Introduction

Hair evidence is readily available at a crime scene as many hairs are shed daily by the average person1,2. Hair shed from the scalp is the most commonly found type of hair evidence3. Scalp hair is often dyed for cosmetic purposes and may be used to disguise an identity. Also, hair colorants give criminals the opportunity to change their appearance.

The market for hair dye is projected to continue expanding to $36.2 billion by 20274. As the market for hair dye grows so does the number of hair colorants available for public use. Due to this projected usage increase, there is a growing demand for forensic methods of hair analyses. Ideally such methods should be robust, reliable, minimally invasive, and non-destructive. At the same time, most of the current methods of analyses, such as high-performance liquid chromatography, gas chromatography, and mass spectrometry, damage hair specimens5,6. PCR analysis can be used to reveal the identity of a suspect through hair evidence. However, this approach requires soft tissue present on hair. It is also destructive to the evidence providing information only about the maternal relatives7,8. It has been proposed that a non-destructive method of UV–visible microspectrophotometry can be used for hair colorant analysis. However, this approach is highly laborious and can be used only after microscopic evaluation of hair9. Lednev group recently showed that Infrared spectroscopy could be used to differentiate between (i) colored and un-colored hair, (ii) different types (permanent vs semi-permanent) and (iii) brands of hair colorants10.

There is a growing body of evidence that surface-enhanced Raman Spectroscopy (SERS) can be used for a confirmatory and minimally destructive analysis of hair colorants11,12. SERS is based on 106–108 enhancement of Raman scattering from the colorants present on hair by noble metal nanostructures13,14. This technique has been broadly used in forensic science to analyse trace amounts of illicit drugs15, bodily samples16, and other trace elements17,18. Recently our group demonstrated that SERS could be used to distinguish more than 30 different colorants, as well as differentiate between different brands and types of colorants19. Finally, the researchers showed that SERS could be used for the automatic identification of hair colors.

Expanding upon this, we investigate the extent to which SERS can detect underlying dyes if the hair was re-colored afterwards. One can envision four different scenarios of hair coloration: (i) overlaying hair dyes with different colors, (ii) overlaying hair colorants of different brands, (iii) overlaying hair colorants of different types, and (iv) overlaying hair colorants that have different brand, type, and color. In the current study, we examine all these combinations using SERS couped to chemometrics to determine the accuracy with which SERS can be used to unravel coloration history of hair.

Results and discussion

SERS-based identification of overlying colorants with different color

We first examined the extent to which SERS can be used to differentiate the underlaying hair colorants of the same brand and type (semi-permanent or permanent) but different colors (blue and purple). For this, hair was first colored by Wella semi-permanent blue (WSBu) and re-colored by Wella semi-permanent purple (WSPu) (WSBuWSPu) that possessed very similar components to WSBu chemical composition. Next, we reversed the order of colorant application by first coloring hair with WSPu and then re-dying it with WSBu (WSPuWSBu). We acquired SERS spectra from all those four samples, Fig. 1. SERS spectra of WSBuWSPu, WSPuWSBu, WSPu and WSBu have distinct vibrational bands at 396, 437, 476, 667, 858, 1050, 1086, 1153, 1267, 1366, 1405, 1475, 1597, and 1645 cm−1. We observed only minor spectral differences that cannot be used for unambiguous differentiation between WSBuWSPu and WSPuWSBu, as well as between SERS spectra acquired from the hair colored with one and two dyes. To overcome this limitation, we used PLS-DA to investigate the accuracy of differentiation between all four classes of SERS spectra.

Figure 1
figure 1

Normalized and baselined SERS spectra of single- and dual-dyes of different colors present on hair. Red trace (Wella semi-permanent blue under Wella semi-permanent purple), green trace (Wella semi-permanent purple under Wella semi-permanent blue), blue trace (Wella semi-permanent blue), pink trace (Wella semi-permanent purple).

Full size image

Our results show that WSBuWSPu, WSPu and WSBu can be identified with 100% accuracy, whereas WSPuWSBu can be predicted with 98% accuracy, Table 1. These results demonstrate that the application of two colorants creates a unique dye appearance on hair that is distinctly different from individual dyes used to color the hair (Figure S1). Our results also show that SERS can be used to identify the order of dyes application on hair, Table 2.

Table 1 Cross-validation matrix of SERS spectra acquired from hair with single and dual-dyes of different colors with TPR of SERS-based identification of hair colored with two dyes of different color.
Full size table
Table 2 Cross-validation matrix of two dyes of different color based on the order of their application on hair with TPR of SERS-based identification of the order of dye application on hair.
Full size table

SERS-based identification of overlying colorants of different brands

We investigated the extent to which SERS can be used to differentiate the underlaying hair colorants of different brands. For this, we colored hair using Ion semi-permanent purple (ISPu) and then applied Wella semi-permanent purple (WSPu) on this hair (ISPuWSPu). We also reversed the order of dye application and first colored hair with WSPu and re-dyed this hair sample with ISPu (WSPuISPu). Next, we acquired SERS spectra from these hair samples, as well as from hair colored by WSPu and ISPu themselves, Fig. 2. SERS spectrum of ISPu has vibrational bands at 364, 437, 531, 576, 761, 927, 973, 1155, 1185, 1324, 1360, 1451, 1479, 1515, 1594, and 1624 cm−1, whereas SERS spectrum of WSPu exhibits distinct vibrational bands at 395, 475, 667, 855, 1048, 1088, 1264, 1366, 1402, and 1646 cm−1.

Figure 2
figure 2

Normalized and baselined SERS spectra of single- and dual-colorants of different brands present on hair. Red trace (Ion semi-permanent purple), green trace (Wella semi-permanent purple), blue trace (Ion semi-permanent purple hair dye under Wella semi-permanent purple), pink trace (Wella semi-permanent purple hair dye under Ion semi-permanent purple hair dye).

Full size image

We found that both WSPuISPu and ISPuWSPu do not exhibit equally intense signatures of WSPu and ISPu. Instead, SERS spectra acquired from hair with two dyes dominate by the spectroscopic signatures of WSPu with no regards whether this colorant was under- or overlaying. These results demonstrate that an application of two colorants that belonged to different dye brands on hair creates the unique dye appearance that is distinctly different from individual colorants used to dye hair (Figure S2).

Utilization of PLS-DA enabled highly accurate identification of all four groups of SERS spectra, Tables 3 and 4. These results demonstrate that SERS can be used to identify coloration history of hair in regard to the brands of colorants used on hair. Our results also show that SERS can be used to identify the order of application of different brands on hair, Tables 3 and 4.

Table 3 Cross-validation matrix of SERS spectra acquired from hair with single and dual-dyes of different brands with TPR of SERS-based identification of hair colored with two dyes of different brands.
Full size table
Table 4 Cross-validation matrix of two dyes of different brand based on the order of their application order with TPR of SERS-based identification of the order of application of colorants of different brands on hair.
Full size table

SERS-based identification of overlying colorants of different types

All hair colorants can be divided into two classes: permanent and semi-permanent. Permanent colorants are based on phenyldiamines that have different substituents around the aromatic ring. Their oxidation by hair developer causes formation of azo polymers, also known as Borowsky bases. These polymers develop strong covalent interactions with keratin on hair, which makes these colorants stay for a long time on hair. Semi-permanent colorants do not require hair developers. These colorants consist of one or several individual dyes that can be easily washed away from the hair.

We first colored hair with Ion semi-permanent blue dye (ISBu) and then applied Ion permanent blue dye (IPBu) on the same hair (ISBuIPBu). We also reversed application of these colorants on hair (IPBuISBu), as well as dyed hair with ISBu and IPBu alone. SERS spectra acquired from these four hair samples are shown in the Fig. 3. SERS spectrum of IPBu exhibit intense bands at 439, 804, 864, and 1493 cm−1, whereas SERS spectrum of ISBu has vibrations at 464, 584, 704, 973, 1049, 1159, 1233 1397, 1450 and 1645 cm−1. SERS spectra of both ISBuIPBu and IPBuISBu exhibited higher intensities at 315, 464, 680, 704, 759, 889, 926, 973, 1159, 1321, 1349, 1622 cm−1, Fig. 3. Our results show that SERS spectra of both ISBuIPBu and IPBuISBu look very similar to ISBu with very little character of IPBu.

Figure 3
figure 3

Normalized and baselined SERS spectra of single- and dual-colorants of different dye types present on hair. Red trace (Ion semi-permanent blue dye under Ion permanent blue dye), green trace (Ion permanent blue dye under Ion semi-permanent blue dye), blue trace (Ion permanent blue), black trace (Ion semi-permanent blue).

Full size image

Utilization of PLS-DA enabled identification of all classes with 100% accuracy, Tables 6 and 7. These results demonstrate that SERS can be used to identify application of different types of colorants on hair, Tables 5 and 6, and Figure S3.

Table 5 Cross-validation matrix of SERS spectra acquired from hair with single and dual-dyes of different types of colorants with TPR of SERS-based identification of hair colored with two dyes of different types.
Full size table
Table 6 Cross-validation matrix of two dyes of different type based on the order of their application on hair with TPR of SERS-based identification of the order of application of colorants of different types on hair.
Full size table

SERS-based differentiation of hair dyes of different color, brand, and type

One may wonder whether SERS can be used to determine hair coloration history of two randomly selected dyes of different color, brand, and type. To answer this question, we first colored hair with Wella semi-permanent purple (WSPu) hair dye that was colored afterwards with L’Oréal permanent red (LPR) hair dye (WSPuLPR). We also altered the order of hair coloration by these two dyes and first colored hair with LPR then re-dying it afterwards with WSPu (LPRWSPu). We also colored hair with just LPR and WSPu. Next, we collected SERS spectra from WSPuLPR, LPRWSPu, LPR and WSPu, Fig. 4. We found that vibrational bands observed in the SERS spectrum of WSPuLPR and LPRWSPu primarily originated from WSPu with very little contribution of LPR (Figure S4).

Figure 4
figure 4

Normalized and baselined SERS spectra of single- and dual-colorants of different brand, type, and color. Red Trace (L’Oréal permanent red), green trace (Wella semi-permanent purple), blue trace (Wella semi-permanent purple hair dye under L’Oréal permanent red hair dye), pink trace (L’Oréal permanent red under Wella semi-permanent purple).

Full size image

PLS-DA was able to identify SERS spectra collected from all four classes with nearly 100% accuracy. The same accuracy was observed for the binary model built for WSPuLPR and LPRWSPu, Tables 7 and 8. These results demonstrate that SERS can be used to unravel hair dying history in regard to the color, brand and type of the colorants used to dye hair.

Table 7 Cross-validation matrix of SERS spectra acquired from hair with single and dual-dyes of different color, brand, and type of colorants with TPR of SERS-based identification of hair colored with two dyes of different color, brand, and type.
Full size table
Table 8 Cross-validation matrix of two dyes of different color, brand, and type based on the order of their application on hair with TPR of SERS-based identification of the order of application of colorants with different color, brand, and type.
Full size table

Conclusion

Our results show that SERS is capable of unravelling coloration history of hair in regard to the dye color, brand, and type that was used to color hair. We also found that spectroscopic fingerprints of re-dyed hair largely represent one of the two dyes used to color hair. This can be explained by different Raman cross-section of colorants in such pairs of dyes. Thus, the colorant with larger Raman cross-section of dyes in it dominates in the SERS spectra acquired from hair with two colorants present on it. Therefore, application of chemometric analysis of spectra is required to reveal the information about the underlying hair colorant. One can expect that forensic application of the discussed above SERS-based approach will required a library of hair colorants with two and three individual colorants simultaneously present on hair to enable robust and reliable determination of hair coloration history.

Materials and methods

Hair coloring procedure

Blonde hair that was never colored prior to the experiments was collected from a hair salon in College Station, Texas from de-identified individuals. Hair was used as received without any pre-treatment or washing. It was cut in small ponytails of approximately the same density and tightened with elastics to minimize hair lost during dying and washing. Six total hair dyes were used to investigate the extent to which SERS could be used to determine the hair dying history, Table 9.

Table 9 Hair colorants used in this study.
Full size table

All semi-permanent dyes were allowed to process approximately 45 min, and permanent hair dye was processed for ~ 60 min according to instructions provided by colorant manufacturers.

Surfaced enhanced Raman spectroscopy

Each hair sample was coated with 5 µl of gold nanoparticles’ suspension (AuNPs). AuNPs were made in the laboratory according to the procedure developed by Esparza and co-workers12. These spherical nanoparticles had ~ 80 nm in diameter. Prior to utilization on hair, the suspension of AuNPs was centrifuged at ~ 5000 g for 10 min to concentrate AuNPs. Next, the pellet of AuNPs was re-suspended in DI water to remove detergent used for the nanoparticle synthesis. SERS spectra were acquired on a TE-2000U Nikon inverted confocal microscope equipped with a 20 × Nikon objective. The objective was used to focus the laser light (λ = 785 nm) generated by continuous wavelength laser on the sample. The same objective was used to collect scattered photons that were directed to a 50/50 light beam splitter and then passed to IsoPlane-320 spectrograph (Princeton Instruments) equipped with a 600 groove/mm grating. A long-pass filter (Semrock, LP-785RS-25) was used to cut off inelastically scattered photons. Laser power at the sample was ~ 1.8 mW. Spectral acquisition times were varying dependent on sample, but all were under 60 s. All reported SERS spectra were normalized and baselined. Spectral resolution was 2 cm−1.

Data analysis

We used Matlab equipped with Partial Least Squares Differentiative Analysis toolbox (Eigenvector Research Inc) for statistical analyses of the collected SERS spectra. All spectra were pre-processed by baselining using a second order automatic weighted least squares, taking the first derivative of spectral intensities with a second polynomial order and filter length of 15. SERS spectra were also area normalized and mean cantered. Partial least squared discriminant analysis (PLS-DA) was used to build all models20,21. Each model had 3–7 principal components. True positive rate (TPR) of the model performance is reported for each model in Tables 1, 2, 3, 4, 5, 6, 7, 8, 9.