Dentine Metabolomics for Forensic Identification: A Pilot Study of the ¹H-NMR Approach to Postmortem Cancer Detection

Abstract

Background: Reliable identification remains a cornerstone of forensic investigations, particularly when encountering degraded remains or suboptimal biological evidence. This study evaluates the potential of dentine metabolomics, utilizing proton nuclear magnetic resonance (¹H-NMR) spectroscopy, to detect cancer-associated metabolic signatures in dental tissues for forensic applications. Methods: Forty-four non-carious second molars were analyzed, comprising 22 samples from deceased individuals with a documented history of cancer and 22 age- and sex-matched controls. Metabolomic profiling was conducted using ¹H-NMR spectroscopy to identify and quantify dentine metabolites. Statistical evaluation included unsupervised principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA), receiver operating characteristic (ROC) curve analysis, and exploratory binary logistic regression. Results: Among the 209 identified metabolites, inosinic acid and 2-ketobutyric acid were identified as the most robust discriminative biomarkers across both multivariate and univariate frameworks. The exploration within-sample predictive model achieved a Nagelkerke R² of 0.822 and an overall classification accuracy of 90.9%, with a specificity of 95.5% and a sensitivity of 86.4%. These key metabolites are fundamentally associated with purine metabolism and oxidative stress pathways frequently dysregulated in oncogenesis. Conclusions: This pilot study suggests that dentine may retain metabolomic information associated with cancer comorbidity under heterogeneous postmortem conditions. However, the findings remain exploratory and require validation in larger cohorts with standardized postmortem variables before practical forensic implementation.

1. Introduction

The identification of individuals is a crucial step in legal and forensic processes, playing a crucial role in determining the identity of deceased persons such as identifying victims in mass casualty events and verifying the age of offenders [1]. While identification is relatively straightforward when a body remains intact and unaffected by decomposition or trauma, the process becomes significantly more challenging when only skeletal remains are available due to advanced decomposition or severe injuries. In such cases, the use of forensic anthropology knowledge to estimate characteristics like sex, ethnicity, age, and height is imperative [2]. However, these methods often provide broad data ranges, which may not be sufficient for precise individual identification. Identifying chronic conditions such as pathological changes from cancer could significantly enhance the specificity and reliability of the identification process.

Cancer is the leading cause of death in Thailand and represents a significant opportunity to enhance forensic identification using disease-specific biochemical markers. According to data from the National Cancer Institute, more than 140,000 new cancer cases are diagnosed annually, with approximately 84,000 cancer-related deaths each year. The most prevalent and fatal cancers among the Thai population include liver and bile duct cancer, lung cancer, breast cancer, cervical cancer, and leukemia [3]. These common forms of cancer underscore the significant health burden they impose on the population while also highlighting the potential utility of cancer-related biochemical markers in forensic applications. By harnessing these markers, researchers and forensic practitioners can develop more precise and reliable methods for identifying individuals in both clinical and forensic investigations.

Metabolomics enables the comprehensive profiling of low-molecular-weight metabolites that reflect physiological and pathological states [4,5]. Among analytical platforms, proton nuclear magnetic resonance ¹H-NMR spectroscopy is increasingly utilized in forensics due to its inherent non-destructive nature, high reproducibility, and minimal sample preparation [6,7]. Recent reviews have emphasized that NMR metabolomics is an essential tool for forensic science, offering unique capabilities in investigating biological traces, estimating the postmortem interval (PMI), and identifying metabolic patterns associated with specific conditions of forensic interest [8]. Unlike mass spectrometry-based methods, which require destructive ionization, ¹H-NMR allows for the preservation of limited forensic specimens for potential re-analysis or complementary examinations.

In forensic metabolomics, NMR-based approaches have predominantly been applied to biofluids such as blood, urine, vitreous humor, and cerebrospinal fluid [8,9,10,11]. Recent validation studies have further explored the use of pericardial fluid for PMI estimation [9] and addressed the complexities of translating metabolomic evidence from animal models to real human forensic scenarios [10]. In contrast, solid mineralized matrices such as bone and teeth remain underexplored despite their superior resistance to decomposition and environmental degradation.

Despite their structural robustness and resistance to decomposition, metabolomic studies on bones and teeth are relatively limited. Teeth, particularly dentine, are well-suited for forensic investigation due to their mixed organic–inorganic composition and low water content. Dentine consists of approximately 70% inorganic material, 20% organic matter, and 10% water [12], characteristics that may support the preservation of metabolic information under adverse postmortem conditions.

The present study is designed as an exploratory pilot investigation applying ¹H-NMR-based dentine metabolomics to evaluate whether discriminative metabolic signals associated with cancer comorbidity can be detected in real-world post-mortem forensic samples. Given the modest sample size and heterogeneous cancer types, the findings are intended to be hypothesis-generating rather than definitive diagnostic evidence.

2. Materials and Methods

2.1. Chemicals

Hydrochloric acid (HCl) was obtained from Merck (Darmstadt, Germany). Deuterium oxide (D₂O) and 3-(trimethylsilyl) propanoic acid sodium salt (TSP) were obtained from Sigma-Aldrich (Oakville, ON, Canada).

2.2. Dentine Sample Collection

The cohort utilized in this study has been described in our preliminary report [13]. Briefly, a total of 44 non-carious permanent second molars were collected (22 from deceased individuals with confirmed cancer diagnoses and 22 from healthy controls). All subjects were 20 years old or older. Samples were obtained from the Department of Forensic Medicine, Faculty of Medicine, Chiang Mai University, under ethical approval (FOR-2566-0323).

The postmortem interval (PMI) of the samples ranged from hours to months; however, the precise PMI data were not consistently available for all cases. In routine forensic casework, the exact time of death is frequently unavailable or unreliable, particularly in cases of unattended death, delayed body discovery, or secondary specimen referral from external facilities [14]. Consequently, PMI could not be incorporated as a stratification variable in the present analysis.

To minimize additional postmortem metabolic variation after sample acquisition, all teeth were promptly cleaned with 0.9% normal saline solution and distilled water to eliminate the remaining blood and stored at −20 °C upon receipt [15]. Nevertheless, PMI related effects prior to sample collection cannot be excluded and are acknowledged as an inherent limitation of real-world forensic metabolomic investigations.

2.3. Dentine Preparation

The dentin samples were washed with saline and distilled water. The tooth samples were horizontally cut into 1 mm thick slices using a precision saw model Isomet 1000 with a diamond blade (Lake Bluff, IL, USA). Subsequently, the finished samples were treated to remove the tooth enamel, using a high-speed diamond drill bit. Both steps of sample preparation were carried out under coolant conditions to prevent heat generation [16].

2.4. ¹H-NMR Spectroscopy

Dentine samples were pulverized using a freezing mill (SPEX CertiPrep 6750 Freezer/Mill, Metuchen, NJ, USA) for 15 min. Approximately 50 mg of dentine powder was digested with 0.6 M HCl at 100 °C for 24 h. The digested solution was lyophilized, reconstituted with 0.1 M TSP in D₂O, and transferred into NMR tubes.

¹H-NMR spectra were acquired at 27 °C on a Bruker AVANCE 500 MHz spectrometer (Rheinstetten, Germany) using a Carr–Purcell–Meiboom–Gill (CPMG) pulse sequence with water suppression. A total of 16 scans were collected per sample. Spectral processing, including phase and baseline correction, was performed using Bruker TopSpin software (version 4.0.7).

2.5. Peak Assignment and Untargeted Metabolite Identification

Metabolite identification was performed using the Human Metabolome Database (HMDB) and the previously published literature [17]. Chemical shift values, coupling constants, and signal multiplicity were analyzed using Bruker TopSpin software (version 4.0.7).

2.6. Data Analysis and Statistical Methods

Chemical compounds were identified using HMDB based on chemical shift matching within ±0.02 ppm. Metabolite concentrations were calculated using the following equation [18]:

$${\mathbf{C}}_{\mathbf{A}}={\displaystyle \frac{{\mathbf{I}}_{\mathbf{A}}}{{\mathbf{I}}_{\mathbf{T}\mathbf{S}\mathbf{P}}}}\times {\displaystyle \frac{{\mathbf{H}}_{\mathbf{T}\mathbf{S}\mathbf{P}}}{{\mathbf{H}}_{\mathbf{A}}}}\times {\mathbf{C}}_{\mathbf{T}\mathbf{S}\mathbf{P}}$$

C_A represents the concentration of the substance.

I_A denotes the intensity value of the substance.

I_TSP signifies the intensity value of TSP.

H_TSP corresponds to the hydrogen atom of TSP.

H_A refers to the hydrogen atom of the substance.

C_TSP represents the concentration of TSP.

Here, C_A is the compound concentration, I_A and I_TSP are the intensities, H_A and H_TSP are the hydrogen atom counts, and C_TSP is the TSP concentration.

2.7. Statistical Analysis for Metabolomic Data

Statistical analysis was conducted using MetaboAnalyst version 6.0. Data were median-normalized, log-transformed (base 10), and auto-scaled prior to analysis. Unsupervised principal component analysis (PCA) was first performed to explore the intrinsic data structure and identify potential outliers prior to supervised modeling [19].

Partial least-squares discriminant analysis (PLS-DA) was subsequently applied to assess metabolic differences between groups. Model performance was evaluated using R² and Q² values obtained through 10-fold cross-validation and permutation testing. Metabolites with VIP scores > 1 were considered contributors to group separation. Volcano plots were generated using absolute fold-change criteria (FC > 2 or FC < 0.5) with p < 0.05. ROC analysis identified metabolites with an AUC > 0.8 as potential candidates [20].

Binary logistic regression was performed as an exploration within simple analysis using IBM SPSS Statistics version 22. The forward stepwise (conditional) method was applied (entry p < 0.05; removal p > 0.10). Given the limited sample size and absence of external validation, regression results were interpreted as exploration and hypothesis-generating rather than definitive predictive performance.

3. Results

3.1. Demographic Data of Participants

A total of 44 cases were analyzed, comprising a healthy control group (n = 22) and a cancer group (n = 22). The cancer group included heterogeneous diagnoses, with unspecified cancer being the most common (n = 4), followed by liver cancer (n = 3). Other cancer types, including hypopharyngeal, pancreatic, buccal, and adrenal gland cancers, were each represented by one case (Figure 1). Diagnoses were confirmed using available medical records and family information. Both groups showed identical sex distributions (12 females and 10 males per group) and comparable mean ages (controls: 54.6 ± 15.0 years; cancer: 55.7 ± 15.7 years), with no statistically significant differences in age or sex (p > 0.05), minimizing demographic confounding effects.

3.2. Unsupervised Principal Component Analysis (PCA)

Unsupervised principal component analysis (PCA) was first performed to explore intrinsic metabolic patterns and detect potential outliers prior to supervised modeling. PCA demonstrated partial separation between cancer and control groups along PC1, which explained 37.1% of the total variance, while PC2 accounted for 10.1% of the variance. Substantial overlap between groups was observed, and no extreme outliers were detected, supporting the exploratory nature of subsequent supervised analyses [14].

The corresponding PCA score plot is shown in Supplementary Figure S1.

3.3. Identification of Cancer-Associated Metabolites by PLS-DA, Volcano Plot, and ROC Analysis

Multivariate datasets were analyzed using Partial Least Squares Discriminant Analysis (PLS-DA) (Figure 2). The first component (PLS-DA Component 1) explained 36.7% of the variance, while Component 2 explained 8.9%. The model revealed partial separation between cancer and control groups. A total of 91 metabolites exhibited variable importance in projection (VIP) scores greater than 1. For clarity and to avoid redundancy, only VIP scores from Component 1 are presented in the main manuscript (

Table 1. Variable importance in projection (VIP) scores of the top 15 discriminant metabolites across principal components.

Metabolites	VIP (Component 1)
Cinnabarinic acid	1.8554
O−phosphotyrosine	1.8373
N−acetyl−L−phenylalanine	1.8314
1−methylhistidine	1.8255
Adenine	1.8191
Deoxyinosine	1.8109
N4−acetylcytidine	1.7956
7−methylguanine	1.7944
Inosinic acid	1.7829
3−methyladenine	1.7663
L−kynurenine	1.755
9−methyladenine	1.7525
L−tryptophan	1.7398
5−HIAA	1.737
Mercaptopurine	1.711

), while additional components are provided in Supplementary Table S2.

The number of metabolites identified in volcano plot analysis reflects the application of combined statistical significance (p < 0.05) and fold-change thresholds (FC > 2 for upregulation or FC < 0.5 for downregulation). Accordingly, both increased and decreased metabolites were considered in subsequent integrative analyses, accounting for the apparent discrepancy between visual density in the volcano plot and the total number of metabolites meeting the selection criteria. In addition, point overlap (overplotting) in dense regions of the volcano plot may visually underestimate the number of metabolites meeting the statistical thresholds.

Among the 209 detected metabolites, 22 metabolites were upregulated in the cancer group (FC > 2), while 41 metabolites were downregulated (FC < 0.5), as illustrated in the volcano plot (Figure 3). The complete list of metabolites meeting the volcano plot criteria is provided in Supplementary Table S1. Univariate ROC analysis identified 56 metabolites with an AUC greater than 0.80. The top 15 discriminatory metabolites are summarized in

Table 2. ROC curve analysis of differential dentine metabolites between cancer and control groups.

Metabolite	AUC (ROC)	p-Value	Fold Change (log2)
O-phosphotyrosine	0.94215	1.81 × 10−9	1.9414
N-acetyl-L-phenylalanine	0.94215	2.19 × 10−9	2.6149
N4-acetylcytidine	0.93595	6.60 × 10−9	2.6602
9-methyladenine	0.92769	2.26 × 10−8	2.6915
1-methylhistidine	0.92562	2.65 × 10−9	2.5540
5-HIAA	0.92149	3.43 × 10−8	1.6964
Cinnabarinic acid	0.91942	1.00 × 10−9	2.5317
Inosinic acid	0.90909	9.57 × 10−9	2.1864
Adenine	0.90702	3.23 × 10−9	2.6200
L-kynurenine	0.90289	2.11 × 10−8	2.1936
7-methylguanine	0.90289	6.83 × 10−9	2.5775
Deoxyinosine	0.89669	4.17 × 10−9	2.6345
Mercaptopurine	0.89463	6.77 × 10−8	2.2505
L-tryptophan	0.89256	3.18 × 10−8	1.7726
3-methyladenine	0.89050	1.54 × 10−8	2.6304

Receiver operating characteristic (ROC) analysis demonstrating the diagnostic performance of the 15 most significantly altered dentine metabolites between cancer (n = 22) and control (n = 22) groups. Metabolites with higher area under the curve (AUC) values exhibit stronger discriminatory power. Among these, O-phosphotyrosine, N-acetyl-L-phenylalanine, N⁴-acetylcytidine, cinnabarinic acid, and 1-methylhistidine showed the highest discriminative performance with an AUC > 0.92 and extremely low p-values (<1 × 10⁻⁸), indicating robust statistical significance.

.

3.4. Identification of Potential Biomarkers Using Integrated Selection Criteria

Potential cancer-associated metabolites were identified using an integrative approach combining multivariate and univariate analyses. Selection criteria included VIP > 1 (PLS-DA), a statistically significant fold change (FC > 2 or FC < 0.5; p < 0.05) from the volcano plot analysis, and an AUC > 0.80 from ROC analysis. The Venn diagram (Figure 4) illustrates the overlap among these methods, identifying 36 metabolites that met all criteria. These candidates included both upregulated (

Table 3. (A) Upregulated metabolites in cancer (FC > 2, p < 0.05, VIP > 1, AUC > 0.80). (B) Downregulated metabolites in cancer (FC < 0.5, p < 0.05, VIP > 1, AUC > 0.80).

A
Metabolite	HMDB ID	VIP	FC (Cancer/Control)	p-Value	AUC
D-serine	HMDB0003406	1.2557	87.923	2.87 × 10−4	0.8636
2-Ketobutyric acid	HMDB0000005	1.0800	9.8283	0.0023	0.8554
Mevalonic acid	HMDB0000227	1.4755	10.941	1.00 × 10−5	0.8161
L-glutamic gamma-semialdehyde	HMDB0002104	1.5092	4.3139	5.44 × 10−6	0.8306
Aminoacetone	HMDB0002134	1.4510	6.7163	1.53 × 10−5	0.8099
Carnosine	HMDB0000033	1.4427	4.7426	1.76 × 10−5	0.8037
L-methionine	HMDB0000696	1.3356	3.4465	9.49 × 10−5	0.8182
Citric acid	HMDB0000094	1.2570	3.5881	2.82 × 10−4	0.8161
D-glutamine	HMDB0003423	1.1504	2.6026	0.00105	0.8161
N-acetylputrescine	HMDB0002064	1.0200	2.3701	0.00420	0.8037
B
O-phosphotyrosine	HMDB0011185	1.8373	0.4268	1.81 × 10−9	0.9422
N-acetyl-L-phenylalanine	HMDB0000512	1.8314	0.2846	2.19 × 10−9	0.9422
N4-acetylcytidine	HMDB0005923	1.7956	0.2880	6.60 × 10−9	0.9359
9-methyladenine	HMDB0140713	1.7525	0.2711	2.26 × 10−8	0.9277
1-methylhistidine	HMDB0000001	1.8255	0.3029	2.65 × 10−9	0.9256
5-HIAA	HMDB0000763	1.7370	0.4686	3.43 × 10−8	0.9215
Cinnabarinic acid	HMDB0004078	1.8554	0.2941	1.00 × 10−9	0.9194
Inosinic acid	HMDB0000175	1.7829	0.3816	9.57 × 10−9	0.9091
Adenine	HMDB0000034	1.8191	0.3049	3.23 × 10−9	0.9070
L-kynurenine	HMDB0000684	1.7550	0.3416	2.11 × 10−8	0.9029
7-methylguanine	HMDB0000897	1.6561	0.3165	6.83 × 10−9	0.9029
Deoxyinosine	HMDB0000071	1.8109	0.3045	4.17 × 10−9	0.8967
Mercaptopurine	HMDB0014958	1.7110	0.3528	6.77 × 10−8	0.8946
L-tryptophan	HMDB0000929	1.7398	0.4241	3.18 × 10−8	0.8926
3-methyladenine	HMDB0011600	1.7663	0.3130	1.54 × 10−8	0.8905
7-methyladenine	HMDB0011614	1.6561	0.4481	2.57 × 10−7	0.8843
Dermatan sulfate	HMDB0000632	1.1369	0.1949	0.00123	0.8698
2-aminobenzoic acid	HMDB0001123	1.6485	0.4430	3.07 × 10−7	0.8657
L-glutamic acid 5-phosphate	HMDB0001128	1.4229	0.3583	2.45 × 10−5	0.8554
Hydroxypropyl-threonine	HMDB0028873	1.4404	0.3778	1.84 × 10−5	0.8554
Adenylsuccinic acid	HMDB0000536	1.5273	0.4577	3.87 × 10−6	0.8450
Phosphoserine	HMDB0000272	1.3837	0.3187	4.60 × 10−5	0.8409
L-cysteine	HMDB0000574	1.4304	0.3270	2.17 × 10−5	0.8326
Phosphocreatine	HMDB0001511	1.3781	0.3404	5.01 × 10−5	0.8285
Norophthalmic acid	HMDB0005776	1.4405	0.3002	1.83 × 10−5	0.8161
L-aspartic acid	HMDB0000191	1.0072	0.3414	0.00476	0.8120

A) and downregulated metabolites (Table 3B), addressing metabolic alterations in both directions.

The selection thresholds applied in this study (VIP > 1, FC > 2 or < 0.5, and AUC > 0.80) were chosen as commonly adopted heuristic criteria in exploratory metabolomics studies to prioritize features with consistent multivariate contribution, substantial effect size, and acceptable discriminatory performance. These thresholds were not intended to define definitive biomarkers but rather to reduce dimensionality and identify candidates for further validation in independent cohorts.

3.5. Exploratory Logistic Regression Analysis

Exploratory binary logistic regression was applied to evaluate the combined discriminatory potential of the 36 candidate metabolites. Using a forward stepwise approach, two metabolites—inosinic acid and 2-ketobutyric acid—were retained in the final model. The model achieved a Nagelkerke R² of 0.822 and a within-sample classification accuracy of 90.9%. The overall classification accuracy of 90.9% was derived from the correct classification of 41 out of 44 samples (20/22 cancer cases and 21/22 control cases). This accuracy value represents a weighted integration of a sensitivity of 86.4% (the true positive rate for identifying cancer cases) and a specificity of 95.5% (the true negative rate for identifying control cases). Together, these metrics provide a comprehensive measure of the model’s discriminative power within the current study population. Detailed regression outputs are provided in Supplementary Table S3.

It is essential to clarify that these metrics reflect within-sample performance, representing the proportion of correctly classified samples within the discovery dataset used for model construction. While these results provide valuable descriptive insights into the model’s behavior facilitated by the balanced group sizes, they remain exploratory in nature. Consequently, these values should not be interpreted as evidence of generalizable predictive performance in external populations. Given the pilot nature of this study, the modest sample size, and the absence of independent external validation, our findings should be considered primarily as hypothesis-generating rather than definitive diagnostic criteria.

4. Discussion

This exploratory pilot study demonstrates the potential of ¹H-NMR-based metabolomics in identifying cancer-associated biochemical signatures within the dentine matrix. Dentine’s inherent durability and resistance to decay make it a promising biological archive for preserving metabolic information, particularly in forensic scenarios where traditional biofluids are degraded or unavailable. By analyzing a cohort of 44 individuals, inosinic acid and 2-ketobutyric acid were identified as significant discriminant metabolites. These candidates were integrated into an exploratory binary logistic regression model that achieved an overall classification accuracy of 90.9%, with 86.4% for cancer cases and 95.5% for non-cancer controls.

Inosinic acid plays a critical role in purine metabolism, supporting DNA and RNA synthesis and energy production—processes frequently dysregulated in cancer. Previous studies have linked its related compound, inosine, to cancer progression, metastasis, and treatment response. For instance, Lee et al. (2021) [21] demonstrated that inosine predicts the metastatic potential of lung squamous cell carcinoma, while Li et al. (2021) [22] reported that elevated inosine levels are associated with esophageal squamous cell carcinoma progression. In bladder cancer [23] and hepatocellular carcinoma [24,25], inosine has been used to distinguish between disease grades and aid in early detection. Interestingly, inosine’s behavior varies across cancer types; in pancreatic cancer, patients exhibit lower inosine levels, which can increase with dietary interventions such as a low-carbohydrate ketogenic diet [26]. These findings emphasize the context-dependent roles of inosinic acid and inosine, influenced by tumor type and systemic metabolic states.

Similarly, 2-ketobutyric acid, a key intermediate in the catabolism of threonine and methionine, contributes to amino acid metabolism and cellular energy production through its role in the tricarboxylic acid (TCA) cycle. [27] This study highlights its potential as a biomarker, particularly in cancer contexts where oxidative stress and amino acid metabolism are disrupted. Elevated levels of related metabolites, such as 2-hydroxybutyric acid, have been observed in lung and colorectal cancers, supporting the relevance of these pathways in tumorigenesis. Dysregulated levels of 2-ketobutyric acid in KRAS-mutant pancreatic cancer further underscore its diagnostic potential and role in metabolic reprogramming [28].

While the present study highlights its potential relevance, observed changes may also reflect generalized physiological stress rather than cancer-specific effects. Cancer is not a single entity but comprises hundreds of distinct diseases, each with unique metabolic reprogramming strategies [29]. This diversity extends even within a single malignancy, such as the distinct metabolomic signatures of luminal A and basal-like breast cancer subtypes [30]. Consequently, a detected metabolic signal may not be cancer-specific but could reflect generalized phenomena like systemic inflammation or cachexia, potentially present in both cancer and control groups.

Several limitations of the study’s design must be acknowledged. Utilizing a modest sample size (n = 44) may not fully capture the metabolic variability of diverse populations or rare cancer subtypes. As an exploratory model, the high predictive values reported must be interpreted with caution until validated through larger, stratified cohorts and independent test sets. Additionally, for forensic reliability, factors such as the post-mortem interval (PMI) and environmental taphonomy must be rigorously investigated. Future research integrating these findings into a systems biology framework combining metabolomics with genomics or proteomics will be essential to enhance the specificity and applicability of this method. This study provides a foundational proof of concept, offering a novel, non-invasive pathway for both precision medicine and forensic identification.