Forensic Science International 340 (2022) 111445

Contents lists available at ScienceDirect

Forensic Science International

journal homepage: www.elsevier.com/locate/forsciint

Dental age estimation based on DNA methylation using real-time

methylation-specific PCR ]]
]]]]]]
]]

⁎
Ayano Ogata a, Masahiro Kondo a, , Masaaki Yoshikawa b, Masaharu Okano a,
Takamichi Tsutsumi a, Hirofumi Aboshi a
a
Department of Legal Medicine, Nihon University School of Dentistry, Tokyo, Japan
b
Division of Anatomical Science, Department of Functional Morphology, Nihon University School of Medicine, Tokyo, Japan

a r t i cl e i nfo a bstr ac t

Article history: Age estimation is crucial for reconstructing the biological profiles of deceased victims in the forensic field.
Received 2 July 2022 DNA methylation, which varies in an age-dependent manner in specific genes, is a candidate biomarker for
Received in revised form 17 August 2022 estimating chronological age. DNA methylation-based models for estimating age have been developed using
Accepted 28 August 2022
various technologies such as pyrosequencing. We recently quantified the methylation levels of elongation of
Available online 31 August 2022
very long chain fatty acids protein 2 (ELOVL2) in teeth using real-time methylation-specific polymerase
chain reaction (RT-MSP) to rapidly assess the methylation value of CpG sites within a CpG island. The
Keywords:
Age estimation methylation levels of ELOVL2 were moderately correlated with chronological age, suggesting the usefulness
CpG island of RT-MSP for age estimation. In this study, we designed eight and five new primer sets for ELOVL2 and
DNA methylation ectodysplasin A receptor-associated death domain (EDARADD), respectively, and selected the best primer
Teeth set. The DNA methylation level was analyzed in 59 tooth samples using the selected primer set. The ELOVL2
Real-time PCR methylation value was positively correlated with age (R2 = 0.50), whereas the EDARADD methylation value
negatively correlated with age (R2 = 0.44). A multiple regression model combining ELOVL2 and EDARADD
showed high accuracy [mean absolute error (MAE) = 6.69], which was verified using 40 test samples (MAE =
8.28). Additionally, the MAE of three age groups showed no significant difference. These results indicate that
the multiple regression model based on the two genes is useful for accurate age estimation across the
human lifespan.
© 2022 Elsevier B.V. All rights reserved.

1. Introduction DNA methylation plays an essential role in various biological

processes such as embryonic development, cellular differentiation,
Age estimation is fundamental for building the biological profile and gene expression regulation [6]. Cytosine methylation of CpG
of unidentified human remains recovered from different forensic islands upstream of specific genes is developmentally regulated in a
contexts and aid in individual identification from a list of potential tissue-specific manner [7,8]. Over the last decade, several groups
candidates. Several methods have been proposed for estimating age have independently evaluated the methylation levels of DNA ex­
based on biochemical markers, such as racemization of aspartic acid tracted from blood and demonstrated a high correlation between
in dentin [1], radiocarbon dating of tooth enamel [2], measurement DNA methylation levels and chronological age [9–16]. Some studies
of the telomere length [3], and determination of the number of focused on the methylation of dental DNA to estimate age
mitochondrial DNA mutations [4,5]. However, most of these [10,17–21]. For example, Bekaert et al. investigated the correlation
methods have not been widely used by the forensic science com­ between the methylation levels of seven CpGs in three genes
munity because of their low accuracy, complicated techniques, and/ [elongation of very long chain fatty acids protein 2 (ELOVL2), phos­
or limited applicability. phodiesterase 4C (PDE4C), and ectodysplasin A receptor-associated
death domain (EDARADD)] and chronological age using DNA ex­
tracted from 29 dentin samples and developed an age estimation
model (R2 = 0.74) with a mean absolute deviation (MAD) of 4.86
years between chronological and predicted ages [10]. Similarly, using
⁎
Correspondence to: Department of Legal Medicine, Nihon University School of 65 tooth samples, Márquez-Ruiz et al. developed an age estimation
Dentistry, 1–8-13 Kanda-Surugadai, Chiyoda-ku, Tokyo 101–8310, Japan.
model based on analysis of DNA methylation of nine CpGs in two
E-mail address: [email protected] (M. Kondo).

https://doi.org/10.1016/j.forsciint.2022.111445
0379-0738/© 2022 Elsevier B.V. All rights reserved.
A. Ogata, M. Kondo, M. Yoshikawa et al. Forensic Science International 340 (2022) 111445

genes (ELOVL2 and PDE4C) with a mean absolute error (MAE) of 5.04 that had been stored in a paper bag under dry conditions following
years [18]. Recently, Zapico et al. reported multivariate regression extraction and preserved in a container to avoid shrinkage due to
models for CpGs in ELOVL2, four and a half LIM domains 2 (FHL2), changes in temperatures and humidity. Samples with large caries,
neuronal pentraxin 2 (NPTX2), Krüppel-like factor 14 (KLF14), and metal prostheses, and root canal treatment, as well as split teeth,
secretagogin (SCGN) from 20 dental pulp samples of wisdom teeth were excluded.
with an MAE of 1.5–2.13 years [19]. Dias et al. reported an accurate
age prediction model for CpGs in ELOVL2 and KLF14, explaining 76.4% 2.2. DNA extraction and bisulfite conversion
of age variation with an MAD of 7.07 years [20]. Overall, various
combinations of genes have been used to develop an optimized age The tooth surface was cleaned with chlorine bleach to remove
estimation model. Some specific genes including ELOVL2 that show a stains. Tartar and soft tissue were carefully removed using a probe or
strong correlation with age estimation and wide range of changes in scaler. Further, DNA was extracted from each tooth (i.e., without
methylation during aging are promising marker candidates for age isolating any part of the tooth, or splitting or grinding of the tooth)
estimation. Thus, evaluation of DNA methylation has received con­ using a DNA extraction kit for hard tissue (TBONE EX Kit; DNA Chip
siderable attention from forensic scientists as a novel method for age Research, Inc., Tokyo, Japan) according to the manufacturer’s pro­
estimation [22]. tocol. The DNA was eluted with 50 µL of EB buffer (QIAamp DNA
Pyrosequencing is the most commonly used method in this field Mini Kit, Qiagen, Hilden, Germany) and quantified using a spectro­
of investigation [23]. A major advantage of pyrosequencing is that photometer (NanoDrop, Thermo Fisher Scientific, Waltham, MA,
quantitative DNA methylation data can be obtained with high ac­ USA). The DNA extracts were bisulfite-converted using an EZ DNA
curacy through direct sequencing of polymerase chain reaction (PCR) Methylation Kit (ZYMO Research, Irvine, CA, USA). The final con­
products [24]. However, this method has some limitations, such as centration was adjusted to 20 ng/µL, and 50 µL of elution buffer from
the low penetration rate of the pyrosequencer and high costs. the kit was used for elution.
Methylation-specific PCR (MSP), which was developed by
Herman et al. [25], is a cost-effective method for rapidly assessing 2.3. Percentage of methylated reference measured using RT-MSP
the methylation status of CpG sites within a CpG island. In particular,
MSP analysis can be performed using a conventional real-time PCR Methylation-specific primers were designed for CpG sites up­
machine. stream of each target gene (ELOVL2 and EDARADD) by referring to the
We recently used the real-time MSP (RT-MSP) method to quan­ MethPrimer website (http://www.urogene.org/cgi-bin/methprimer/
tify the methylation levels of three CpGs in ELOVL2 extracted from 29 methprimer.cgi); the primer sequences are listed in Table 2. The
tooth samples. The methylation levels of ELOVL2 showed a moderate primers targeting ELOVL2 and EDARADD contained six CpG sites
correlation with chronological age, suggesting that RT-MSP can be (ELOVL2: chr6:11044644, chr6:11044647, chr6:11044655,
applied in age estimation [26]. The aim of this study was to measure chr6:11044661, chr6:11044711, and chr6:11044727 of GRCh38/hg38;
the methylation levels of 2 genes, ELOVL2 and EDARADD, using RT- EDARADD: chr1:236348188, chr1:236348190, chr1:236348193,
MSP and to develop a multiple regression model for age estimation. chr1:236348272, chr1:236348276, and chr1:236348280 of GRCh38/
hg38). The human Alu sequence was used as the reference primer
2. Materials and methods [27,28]. The bisulfite-converted DNA was amplified using an Epi­
Scope® MSP Kit (TAKARA Bio, Shiga, Japan) and 0.3 µM primers. RT-
2.1. Sample selection MSP was performed as previously described [26]. Briefly, the mean
Ct value for each sample was calculated using a standard curve. The
Ninety-nine teeth extracted from Japanese individuals (aged standard solutions, 100% methylated human DNA (EpiScope® Me­
20–85 years); these teeth had been stored in the laboratory and thylated HCT116 gDNA, TAKARA Bio), and 100% unmethylated
were used in this study after obtaining approval from the Ethics human DNA (EpiScope® Unmethylated HCT116 DKO gDNA, Epi­
Committee of Nihon University School of Dentistry (approval Scope® Unmethylated HCT116 DKO gDNA, TAKARA Bio) were bi­
#EP19D007). sulfite-treated in the same manner as the sample, and the
Fifty-nine of the 99 teeth were selected as training samples, and methylation rates were adjusted to 0%, 10%, 25%, 50%, 75%, 90%, and
the remaining 40 samples were used as test samples. The sample 100%. PCR was performed for 45 cycles using a Thermal Cycler Dice
distribution according to age group is shown in Table 1. All samples Real Time System II (TP900; TAKARA Bio). The annealing tempera­
were permanent teeth (87 molars, 7 premolars, and 5 anterior teeth) tures were 55 ℃ for ELOVL2 and 60 ℃ for EDARADD. After the re­
action, the percentage of methylated reference (PMR) was calculated
Table 1 from the fluorescence level (relative quantity) of each sample using
Number of samples and sex distribution per age group. a: training sample (n = 59). b: the following equation:
test sample (n = 40). *One sample in the ≥ 70 group was of unknown sex.

Age (years) Frequency Men Women PMR (%) = {[(target gene†/Alu) mean value sample]/[(target gene†/
a Alu) mean value universal methylated human DNA]} × 100
20–29 14 6 8
30–39 12 6 6 † represents target ELOVL2 or EDARADD.
40–49 12 8 4
50–59 8 7 1
2.4. Statistical analysis
60–69 7 5 2
≥ 70 6* 1 4
Total 59 33 25 The correlations between PMR and chronological age were ex­
b amined using 59 teeth in the training sample, and regression
20–29 8 3 5
equations were obtained (Microsoft Excel 2019, Microsoft, WA, USA).
30–39 9 4 5
40–49 8 7 1 To compare the residuals of the regression equations among age
50–59 6 3 3 groups, the samples were divided into three age groups (young
60–69 5 4 1 adults: 20–34 years; middle-aged adults: 35–54 years; and older
≥ 70 4 3 1 adults: ≥55 years), and the MAE of each group was calculated and
Total 40 24 16
tested for significant differences between groups [one-way analysis

2
A. Ogata, M. Kondo, M. Yoshikawa et al. Forensic Science International 340 (2022) 111445

Table 2
Sequences of methylation-specific and reference primers used in methylation-specific polymerase chain reaction.

Target Primer Sequence (5′−3′)

ELOVL2 ELOVL2-F GCGGCGGTTTAACGTTTAC

ELOVL2-R CACGATACTACTTCTCCCCG
EDARADD EDARADD-F GTAGATGTTAGGCGCGGC
EDARADD-R CCCTACCTTACGATCGTCCG
Alu ALU-F GGTTAGGTATAGTGGTTTATATTTGTAATTTTAGTA
ALU-R ATTAACTAAACTAATCTTAAACTCCTAACCTCA

*Methylation specific primers: ELOVL2-F, ELOVL2-R, EDARADD-F, and EDARADD-R; reference primers: ALU-F and ALU-R.
†ELOVL2, elongation of very long-chain fatty acids protein 2; EDARADD, ectodysplasin A receptor-associated death domain.
Figure legends

of variance (ANOVA) with post-hoc Tukey–Kramer test in GraphPad 3.2. Validation of regression equations using the training sample
Prism 9 software (GraphPad, Inc., San Diego, CA, USA). Multiple re­
gression analysis was performed between chronological age and In each regression equation, we obtained MAEs of 9.59 years
PMR of ELOVL2 and EDARADD (Microsoft Excel 2019). The training (ELOVL2) and 10.12 years (EDARADD) for the training sample. In the
samples were divided into men and women, and multiple regression regression equations for the PMR of ELOVL2 and EDARADD, the re­
equations were obtained for each sex. The significance of the dif­ siduals were biased toward positive values for individuals in their
ference between the correlation coefficients was analyzed using 20 s and 30 s and toward negative values for those in their
VassarStats. Forty teeth were used to verify the accuracy of the 60 s (Fig. 2).
multiple regression equations. To compare the test sample residuals In the regression equations for ELOVL2, the largest residual was
from the multiple regression equation among age groups, the MAE of 12.67 years, which was observed in older adults (≥55 years), and the
each age group was calculated, and a significance test among age smallest residual was 8.02 years, which was observed in young
groups was performed (one-way ANOVA with post-hoc adults (20–34 years; Fig. 3a). In contrast, in the regression equations
Tukey–Kramer test, GraphPad Prism 9). Statistical significance was for EDARADD, the largest residual was 12.20 years, which was ob­
set at p < 0.05. served in older adults (≥55 years), and the smallest residual was 8.00
years, which was observed in middle-aged adults (35–54 years;
Fig. 3b). There was no significant difference between any of the age
3. Results
groups for ELOVL2 and EDARADD (one-way ANOVA with post-hoc
Tukey–Kramer test).
3.1. Correlation between age and PMR of ELOVL2 and EDARADD

The correlation between the PMR of the two genes (ELOVL2 and 3.3. Multiple regression analysis using the PMR of ELOVL2 and
EDARADD) and chronological age was assessed for the 59 training EDARADD and accuracy verification using a test sample
samples (Fig. 1). The PMR of ELOVL2 was strongly positively corre­
lated with age, whereas that of EDARADD was negatively correlated PMR values for ELOVL2 and EDARADD were used in a multivariate
with age. A quadratic regression was a better fit for the relationship regression model to obtain an age estimation formula. The multiple
between age and PMR of ELOVL2 (R2 = 0.50) than a linear (R2 = 0.42) regression equation was as follows: Y = −0.041X12 + 2.243X1 –
or logarithmic (R2 = 0.42) regression. For EDARADD, logarithmic re­ 10.815ln(X2) + 66.538 (R2 = 0.74; X1: PMR of ELOVL2, X2: PMR of
gression was a better fit (R2 = 0.44) than linear (R2 = 0.40) or EDARADD). Age was estimated using the multiple regression equa­
quadratic (R2 = 0.42) regression. The regression equations were as tion (Fig. 4). We obtained an MAE between the estimated and
follows: Y = −0.0784X2 + 3.3587X + 23.529 for ELOVL2 and Y chronological age of 6.69 years for the training samples (n = 59). The
= −13.91ln(X) + 92.259 for EDARADD. correlation between the PMR and chronological age did not

Fig. 1. Correlation between the percentage of methylated reference (PMR) and chronological age in the training samples (n = 59). a: PMR of ELOVL2 positively correlated with age.
b: PMR of EDARADD negatively correlated with age.

3
A. Ogata, M. Kondo, M. Yoshikawa et al. Forensic Science International 340 (2022) 111445

Fig. 4. Relationship between estimated and chronological age of the training and test
samples using multiple regression equations. In the training sample (n = 59), the
mean absolute deviation (MAE) was 6.69 years. In the test sample (n = 40), the MAE
was 8.28 years.

4. Discussion

Fig. 2. Residuals of the training sample (n = 59) using a simple regression equation. a: Over the last three decades, various molecular markers for age
Mean absolute deviation (MAE) was 9.59 years. b: MAE was 10.12 years. estimation have been identified for use in the forensic field.
Particularly, racemization of aspartic acid in a dental protein was
tested for forensic applications [1]. However, aspartic acid racemi­
significantly differ between men and women (two correlation zation is affected by temperature, which is a limitation of this ana­
coefficients test, Z = −1.18, P = 0.24, two-tailed). In addition, the lysis. Alternatively, DNA methylation is relatively stable at different
multiple regression equation was validated using a test sample temperatures, suggesting that it can be used to estimate the
(n = 40), for which we obtained an MAE of 8.28 years. The MAE was chronological age of burnt remains.
7.61 years for young adults (n = 14), 7.48 years for middle-aged Epigenetic age estimation using pyrosequencing [10,11,19], Epi­
adults (n = 15), and 10.21 years for older adults (n = 11) (Fig. 5). No Typer [17,29], SNaPshot [20,21,30,31], and methylation-sensitive
significant differences were observed between any age of the groups high-resolution melting (MS-HRM) [13] have been tested for prac­
(one-way ANOVA with post-hoc Tukey–Kramer test). tical applications in the forensic field [32].

Fig. 3. Mean absolute deviation (MAE) per age group for the training sample (n = 59) using the simple regression equation (young adults, middle-aged adults, and older adults;
n = 22, 21, and 16, respectively). a: No significant difference was observed in ELOVL2 between any age groups. b: No significant difference was observed in EDARADD between any
age groups.

4
A. Ogata, M. Kondo, M. Yoshikawa et al. Forensic Science International 340 (2022) 111445

The final primer sets for ELOVL2 and EDARADD each contained six
CpGs. Single regression analyses of individual genes revealed mod­
erate correlations (R2 = 0.50, ELOVL2; R2 = 0.44, EDARADD). Moreover,
the final multiple regression model combining six CpGs in ELOVL2
and six CpGs in EDARADD (i.e., 12 CpGs in total) showed a high
correlation (R2 = 0.74) and an MAE of 6.69 years.
In the narrow ELOVL2 upstream region, dozens of consecutive CpG
sites have been identified to be specifically correlated with age [11,38].
Interestingly, the CpG showing the highest correlation with age differed
in each analysis, even when the same tissue (blood) and same method
(pyrosequencing) were used. For example, Zbieć-Piekarska et al. [11]
found that two of the seven CpGs (chr6:11,044,642 and
chr6:11,044,634) were strongly correlated with chronological age; they
also developed an age estimation model (R2 = 0.859, MAD = 5.03) using
blood samples. In contrast, Bekaert et al. [10] showed that three CpGs
(chr6:11,044,661, chr6:11,044,640, and chr6:11,044,625) in ELOVL2
were correlated in their age estimation model, with chr6: 11,044,640 as
the most informative CpG in the blood sample. Park et al. [15] found a
strong correlation between age and the CpG (chr 6: 11,044,894) in
ELOVL2. Similar to the results observed in blood, CpG sites showing the
Fig. 5. Mean absolute deviation (MAE) per age group for the test sample (n = 40) using best correlation with age differed in each model in teeth [10,18,20].
a multiple regression equation. No statistically significant difference was observed
Bekaert et al. [10], Márquez-Ruiz et al. [18], and Dias et al. [20] used five
between any age groups (young adults, middle-aged adults, and older adults; n = 14,
15, and 11, respectively). CpGs (chr6:11,044,655, chr6:11,044,640, chr6:11,044,634,
chr6:11,044,628, and chr6:11,044,625), six CpGs (chr6:11,044,617;
chr6:11,044,631; chr6:11,044,642; chr6:11,044,644; chr6:11,044,649;
MS-HRM, a PCR-based method, enables rapid and relatively in­ chr6:11,044,661), and one CpG (chr6:11,044,628), respectively. The
expensive quantification of methylation levels. DNA methylation difference in the optimal CpG sites in each model may be attributed to
levels are estimated by comparing the melting profiles of an un­ the stochastic phenomenon of DNA methylation. In other words, the
known PCR product with those of a PCR product derived from a best CpG site may vary when samples are added to the analysis. Three
standard material with a known methylation ratio [33]. Recently, (chr6:11,044,655, chr6:11,044,647, chr6:11,044,644) of the six CpGs
Hamano et al. used MS-HRM to quantify the methylation levels of contained in the primer set for ELOVL2 used in this study were pre­
ELOVL2 and FHL2 in blood-derived DNA and developed a multiple viously used for age estimation analysis, suggesting the usefulness of
regression model with an MAD of 7.44 years [13]. the region containing nine CpGs.
RT-MSP, a PCR-based methylation analysis similar to MS-HRM, In contrast, a minimum of six CpGs in EDARADD has been used to
has been widely used in medical sciences to diagnose cancer and develop epigenetic age estimation models [10,36,37]. To the best of
some syndromes [25,34]. This method can also be used to rapidly our knowledge, the six CpGs in our EDARADD primers have not been
assess the methylation values of virtually any group of CpG sites previously used for age estimation analyses.
within a CpG island using methylation-specific primer sets for the Blood is relatively easy to obtain and has been used for DNA
sequence of interest. The greatest advantage of this method is that methylation-based age estimation [5]. Compared with blood, teeth
the analysis can be performed using a conventional real-time PCR are a stable source of DNA, even in severely damaged cadavers (e.g.,
machine, without requiring specific equipment such as a pyr­ highly decomposed corpses, burned corpses, and white skeletons).
osequencer. Therefore, this method can be used in primary screening The molars are the most protected teeth in the jaw, and thus are
for personal identification, such as following a large-scale disaster, useful for assessing age by analyzing their methylation levels.
where a large number of cadavers must be analyzed immediately. In this study, we used teeth that had not been crushed or split to
Here, we showed that RT-MSP can be used to quantify CpG methy­ extract DNA, which is required for rapid processing in the forensic
lation levels and estimate chronological age. field. Therefore, the origin of the DNA is expected to be from dentin
The selection of appropriate genes is crucial for developing a (odontoblasts), dental pulp (odontoblasts, fibroblasts, defense cells,
highly accurate age estimation model. To date, at least 10 genes and undifferentiated mesenchymal cells), and cementum (ce­
derived from teeth have been considered as targets for developing mentocytes) [39].
age estimation models [5,10,14,17–21]. ELOVL2 is the most promising Giuliani et al. divided the tooth into three parts, dentin, pulp, and
gene for creating an age estimation model because its methylation cementum, and built an age estimation model for each. The MAD
values are strongly positive correlated with age [10–12,15]. In addi­ between the estimated age and chronological age was better in the
tion, this correlation is highly conserved in multiple tissues [35]. In model from the pulp (MAD = 2.25) or cementum (MAD = 2.45)
contrast to ELOVL2 methylation, EDARADD methylation is negatively compared to that from the dentin (MAD = 7.07) [17]. Recently, Zapico
correlated with age [10,36,37]. EDARADD has also been used for et al. developed excellent multiple regression models (MAE =
methylation-based age estimation in tooth samples [10,21]. There­ 1.5 −2.13 years) by quantifying DNA methylation in dental pulp [19].
fore, we used ELOVL2 and EDARADD to develop an age estimation The results suggested that using dental pulp can provide the most
model. accurate estimate of age. Further studies are needed to quantify the
Eight and five sets of primers were designed for ELOVL2 and methylation levels of DNA extracted from dental pulp using RT-MSP.
EDARADD, respectively. Each primer contained 2–4 CpGs of 161 In contrast, when collecting DNA, a certain amount of time is
(ELOVL2) and 136 (EDARADD) CpG sites on CpG islands, with each required to divide the teeth. Given the rapid processing needs in the
primer set consisting of forward and reverse primers, containing 4–8 forensic field, such as following large-scale disasters, DNA extraction
CpGs with various combinations. The best primer set for each gene from the whole tooth is a suitable method and is used in our la­
was selected based on a single peak in the dissociation curve and boratory. In addition, as the small amount of DNA commonly found
correct calibration curve (R2 = 0.89, ELOVL2; R2 = 0.89, EDARADD). in forensic cases increases the margin of error when determining

5
A. Ogata, M. Kondo, M. Yoshikawa et al. Forensic Science International 340 (2022) 111445

DNA methylation levels, extracting DNA from the whole tooth may 2021) and the Sato Fund, Nihon University School of
increase the total amount of DNA collected. Dentistry (2020).
Furthermore, to develop a highly accurate regression model, it is
important to consider a sufficient sample size or population varia­ References
bility in each age group. However, in most previously reported ex­
periments, the number of tooth samples used was only 20–30, and [1] S. Ohtani, K. Yamamoto, Estimation of age from a tooth by means of racemiza­
tion of an amino acid, especially aspartic acid–comparison of enamel and dentin,
the sample distribution in each age group was not always uniform. J. Forensic Sci. 37 (4) (1992) 1061–1067 〈https://www.astm.org/jfs13292j.html〉.
In contrast, we used 99 samples (59 training samples and 40 test [2] K.L. Spalding, B.A. Buchholz, L.E. Bergman, H. Druid, J. Frisén, Forensics: age
samples) and observed a low level of sample variability in each age written in teeth by nuclear tests, Nature 437 (7057) (2005) 333–334, https://doi.
org/10.1038/437333a
group, indicating that our experimental conditions are suited for [3] A.O. Karlsson, A. Svensson, A. Marklund, G. Holmlund, Estimating human age in
developing an age estimation model. forensic samples by analysis of telomere repeats, Forensic Sci. Int. Genet. Suppl.
Both highly accurate age estimation models and homogeneous Ser. 1 (1) (2008) 569–571, https://doi.org/10.1016/j.fsigss.2007.10.153
[4] C. Meissner, N. von Wurmb, M. Oehmichen, Detection of the age-dependent
estimation accuracy across generations are important in the forensic
4977 bp deletion of mitochondrial DNA, A Pilot Study Int. J. Leg. Med. 110 (5)
field. Previous studies showed that age estimation accuracy de­ (1997) 288–291, https://doi.org/10.1007/s004140050089
creases in an age-dependent manner [10,18]. In the present study, no [5] A. Vidaki, B. Daniel, D.S. Court, Forensic DNA methylation profiling–potential
opportunities and challenges, Forensic Sci. Int. Genet. 7 (5) (2013) 499–507,
statistically significant difference was observed in the MAE of each
https://doi.org/10.1016/j.fsigen.2013.05.004
age group using the multiple regression model for age estimation [6] A.M. Deaton, A. Bird, CpG islands and the regulation of transcription, Genes Dev.
(Fig. 5). These results suggest that our model is stable across the 25 (10) (2011) 1010–1022, https://doi.org/10.1101/gad.2037511
human lifespan. Therefore, this method may be useful for identifying [7] D. Schübeler, Function and information content of DNA methylation, Nature 517
(7534) (2015) 321–326, https://doi.org/10.1038/nature14192
missing people (e.g., in criminal cases), which mostly includes adults [8] S. Horvath, DNA methylation age of human tissues and cell types, Genome Biol.
and the elderly. 14 (10) (2013) R115, https://doi.org/10.1186/gb-2013-14-10-r115
Age estimation, which is based on various biological or molecular [9] C.I. Weidner, Q. Lin, C.M. Koch, L. Eisele, F. Beier, P. Ziegler, D.O. Bauerschlag,
K.H. Jöckel, R. Erbel, T.W. Mühleisen, M. Zenke, T.H. Brümmendorf, W. Wagner,
characteristics including DNA methylation, is a scoring method used Aging of blood can be tracked by DNA methylation changes at just three CpG
as part of individual identification. The final personal identification sites, Genome Biol. 15 (2) (2014) R24, https://doi.org/10.1186/gb-2014-15-2-r24
is performed by comparing postmortem and antemortem dental [10] B. Bekaert, A. Kamalandua, S.C. Zapico, W. Van de Voorde, R. Decorte, Improved
age determination of blood and teeth samples using a selected set of DNA me­
charts and/or through DNA identification. In this study, we used RT- thylation markers, Epigenetics 10 (10) (2015) 922–930, https://doi.org/10.1080/
MSP to evaluate DNA methylation in tooth samples and developed a 15592294.2015.1080413
multiple regression model for age estimation. Our model showed a [11] R. Zbieć-Piekarska, M. Spólnicka, T. Kupiec, Z. Makowska, A. Spas, A. Parys-
Proszek, K. Kucharczyk, R. Płoski, W. Branicki, Examination of DNA methylation
slightly higher MAE and was less accurate than other age estimation
status of the ELOVL2 marker may be useful for human age prediction in forensic
methods. However, as described above, the method can be easily science, Forensic Sci. Int. Genet. 14 (2015) 161–167, https://doi.org/10.1016/j.
performed at low costs using conventional real-time PCR. Therefore, fsigen.2014.10.002
[12] R. Zbieć-Piekarska, M. Spólnicka, T. Kupiec, A. Parys-Proszek, Z. Makowska,
this method can be used in primary screening for personal identi­
A. Pałeczka, K. Kucharczyk, R. Płoski, W. Branicki, Development of a forensically
fication in situations such as a large-scale disasters, where large useful age prediction method based on DNA methylation analysis, Forensic Sci.
numbers of cadavers must be analyzed immediately. Int. Genet. 17 (2015) 173–179, https://doi.org/10.1016/j.fsigen.2015.05.001
[13] Y. Hamano, S. Manabe, C. Morimoto, S. Fujimoto, M. Ozeki, K. Tamaki, Forensic
age prediction for dead or living samples by use of methylation-sensitive high
5. Conclusions resolution melting, Leg. Med. (Tokyo). 21 (2016) 5–10, https://doi.org/10.1016/j.
legalmed.2016.05.001
[14] H.Y. Lee, S.D. Lee, K.J. Shin, Forensic DNA methylation profiling from evidence
We assessed the methylation of ELOVL2 and EDARADD in 59 tooth material for investigative leads, BMB Rep. 49 (7) (2016) 359–369, https://doi.org/
samples using RT-MSP and developed a multiple regression model, 10.5483/bmbrep.2016.49.7.070
the MAE value of which was 6.69 years. The accuracy of the age [15] J.L. Park, J.H. Kim, E. Seo, D.H. Bae, S.Y. Kim, H.C. Lee, K.M. Woo, Y.S. Kim,
Identification and evaluation of age-correlated DNA methylation markers for
estimation model was verified using an additional 40 extracted forensic use, Forensic Sci. Int. Genet. 23 (2016) 64–70, https://doi.org/10.1016/j.
teeth; the obtained MAE value was 8.28 years, which was close to fsigen.2016.03.005
that obtained for the training samples. This result indicates that our [16] Z. Thong, X. Liang Shun Chan, J. Ying Ying Tan, E. Shuzhen Loo, C. Kiu ChoongSyn,
Evaluation of DNA methylation-based age prediction on blood, Forensic Sci. Int.
multiple regression model can be used for dental age estimation. Genet 6 (Suppl. Ser) (2017) e249–e251, https://doi.org/10.1016/J.FSIGSS.2017.09.
095
[17] C. Giuliani, E. Cilli, M.G. Bacalini, C. Pirazzini, M. Sazzini, G. Gruppioni,
CRediT authorship contribution statement
C. Franceschi, P. Garagnani, D. Luiselli, Inferring chronological age from DNA
methylation patterns of human teeth, Am. J. Phys. Anthropol. 159 (4) (2016)
Ayano Ogata: Investigation, Formal analysis, Writing – original 585–595, https://doi.org/10.1002/ajpa.22921
draft, Writing – review & editing, Visualization. Masahiro Kondo: [18] A.B. Márquez-Ruiz, L. González-Herrera, J.D. Luna, A. Valenzuela, DNA methy­
lation levels and telomere length in human teeth: usefulness for age estimation,
Conceptualization, Formal analysis, Writing – original draft, Writing Int. J. Leg. Med. 134 (2) (2020) 451–459, https://doi.org/10.1007/s00414-019-
– review & editing, Project administration, Funding acquisition. 02242-7
Masaaki Yoshikawa: Formal analysis, Writing – review & editing. [19] S.C. Zapico, Q. Gauthier, A. Antevska, B.R. McCord, Identifying methylation pat­
terns in dental pulp aging: application to age-at-death estimation in forensic
Masaharu Okano: Writing – reviewing and editing. Takamichi anthropology, Int. J. Mol. Sci. 22 (7) (2021) 3717, https://doi.org/10.3390/
Tsutsumi: Writing – review & editing. Hirofumi Aboshi: ijms22073717
Supervision, Writing – review & editing. [20] H.C. Dias, F. Corte-Real, E. Cunha, L. Manco, DNA methylation age estimation
from human bone and teeth, Aust. J. Forensic Sci. 54 (2) (2022) 163–176, https://
doi.org/10.1080/00450618.2020.1805011
Conflicts of interest [21] H.C. Dias, L. Manco, F. Corte-Real, E. Cunha, A blood-bone-tooth model for age
prediction in forensic contexts, Biol. (Basel) 10 (12) (2021) 1312, https://doi.org/
10.3390/biology10121312
The authors have no conflicts of interest to declare. [22] W. Parson, Age estimation with DNA: From forensic DNA fingerprinting to for­
ensic (epi)genomics: a mini-review, Gerontology 64 (4) (2018) 326–332, https://
doi.org/10.1159/000486239
Acknowledgments [23] A. Freire-Aradas, C. Phillips, M.V. Lareu, Forensic individual age estimation with
DNA: From initial approaches to methylation tests, Forensic, Sci. Rev. 29 (2)
(2017) 121–144.
We would like to thank Editage (www.editage.com) for English
[24] J. Tost, I.G. Gut, Analysis of gene-specific DNA methylation patterns by pyr­
language editing. This work was supported in part by grants from the osequencing technology, Methods Mol. Biol. 373 (2007) 89–102, https://doi.org/
Dental Research Center, Nihon University School of Dentistry (2020, 10.1385/1-59745-377-3:89

6
A. Ogata, M. Kondo, M. Yoshikawa et al. Forensic Science International 340 (2022) 111445

[25] J.G. Herman, J.R. Graff, S. Myöhänen, B.D. Nelkin, S.B. Baylin, Methylation-specific [32] A. Vidaki, M. Kayser, Recent progress, methods and perspectives in forensic
PCR: a novel PCR assay for methylation status of CpG islands, Proc. Natl. Acad. epigenetics, Forensic Sci. Int. Genet. 37 (2018) 180–195, https://doi.org/10.1016/j.
Sci. U. S. A. 93 (18) (1996) 9821–9826, https://doi.org/10.1073/pnas.93.18.9821 fsigen.2018.08.008
[26] M. Kondo, H. Aboshi, M. Yoshikawa, A. Ogata, R. Murayama, M. Takei, S. Aizawa, [33] T.K. Wojdacz, A. Dobrovic, L.L. Hansen, Methylation-sensitive high-resolution
A newly developed age estimation method based on CpG methylation of teeth- melting, Nat. Protoc. 3 (12) (2008) 1903–1908, https://doi.org/10.1038/nprot.
derived DNA using real-time methylation-specific PCR, J. Oral. Sci. 63 (1) (2020) 2008.191
54–58, https://doi.org/10.2334/josnusd.20-0138 [34] Y. Pan, G. Liu, F. Zhou, B. Su, Y. Li, DNA methylation profiles in cancer diagnosis
[27] D.J. Weisenberger, M. Campan, T.I. Long, M. Kim, C. Woods, E. Fiala, M. Ehrlich, and therapeutics, Clin. Exp. Med. 18 (1) (2018) 1–14, https://doi.org/10.1007/
P.W. Laird, Analysis of repetitive element DNA methylation by MethyLight, s10238-017-0467-0
Nucleic Acids Res 33 (21) (2005) 6823–6836, https://doi.org/10.1093/nar/gki987 [35] R.C. Slieker, C.L. Relton, T.R. Gaunt, P.E. Slagboom, B.T. Heijmans, Age-related
[28] T. Niwa, T. Tsukamoto, T. Toyoda, A. Mori, H. Tanaka, T. Maekita, M. Ichinose, DNA methylation changes are tissue-specific with ELOVL2 promoter methylation
M. Tatematsu, T. Ushijima, Inflammatory processes triggered by Helicobacter as exception, Epigenetics Chromatin 11 (1) (2018) 25, https://doi.org/10.1186/
pylori infection cause aberrant DNA methylation in gastric epithelial cells, s13072-018-0191-3
Cancer Res 70 (4) (2010) 1430–1440, https://doi.org/10.1158/0008-5472.CAN- [36] S. Bocklandt, W. Lin, M.E. Sehl, F.J. Sánchez, J.S. Sinsheimer, S. Horvath, E. Vilain,
09-2755 Epigenetic predictor of age, PLoS One 6 (6) (2011) e14821, https://doi.org/10.
[29] S. Kunze, Quantitative region-specific DNA methylation analysis by the 1371/journal.pone.0014821
EpiTYPER™ technology, Methods Mol. Biol. 2018 (1708) 515–535, https://doi. [37] A. Freire-Aradas, C. Phillips, L. Girón-Santamaría, A. Mosquera-Miguel,
org/10.1007/978-1-4939-7481-8_26 A. Gómez-Tato, M.A. Casares de Cal, J. Álvarez-Dios, M.V. Lareu, Tracking age-
[30] H.Y. Lee, S.E. Jung, Y.N. Oh, A. Choi, W.I. Yang, K.J. Shin, Epigenetic age signatures correlated DNA methylation markers in the young, Forensic Sci. Int. Genet. 36
in the forensically relevant body fluid of semen: a preliminary study, Forensic (2018) 50–59, https://doi.org/10.1016/j.fsigen.2018.06.011
Sci. Int. Genet. 19 (2015) 28–34, https://doi.org/10.1016/j.fsigen.2015.05.014 [38] P. Sukawutthiya, T. Sathirapatya, K. Vongpaisarnsin, A minimal number CpGs of
[31] S.E. Jung, S.M. Lim, S.R. Hong, E.H. Lee, K.J. Shin, H.Y. Lee, DNA methylation of the ELOVL2 gene for a chronological age estimation using pyrosequencing, Forensic
ELOVL2, FHL2, KLF14, C1orf132/MIR29B2C, and TRIM59 genes for age prediction Sci. Int. 318 (2021) 110631, https://doi.org/10.1016/j.forsciint.2020.110631
from blood, saliva, and buccal swab samples, Forensic Sci. Int. Genet. 38 (2019) [39] D. Higgins, J.J. Austin, Teeth as a source of DNA for forensic identification of
1–8, https://doi.org/10.1016/j.fsigen.2018.09.010 human remains: a review, Sci. Justice 53 (4) (2013) 433–441, https://doi.org/10.
1016/j.scijus.2013.06.001