Review: Differentiation of Hair Using ATR FT-IR Spectroscopy: A Statistical Classification of Dyed and Non-Dyed Hairs

Emily C. Lennert

 

Category

Chemistry

Keywords

hair, dye, differentiation, classification, attenuated total reflection, Fourier transform, infrared spectroscopy, ATR, FT-IR

Article Reviewed

1. Boll, M. S.; Doty, K. C.; Wickenheiser, R.; Lednev, I. K. Differentiation of hair using ATR FT-IR spectroscopy: A statistical classification of dyed and non-dyed hairs. Forensic Chemistry. 2017, 6, 1-9.

Disclaimer

The opinions expressed in this review are an interpretation of the research presented in the article. These opinions are those of the summation author and do not necessarily represent the position of the University of Central Florida or of the authors of the original article.

Summary

Hair is a commonly encountered form of trace evidence. Hair can be analyzed based on several class characteristics, e.g. color or texture, which may help to identify suspects. Current methods of hair analysis may be subjective, as they are primarily based in microscopy. The quality of examination may be dependent on the examiner’s experience and training, the hair condition, and the analytical capabilities of the instrument used. This study proposes the use of attenuated total reflection Fourier transform – infrared spectroscopy (ATR FT-IR) for the chemical analysis of hair samples.

Hair samples were obtained from 11 volunteers; age, biological sex, and race were recorded. All samples were natural in color, i.e. un-dyed. Several hairs were collected to allow for the analysis of dyed and un-dyed hairs from each volunteer. Two colors of hair dye from three manufacturers were selected for this study; medium brown and black, from Revlon, Clairol, and Just For Men®. Samples were dyed according to manufacturer instructions in individual petri dishes to eliminate the possibility of cross contamination and then the hair was allowed to sit for one week in ambient conditions prior to analysis. For analysis, hairs were placed on the FT-IR crystal and compressed with the ATR attachment. This was repeated ten times per hair sample, at different points along the hair shaft, for a total of ten trials per sample. Each trial consisted of ten spectra that were averaged to produce one trial.

Partial least squares discriminant analysis (PLSDA) was employed for the classification and differentiation of hair samples. Three types of analysis were carried out. Classification models were created to determine if the sample was dyed or un-dyed, to distinguish between dye brands, and to distinguish between dye colors. 8 volunteers’ samples were used as the calibration set, e.g. to develop the model, and the hair from the remaining 3 volunteers were used as the validation set, e.g. to test the model’s ability to accurately determine which class the hair should be assigned to.

Distinguishing Dyed and Un-Dyed Hairs

Hair dyes alter the chemistry of hair by oxidizing the hair, which allows the dye to bind to the hair. Based on this, it is expected that spectral differences would be observed between dyed and un-dyed hairs. In fact, significant spectral differences were reported which lead to the differentiation of dyed and un-dyed hairs. Differences were observed in the sulfate stretching region, 662-673 cm-1, and in the ester stretching region, 1250-1270 cm-1. Additionally, due to hair oxidation, an intensity difference was observed at 1040 cm-1. PLSDA was used to generate a model incorporating five latent variables. Latent variables are variables that result from the reduction of hundreds or thousands of variables down to a few significant variables and illustrates where most of the variation is that differentiates samples within the data set. Latent variable 1 accounted for the most variance compared to the others, and showed large differences between dyed and un-dyed for the sulfate and ester stretching. Ester intensity changes are expected, due to common esters present in the ingredients of the dyes, such as erythorbic acid. Sulfate signal changes are due to the chemical process of hair dyes, in which oxidizers break disulfide bonds within hair so that new bonds can be made with color molecules. The breaking of sulfide bonds results in the release of sulfur, which creates a change in the sulfate signal.

An internally cross-validated prediction plot was created to determine how well new hairs could be classified as dyed or un-dyed. Only one dyed hair samples spectra failed to classify as dyed, otherwise all classifications were correct, for a prediction accuracy of 99.6%. Then, 60 spectra that were not used to create a PLSDA model were used as an external validation set. Of this set, three spectra were misclassified, resulting in a prediction accuracy of 95%, which is also a very good prediction rate.

Distinguishing Dye Brands

Dyed samples were further analyzed statistically to determine whether the brand could be differentiated. Each brand contained characteristic ingredients that were consistent between the two colors studied. Revlon dyes contained oleic acid and sodium benzotriazolyl butylphenol sulfonate. Clairol dyes both contained ammonium hydroxide and disuccinates. Just For Men® had the common ingredient of petrolatum between dyes.

While there were no differences in the location and quantity of IR peaks observed between brands, peak intensity differences were observed. Significant intensity differences were observed in the sulfate and ester stretching regions as well as at peaks at 794 and 819 cm-1 (erythorbic acid) and at 1040 cm-1 (cysteine oxidation). A PLSDA model was created for the prediction of samples into brands. Of the 240 spectra calibration set, three were considered unclassified. Unclassified is considered a better output than misclassification, and is analogous to reaching an inconclusive result. Of the 30 validation spectra tested against the model, only one was predicted as unclassified and the remaining 29 were correctly classified, for a prediction accuracy of 96.7%. The authors noted that, although the model did not correctly predict the brand for all samples, it did not produce any misclassifications.

Distinguishing Dye Colors

Spectra were then classified as either medium brown or black, regardless of brand. It was expected that this classification would be less successful than the previous two. However, for the calibration of the model only two samples were misclassified into the other color group. Additionally, for the validation spectra set, no misclassifications were observed, for a prediction accuracy of 100%. The authors attribute this accuracy to the presence of specific ingredients that are present to produce the desired color.

Scientific Highlights

• Hair was successfully classified as dyed or non-dyed based on PLSDA of ATR FT-IR spectral data.
• Dyed hair was further classified by brand and by color based on PLSDA of ATR FT-IR spectral data.
• ATR FT-IR spectroscopy was successfully implemented in the analysis of dyed and non-dyed hairs.

Relevance

ATR FT-IR offers an additional means of hair analysis that is rapid, provides chemical information, and requires no sample preparation.

Potential Conclusions

ATR FT-IR analysis may be used to differentiate dyed and non-dyed hairs, as well as to differentiate hair dye brands and colors from one another.