This study extends a previously modified method for human sex determination using electrothermal vaporization inductively coupled plasma optical emission spectrometry on chemically altered (dyed) hair. The impact of hair dye on elemental composition was evaluated, and model performance was tested under these conditions. Plasma loading effects were mitigated through argon-based internal standardization. This was followed by background correction, peak area integration, and multivariate analysis (principal component analysis and linear discriminant analysis (LDA)). Using new predictor elements (Ca, Fe, Li, and Sb), LDA accurately classified the sex of 10 dyed hair and 7 undyed hair samples with a model trained on 10 undyed samples analyzed under different conditions. These findings demonstrate the potential of the method for sex classification in chemically altered hair. This study serves as a proof-of-concept based on a limited and imbalanced dataset, establishing a basis for the future validation of a conclusive forensic model. The resulting model appears promising within the constraints of the dataset and adheres to green chemistry principles by employing polytetrafluoroethylene powder as a cost-effective, environmentally friendly carrier agent in place of greenhouse gas.

Forensic chemistry is crucial in criminal investigations, providing objective evidence through the analysis of trace evidence.¹ Although deoxyribonucleic acid (DNA) analysis is a powerful tool in identifying suspects and victims, its use in hair is limited by poor yield and degradation, often leaving insufficient genetic material for identification.^2–3

Hair is valuable in forensic chemistry due to its durability, non-invasive collection, and ability to retain a record of trace elements over time. Because of these properties, elemental composition has emerged as an alternative marker for sex determination when DNA is unavailable; however, hair reflects both endogenous (metabolism, physiology) and exogenous (environmental exposure, diet, cosmetic treatments) exposures, complicating interpretation.^4–6

A significant exogenous factor is hair dye, which is increasingly relevant in forensic contexts as criminals may use it to alter their appearance. Hair dye changes elemental profiles by introducing and depleting specific elements. The concentrations of Mn, Fe, Ni, Cu, Cd, and Sb are often higher in dyed hair, while levels of As, Cr, Zn, Ag, Pb, and Hg are frequently lower compared to undyed hair^7,8. These changes can significantly affect forensic analysis, highlighting the need for models that account for both dyed and undyed hair samples.^7–9

Recent advances in forensic chemistry have used electrothermal vaporization (ETV) coupled to inductively coupled plasma optical emission spectrometry (ICP-OES) to classify sex from trace elements in hair. This technique enables direct solid sample analysis with high sensitivity, minimal sample size (2 to 5 mg), low detection limits, and good reproducibility, while eliminating the need for sample digestion.¹⁰

Several environmentally driven advancements have improved this ETV-ICP-OES method. Originally, dichlorodifluoromethane was used as the ETV gas modifier in 2014.¹¹ In 2022, this was replaced with carbon tetrafluoride.¹² Most recently, the method was modified to use polytetrafluoroethylene (PTFE) powder mixed with the sample instead of a CF4 gas modifier, in alignment with green chemistry principles.¹³ This switch minimizes environmental impact without compromising analytical performance. Each model successfully enables 100% accuracy in human sex determination using Mg, Sr, S, and Zn as predictor elements for classification of sex from undyed hair; however, application of the 2022 CF4-based method to dyed hair required Cd, Ce, Fe, and Sn as new predictor elements for accuracy across both dyed and undyed hair samples.¹⁴

Given that hair dye is becoming more frequently used, it is essential that forensic models account for the exogenous changes introduced by such treatments. This study serves as a proof-of-concept, expanding upon the latest PTFE-modified ETV-ICP-OES method, evaluating its robustness for sex determination across both dyed and undyed hair samples. The objective is to test the method's potential accuracy and robustness under altered conditions, laying the groundwork for future refinement and broader forensic applicability. At this stage, the method is intended for investigative intelligence when DNA is unavailable, rather than evidentiary use in formal forensic proceedings.

Experimental

Instrumentation

A lateral plasma view ARCOS ICP-OES (SPECTRO Analytical Instruments, Kleve, Germany) instrument and an ETV system (ETV 4000C, Spectral Systems, Fürstenfeldbruck, Germany) were used throughout this work. Table 1 summarizes the operating conditions and ETV temperature programs (for sample analysis and cleaning the graphite boat between samples to minimize memory effects). To demonstrate robustness, a set of undyed hair samples analyzed with the PTFE-based method was used to build the predictive model, and a separate set of both dyed and undyed hair analyzed by the CF4-based method provided blind samples for testing model accuracy. The only differences were in the chemical modifier, flow rates of the carrier and by-pass gases, and ETV temperature during the pyrolysis step.

In all cases, an analytical balance was used for directly weighing samples into pyrolytically coated graphite boats. Each boat was manually inserted into the ETV furnace using tweezers for analysis. Vaporized samples were carried by a flow of Ar gas through a 1 m-long Teflon tube connecting the outlet of the ETV system to the ICP torch.

Samples and Reagents

All head hair samples (Table 2) were 10–15 cm long and stored in Ziploc bags. Undyed hair samples were obtained from random adults in Kingston, ON, Canada. Dyed hair samples were obtained from Salon 296 and Maison Paul Coiffure located in Kingston, ON. Two different salons were approached to account for any differences in the dyeing process or commercial dye brands. Additional samples were collected from randomly selected individuals with dyed hair. Based on information provided by the hair salons at the time of sample collection, most samples originated from individuals of Caucasian background. Exceptions include individuals of East-Asian (S2 and S6) and South-Asian (S5) descent. Table 2 shows that the training set consisted of 10 undyed hair samples (6 female, 4 male), while the validation set comprised 10 dyed hair samples (8 female, 2 male) and 7 undyed hair samples (3 female, 4 male). The uneven distribution of dyed male samples reflects limited sample availability and represents a limitation of the current study.

Hair samples were washed with doubly deionized water (DDW) with a resistivity of 18 MΩ·cm (Arium Pro UV/DI System, Sartorius Stedim Biotech, Göttingen, Germany). For ETV analyses of samples in the training set, PTFE powder, 1 micron, M-Clarity™ quality level = MQ100 (Sigma Aldrich, Saint Louis, MO, USA) was used.

Sample Preparation

Prior to analysis, the hair samples were cleaned to remove contaminants, dirt, product, or build-up that may have affected the results. Each sample used to build the model (Table 2) was washed with three portions of double deionized water (DDW), air dried on a tissue, and then ground into a fine powder using a mortar and pestle. Samples were stored in glass vials at room temperature. A 2.0 ± 0.2 mg sample was then weighed directly in the graphite boats and mixed with 2.0 mg of PTFE powder using a stainless-steel spatula. For each sample included in the model, five replicates were analyzed under the operating conditions presented in Table 1, optimized for the use of PTFE powder.

Dyed hair samples and undyed hair samples used for testing the model (Table 2) were washed three times with 20 mL DDW followed by three 20 mL hexane washes. Data were collected from three replicates, with sample masses ranging from 2.0 to 4.0 mg. The blind hair samples were analyzed under slightly different conditions using CF4 as chemical modifier (Table 1).

Data Processing

Point-by-point internal standardization was conducted by determining the ratio of the analyte's emission intensity to that of the internal standard, Ar 430.01 nm, at each measurement point. Internal standardization has been previously shown to correct for signal drift and sample loading effects on the plasma.^11–14 Background correction was applied by calculating the average signal before and after the analyte peak and subtracting it from each data point within the peak. Peak areas were then integrated across the vaporization step. The integrated peak areas were then normalized by dividing them by the mass of the sample.

Using the calculated peak areas, multivariate analysis was performed using linear discriminant analysis (LDA) and principal component analysis (PCA) in Minitab Statistical Software™ (V.21) to reduce dimensionality and simplify the data while preserving patterns in elemental contents. Emission lines were selected using Student's t-test at the 95% confidence interval applied to both the training set of undyed hair analyzed with PTFE (TS1-TS10) and the dyed test set (D1-D10) analyzed with CF4 to identify elements exhibiting statistically significant sex-based differences while showing minimal sensitivity to dyed treatment. Elements showing consistent behavior across both datasets were retained as predictor elements for LDA. Undyed hair test samples (S1-S7) were not examined during the feature selection.

Leave-one-out cross-validation was subsequently used to assess the stability of the LDA classification model, which tests each data point while training the model with the remaining data. Although this approach makes efficient use of limited sample sizes and is known to exhibit low bias in theoretical settings, its application following supervised predictor selection may result in optimistic performance estimates, particularly for small data sets.¹⁵ Consequently, the reported classification accuracy should be viewed as descriptive of the present data set rather than as an unbiased estimate of real-world predictive accuracy and is intended to serve as a preliminary benchmark for future validation using larger datasets.

Results and Discussion

Addition of Dyed Hair Samples

Given the widespread use of hair dye, its inclusion in forensic analysis is essential. Dye treatments alter hair by degrading the cuticle and by introducing or leaching trace elements,⁸ potentially obscuring the endogenous elemental patterns that are critical for accurate classification. Failing to account for such exogenous influences risks compromising the reliability of forensic models. Therefore, integrating dyed hair into analytical validation for sex ensures the predictive model remains robust under real-world conditions, ultimately enhancing its predictive accuracy.

To assess predictive capability, 10 dyed hair samples were added to the original model. While the initial model (excluding dyed hair) showed distinct male and female clustering in PCA and 100% accuracy in LDA,¹³ introducing dyed samples caused overlap in PCA (Figure 1) and reduced LDA classification accuracy, with 5 of 20 predictions falling below 0.600 probability. This suggests that dye treatment alters the elemental composition of hair in a manner that can interfere with sex-based classification when predictors are sensitive to chemical treatment. Apart from Zn 213.856 nm, all original predictor elements showed statistically significant differences between dyed and undyed samples in either the male or female subset. These discrepancies likely explain the observed reduction in model performance after including dyed hair samples; however, the imbalance between dyed male and dyed female samples may also bias the discriminant boundaries toward the majority class.

Selection of New Predictor Elements

Of the 25 elements and over 65 emission lines measured, optimal predictor elements were those that distinguished between sexes while remaining unaffected by dye. At the 95% confidence level, Ba, Ca, Li, Mg, Mo, and Sb differed significantly (p<0.05) between male and female samples. When comparing dyed and undyed hair samples (ignoring sex), Be, Cr, Cu, K, Li, Na, Pb, S, Sb, and Zn showed no significant differences (p>0.05) in content; however, due to the limited dyed sample set, further validation is warranted.

Only Li and Sb satisfied both selection criteria, exhibiting statistically significant sex-based differences while showing minimal sensitivity to hair dye treatment. They were the least dye-sensitive predictors among the elements examined. However, Li and Sb alone were insufficient to achieve clear sex-based separation.

Although Ca and Fe exhibited borderline statistical differences between dyed and undyed hair samples (p = 0.0419 and p = 0.029, respectively), their inclusion substantially improved class separation within the present dataset. Their incorporation was further supported by established literature relevance and strong sex-based differentiation.^16–18 Whether these predictor elements remain sex-discriminatory across broader demographic groups is not yet known. Future studies incorporating wider ethnic, environmental, physiological, and age variation will be required to assess their universality.

The final predictor set (Ca, Fe, Li, and Sb) enabled 100% classification accuracy by LDA (Table 3) for the 27 samples in Table 2. Eigenanalysis confirmed effective dimensionality reduction, with the first three principal components accounting for 95.6% of total variance, preserving the dataset's structure with minimal information loss.

Further Validation

To further validate the model, an additional set of undyed samples (S1-S7) was introduced as a test set. Their inclusion did not alter the observed trends. While some overlap is visible in the PCA score plot (Figure 2), sex-based clustering remains apparent, with male samples primarily occupying the left quadrants and female samples the right. Future work may benefit from investigating 3-dimensional plots to enhance this separation. PC1 was driven mainly by Li, Sb, and Ca, with loadings of 0.652, 0.561, and 0.509, respectively. These high contribution values, even after the addition of 17 new samples, reinforce their importance as key predictors of sex differentiation.

Using the 10 undyed PTFE-analyzed samples (TS1-TS10) as the training set, LDA correctly classified all 17 samples within the present test dataset (Table 4), which had been prepared differently and analyzed using CF4 gas with a different ETV program (D1-D10 and S1-S7). Table 4 shows that 13 samples were predicted with 1.000 probability and the remaining 4 were predicted with probabilities above 0.98, demonstrating encouraging separation within the limited and imbalanced dataset. Leave-one-out cross-validation further confirmed this performance, with LDA achieving 100% accuracy and predicting the sex of all samples with probabilities of 1.000. Such results highlight the strength of LDA in maintaining full classification accuracy by leveraging class labels, which is crucial in forensic analysis, where accurate classification is the primary goal. Although no misclassifications were observed in this study, the small size of the dataset prevents accurate estimation of false-positive or false-negative rates. Performance will likely be lower in larger and more diverse populations, reinforcing the necessity for further validation.

Conclusions

This study demonstrates that hair dye alters elemental composition and must be considered in elemental-based forensic analyses of hair. The revised and “greener” model using PTFE powder was evaluated as a proof-of-concept for sex classification across dyed and undyed hair samples. LDA, and to a lesser extent PCA, were applied using a revised set of predictor elements (Ca, Fe, Li, and Sb), which enabled clear sex-based separation within the present dataset despite variations in sample preparation, instrumental conditions, and data processing approaches. The inclusion of samples collected and analyzed over a span of more than 10 years further suggests the potential applicability of the approach for exploratory analysis of archived forensic samples.

As a proof-of-concept, this study provides a foundation rather than a conclusive forensic model. The findings highlight the potential use of the method for investigative purposes, offering probabilistic sex classification when DNA analysis is unavailable or compromised. Given the limited and imbalanced dataset, particularly with respect to dyed male hair, the results should not be interpreted as definitive for evidentiary use. Prior to forensic deployment, further validation will be required, including expansion to larger and more balanced datasets, interlaboratory reproducibility studies, evaluation of environmental and biological factors, and assessment against established forensic frameworks. Furthermore, this validation must be conducted under conditions that reflect real-life casework.

Future work will also focus on incorporating hair samples from individuals of diverse racial and ethnic backgrounds to assess the model's robustness. Both endogenous factors (genetic differences influencing metabolism and hair growth) and exogenous factors (cultural practices, environmental exposure, hair care routines, and dietary habits) associated with population diversity can influence the elemental composition of hair.¹¹ Expanding the dataset to better reflect Canada's diverse population is therefore an essential next step toward developing a broadly applicable forensic tool for sex determination using hair.

Acknowledgements

The authors gratefully acknowledge research funding from the Natural Sciences and Engineering Research Council of Canada (Grant No. RGPNM 39487-2018). CW also thanks Queen's University School of Graduate Studies and Postdoctoral Affairs for a graduate award.

References

1. Jackson, A. R.; Jackson, J. M. Forensic Science, 3rd ed.; Pearson Education Limited, 2011.

2. Alberts, B.; Johnson, A.; Lewis, J.; Morgan, D.; Raff, M.; Roberts, K.; Walter, P. Molecular Biology of the Cell, 6th ed.; Garland Science, 2015.

3. Leskovar, T.; Mlinšek, T.; Počivavšek, T.; Zupanič Pajnič, I. Comparison of Morphological Sex Assessment and Genetic Sex Determination on Adult and Sub-Adult 17th–19th Century Skeletal Remains. Genes 2023, 14 (8), 1561. DOI: 10.3390/genes14081561

4. Rodushkin, I.; Axelsson, M. D. Application of Double Focusing Sector Field ICP-MS for Multielemental Characterization of Human Hair and Nails. Sci. Total Environ. 2000, 250 (1–3), 83–100. DOI: 10.1016/s0048-9697(00)00369-7

5. Wilhelm, M.; Hafner, D. Monitoring of Cadmium, Copper, Lead and Zinc Status in Young Children Using Toenails: Comparison with Scalp Hair. Sci. Total Environ. 1991, 103 (2–3), 199–207. DOI: 10.1016/0048-9697(91)90145-5

6. Saber-Tehrani, M.; Tadayon, F. Endogenous and Exogenous Elemental Speciation in Human Hair by Various Washing Procedures and Determination by Flame and Electrothermal Atomic Absorption Spectrometry. Asian J. Chem. 2009, 21 (5), 3660–3670.

7. Wei, Z. L.; Rui, Y. K.; Shen, L. Effects of Hair Dyeing on the Heavy Metals Content in Hair. Guang Pu Xue Yu Guang Pu Fen Xi 2008, 28 (9), 2187–2188. PMID: 19093590.

8. Yoshinaga, J.; Yanagisawa, Y.; Morita, M.; Okamoto, K. Hair-to-Blood Ratio of Mercury (Hg) and Other Elements in Residents of Niigata, Japan. Arch. Environ. Health 2003, 58, 437–443.

9. Galizia, A.; Falchi, L.; Laquinta, F.; Machado, I. Biomonitoring of Potentially Toxic Elements in Dyed Hairs and Its Correlation with Variables of Interest. Biol. Trace Elem. Res. 2024, 202 (8), 3529–3537. DOI: 10.1007/s12011-023-03964-x

10. Hughes, D. M.; Chakrabarti, C. L.; Goltz, D. M.; Grégoire, C. D.; Sturgeon, R. E.; Byrne, J. P. Seawater as a Multi-Component Physical Carrier for ETV-ICP-MS. Spectrochim. Acta Part B 1995, 50 (4–7), 425–440. DOI: 10.1016/0584-8547(94)00149-P

11. 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 (7), 1228–1232. DOI: 10.1039/C4JA00071D

12. MacConnachie, M.; Lu, S.; Wang, Y.; Williams, J.; Beauchemin, D. Sex Determination of Mummies through Multi-Elemental Analysis of Head Hair Using Electrothermal Vaporization Coupled to Inductively Coupled Plasma Optical Emission Spectrometry. RSC Adv. 2022, 12 (42), 27064–27071. DOI: 10.1039/D2RA05654B

13. Wheeler, C.; Beauchemin, D. Verification of a Greener Method of Sex Determination through the Multi-Elemental Analysis of Hair Using Electrothermal Vaporization Coupled to Inductively Coupled Plasma Optical Emission Spectrometry. At. Spectrosc. 2024, 45 (5), 391–399. DOI: 10.46770/AS.2024.169

14. MacConnachie, M.; Beauchemin, D. Validation and Expansion of Sex Determination Method through Analysis of Human Hair Using Electrothermal Vaporization Coupled to Inductively Coupled Plasma Optical Emission Spectrometry. J. Anal. At. Spectrom. 2023, 38, 1394–1401. DOI: 10.1039/D3JA00165B

15. Smith, G. C. S.; Seaman, S. R.; Wood, A. M.; Royston, P.; White, I. R. Correcting for Optimistic Prediction in Small Data Sets. Am. J. Epidemiol. 2014, 180 (3), 318–324. DOI: 10.1093/aje/kwu140

16. Sera, K.; Goto, S.; Hosokawa, T.; Saitoh, Y.; Kumagai, R.; Yamazaki, K.; Yamaya, K. Sex-Specific Differences in Elemental Movement from Blood to Hair. Int. J. PIXE 2018, 28 (3–4), 107–120. DOI: 10.1142/S0129083518500201

17. Suliburska, J. A Comparison of Levels of Select Minerals in Scalp Hair Samples with Estimated Dietary Intakes of These Minerals in Women of Reproductive Age. Biol. Trace Elem. Res. 2011, 144 (1–3), 77–85. DOI: 10.1007/s12011-011-9034-9

18. Rushton, D. H.; Barth, J. H. What Is the Evidence of Gender Differences in Ferritin and Haemoglobin? Crit. Rev. Oncol. Hematol. 2010, 73 (1), 1–9. DOI: 10.1016/j.critrevonc.2009.03.010

About the Authors