X. Wei, K. Yu, H. Wu et al.
Forensic Science International 333 (2022) 111205
2. Materials and methods
2.1. Sample preparation
10 human wisdom teeth (5 males and 5 females) were collected
from Hospital of Stomatology Xi’an jiaotong University. The effect of
oral diseases on teeth was not considered in this study, so we chose
the healthy donor and the wisdom teeth which were not decayed.
10 teeth samples from each animal species, including bovine
(Simmental), dog (Beagle), rat (Sprague-Dawley), rabbit (New
Zealand White Rabbit). Each animal species donor consists of five
males and five females. One permanent Incisor was collected from
each bovine, rat and rabbit, one canine tooth from each dog. All the
animal donors were provided by the Laboratory Animal Research
Center of Xi’an Jiaotong University (Xi’an, China). The collection of
teeth samples has followed the guidance of the Department Ethics
Committee of Xi’an Jiaotong University and complied with the re­
quirements of local laws and regulations.
For the purpose of simulating an actual forensic case, all teeth
were placed in disposable medical urine cups separately with no
cover in a room for 100 days. The environmental conditions of the
room were not controlled. The temperature was varied from 15° to
35°C and the humidity was from 30% to 70%. Finally, a set of 50 teeth
samples were prepared for the subsequent experiment.
2.2. FT-IR spectral acquisition and data preprocessing
First, the part of dental crown of teeth samples was ground into
powder by metal file and the resulting powder was collected. The
powder was then passed through a No. 200 mesh with sieve of
74 µm, and the filtered portion was used as the stock powder. Next,
1.0 mg ( ± 0.1) of stock powder was mixed with approximately
100 mg of KBr powder and ground in an agate mortar thoroughly.
Subsequently, the mixture was placed in a 13-mm sample pellet and
then compressed with a hydropress at a pressure about 7845 kPa for
1 min. Three replicate pellets were prepared for each sample in
order to guarantee the stability and repeatability of the experiment.
It should be noted that similar data results could be obtained
using non-invasive and non-destructive FTIR methodologies, such as
reflection-Micro-FTIR. Due to sample processing and data analysis
issues, the experiment in this paper is only involved traditional
transmission-FTIR spectroscopy by preparing KBr pellets. The re­
search using Micro-FTIR is also in progress.
All spectra in this study were obtained by a Thermo Nicolet iS50
FTIR spectrometer (Thermo Fisher Scientific, Waltham, WA, USA)
and controlled by OMNIC™ Software version 8.2. Before each mea­
surement, the spectrum of pure KBr pellets was obtained as back­
ground reference spectrum. All spectra were recorded in the range of
4000–400 cm−1 and the resolution of 4 cm−1, 32 scans. Similarly, to
ensure the repeatability of the method, three replicate measure­
ments were made for each pellet. Briefly, there were a total of nine
spectra for each sample, the average of which was used as dataset.
In this case, only the region from 2000 to 450 cm−1 was chosen
for further analysis which called ‘fingerprint region’ and contained
the fundamental spectral information of many biomolecules. In
order to reduce the influence of light scattering and particle size, the
standard normal variate (SNV) was applied to all of the selected
spectral regions. At last, mean centering was performed to the pre­
processed spectral dataset.
2.3. Spectral analysis
In this work, all analytical methods were performed in MATLAB
R2020b (The MathWorks, MA, USA) equipped with PLS_Toolbox 9.0
(Eigenvector Research, Inc., USA).
PCA is a frequently used unsupervised data analysis method. PCA
can achieve data simplification by turning the original spectral
variables from high dimensions into substantially lower dimensions
[21]. Significantly, the most valuable information in the original data
set is stilled retained in the new set of variables after dimensionality
reduction. Correlated variables were extracted from the original data
and then converted into a smaller set of uncorrelated variables
which called principle components (PCs). The corresponding loading
plots could show the contribution of each PC.
PLS-DA is a supervised linear classification method that com­
bines the properties of partial least squares regression with the
discrimination ability. This method can decompose the X-variable
with the guidance of the Y-variable to find the latent variables (LVs)
[22]. One of the most significant steps in establishing model is
choosing the number of latent variables which is determined by
cross-validation (CV). In this present work, a 20-fold CV was per­
formed with the Venetian blinds procedure during the validation
step. Finally, according to the correct classification rate of CV, we
chose the appropriate LVs to build the final predictive models.
3. Results and discussion
Teeth tissue is a composite material with a hierarchical structure
which includes enamel and dentin. The chief inorganic constituent of
enamel and dentine is calcium phosphate which was in the form of
hydroxyapatite. The organic portion mainly consists of collagen and
non-collagenous proteins623. The average spectra of the fingerprint
region of teeth from human, bovine, dog, rat and rabbit are shown in
Fig. 1. The organic phase includes amide I (1700–1600 cm-1), amide II
(1600–1500 cm-1) and amide III (1240 cm-1). The inorganic phase
includes phosphate (PO43- ν1: 960 cm-1, PO43-ν3: 1200–1000 cm-1,
PO43- ν4: 660–510 cm-1) and carbonate (CO32- ν2: 850–900 cm-1,
CO32- ν3: 1600–1300 cm-1) [22–25].
From the average spectrogram, it can be seen that the types of
chemical components in the teeth of different species are basically
the same, but the proportion of components is not exactly the same,
which provides the possibility of species identification.
3.1. PCA analysis of teeth from different species
Fig. 2 shows the 3D-PCA score plots of the teeth from human and
four animal species. Along the direction of PC1, there is a separation
between human teeth and the three animal teeth (dog, rabbit and
rat). The rabbit and rat are also different from human along the di­
rection of PC2. The difference between human and bovine is mainly
reflected in the direction of PC3. In order to explain the PCA results
of species differences in spectra more clearly, we combined the load
plot (Fig. 3) to further analyze. Along PC1(Fig. 3a), the discriminating
positively correlated loading representative of teeth from dog and