Isaac
Juarez
a and
Dmitry
Kurouski
*abc
aDepartment of Biochemistry and Biophysics, Texas A&M University, College Station, Texas 77843, USA. E-mail: dkurouski@tamu.edu
bDepartment of Biomedical Engineering, Texas A&M University, College Station, Texas 77843, USA
cInstitute for Advancing Health Through Agriculture, Texas A&M University, College Station, Texas 77843, USA
First published on 15th August 2023
Trace evidence found at crime scenes is rarely in an unsullied condition. Surface-enhanced Raman spectroscopy (SERS) is a modern analytical technique that can be used for the detection of artificial hair colourants (S. Higgins and D. Kurouski, Surface-Enhanced Raman Spectroscopy Enables Highly Accurate Identification of Different Brands, Types and Colors of Hair Dyes, Talanta, 2022, 251, 123762). However, contaminants pose a problem to collecting accurate spectra from the dyes. In this study, we sought to analyze how the different physical properties of contaminants can influence the collected spectra. We utilized 11 household substances of varying viscosity and opacity to contaminate hair dyed with permanent black or semi-permanent blue dyes. We discovered that contaminant opacity generally does not affect the spectral quality but that high contaminant viscosity does and that acidic substances could destroy the colourant's spectral identity altogether. Cleaning the contaminated hair with a water rinse allowed the underlying colourant to be identified in 21 out of 22 cases. Overall, this study provided a clearer understanding of the capabilities and limitations of SERS in forensic hair analysis.
This has catalysed a demand to advance the techniques available to trace evidence experts.6 Several new methods are now used for hair analysis including high-performance liquid chromatography coupled with mass spectroscopy (HPLC-MS) and PCR-based nuclear DNA analysis. The former is commonly used to identify explosives and illicit substances both within and atop the hair.7,8 Still this method requires a large amount of hair, along with being time-consuming and destructive. Nuclear DNA analysis has been the preferred method by the FBI since 1996 but comes with the issue of being similarly laborious and expensive. Nuclear DNA is also only found in the hair root which is rarely attached to the hair shaft at a crime scene.9 While mtDNA can be found in the shaft, results are limited since mtDNA cannot differentiate between individuals with the same maternal genotype.10 Lastly, neither of the aforementioned techniques can be used to determine if hair has been artificially coloured. Around 48% of people dye their hair today, a variable which forensic experts must now take into consideration when examining hair evidence.11
Surface-enhanced Raman spectroscopy (SERS) offers a viable alternative to these issues. Previous studies have shown the ability of SERS to identify artificial colourants present on hair strands.12 SERS is able to detect the chemical fingerprint of these colourants by utilizing the excitation of localized surface plasmon resonances (LSPRs) on metal nanoparticles to enhance their electromagnetic field. This in turn amplifies the Raman signal strength of the colourant molecules adsorbed to the surface of the nanoparticles.13 Further studies have expanded knowledge of SERS' hair analysis capabilities, such as its ability to detect an underlying colourant in redyed hair, or in its robustness, through the differentiation of 30 distinct dyes by brand, colour, and permanence.14,15
While the foundation of hair analysis by SERS is today is well known, crime scenes will seldom have the clean conditions available to us in the laboratory setting. Biological and non-biological contaminants are ubiquitous at a crime scene, so how these scene contaminants affect one's ability to detect the hair colourant's Raman signature is of key importance to study. The Kurouski group showed in 2022 that the ability to detect colourants on contaminated hair varies from contaminant to contaminant.16 Some contaminants such as saliva and dirt did not block the underlying Raman signal, while others such as blood did. Some contaminants, like bleach, destroyed the possibility of detecting the underlying colourant even after the hair was rinsed. This raises the question as to whether different physical or chemical characteristics of crime scene contaminants affect one's ability to detect the hair colourant. Many household substances, from food items to chemical products, could easily contaminate a crime scene. In this study, we sought to examine the effects that a variety of domestic substances could have on the identification of an underlying hair colourant.
Gold nanoparticles (AuNPs) were used as our plasmonic material and were synthesized in lab following a procedure reported by Nikoobakht and El-Sayed with modifications.17 Au seeds were first synthesized by adding 1 mL of 0.01 M HAuCl4, 1 mL of 0.01 M Na3Cit, 36 mL of DI H2O, and 1 mL of 0.1 M NaBH4 together. We then allowed the solution to sit for 2 hours. Meanwhile, 3 growing solutions were prepared using 9 mL of 0.05 M CTAB, 0.25 mL of 0.01 M HAuCl4, 0.05 mL of 0.1 M NaOH, 0.05 mL of 0.01 M KI, and 0.05 mL of 0.1 M ascorbic acid. The first 2 growing solutions used this formula, while the third growing solution used the same volumes multiplied by 10. All growing solutions were stirred at 500 rpm. After the Au seeds were synthesized, 0.5 mL of the Au seed solution was added to growth solution 1. After a few minutes, 1 mL of growth solution 1 was added to growth solution 2. Finally, all of growth solution 2 was added to growth solution 3. The stir bar was turned off after 3 hours of stirring and the final solution was allowed to grow overnight. Nanoparticles were characterized using UV-vis absorption. The nanoparticles were suspended in 10 mL of ethanol and then their Raman signature was acquired for background subtraction.
1 inch hair strands were cut from the two dyed groups of hair to be placed on microscope slides. 80 μL of each contaminant was then applied to clean, dry hair samples from both dyes. The contaminant was allowed to dry for 24 h on the hair strands. To prep a hair strand for Raman spectrum acquisition, it was first removed from a contaminated hair sample on a microscope slide and placed onto a clean slide. 5 μL of AuNP solution was then pipetted onto the contaminated strand of hair. The hair strand then had its spectrum recorded by the Raman spectrophotometer. Separately, we also washed hair strands from each group of contaminated hair samples by shaking them in individual test tubes with Millipore DI water until no contaminant visibly remained on the hair strand.18 The now cleaned hair strands were allowed to air dry before applying AuNPs and having their spectrum recorded.
![]() | ||
Fig. 1 SERS spectra acquired from SpBlu- (left) and PBlk- (right) coloured hair exposed to different contaminants. |
We observed similar changes in the SERS spectra acquired from PBlk-colored hair that was subjected to the same set of contaminants (Fig. 1). The overall shape of the averaged spectra was more consistent within the contaminated PBlk hair, however there was a notable lack of consistency in peak location and peak height relative to expectations set by the uncontaminated PBlk spectra. SERS spectra acquired from both molasses- and clear glue-contaminated hair were lacking any chemical signatures, while SERS spectra collected from wine-contaminated hair had almost unidentifiable peaks compared to spectra collected form hair exposed to other contaminants. Lastly, the overall shape of the SERS spectra collected from soy sauce-contaminated hair was noisy in the 400 cm−1 to 800 cm−1 region, while having very little signal in the 1200 cm−1 to 1600 cm−1 region. This further confirms our conclusion that household contaminants could substantially lower the feasibility of SERS-based identification of colourants on hair. Our results also show that for both SpBlu- and PBlk-colored hair, highly viscous substances, such as clear glue, corn syrup, honey, and molasses significantly stronger reduced the probability to detect the colourants compared to other contaminants. One can speculate that upon the presence of these contaminants on the coloured hair, AuNPs could not reach the hair surface and, consequently, enhance Raman scattering from the colourants present on hair.
In the previous study, we demonstrated that contaminants could be removed from hair by application of numerous copies of water. Once the contaminant was removed, SERS could be used to detect and identify hair colourants.16 In the current study, we investigate the extent to which this simple washing procedure could be used to remove the discussed above contaminants from hair. We found that SERS spectra acquired from cleaned SpBlu-colored hair spectra were nearly identical to the spectra acquired from uncontaminated hair (Fig. 2). Analysis of the clean PBlk spectra differed as the results did not fully match the success of the clean SpBlu spectra. It is important to first recognize that cleaning the contaminated PBlk hair resulted in overall better spectra. The location of peaks was much more consistent between averaged spectra and the uncontaminated control spectrum in the cleaned groups. Also, peak height relative to the baseline was much more pronounced post-rinse than in the contaminated spectra. However, closer examination reveals inconsistencies, such as the peak at 1400 cm−1 being strongly pronounced in several of the spectra as opposed to the uncontaminated spectra where the peak at 1500 cm−1 is most pronounced (Fig. 2). This is most easily visible in the cleaned olive oil, kerosene, and honey spectra. More obvious is the entire 1400 cm−1 peak missing in the cleaned half-and-half spectrum. Finally, the spectrum of cleaned red wine contained minimal chemical information. Although the red wine spectrum stands out, observations from the clean PBlk raw data show relatively low intensities for white wine and orange juice when compared to the rest of the clean PBlk spectra. The raw spectra of these three each have very strong background noise and very low intensity (Fig. S1†). All three of these contaminants have a low pH in the range of 3.0 to 4.0, indicating that pH could also play an important role in our ability to detect the colourant. Healthy scalp pH is around 5.5, and so artificial colourants have a pH of ≥6.0 to promote the cuticle opening for the color to penetrate.20–22 This acidic pH could plausibly oxidize the hair dye, destroying the molecule after a period, explaining the observation we see in the clean PBlk raw data.
![]() | ||
Fig. 2 SERS spectra acquired from cleaned SpBlu- (left) and PBlk- (right) coloured hair exposed to different contaminants. |
Two possible explanations for the inconsistencies between the effects of decontamination in one dye versus another are either the colourant's permanence or hue. While the data in the graphs have been normalized, the raw data shows massive differences in average intensity between SpBlu spectra and PBlk spectra. Even when contaminated, the underlying SpBlu spectra was much more pronounced in peak height relative to the baseline. For SpBlu, peak height was only really reduced in the presence of high viscosity substances. This compared to PBlk which experienced spectral distortion from nearly every contaminant.
We corroborated our visual examination with chemometric analysis using PLS-DA. Machine learning has a growing role in being able to classify forensic pattern-evidence, thereby removing the subjective human aspect.23 PLS-DA uses a supervised learning algorithm to learn how to classify a spectrum based on connections it draws between labels and spectral patterns.24 In this experiment, we provided the learning algorithm with 50 contaminated spectra for each group and then tested the model's ability to correctly differentiate the spectra. The results for SpBlu indicated that the model was able to accurately classify spectra with a 92% TPR or greater for 10 out of 12 contaminant groups, (Table 1). The model did get confused with kerosene (64% TPR) and with molasses (78% TPR). This was consistent when we compared the 3-dimensional score plot for SpBlu; kerosene was more dispersed than other groups are, and molasses was clustered with soy sauce and red wine.
Actual Class ≫ | WW | UC | SS | RW | OO | OJ | MO | HH | KE | HO | CS | CG | TPR, % | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
a White Wine (WW), Uncontaminated (UC), Soy Sauce (SS), Red Wine (RW), Olive Oil (OO), Orange Juice (OJ), Molasses (MO), Half-and-Half (HH), Kerosene (KE), Honey (HO), Corn Syrup (CS), Clear Glue (CG). | ||||||||||||||
Predicted class | WW | 47 | — | — | — | — | — | — | — | — | — | — | — | 94 |
UC | — | 50 | — | — | — | — | — | — | — | — | — | — | 100 | |
SS | — | — | 48 | — | — | — | 7 | — | — | — | — | — | 96 | |
RW | — | — | 1 | 50 | — | — | 1 | — | — | — | — | — | 100 | |
OO | — | — | — | — | 46 | — | — | — | — | — | — | — | 92 | |
OJ | — | — | — | — | — | 50 | — | — | — | — | — | — | 100 | |
MO | — | — | 1 | — | — | — | 39 | — | — | — | — | — | 78 | |
HH | 3 | — | — | — | 1 | — | — | 50 | — | — | — | — | 100 | |
KE | — | — | — | — | 2 | — | — | — | 32 | — | — | 2 | 64 | |
HO | — | — | — | — | — | — | 2 | — | 3 | 49 | — | — | 98 | |
CS | — | — | — | — | — | — | 1 | — | 7 | — | 50 | — | 100 | |
CG | — | — | — | — | — | — | — | — | 8 | 1 | — | 48 | 96 |
The model built for PBlk had similar accuracy to SpBlu's model and had no TPRs below 82% with 9 out of 12 contaminant groups having a TPR of 92% and above, (Table 2). The confusion matrix for PBlk model showed that most spectra that were misclassified were incorrectly identified as being contaminated with clear glue. The lower scores for red wine (82% TPR) and kerosene (82% TRP) contamination were also consistent with their 3-dimensional score plot for PBlk since both contaminants had their spectra distributed across a wider area. This coupling of SERS with PLS-DA provides strong evidence for the distinct effects that different crime scene contamination can have upon hair colourants' identification. In the future, quantification of the physical differences between contaminants and their effect on the collected spectrum could lead to an ability to predict what spectrum we should expect for a contaminant with known physical properties.
Actual class ≫ | WW | UC | SS | RW | OO | OJ | MO | HH | KE | HO | CS | CG | TPR, % | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
a White Wine (WW), Uncontaminated (UC), Soy Sauce (SS), Red Wine (RW), Olive Oil (OO), Orange Juice (OJ), Molasses (MO), Half-and-Half (HH), Kerosene (KE), Honey (HO), Corn Syrup (CS), Clear Glue (CG). | ||||||||||||||
Predicted class | WW | 46 | — | — | — | — | 2 | — | — | — | — | — | — | 92 |
UC | — | 50 | — | — | — | — | — | — | — | — | — | — | 100 | |
SS | — | — | 49 | — | — | — | 3 | — | — | — | — | — | 98 | |
RW | — | — | 1 | 41 | — | — | — | — | — | — | — | 2 | 82 | |
OO | — | — | — | — | 48 | — | — | — | — | — | — | — | 96 | |
OJ | 4 | — | — | — | — | 48 | — | — | — | — | — | — | 96 | |
MO | — | — | — | — | — | — | 44 | — | — | — | — | 1 | 88 | |
HH | — | — | — | — | — | — | — | 50 | — | — | — | — | 100 | |
KE | — | — | — | — | — | — | — | — | 41 | — | — | — | 82 | |
HO | — | — | — | — | — | — | — | — | 5 | 45 | 1 | — | 90 | |
CS | — | — | — | — | — | — | — | — | 3 | — | 49 | — | 98 | |
CG | — | — | — | 9 | 2 | — | 3 | — | 1 | 5 | — | 47 | 94 |
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d3ay01219k |
This journal is © The Royal Society of Chemistry 2023 |