Introduction

Ivory has been esteemed since ancient times as one of the most aesthetically appealing and valuable biological materials. It can be carved into a nearly limitless number of beautiful shapes and objects that possess significant economic, cultural, and collector value. This high demand has precipitated rampant poaching, with countless individuals targeting elephants for their ivory, leading to a dramatic decline in elephant populations globally1,2,3. If current trends continue unabated, wild elephant populations could face extinction within the foreseeable future4,5. In response to this pressing crisis and to protect the remaining elephant populations, the Chinese government enacted a comprehensive and stringent ban on the ivory trade in 2018, prohibiting both the commercial import and domestic sale of ivory products6.

Due to rising global temperatures, the thawing of Siberian permafrost in recent years has exposed numerous specimens of the mammoth (Mammuthus primigenius), which became extinct during past civilizations7. The color and texture of mammoth ivory found in permafrost closely resemble contemporary elephant ivory, allowing the trade of mammoth ivory to remain unregulated under the Convention on International Trade in Endangered Species of Wild Fauna and Flora (CITES). Driven by significant economic incentives and market demand, to bypass the elephant ivory ban, traders sometimes mix the two types of ivory, a practice that can stimulate demand for illegal ivory and indirectly contribute to the poaching of African elephants. This phenomenon also complicates efforts to enforce regulations, as the difficulty in visually distinguishing between the two types of ivory creates significant challenges for frontline law enforcement authorities8. Therefore, an urgent and critical need exists to develop an accurate, rapid, economical, and user-friendly method for identifying ivory products.

Driven by significant economic incentives and market demand, traders sometimes mix the two types of ivory to bypass the elephant ivory ban. This practice can stimulate demand for illegal ivory and indirectly contribute to the poaching of African elephants. This phenomenon also complicates efforts to enforce regulations, as the difficulty in visually distinguishing between the two types of ivory creates significant challenges for frontline law enforcement authorities. Therefore, an urgent and critical need exists to develop an accurate, rapid, economical, and user-friendly method for identifying ivory products.

Common methods for distinguishing between elephant and mammoth ivory include morphometric and molecular biology techniques. Morphometric methods rely on differences in the Schreger angles of the two types of ivory9,10; however, the accuracy diminishes when the ivory is carved or damaged. Molecular biology has long been regarded as the ‘gold standard’ for species identification11,12; its drawbacks include high costs, time consumption, complex procedures, large sample requirements, and the potential for damage to valuable or small samples. Additionally, several studies have focused on identification methods based on the inorganic components of ivory, such as isotope analysis13, X-ray fluorescence14, near-infrared spectroscopy15, and Raman spectroscopy16, etc. However, these techniques also face challenges, including high costs or reduced accuracy due to factors like sample color, texture, preservation conditions, and quality.

Ivory predominantly consists of dentin, with a thin cementum layer on the outer surface of the tusk, which contains inorganic components, primarily hydroxyapatite, embedded in an organic matrix composed mainly of collagen, along with various other organic compounds found in dentin and cementum17,18. The chemical composition of ivory is influenced by the elephant’s physiology, diet, and environment, resulting in distinct characteristics of ivory from different regions and species19,20,21. Additionally, these chemical components are permanently retained and stabilized as the ivory grows13. These characteristics enable the accurate distinction and identification of ivory products. The identification of unique and discriminative chemical features from the perspective of organic compounds offers a promising new direction for wildlife product identification. However, to date, few studies have focused on organic compounds in ivory identification, and only one study has used two-dimensional gas chromatography-time-of-flight mass spectrometry (GC × GC-TOFMS) to analyze and differentiate between ivory, dentin, and bone samples based on volatile compounds22; however, it reported an accuracy rate of only 71–75%, which is insufficient to meet the needs of frontline law enforcement agencies for reliable identification. Given this, there is an urgent need for more reliable and effective identification methods in wildlife product research.

Previous tests using fluorescence spectrophotometry on contemporary ivory and mammoth ivory indicated that while there were some differences, the total fluorescence values of the two types of ivory were not significantly distinct. This led to the hypothesis that further separation of the substances within each type of ivory, followed by detailed detection of individual fluorescent compounds, might reveal more pronounced differences. Ultra Performance Liquid Chromatography (UPLC), combined with a fluorescence detector, offers a solution by allowing for the pre-separation of samples through a column, followed by detection. This technique has gained prominence recently for analyzing various complex natural substances due to its high sensitivity, effective separation capabilities, and reusability23,24,25,26. Additionally, the advantages of UPLC—such as rapid analysis, low costs, robust automation, and widespread accessibility—make it a promising method for identifying ivory products, facilitating its adoption in frontline law enforcement efforts against illegal trade.

This study aims to utilize Ultra Performance Liquid Chromatography coupled with a fluorescence detector (UPLC-FLR) to isolate fluorescent characteristic compounds and construct a comprehensive library of UPLC fingerprints for both living ivory and mammoth ivory. This approach will reveal the organic compounds present in ivory and facilitate the identification of characteristic compounds with discriminatory and indicative properties. Ultimately, the goal is to develop a novel method for detecting and identifying ivory products, enhancing the accuracy and efficiency of ivory analysis in the context of wildlife protection and law enforcement.

Materials and methods

Elephant and mammoth ivory samples

All African elephant and mammoth ivory samples used in this study were ivory blocks provided by the National Forestry and Grassland Administration Detecting Center of Wildlife at Northeast Forestry University for scientific research. A total of 72 African elephant ivory samples—comprising raw ivory from African savanna elephants (Loxodonta africana, n = 64) and African forest elephants (L. cyclotis, n = 8)—and 69 woolly mammoth ivory samples were examined (Supplementary Table S1). All woolly mammoth ivory samples originated from the Sakha Republic (Yakutia) in the Russian Federation. DNA extraction from ivory samples was performed using the total demineralization method27. The samples were identified through morphometric techniques or molecular DNA analyses28,29.

Sample preparation

Before analysis, the surfaces of ivory blocks were cleaned using a sterile, moistened cotton ball to remove contaminants. The dentin layer of each ivory sample was then ground into a fine powder using a SPEX-6770 liquid nitrogen grinder (SPEX SamplePrep, USA) with a protocol of 2 min pre-cooling followed by 3 min of grinding. The resulting ivory powder was transferred to 5 mL brown glass vials and stored at room temperature for subsequent analyses.

For extraction, 100 mg of ivory powder was placed into a 2 mL centrifuge tube, followed by the addition of 500 μL of a 0.1 M 70% methanolic hydrochloric acid extraction reagent. The mixture was shaken at 1500 rpm for 5 min and then subjected to ultrasonic treatment in a water bath at room temperature for 10 min. Samples were centrifuged at 12,000 rpm for 5 min at 4 °C. A 400 μL aliquot of the supernatant was filtered through a 0.22 μm membrane filter and subsequently transferred for Ultra Performance Liquid Chromatography (UPLC) analysis.

UPLC data acquisition

A Waters Acquity Ultra Performance Liquid Chromatography system (Milford, MA, USA) was used for analysis. This UPLC system included a binary solvent manager, auto-sampler, column oven, and fluorescence detector. Data acquisition and processing were performed using Waters Empower 3 software.

Samples were analyzed in triplicate, with the software calculating the average values. Analytical conditions were as follows: Shiseido Capcell CORE C18 column (2.1 mm × 100 mm, 2.7 µm); column temperature: 30 °C; flow rate: 0.4 mL/min; injection volume: 2 μL. The elution gradient (eluent A: 0.1% formic acid in water; B: 0.2% formic acid in methanol) was set as follows: 5% B initially, 5–25% B from 0 to 10 min, 25%–45% B from 10 to 15 min, 45–60% B from 15 to 16 min, after which the system rapidly returned to initial conditions. The fluorescence detector (FLR) was set to λ_ex = 280 nm and λ_em = 325 nm.

Data processing

Peaks in the chromatograms were identified and integrated, with both peak heights and areas calculated using the ApexTrack method in Waters Empower 3 software (version 3.6.1, https://www.waters.com/empower3). The processed data were exported for retention time correction and peak alignment using the Correlation Optimized Warping (COW)30 method in PLS_Toolbox (version 7.5.2, https://eigenvector.com/software/pls-toolbox) within MATLAB (version R2013a, https://www.mathworks.com/products/matlab.html). The COW method, a time alignment technique, improves chromatographic peak alignment by minimizing retention time discrepancies across samples. This approach is particularly effective in addressing chromatographic shifts caused by experimental variability. The algorithm optimizes the correlation between the reference and target chromatograms, enhancing peak alignment accuracy while reducing noise.

Mean chromatograms for African elephant and mammoth ivory were generated by averaging the data. Error bars representing the standard deviation were included to visualize variability in the chromatographic profiles. This enabled a clear demonstration of the average chromatographic features and highlighted the differences in characteristic compounds between the two types of ivory.

Peaks accounting for less than 1% of the total peak area and height were excluded, as were those detected in fewer than 10% of samples within each group. Shared peaks between the two ivory sample groups were identified separately based on their respective retention times, with these common peaks designated as characteristic compounds of ivory. The concentration levels of these compounds were represented by their peak areas and expressed as mean ± standard deviation. Differences in the characteristic compounds between African elephant and mammoth ivory were assessed using the Mann–Whitney U test, with statistical significance defined as p < 0.05. A box plot was generated to provide a visual representation of these differences. Species-categorized data and all identified characteristic compounds were further analyzed using principal component analysis (PCA), with proportions displayed using proportional chord plots. All statistical analyses were conducted using R software (version 4.4.0, https://www.r-project.org).

For species classification, the data were standardized and normalized using log10 transformation and the Z-Score method. Six machine learning models—K-Nearest Neighbor (KNN), Random Forest (RF), Support Vector Machine (SVM), Gaussian Naive Bayes (GNB), Logistic Regression (LR), and Decision Tree (DT)—were utilized to differentiate the ivory samples. The dataset was split into an 80% training set and a 20% testing set. Model performance was rigorously evaluated through confusion matrices, ROC curves, and comprehensive performance metrics, including accuracy, precision, recall, F1 score, and AUC, to assess the models’ effectiveness. To ensure robustness and generalizability, cross-validation was performed using both K-Fold and Leave-One-Out (LOO) methods. All machine learning calculations were carried out using scikit-learn, a Python library (version 3.9.0, https://www.python.org).

Results

The UPLC compound profiles of ivory samples were analyzed to evaluate the potential of UPLC-based methods as a forensic screening tool for identifying ivory contraband. A UPLC compound library was constructed using 72 known African elephant ivory samples and 69 mammoth ivory samples. Initial characterization of each ivory sample’s compound profile was completed using average UPLC chromatograms (Fig. 1a–d). These average chromatograms illustrate the data obtained from the analysis and reveal similarities and differences in the compound “fingerprints” of ivory from different species. The average chromatograms for each sample indicate that the main organic compound components are generally similar. However, distinct visual patterns were observed, highlighting differences in the presence of certain high-concentration compounds and a range of low-concentration compounds between the two types of ivory (Fig. 1b–d).

Fig. 1
figure 1

Average UPLC chromatograms of ivory samples from African elephant and Woolly mammoth. (a) Average UPLC chromatograms from the ivory samples of African elephant (deep yellow) and woolly mammoth (sky blue), with error bars representing standard deviation. Error bars for African elephants are shown in light yellow, while those for woolly mammoths are light blue. (b)–(d) are magnified sections of the chromatogram in (a), each highlighting lower-intensity peak regions. These areas are marked with green boxes in (a) to emphasize key distinctions in the chemical compositions of the two ivory types.

Full size image

Additionally, the quality of the chromatograms for mammoth ivory, along with the comparative results of the average UPLC chromatograms, indicates that the mammoth ivory samples used in this study are well-preserved, as evidenced by the presence of numerous prominent compound peaks in the chromatograms.

To further assess the similarities and differences between species, 85 characteristic compounds were identified as key indicators of ivory based on peaks shared in over 10% of each sample category. These compounds were named according to their retention times observed during UPLC analysis. The peak area, representing the concentration of each compound, was used as the primary feature for constructing the model. Subsequent analyses and calculations for these 85 compounds were all performed using their peak area values.

A PCA analysis was conducted on these 85 characteristic compounds (Fig. 2) to visualize any groupings or separations in the dataset. The PCA results demonstrated that the sample groups could be effectively distinguished from one another. The two groups—woolly mammoth and African elephant—were well-separated along the principal components (PC1 and PC2), indicating clear distinctions between the species based on the measured variables. Specifically, PC1 accounted for 48.75% of the variance, while PC2 accounted for 11.18%, collectively explaining a substantial portion of the data’s variability.

Fig. 2
figure 2

Principal component analysis of each ivory sample of African elephant (deep yellow) and Woolly mammoth (sky blue).

Full size image

This clear separation of the two groups, along with the observed variability patterns, underscores the effectiveness of PCA in distinguishing between mammoth and African elephant ivory. The PCA and marginal density plots (mountain-like shapes) on either side reveal distinct clustering patterns and data spread for the two ivory sample groups. The African elephant ivory samples appear more tightly clustered, particularly along PC1 and PC2. This suggests that the variation within this group is relatively low, indicating a more homogeneous chemical profile among the African elephant ivory samples. In contrast, the mammoth ivory samples exhibit more dispersed clustering, especially along PC2, indicating more significant variability within this group.

The chord diagram (Fig. 3) illustrates species-specific and shared chemical indices between African elephants and mammoth ivory. From the analysis, we found 73 characteristic compounds in African elephant ivory and 49 in mammoth ivory, with 37 compounds shared between the two types. Mammoth ivory has 12 unique compounds, but only one is universally present in all mammoth samples. In contrast, African elephant ivory features 36 specific compounds, with 10 universally found across all samples.

Fig. 3
figure 3

The circular chord diagram of characteristic compounds by percent of African elephant (deep yellow) and Woolly mammoth (sky blue).

Full size image

Species categorized each ivory sample’s distribution of characteristic compounds (Fig. 4). Visual comparisons indicated significant differences in the proportion and distribution of characteristic compounds between African elephant and mammoth ivory samples. Additionally, there were notable similarities and differences in the proportions and distributions of characteristic compounds within each species.

Fig. 4
figure 4

Characteristic compounds profiles by percent compound class of each ivory sample of African elephant (deep yellow) and Woolly mammoth (sky blue).

Full size image

To further reveal the differences between species, statistical analysis using the Mann–Whitney U test was conducted to compare the compound classes of the different species and determine whether characteristic ivory compounds could reliably distinguish between elephant ivory and mammalian teeth. A differential statistical table (Table S2, Supplementary Information) and box plots for the differential tests (Fig. S1) were created for the 85 characteristic ivory compounds. Among these 85 compounds, 5 showed significant differences (p < 0.05), 68 exhibited very significant differences (p < 0.01), and 12 showed no significant differences. The Mann–Whitney U test results highlight the distinct chemical profiles between the two species, providing robust statistical evidence for differentiating elephant ivory from mammalian teeth based on their characteristic compounds.

Machine learning and discriminant analyses

Binary classification models were evaluated to distinguish between mammoth and African elephant ivory samples. From an initial set of 85 characteristic compounds, 11 compounds—10 unique to African ivory and 1 unique to mammoth ivory—were identified and selected based on their significant differences (Index68.2, Index76.8, Index101.3, Index117.7, Index119, Index120, Index139.8, Index140.2, Index142.7, Index151.1, Index156.1). Discriminant analysis was performed using six common machine learning models: K-Nearest Neighbor (KNN), Random Forest (RF), Support Vector Machine (SVM), Gaussian Naive Bayes (GNB), Logistic Regression (LR), and Decision Tree (DT). The dataset was split into an 80% training set and a 20% testing set to evaluate the performance of the models.

The classification results were outstanding: all six machine learning models achieved perfect performance metrics, including accuracy, precision, recall, F1 score, and ROC-AUC values 1.0 (Table S3). Additionally, K-fold cross-validation and leave-one-out (LOO) cross-validation were performed to assess the models’ robustness and generalizability. Both validation methods yielded mean accuracy values of 1.0 with standard deviations of 0, further confirming the models’ exceptional performance and stability (Table S3).

All 14 mammoth and 15 African ivory samples in the test set were correctly classified without any errors (Fig. 5a). Furthermore, the ROC curve confirmed the models’ effectiveness, displaying an AUC (Area Under the Curve) of 1.0, which indicates perfect classification accuracy (Fig. 5b). This translates to a 100% true positive rate (sensitivity) and a 0% false positive rate, demonstrating that the models successfully distinguished between mammoth and African ivory with no misclassifications.

Fig. 5
figure 5

The result of machine learning and discriminant analyses. (a) The confusion matrix generated by machine learning for the ivory samples, where “E” represents African elephant and "M" represents Woolly mammoth; (b) The ROC curve of machine learning. Notes: ROC curves for all machine learning models. Since all models achieved identical classification performance (AUC = 1.00), the curves overlap entirely, with only the color of the final plotted curve being visible.

Full size image

These results demonstrate the models’ strong discriminatory power and flawless classification accuracy in the test set. This provides reliable evidence that the selected characteristic compounds can effectively differentiate ivory types. The ability to use this method to identify mammoth and African elephant ivory in the future clearly has significant practical implications, particularly for wildlife product identification and ivory regulation. However, we emphasize that while the compound identified in mammoth ivory exhibited strong consistency across the test samples in this study, future research should expand the sample size to include a broader range of elephant species and geographic locations. This will further enhance the applicability and robustness of the method, ensuring its reliability across diverse scenarios.

Discussion

This study developed a UPLC-FLR method for identifying illegal wildlife products, specifically to differentiate ivory samples of unknown origin. The findings demonstrate that liquid chromatography is highly effective in distinguishing between various types of ivory. Preliminary comparisons of the chemical profiles of mammoth and African elephant ivory revealed similarities and differences. The results from the PCA plot and accompanying density plots clearly illustrate distinct clustering and data distribution patterns between the two ivory sample groups. African elephant ivory samples exhibit tighter clustering, indicating relatively low variability within the group and suggesting a more homogeneous chemical composition.

In contrast, mammoth ivory samples display more dispersed clustering, particularly along PC2, indicating more significant variability in their chemical compounds. This increased variability in mammoth ivory may be attributed to differences in preservation conditions. Often buried in Siberian permafrost for thousands of years, mammoth tusks experience varying degradation rates of organic compounds like collagen, leading to the observed chemical diversity among the samples.

The study identified 85 key characteristic compounds in ivory, highlighting a clear distinction in the chemical profiles between African elephant and mammoth ivory. The compound distribution chart reflects significant differences between the two ivory types while revealing some variability in compound distribution among individual samples. This variation may be attributed to the African elephant samples originating from different African regions, as the museum-held samples likely represent elephants from various habitats. These elephants may have been exposed to differing environmental conditions and possibly varied diets. This underscores the importance of expanding the sampling of ivory from diverse regions to develop a comprehensive ivory compound database further.

Only 85 characteristic compounds in ivory were identified, which can be attributed to two main factors. First, ivory is a mineralized connective tissue primarily composed of hydroxyapatite (a form of calcium phosphate) tightly bound with collagen, resulting in a dense structure that complicates compound extraction. The research employed a 0.1 M 70% methanolic HCl solution for extraction; the hydrochloric acid partially breaks down ivory’s tightly packed structure, while the 70% methanol helps dissolve and extract as many compounds as possible. The second factor pertains to the detection method employed. The study used a liquid chromatography-fluorescence detector, which primarily detects fluorescent compounds, such as polycyclic aromatic hydrocarbons. This means that this method did not detect non-fluorescent compounds like phenols. If a different detection system were used—such as liquid chromatography-mass spectrometry (LC–MS) or liquid chromatography-time-of-flight mass spectrometry (LC-QTOF)—the number of detectable compounds could potentially increase dramatically, possibly reaching into the hundreds or even thousands.

Machine learning and discriminant analyses

Machine learning models were employed using a carefully selected set of 11 characteristic compounds to achieve precise classification between mammoth and African elephant ivory samples. This selection aimed to streamline the identification process by reducing the number of indicators required, thereby minimizing analysis time and workload. Among these compounds, 10 were unique to African elephant ivory, and 1 was specific to mammoth ivory.

The rationale for selecting 11 compounds, rather than fewer, was guided by the chromatographic profiles illustrated in Fig. 1. These profiles reveal that compounds with higher concentrations often exhibited greater similarity between the two types of ivory, whereas compounds with stronger distinguishing characteristics were typically present at lower concentrations. All 10 unique compounds specific to African elephant ivory were included as markers to maximize classification accuracy, ensuring clear differentiation between the two species.

Compared to DNA-based identification methods, liquid chromatography offers several notable advantages. It is significantly more cost-effective and provides rapid results, with sample collection, processing, and classification taking less than an hour. In contrast, DNA-based methods often require one to two days, including time for decalcification, DNA extraction, and sequencing. Liquid chromatography also reduces sample requirements by 50% and costs by 90% while needing minimal material, thus causing minimal damage to delicate items like jewelry and carved ivory. Consequently, this method shows great potential as an efficient tool for accurately identifying and differentiating ivory products.

To enhance the liquid chromatography identification model’s accuracy, reliability, and comparability with DNA testing—the current gold standard in wildlife forensics—it is crucial to validate this method using a larger and more diverse sample set. Future samples should include African forest elephants (L. cyclotis), African savanna elephants (L. africana), and Asian elephants (E. maximus), as well as other species that appear in the ivory trade, such as walruses (Odobenus rosmarus), killer whales (Orcinus orca), narwhals (Monodon monoceros), hippopotamuses (Hippopotamus amphibious), and warthogs (Phacochoerus africanus). In addition, incorporating other materials historically used as ivory substitutes, such as bone and synthetic alternatives (e.g., ivory nut, plastic, resin), would be beneficial. Further refinement of the model will require additional bone samples from various species and synthetic ivory substitutes to broaden its applicability and effectiveness in wildlife product identification.

Creating a comprehensive reference database with the chemical fingerprints of each species would be a vital resource. Incorporating these characteristic compounds into established databases, like the CITES trade database or the UNODC Ivory Identification Guide, would allow for rapid identification of unknown ivory samples at global customs checkpoints. Such an efficient and accessible identification method could be a powerful tool in combating illegal ivory trade, bolstering monitoring and deterrence efforts at borders, and aiding in conserving endangered and critically endangered elephant species.

Conclusions

This study developed a UPLC-FLR-based method for identifying ivory products, focusing on classifying confiscated suspected ivory samples by analyzing ivory-specific characteristic compounds. Schreger morphological methods and DNA analysis were employed to confirm classifications for all African and mammoth ivory specimens. Utilizing liquid chromatography with a fluorescence detector, chemical fingerprints of ivory were established, resulting in a database of 85 characteristic ivory compounds derived from 72 African elephant and 69 mammoth ivory samples. African ivory exhibited 73 characteristic compounds, while mammoth ivory had 49. Notably, mammoth ivory contained 12 unique compounds, with only one present in all mammoth samples. In contrast, African ivory featured 36 unique compounds, with 10 universally present across all African samples.

Principal Component Analysis (PCA) clearly separated African and mammoth ivory, indicating substantial classification potential. Among the 85 characteristic compounds, 5 showed significant differences, 68 exhibited highly significant differences, and 12 revealed no differences. Machine learning models were employed using a carefully selected set of 11 characteristic compounds to achieve precise classification between mammoth and African elephant ivory samples, with all models achieving perfect classification and 100% accuracy. This underscores the considerable potential of liquid chromatography in identifying and regulating ivory products.

Due to its specificity and separation capabilities, ultra-performance liquid chromatography (UPLC) provides robust analysis for classifying the species identity of confiscated ivory based on compound profiles. Furthermore, the liquid chromatography-based characteristic compound analysis method for wildlife specimens offers speed, cost-effectiveness, convenience, and accuracy advantages. This method could serve as a new and effective screening tool for ivory and other wildlife products at the forefront of conservation efforts.