A head for identification

There is a need for techniques that can assist crime scene investigators in narrowing down the number of suspects when DNA analysis is not possible or does not provide any match – one team from Canada think that hair analysis can give useful information of gender and ethnicity

The analysis of deoxyribonucleic acid (DNA) is a powerful analytical method for the identification of individuals; but it can only lead to identification if the DNA in question is included in a database. Furthermore, DNA analysis is not always possible. For instance, some organic materials, such as urine or sweat, can degrade over a small period of time. And fresh samples of such materials only provide information on what the person consumed over the past few days at the most. On the other hand, some other organic materials, such as fingernails and hair, are stable. Hair in particular remains unchanged for years and can provide information on what a person has consumed or was exposed to over months or even years, depending on their length. Fingernails and hair have been used in the medical field to help diagnose patients.  Inductively coupled plasma (ICP) mass spectrometry (MS) is an example of a technique that can be used to detect high aluminum concentration in hair, which can be indicative of pre-senile dementia or renal problems¹.

There is a need for techniques that can assist crime scene investigators in narrowing down the number of suspects when DNA analysis is not possible or does not provide any match. This article describes one technique that shows great potential for extracting information from head hair: electrothermal vaporisation (ETV) coupled to ICP optical emission spectrometry (OES). The sample is simply weighed into a graphite boat, which is then inserted into the ETV graphite furnace. The latter is resistively heated following a temperature program that typically involves four steps: desolvation (for drying the sample), pyrolysis (to remove as much of the matrix as possible), vaporisation, and cleaning (removal of any residue)². When a solid is analysed, the drying step may be omitted, which further shortens the analysis. For the analysis of hair, an 85-second temperature program, including 30 seconds for the vaporization step, is sufficient. During the vaporisation step, the vapour, along with particulates formed through condensation on the way to the ICP, is carried into the ICP by a flow of argon. A reaction gas, dichlorodifluoromethane (CCl₂F₂) in this case, is often used to transform elements into volatile fluorides or chlorides so as to maximise the fraction of analytes (i.e. elements being determined) reaching the ICP.

The temperature in the argon ICP, which consists in partially ionised argon, ranges from 6000 to 10000 K and is sufficient to not only atomise vapours and particulates but also ionize the resulting atoms, and excite atoms and ions, which ultimately emit at characteristic wavelengths upon relaxing to their ground states. The emission spectrum thus constitutes a fingerprint of the elements that were present in the sample and the intensity of analyte emission is proportional to analyte concentration. Although ICP-OES was developed for the analysis of solutions, coupling ETV to it drastically increases the range of samples that can be analysed, as it enables the analysis of slurries, viscous samples and solids. The possibility of directly analysing solids presents some significant advantages, as it drastically decreases sample preparation, which may be limited to grinding, and thereby minimises contamination. This eliminates the need for time-consuming digestion or fusion methods that also require expensive reagents, must be carried out carefully and are prone to contamination or analyte loss.

However, using ETV complicates data processing.  Indeed, a transient signal, which only lasts a few seconds, results rather than the steady-state signal observed during the continuous nebulisation of solutions, which can simply be integrated during a much longer time.  Furthermore, because the introduction of solid sample has a visible effect on the plasma, which depends on the amount of sample, internal standardisation must first be carried out using an argon emission line (such as that at 763.511 nm, which is sensitive and stable) to compensate for sample loading effects.  This approach, which does not require any addition to samples, simply consists in calculating the point-by-point analyte-over-argon intensity ratio.  Before integrating the area under the resulting analyte transient signal (i.e. peak), blank subtraction must also be done using empty graphite boats to remove contributions from potential contaminants in the graphite.

 For the analysis of head hair, each sample is washed with three 30 mL aliquots each of water and hexane to remove exogenous contamination. It is then dried, cut, and grinded into a fine powder.  For analysis, 4-mg aliquots are weighed onto graphite boats, which are then placed on the turntable of the autosampler, where automated tweezers then pick up a boat and insert it into the ETV furnace. At the end of the temperature program, these tweezers remove the boat and insert the next one. While quantification of elements can be done by external calibration with increasing amounts of hair certified reference material, such calibration is unnecessary, as multivariate statistical tests looking at the relative proportions of elements in the sample are sufficient to ascribe gender and general ethnicity.

Two multivariate statistical tests were considered: principal component analysis (PCA) and linear discriminant analysis (LDA). The former looks for the greatest variation between the linear combination of multiple variables to match samples together. For example, it has been used to classify different narcotics (cocaine, heroin, and ecstasy) at different purity levels by looking at complex variations in their Raman spectra⁴. By comparing the two principal components and finding the greatest variations between the factors, it is capable of discrimination and grouping, using positive and negative correlations of the elements between individuals.  PCA has also been used in forensic analysis for infrared spectra recognition and for 3-dimensional spatial recognition in methods looking at skull and facial features.  Although PCA provides a nice visual representation of how samples group together, it was not capable of accurate classification of red automotive paint².

In contrast to PCA, LDA classifies the variables based on an output linear function, which acts as a “rule” to indicate how closely each subject follows that function. The basis of LDA involves building a training data set or ground truth based on known group memberships to provide a combined linear function. Each unknown sample can then be used to “test” to what extent the function is close to it (i.e. the probability of a match)⁵. With cross validation, LDA can even estimate misclassification probabilities, thereby testing the efficiency of the program.  However, this option may not be possible if the database is too large for input (i.e. crashes the software).  Nonetheless, the advantage of LDA versus other common discriminant analysis techniques, such as the quadratic one, lies in the assumption of equal levels of variance among the predictors (i.e. the peak areas obtained by ETV-ICP-OES for each element), which can be verified by PCA. As long as the variance-covariance matrices built between the subjects and elements are homogenous (i.e. are the same throughout), then simple classification is sufficient. This also allows large data computations in which multiple dependents (such as the different elements) can be assessed simultaneously.

A correlation of hair to biological gender and generalised ethnicities comes from the genetic make-up and how the body produces components to make hair. A large amount of elements are absorbed endogenously by the human hair from sweat glands, which release salts and minerals as the hair is growing, and such release is dependent on a person’s gender, ethnicity, as well as on the person’s diet and environment³. Hence, to isolate this information, an adequate cleaning step must be used to strip the hair samples from exogenously absorbed elements. Going through each element’s LDA output for each categorisation (gender and ethnicity) and carefully selecting elements with high correlation may also help minimise the influence from external factors.

In this work, 13 hair samples – 12 of which were collected in Kingston, Ontario, Canada – were used to build the model. By going through each element’s LDA, one group of elements (Li, Mo, S, Sr, Cr, K, Ni, Zn, and Pb) – when applied to duplicate samples – correctly identified all 13 samples by general ethnicity. Another group of elements (Mg, S, Sr, and Zn – three of which are common with the elements used for predicting general ethnicity) allowed the correct identification of all 13 samples for gender. The smaller number of elements required in the latter case is commensurate with there being only two biological genders, whereas there are many ethnicities.  To further verify the validity of the model, it was applied to two blind tests, i.e. hair from individuals leaving in cities located 430 km west and 270 km east of Kingston.  Both were correctly categorised in terms of both gender and general ethnicity, despite the different environment where these individuals live³.

These surprisingly impressive results, given the little number of samples used to build the model, are very encouraging. More work will thus be done to improve the method by refining the ethnicity (such as categorising Chinese, Japanese, etc. instead of just East-Asian). As well, means to improve its reproducibility will be investigated.  Indeed, one duplicate sample had only 59.70% probability of correct assignment as Caucasian and could have been misclassified as East-Asian 40.30% of the time. While the effect of diet and other external factors may be at play, this may also indicate that mixed races can be identified (this individual is from German-Israeli descent), which could further help crime scene investigators. More hair samples must thus be collected and analysed to create a truer output linear function “rule” to decrease the probabilities of misclassification while providing more information.

References:

1. Doctor’s Data Inc. 2006. Hair elements report. Lab number H0000000-0000-0 Sample patient. Client 12345. http://www.doctorsdata.com/repository.asp?id=1270.

2. Asfaw, A., Wibetoe, G., Beauchemin, D. Solid sampling electrothermal vaporization inductively coupled plasma optical emission spectrometry for discrimination of automotive paint samples in forensic analysis. J. Anal. At. Spectrom. 2012 27 1928-1934.

3. Huang, L., Beauchemin, D. Ethnic background and gender identification using electrothermal vaporization coupled to inductively coupled plasma optical emission spectrometry for forensic analysis of human hair. J. Anal. At. Spectrom. 2014 29 1228-1232.

4. Ryder, A.G. Classification of narcotics in solid mixtures using principal component analysis and Raman spectroscopy. J. Foren. Sci. 2002 47 (2) 275-284.

5. French, A., Macedo, M., Poulsen, J., Waterson, T., Yu, A. Multivariate analysis of variance (MANOVA). San Francisco State University June 4, 2008.

The authors:

Lily Huang and Diane Beauchemin of Queen’s University, Department of Chemistry, Kingston, ON, Canada