Page 12 of 18
Page 13 of 18
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Analytical Chemistry
References (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (16) (17)
(18) (19) (20) (21) (22) (23) (24) (25) (26) (27) (28)
(29)
Inouel, H.; Takabe, F.; Takenaka, O.; Iwasa, M.; Maeno, Y.; Int. J. Legal Med. 1990, 104, 9-12. Andrasko, J.; Rosén B. J. Forensic Sci. 1994, 39, 1018-25. Espinoza E. O.; Kirms M. A.; Filipek M. S. J. Forensic Sci. 1996, 41, 804-11. Espinoza, E. O., Lindley N. C.; Gordon K. M.; Ekhoff J. A.; Kirms M. A. Anal. Biochem. 1999, 268, 252-61. Steendam, K.; De Ceuleneer, M.; Dhaenens, M.; Van Hoofstat, D.; Deforce, D. Int. J. Legal Med. 2013, 127, 287-298. Gebel, E. Anal. Chem. 2009, 81, 7862. Virkler, K.; Lednev, I. K. Forensic Sci. Int. 2008, 181, e1–e5. Virkler, K.; Lednev, I. K. Anal. Bioanal. Chem. 2010, 396, 525-34. Sikirzhytskaya, A.; Sikirzhytski, V.; Lednev, I. K. J. Biophotonics 2014, 7, 59-67. Virkler, K.; Lednev, I. K. Forensic Sci. Int. 2009, 193, 56-62. Virkler, K.; Lednev, I. K. Analyst 2010, 135, 512-517. McLaughlin, G.; Sikirzhytski, V.; Lednev, I. K. Forensic Sci. Int. 2013, 231, 157-166. Sikirzhytski, V.; Sikirzhytskaya, A.; Lednev, I. K. Anal. Chim. Acta 2012, 718, 78-83. Sikirzhytski, V.; Sikirzhytskaya, A.; Lednev, I. K. Appl. Spectrosc. 2011, 65, 1223-1232. Sikirzhytskaya, A.; Sikirzhytski, V.; Lednev, I. K. Forensic Sci. Int. 2012, 216, 44-48. Virkler, K.; Lednev, I. K. Forensic Sci. Int. 2009, 188, 1-17. Nafie, L.A. In Handbook of Raman spectroscopy: From the Research Laboratory to the Process Line; Lewis, I. R.; Edwards, H. G. M., Ed.; Marcel Dekker: New York, 2001; pp 1-10. De Wael, K.; Lepot, L.; Gason, F.; Gilbert, B. Forensic Sci. Int. 2008, 180, 37–42. Virkler, K.; Lednev, I. K. Anal. Chem. 2009, 81, 7773-7777. McLaughlin, G.; Lednev, I. K. Am. J. Anal. Chem. 2012, 3, 161-167. Sikirzhytski, V.; Virkler, K.; Lednev, I. K. Sensors 2010, 10, 2869-2884. McLaughlin, G.; Doty, K. C.; Lednev, I. K. Forensic Sci. Int. 2014, 238, 91-95. Varmuza, K.; Filzmoser, P. Introduction to Multivariate Statistical Analysis in Chemometrics; CRC Press: Boca Raton, 2009; pp 195-207. Wise, B. M.; Gallagher, N. B.; Bro, R.; Shaver, J. M.; Windig, W.; Koch, R. S. PLS Toolbox; Eigenvector Research, Inc.: Wenatchee, 2005. Premasiri, W. R.; Lee, J. C.; Ziegler, L. D. J. Phys. Chem. B 2012, 116, 9376-9386. Lemler, P.; Premasiri, W. R.; DelMonaco, A.; Ziegler, L. D. Anal Bioanal Chem 2014, 406, 193. Altman, P. L. Blood and Other Body Fluids; Federation of American Societies for Experimental Biology: Washington, D.C., 1961; pp 109-118. Zuckerkandl, E.; Pauling, L. In Linus Pauling: Selected Scientific Papers. Biomolecular Sciences, Vol. 2; Kamb, B., Pauling Kamb L., Pauling, P.J., Kamb, A., and Pauling Jr., L., Ed.; World Scientific: River Edge, 2001; pp 1305-1306. Izake, E. L. Forensic Sci. Int. 2010, 202, 1-8.
13 ACS Paragon Plus Environment
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60
Figure captions Figure 1. Comparison of the preprocessed averaged Raman spectra from the 12 species analyzed. Each trace is the mean spectrum (after baselining and normalizing by area) for the particular species noted. Figure 2. A 2D scores plot showing separation of the human class from all other animal classes based on LVs #1 and #6. Green dotted line is demonstrating human class separation. Figure 3. Prediction scores for human class using the species specific PLS-DA model with cow (red triangles), internal validation (red stars), and external human (green circles) spectra loaded as predictions. Red dotted line represents the default classification threshold.
14 ACS Paragon Plus Environment
Page 14 of 18
Page 15 of 18
Rat
Raccoon
Rabbit
Pig
Relative Intensity (arbitrary units)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44
Analytical Chemistry
Opossum
Mouse
Horse
Dog
Cow
Chicken
Cat
Human 1600
1400
1200 1000 800 Raman shift (cm-1)
600
400
Figure 1. ACS Paragon Plus Environment
Analytical Chemistry
-3
3 x 10
Scores on LV 6 (2.48%)
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
Page 16 of 18
Cat Dog Horse Pig Rabbit Rat Mouse Opossum Raccoon Chicken Human Cow (external) Internal unknown External human
#14 #12 #3 #5
#4
#10 #9
#8
#13 #11 #7
0 #6
#2
#1
-3
-3 x 10 -0.015
0
Scores on LV 1 (50.25%)
Figure 2. ACS Paragon Plus Environment
0.015
Page 17 of 18
1.2
0.8
Predicted Class Human
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
Analytical Chemistry
Cat Dog Horse Pig Rabbit Rat Mouse Opossum Raccoon Chicken Human Cow Internal unknown External human
#10 #9
#5
#8 #3
#14 #13 #12
#11
0.4
#7 #6 #4
0
-0.4
#2 #1
20
40
60
Sample
80
100
Figure 3. ACS Paragon Plus Environment
120
Analytical Chemistry
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26
Page 18 of 18
Pig Dog Raccoon Rat Cow
Rabbit Cat Opossum Human Horse Chicken Mouse
ACS Paragon Plus Environment
