Decoding Nacre: Uncovering Unique Identifiers of Pearl Provenance

Decoding Nacre: Uncovering Unique Identifiers of Pearl Provenance | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Article Decoding Nacre: Uncovering Unique Identifiers of Pearl Provenance Admir Masic, Dahyun Kyung, Ruiqi Zhang, Alice Zehner, Ali Alatawi, and 5 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6874927/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Pearls are nacreous biogenic products that can be classified by whether they are natural or cultured, what species of mollusk produced them, and what environment they were grown in. Due to the subtle compositional and morphological differences between pearl types, determining pearl provenance can be problematic. To address these challenges, in this study we introduce a pearl identification workflow combining stable isotope analysis of pearl carbonate minerals to identify their geographic origin, and a Raman spectroscopy-based machine learning model to determine pearl species. Stable isotope data reveal clustering in δ18O based on the geographic origin of saltwater pearls. Additionally, pearl oyster shells from the Arabian Gulf were used as a tractable model system to investigate links between seawater geochemistry and carbonate stable isotope signatures. Paralleling these studies, Raman spectra of pearls formed by P. radiata, P. maxima, and P. fucata oysters were utilized to train a support vector machine classifier to predict pearl species with 96.4% accuracy. The combined results from these investigations demonstrate their utility in tracing pearl provenance by identifying geography and species of origin, which could be employed at a larger scale to improve pearl classification and provide insights into the formation mechanisms of biogenic carbonates. Physical sciences/Materials science Physical sciences/Materials science/Techniques and instrumentation/Characterization and analytical techniques Physical sciences/Materials science/Biomaterials Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction Pearls are unique among precious gems in that they are produced through biological processes. The outer surfaces of pearls are primarily composed of nacre (Fig. 1), a hierarchically structured brick and mortar-like nanocomposite material consisting of aragonite (CaCO 3 ) tablets separated by thin organic inter-layers 1–3 . Within the broad range of commercially available pearl types 4,5 , one major classification distinction is based on whether a pearl is natural or cultured. Natural pearls are formed without any human intervention and rarely occur 4 , while in cultured pearls, pearl formation is artificially induced through the implantation of a piece of tissue from a donor mollusk, oftentimes along with an inorganic bead, into a host mollusk 6,7 . In addition, pearls can be further subdivided based on whether they were formed by saltwater or freshwater mollusks, and further distinctions are made between pearls produced in different geographical locations and by different mollusk species 4,6,8 . Despite this incredible diversity in the employed mollusk species, production method, and geographical origin, all of which have a profound impact on pearl economic value, the ability to accurately classify pearls based on visual inspection alone has been historically problematic since the unique identifying features of different pearl types can be extremely subtle or ambiguous. To help augment these visual inspection-based workflows, gemologists can also employ a diverse set of analytical tools to classify pearls by structure and chemistry, some examples of which are shown in Fig. 1 4,5,9 . For example, cultured and natural pearls can be differentiated using x-ray imaging techniques such as x-ray microradiography and micro-computed tomography (micro-CT), which can non-destructively visualize internal structural features such as pearl nuclei, growth rings, and voids 9-11 . Freshwater and saltwater pearls can be routinely distinguished using x-ray fluorescence (XRF) techniques, as freshwater pearls have a higher manganese content than saltwater pearls and thus fluoresce under x-ray irradiation 12 . Additionally, spectroscopy methods including photoluminescence 13,14 , ultraviolet-visible-near infrared 13,15–18 , Raman 15,19,20 , and fluorescence spectroscopy 8 have been used to identify organic compounds (e.g., pigments) in pearl nacre. The unique combinations and quantities of these organic compounds in nacre have been linked to different mollusk species, allowing for the identification of specific spectral features to be attributed to pearls produced by distinct species and enabling the identification of pearls that have been artificially enhanced through color treatments. As with visual inspection-based classification workflows, however, the utilization of specific spectroscopic features for pearl classification may not always present themselves clearly and can be difficult to interpret 5,14,21,22 . Furthermore, previous studies have been largely correlative and have not been widely employed for pearl identification purposes across the broad range of commercially available pearl types. There thus remains a need for the development of methods that can enhance the existing suite of characterization tools by providing useful information on the different aspects of a pearl, which would consequently allow for improved provenance tracing and economic valuation of pearls of uncertain origin. Toward achieving this goal, in the present study we present two techniques—stable isotope analysis of carbonate minerals and Raman spectroscopy—for the identification of pearl geographic provenance and species, respectively (Fig. 1). A summary of the analyzed samples can be found in Supplementary Table 1, with additional details in Supplementary Tables 2 and 3. Results Stable Isotope Analysis of Carbonates Carbonate stable isotope analysis is a well-documented technique which is commonly utilized for paleoclimate reconstructions, as the carbon and oxygen isotopic compositions of biogenic carbonate minerals are reflective of the conditions in which they were formed. Carbonate δ 13 C and δ 18 O are measures of the ratio of 13 C to 12 C and 18 O to 16 O, respectively, relative to a standard (Vienna Pee Dee Belemnite or VPDB) and expressed in per mil (‰). Correlations between mineral δ 18 O, temperature, and water δ 18 O (standardized relative to Vienna Standard Mean Ocean Water or VSMOW), have long been studied and form the basis of a commonly applied palaeothermometry method 23–25 . Carbonate δ 18 O is linked to both mineralization temperature and hydrological processes such as evaporation and precipitation. Temperature exerts a direct control on mineral δ 18 O through the temperature-dependent equilibrium fractionation between water and dissolved carbonate: at lower temperatures, an increased proportion of 18 O is incorporated into dissolved carbonate, which is subsequently captured by the carbonate mineral 23,24 . Hydrological processes are a primary driver of water δ 18 O, which is in turn recorded in the δ 18 O of biogenic carbonates 24,25 . As illustrated in Fig. 2a, the evaporation of seawater leads to enrichment of the heavier 18 O isotope in the source water as the lighter 16 O isotope preferentially enters the vapor phase, which then becomes the water source for subsequent precipitation. Since freshwater bodies are primarily supplied and replenished by meteoric water, they are consequently more abundant in the lighter 16 O isotope than the evaporative source (i.e., seawater) 26 . Environmental factors such as water depth, circulation, and salinity can additionally influence water δ 18 O 26–28 and, by consequence, mineral δ 18 O. To illustrate the compounding effects of these processes on a global scale, maps of temperature 29 and seawater δ 18 O 30 are shown in Fig. 2b. Beyond the value of measuring biomineral δ 18 O values, complementary information can also be obtained from C isotopic analysis, since the δ 13 C of biogenic carbonates has been linked to metabolic processes such as respiration and feeding, and the δ 13 C of dissolved inorganic carbon in the water 31,32 . Motivated by these observed variabilities in global isotopic concentrations and their impact on precipitated biominerals, analyses of stable C and O isotopes in pearl nacre were conducted to determine whether carbonate isotopic composition sufficiently captures geographic differences for accurate identification of pearl origins. The plot of δ 13 C against δ 18 O in Fig. 2c shows a clear distinction between freshwater and saltwater pearl samples, with the saltwater pearls being more enriched in both 13 C and 18 O. There is also considerably wider spread in the freshwater data and no clear trends related to geographic origin. However, the saltwater pearls demonstrate separation, mainly along the δ 18 O axis, based on where they were formed. To assess the possibility of automating the task of identifying the geographic provenance of saltwater pearls, a logistic regression model was constructed to separate the saltwater pearl isotope data. The logistic regression model produced an accuracy score of 88%, demonstrating that the geographic origin of saltwater pearls can indeed be classified based on pearl carbonate stable isotopic composition. The logistic regression decision boundaries are shown in the rightmost panel of Fig. 2c. A confusion matrix for this model can be seen in Supplementary Fig. 1, along with decision boundaries and confusion matrices for support vector machine and k-nearest neighbors models, for both of which we report accuracy scores of 85%. In agreement with these observed results, previous comparisons of marine and non-marine carbonates show similar trends in δ 13 C and δ 18 O in numerous types of geologic and biogenic samples 33–35 . The relative enrichment of saltwater pearls in 18 O compared to freshwater pearls aligns with the differences in freshwater and saltwater δ 18 O that are primarily driven by the hydrological processes referenced above (Fig. 2a) 26,28 . The difference in carbon isotope ratios may be accounted for by the sources of carbon in the environment where the pearl was grown. In freshwater systems, for example, δ 13 C values can be strongly influenced by organic carbon cycling; the remineralization of abundant, isotopically light organic matter can dominate the isotopic composition of dissolved inorganic carbon utilized in pearl mineralization 36,37 . There may also be differences in pearl biomineralization processes between freshwater and saltwater species that affect isotopic fractionation, as variations in fractionation among carbonate minerals from different taxa have been reported 38,39 . Due to limited prior data, however, it remains unknown whether pearls exhibit similar species-specific controls on precipitation kinetics that might affect isotopic equilibrium. The considerable spread in the freshwater pearl data in Fig. 2c, which was collected from a limited number of locations, prevents any definitive conclusions on δ 13 C and δ 18 O sensitivity to geographic provenance. The wide range of freshwater pearl δ 13 C and δ 18 O values may be attributed to local environment dependencies and larger seasonal changes in lakes and rivers, which lead to greater variability in freshwater chemistry compared to their marine counterparts 36,37 . For example, seasonal shifts have been previously observed by measuring δ 18 O across transects of freshwater pearls 40,41 , which in one study were found to span a range upwards of 4 ‰ 41 . Conversely, the saltwater data (as seen in Fig. 2c) broadly cluster by geographic region along the δ 18 O axis. This trend suggests that the δ 18 O values are primarily influenced by the conditions in which the pearls were grown. While the literature on pearl carbonate isotopes is limited, δ 18 O values of saltwater pearls from Japan 42 and South Korea 43 have been previously linked to seawater temperature. On a larger spatial scale, as shown by the global seawater temperature (Fig. 2b, top left) and δ 18 O (Fig. 2b, bottom left) maps, water δ 18 O also plays an important role in determining saltwater pearl δ 18 O. Of the source countries in our analysis of saltwater pearls, Japan has the lowest average sea surface temperature and seawater δ 18 O. The δ 18 O values of pearls from Japan are higher than the δ 18 O values of pearls from (northern) Australia. However, the δ 18 O values of pearls from Bahrain and Qatar are higher than the δ 18 O values of pearls from Japan. While Bahrain and Qatar have temperatures that fall between those of Australia and Japan, they have the highest seawater δ 18 O values due to high rates of evaporation in the Arabian Gulf region. These results suggest that there is a combined importance of water δ 18 O and temperatures in determining pearl δ 18 O values. It should still be noted, however, that there remains some overlap in δ 18 O values of pearls from different locations, particularly between Bahrain and Qatar, where the isotopic compositions of pearls are difficult to distinguish. Of the source locations, Bahrain and Qatar have the greatest proximity to one another, meaning that the water δ 18 O and temperatures in these locations (and consequently pearl δ 18 O values) may be similar, especially relative to the other source locations. Pearls from Qatar were most commonly misclassified as pearls from Bahrain by the logistic regression, support vector machine, and k-nearest neighbors models, and we note that grouping pearls from Bahrain and Qatar into the same class noticeably increases accuracy scores (Supplementary Fig. 1). Furthermore, due to material availability limitations, the number of pearls analyzed in this study are not consistent across classes (see Supplementary Table 1), and as such, we expect that having a more balanced dataset may produce more reliable performance metrics for the models. Moreover, saltwater pearl δ 13 C has been associated with feeding and organic matter degradation 42 and has not been concluded to significantly change with temperature. Saltwater pearl δ 13 C data in Fig. 2c appear to remain within the same range of ca. –1 to 2 ‰ for all groups, with the pearls from Australia showing a more limited span. Given the available source locations for this study, these results suggest that saltwater pearl δ 13 C may not be sensitive to location or temperature in the way δ 18 O appears to be, and remains within a specific range regardless of growth environment, and possibly even species (pearls from Bahrain, Qatar, and Japan were formed by oysters from the ( Pinctada ) P. fucata/martensii/radiata/imbricata species complex, while pearls from Australia were formed by P. maxima oysters). It is possible, however, that pearls from other locations may exhibit distinct δ 13 C values depending on regional circulation patterns and rates of organic matter remineralization and primary productivity. To investigate the link between environmental conditions and the δ 18 O values of carbonates formed by pearl oysters on a more local scale, isotopes from pearl oyster shells in the Arabian Gulf (Fig. 3a) were also examined. Aragonite from shells in Bahrain, Kuwait, and Qatar were analyzed and compared to annual mean surface temperature (Fig. 3b, left) and salinity (Fig. 3b, right) maps of the Arabian Gulf 44 . Salinity is controlled by much of the same evaporative processes as water δ 18 O 28 , and the calculation of regional slopes correlating these two variables have been previously used to construct seawater δ 18 O datasets 30 . As there is better availability and spatial resolution of salinity data in the Arabian Gulf compared to seawater δ 18 O data, the salinity maps are shown in Figure 3. While overall there is considerable overlap in the δ 13 C and δ 18 O values of the shell samples from Qatar with the shell samples from Bahrain and Kuwait (Fig. 3c), some separation can be observed between the isotopic signatures of shell samples from different locations within Qatar, as well as between samples from Bahrain and Kuwait (see Fig. 3d). To better understand potential trends in the shell and pearl datasets, additional calculations showing the offset of these results with expected δ 13 C and δ 18 O values of inorganic aragonite and calcite precipitated at equilibrium 39 using available local temperature datasets are provided in Supplementary Fig. 2. The different biomineralization products of pearl oysters were additionally examined: comparisons between shell aragonite, shell calcite, and pearl aragonite in oysters collected in Bahrain are shown in Fig. 3e, revealing that shell calcite is depleted in 13 C compared to its aragonitic counterparts. In contrast, pearl and shell aragonite display more similar δ 13 C values, with slightly lower δ 18 O values on average in pearl samples compared to shell samples as demonstrated in the δ 18 O violin plot in Fig. 3f (a δ 13 C violin plot can be seen in Supplementary Fig. 3). The Arabian Gulf is a unique and relatively isolated water body characterized by high evaporation rates, elevated salinity, shallow bathymetry, and large seasonal temperature variability 45–47 . The analysis of carbonates grown in these distinctive conditions could thus offer insights on the relationship between the isotopic composition of carbonates and the environments in which they were formed. Figure 3c shows that the δ 18 O values of shell aragonite from the Arabian Gulf span a range of ca. -1 to 2 ‰. More variability is seen in δ 13 C values compared to δ 18 O values, and a wider overall range of shell δ 13 C values are observed than in the pearl samples seen in Fig. 2c. The shells collected in Kuwait came from the coldest and lowest salinity regions in the Arabian Gulf, owing to the Shatt Al Arab river, which is the primary freshwater input for the Gulf 47 . As such, these δ 13 C and δ 18 O values cluster close to the origin, while shells from Bahrain have high δ 18 O values and variable δ 13 C values (Fig. 3c and 3d). Shells from Qatar, which come from a region of the Arabian Gulf with higher average temperatures and salinity, exhibit δ 18 O values spanning most of the range of δ 18 O values observed in the samples from both Bahrain and Kuwait. In particular, shells from Al Wakrah and Al Safliya Island, Qatar (light green and light blue diamonds, respectively, in Fig. 3c and 3d), which experience the highest average annual temperature and salinity of all the source locations, have lower δ 18 O values compared to other locations in Qatar and more closely resemble those from Kuwait. These observations suggest that elevated temperature, which drives the δ 18 O composition toward lower values, exerts a stronger influence than evaporation. Overall, correlations between δ 18 O values and the average temperature and salinity maps shown in Fig. 3b are not always straightforward, in part because the two main controls—evaporation and elevated temperature—exert opposing effects on δ 18 O. The results in Fig. 3c generally suggest a greater relative role of local evaporation in shell aragonite δ 18 O within the Arabian Gulf, although the influence of mineralization temperature appears to become stronger in locations with higher average sea surface temperatures. As suggested from the data presented in Fig. 2 that water δ 18 O noticeably influences the δ 18 O values of pearls from Bahrain and Qatar, the observation that temperature could still play a dominant role in determining δ 18 O values of shells within the Arabian Gulf signifies that the relative effects of temperature and water δ 18 O on the δ 18 O values of minerals produced by pearl oysters may be sensitive to context and scale. As there is substantial overlap between the δ 18 O values of shells sourced from different countries in Fig. 3c (particularly with regard to Qatar), it may be important to additionally consider that the proximity of the source locations within Bahrain and Qatar may render a country-based separation of isotope data less effective. Previous mapping of water data in Qatar has shown considerable differences in salinity between the East and West coast 48 , which is not fully captured in the salinity map shown in Fig. 3b due to differences in resolution between the datasets. Given such variations that are highly sensitive to locality, trends in mineral δ 18 O may be more readily seen based on the specific water conditions of the source locations and on their proximity to one another. Furthermore, while the salinity-δ 18 O correlation has been well-studied, it has been shown to exhibit spatiotemporal sensitivity 30,49 , and given that the Arabian Gulf is isolated and evaporative, definitive conclusions about the primary influences on mineral δ 18 O within this region using salinity data would require a nuanced understanding of the local relationship between salinity and water δ 18 O. In addition, the maps do not account for the extent of seasonal variations; furthermore, variables such as circulation and water depth may also help to paint a more complete picture of the observed δ 18 O 50 , and improved temporal and spatial resolution of seawater δ 18 O measurements in the Arabian Gulf could help resolve the relative importance of these controls. Pearls are primarily composed of aragonite in the form of nacre, which is the same material that constitutes the innermost layer of pearl oyster shells 3 . Above this nacreous shell layer is a layer of prismatic calcite (another polymorph of CaCO 3 ) 3 . In Fig. 3e it is observed that shell aragonite and pearl aragonite fall in similar regions of the plot, with shell aragonite being more variable in δ 13 C, whereas shell calcite is visibly more depleted in 13 C (which can be more readily observed in Supplementary Fig. 3a). Additionally, δ 18 O in shell calcite is slightly higher than in shell aragonite and pearl aragonite, with δ 18 O values of pearls being slightly lower compared to δ 18 O values of shells (Fig. 3f). The general trends in δ 13 C values are in line with expectations of distinct carbon isotope equilibrium fractionation factors for calcite and aragonite 51–53 , where shell aragonite was previously found to be enriched in 13 C relative to calcite 52 . δ 13 C values for pearl and shell aragonite, while similar on average, appear more variable in shell aragonite (see Supplementary Figure 3), which may be a result of greater heterogeneity in the shell samples compared to the pearl samples. Regarding oxygen isotopes, our results show that pearl δ 18 O values appear slightly lower than δ 18 O values for both shell aragonite and shell calcite in samples from Bahrain, although the reason for this remains unclear. The process of pearl mineralization may indeed be distinct from shell nacre mineralization, resulting in differences in oxygen isotope fractionation, or there may be certain seasons or environmental conditions in which pearls are more likely to form, which may put pearl growth out of phase with shell nacre growth. With the large suite of analyzed materials from a wide range of locations, our results present isotope ratio mass spectrometry as a promising method for improving pearl classification as well as an advancement of insights on isotope fractionation and biomineralization processes in shell and pearl minerals produced by pearl oysters. However, there are some considerations to be made for the practical implementation of this tool. The collection of samples from cultured pearls with a large beaded nucleus and a thin layer of native nacre must be performed with caution—drilling into the nucleus should be avoided as this would certainly affect the results 43 . Most of the saltwater pearls analyzed in this study did not have a beaded nucleus—the pearls from Bahrain, Qatar, and Australia were determined to be natural, as these pearls were either collected firsthand by the authors and/or were verified with x-ray imaging. The pearls from Japan, however, included both non-beaded and beaded pearls. For these samples drilling was performed to a shallow depth (ca. 1 mm), and no notable outliers were observed in the isotope data—from this, it is expected that there was no unintentional drilling of the nuclei, which are typically made from freshwater shells 7 and would therefore likely produce a distinct isotopic signature from saltwater nacre. Additionally, when considering a real-world use case for carbonate isotope analysis, it must be noted that the method requires some destruction of the sample. While a relatively small sample mass is required (around 50-100 µg), this may be difficult to acquire from smaller pearls. In these cases, a subsampling of entire pearls from a large batch may be required (since batches of small wild pearls routinely come from a single locality). Furthermore, to improve the robustness of utilizing mass spectrometry data as a means of geographic provenance identification, a larger dataset of the isotopic composition of pearls from all known locations of natural and cultured pearls should be further developed. For example, the pearls from Qatar in Fig. 2c were only sourced from one specific location (Al Aaliya Island), which exhibit a different range of δ 13 C and δ 18 O values compared to the shells from Qatar, which were sampled from more locations. If we assume that the isotopic composition of shell aragonite, as seen in Fig. 3c, is similar to that of the pearls formed in the same environment, the analysis of pearls from different locations in Qatar may affect the performance of the classification models in distinguishing between pearls from Bahrain and Qatar as presented in Fig. 2c. Although shell aragonite—which would be more convenient to sample from than pearls for expanding the dataset—could be a sufficiently suitable proxy for pearl aragonite, the observed differences in the isotopic composition of pearls and shell aragonite (Fig. 3e) may affect the accuracy of the dataset and should thus be subject to further analysis and consideration. From preliminary results, we show that machine learning algorithms such as logistic regression (Fig. 2c) could be applied to predict geographic origin from pearl isotope data. We expect that dataset expansion may be able to improve the accuracy of such algorithms and hopefully resolve ambiguities pertaining to some areas of overlap in the δ 13 C and δ 18 O values of pearls from different locations. Clumped isotope analysis, which measures the extent of ‘clumping’ of heavy isotopes (e.g., 13 C- 18 O bonds) rather than measuring isotope ratios of individual elements, could be additionally investigated to reconstruct seawater temperature from pearl Δ 47 54,55 . This method requires a larger sample mass than the isotope ratio measurements used in this study, but may be useful in improving the dataset as it could help clarify the temperature-mineral δ 18 O correlation and isolate vital effects (i.e., kinetic and metabolic contributions that result in an isotopic offset between biogenic and inorganic carbonate when the minerals are precipitated in the same conditions) 56 . Raman Spectroscopy and Support Vector Machine (SVM) Prediction Since species of origin is another important factor in pearl classification, and considering previous investigations that have utilized Raman spectroscopy for the identification of species-specific organic compound content in pearl oyster nacre 15,19,20 , we next explored the robustness of this technique for such purposes. In this study, we utilized Raman spectroscopy data from the surface of 771 pearls to train a machine learning (ML) algorithm to differentiate pearl species based on Raman-collected chemical features. While forming the training dataset for the ML model, it was essential to ensure consistency between independent Raman measurements from all analyzed pearls. We thus pre-processed each measured Raman spectrum so that it was constrained to a wavenumber range of 100 cm -1 to 2000 cm -1 , as shown in Fig. 4a. To maintain uniform data intervals and ensure an equal number of data points, each processed Raman spectrum was resampled to 250 equally spaced data points within this range. For the sampled data points that did not precisely align with the original measurement, linear interpolation was applied using the two nearest neighboring points in the raw spectrum (Supplementary Fig. 4). To retain information potentially contained within fluorescent signals and signals from high-energy organic components for the ML algorithm, no background subtraction was carried out on the raw Raman spectra. In total, 700, 170, and 51 Raman spectra of pearls from P. maxima , P. radiata , and P. fucata saltwater oysters, respectively, were collected and analyzed. The collected samples were then randomly partitioned into two groups: a training dataset and testing dataset, followed by a train-to-test ratio of 90:10, as illustrated in Fig. 4a. A Support Vector Machine (SVM) model was developed to classify the three pearl-forming oyster species based on their Raman spectral features. Hyperparameter optimizations were performed to enhance the SVM classification performance: a linear kernel matrix was applied with a kernel coefficient (gamma) value of 0.5 and a regularization parameter (C) of 1.0. The training weights assigned to each Raman feature are presented in Fig. 4b. Notably, data points around 1080, 700 and 200 cm -1 wavenumbers exhibit strong contributions to specimen classification prediction. These highly weighted Raman bands correspond to the vibrational modes of aragonite, which is the primary crystalline component of the pearls. Specifically, bands at 1080 and 703 cm -1 correspond to the v 1 symmetric and v 4 in-plane bending modes of the aragonite carbonate CO 3 2- , whereas the wavenumber at ca. 230 cm -1 corresponds to the vibration of the aragonite lattice 15 . In addition, the highly weighted data points at wavenumbers above 1250 cm -1 are attributed to different organic components within the pearls. In particular, Raman band peaks at ca. 1260, 1350, 1410, and 1560 cm -1 are indicative of the in-plane deformation of C-H bonds, biological pigment vibrations, hexagonal carbon ring stretching, and C=C bond stretching, respectively 15 . The consistency between these dominantly weighted Raman features with known pearl spectral signatures validates the capacity of the SVM model to extract chemically relevant classification features when making a prediction. Moreover, the SVM model is capable of capturing subtle spectral variations that may not be directly attributed to typical Raman features of pearls. This model also accounts for background fluorescence, which can vary according to the structure and composition of individual pearls. Unlike conventional physical measurements or manual baseline correction of the spectra, the SVM model can autonomously learn and integrate these complex and overlooked spectral perturbations, thereby enhancing the correlations between the pearl Raman spectral data and the prediction outcomes, and ultimately improving the classification accuracy. Each testing dataset was randomly selected according to the overall data distribution (55 P. maxima , 15 P. radiata , and 5 P. fucata spectra per trial, for a total of 75 testing data, which is 10% of the overall collected data). Across the 3000 independent testing trials, this yields a total of 2200, 600, and 200 samples for P. maxima , P. radiata , and P. fucata , respectively. Two statistical confusion matrices summarizing the prediction results are shown in Fig. 4c. Each entry in the % error confusion matrix was calculated by dividing the number of misclassifications of that species by its total number of test samples. For example, within 2200 P. maxima test samples, 42 samples were misclassified as P. radiata , giving an error of (42/2200)´100% = 1.91%. The three diagonal elements of each matrix specify an accurate pearl specimen prediction through SVM, whereas the off-diagonal elements in the matrix represent a misclassification. Among the three species, P. maxima pearls exhibited the highest classification accuracy, whereas P. radiata and P. fucata pearls are more frequently misclassified as P. maxima by the model. This predicted classification bias is likely due to the imbalanced distribution of data in the training dataset, wherein spectra from pearls formed by P. maxima account for 73% of the total samples, compared to the 21% for P. radiata pearls and 6% for P. fucata pearls. However, despite this imbalance in data distribution, an overall high prediction accuracy of 96.4% was achieved over the 3000 testing samples across the three pearl species (Supplementary Fig. 5, Supplementary Table 4). To further investigate the influence of specific Raman spectral regions on the prediction accuracy, we divided the full Raman spectrum into halves, with the first half containing data from a wavenumber range of 100 cm -1 to1050 cm -1 and the second half containing data from a range of 1050 cm -1 to 2000 cm -1 . Each separated spectrum consisted of 125 data points (125 data points each from the first and second halves of the spectra from the original 250 Raman data points; this is detailed in Supplementary Fig. 6). The same hyperparameter-optimized SVM model was retrained using only one spectral region at a time for over 3000 randomly selected testing pearl spectra. Fig. 5a presents the accuracy confusion matrices of the prediction results for the halved spectral regions (same procedure as previous; see Supplementary Tables 5 and 6 for details). Compared to the original Raman spectrum input, using only half of the spectral data led to a decrease in the overall classification accuracy, particularly for P. radiata and P. fucata . This decline in accuracy can be attributed to the loss of data variations and spectral information, including critical vibrational features associated with the pearl organic matrix components and lattice vibrations, thus reducing the prediction power of the model. A statistical box plot overlaid with a scatter plot of accuracies from individual trials further illustrates the variations in classification accuracy across different trials, as shown in Fig. 5b. The predictions from the full-spectrum SVM model exhibits lower error variability and superior predictive performance compared to the models using data points from partial spectra. Given that the majority of the Raman features exhibit a minimal impact weight in making the SVM prediction, a refined feature selection approach was implemented. Indicated by the highly weighted Raman features obtained in Fig. 4b, we extracted a subset of 100 highly weighted Raman data points, ranging from the spectral regions of 100-400 (lattice vibration), 670-750 (CO 3 2- bending modes), 1050-1120 (CO 3 2- stretching), 1280-1400 (organic pigments), 1510-1580 (C=C bond vibrations), and 1850-1960 cm -1 (triple-bonding organic components and overtones). After training the SVM algorithm with these 100 highly weighted data points with the same model criteria, we extracted the corresponding feature importance weights, shown in Fig. 5c. A similar result was obtained as with the previous extraction in Fig. 4b, which matches the characteristic band peak of the Raman spectrum. To evaluate model performance with these 100 highly weighted data points, the species for 3000 randomly selected pearl spectra were again predicted, as shown in Fig. 5d. Compared to the half-spectrum input dataset, the dataset using the 100 highly weighted data points achieves a comparable prediction accuracy with respect to the original full-spectrum model, except for a notable drop in classification accuracy for P. fucata . Overall, SVM-based classification models can predict pearl species from their chemical signatures obtained through Raman spectroscopy, with our models yielding a prediction accuracy of 96.4%. By analyzing the importance of Raman features, we identify the key Raman band wavenumbers that dominate pearl species classification, which consistently align with known pearl aragonite and organic spectral features. Furthermore, by optimizing feature selection, we can reduce the complexity of input data while maintaining a high prediction accuracy. These findings demonstrate that SVM can be an effective tool for pearl identification and classification, which leverages Raman spectroscopy as a non-destructive analytical method that fits in the as-developed numerical ML method. The demonstrated models provide a foundational proof-of-principle framework that can serve as a basis for future pearl identification studies. In summary, we demonstrate that stable isotope analysis and Raman spectroscopy, respectively, are capable of answering two major questions in identifying pearl provenance—geography and species. We found that saltwater pearls were enriched in both 13 C and 18 O relative to freshwater pearls, and δ 18 O values of saltwater pearls were found to further separate by country of origin, while δ 13 C data remained within a similar range of values for all saltwater pearl groups. Further isotopic analysis of shell aragonite samples from the Arabian Gulf revealed that both temperature and evaporation could play an important role in δ 18 O values of carbonates produced by pearl oysters, with the influence that these controls exert being variable from location to location. A comparison of the isotopic composition of shell aragonite, shell calcite, and pearl in oysters from Bahrain revealed that pearls have slightly lower δ 18 O values than shell samples, while shell calcite has lower δ 13 C values than pearl and shell aragonite. Our results show that the isotopic analysis of saltwater pearls could be utilized as a powerful tool for identifying geographic provenance of pearls with unknown origins, and there is a rich body of existing work on the δ 18 O values of minerals formed by bivalves and their link to temperature and hydrological processes in support of this proposal. However, there are uncertainties and limitations with this tool that remain to be examined and further addressed in future studies. Forming a clearer understanding of the factors influencing pearl δ 18 O by acquiring more environmental data on source locations and/or conducting additional isotopic analysis will be key to addressing overlaps in isotopic composition between pearls from different locations and improving the use of mass spectrometry for pearl classification. For species identification, Raman spectra of pearls produced by P. maxima , P. radiata , and P. fucata oysters were used to develop an SVM model that could predict species with an accuracy of 96.4%. SVM training weights were the highest in areas of the spectra corresponding to the characteristic Raman features of aragonite and organic compounds found in pearls. The best model performance was observed using full Raman spectra for training and prediction, with accuracy scores dropping when using a dataset containing only the first or second halves of the Raman spectra. A model using 100 points from the Raman spectra corresponding to the highest SVM training weights achieved similar results as the full-spectrum model but produced a considerably lower accuracy score for P. fucata pearls. Together, these results show that stable isotope analysis and Raman spectroscopy reveal chemical information encoded into pearl nacre that can help uncover important insights on provenance. These techniques thus represent a robust strategy for improving aspects of the pearl classification process, while simultaneously highlighting their value in future studies aimed at addressing underlying questions on the mechanisms of biogenic carbonate formation. Methods Sample Acquisition Samples of pearls and shells were acquired from diverse geographic locations; specimens of Pinctada radiata were collected from the Arabian Gulf (Bahrain, Qatar, and Kuwait), whereas Pinctada fucata were sourced from pearl culturing farms and markets in Japan. Pinctada maxima specimens were obtained from the northern oceans of Australia, whereas the freshwater pearls were collected from Scotland and markets in Japan. The samples were categorized according to varying levels of acquisition confidence ranging from 1-5 in decreasing order of confidence (Supplementary Table 2). Most of the samples were obtained directly from the source by the Bahrain Institute of Pearls and Gemstones (DANAT) field research team (i.e., the team went into the sea to collect samples); this direct acquisition corresponds to a sample confidence level of 1. Samples from Scotland were gathered for DANAT by a trusted external entity without paper trail evidence (Confidence Level 3). The remainder of the samples were acquired from wholesale pearl merchants from the pearl market in Kobe, Japan (Confidence Level 5). Supplementary Table 3 displays samples from different species and their corresponding confidence levels. All natural samples were considered to be untreated raw materials, whereas cultured pearls were regarded as having potentially undergone various treatments, including bleaching. Mass Spectrometry For carbon and oxygen isotopic analysis, powder samples were collected from both pearls and shells. The majority of the analyzed pearls were drilled using a jewelry drill press with a 0.6 mm carbide bur drill bit, whereas pearls that were too small for drilling (i.e., similar to or smaller than the size of the drill bit) were instead ground with an agate mortar and pestle. Powdered shell nacre was collected by passing a Dremel with a rough conical bit over an approximately 1.5 cm strip of nacre at the center of the inner surface of the shell until a 1-2 mm deep groove was formed. For calcite collection, the outer surface of the shell was blasted with compressed air and scraped with a metal spatula to remove larger pieces of debris. An approximately 0.5 cm ´ 2 cm piece of prismatic calcite from the growth edge of the shells was then broken off using pliers and ground using an agate mortar and pestle. Isotope analysis was conducted using a Nu Perspective isotope ratio mass spectrometer. Carbonate samples weighing 50–100 μg were digested in a Nu Carb automated sample preparation unit for 25 minutes in individual glass vials with 150 μL orthophosphoric acid (ρ = 1.93 g/cm³), and the evolved CO₂ gas was purified cryogenically. Purified sample gas and reference gas of known composition were alternately measured on six Faraday collectors (m/z 44–49) in 6 cycles, each with a 30-second integration time (3 minutes total integration time). Each session of approximately 50 individual analyses began with two ETH anchors, then alternated between blocks of six to eight unknowns and two ETH anchors, totaling twelve anchors per run. Data were processed using the “D47crunch” Python package 57 with IUPAC 17 O parameters and 70°C 18 O acid fractionation factors of 1.00871 for calcite and 1.009091 for aragonite 58 . Raw measurements were converted to Vienna Pee Dee Belemnite (VPDB) using a pooled regression approach 59 that used the ETH anchor values from Bernasconi et al. 60 . Nominal anchor values for δ 13 C and δ 18 O (in ‰ VPDB), respectively, are ETH-2: -10.17, -18.69; and ETH-3: 1.71, -1.78 61 . The δ 13 C and δ 18 O values of a portion of the pearl samples were obtained with a set of run parameters optimized for clumped isotope spectrometry (see Supplementary Table 7), which are described in Anderson et al . 62 . Standard errors on the anchor measurements can be seen in the Supplementary Information as a metric of accuracy (Supplementary Table 8). Seawater Temperature, δ 18 O, and Salinity Maps Global sea surface temperature data for Fig. 2b (top left) were sourced from the World Ocean Atlas 2023 using the annual statistical mean on 1/4° grid for all decades 29 . The data were imported into ArcGIS Pro and the Empirical Bayesian Kriging tool (Geostatistical Analyst Tools) was applied to the surface Z value field for interpolation. The resulting data were classified in equal intervals to generate the map. Global seawater δ 18 O data for Fig. 2b (bottom left) were sourced from LeGrande and Schmidt 30 and imported into ArcGIS Pro, where it was classified in equal intervals to generate the map. Temperature and salinity data of the Arabian Gulf as presented in Fig. 3b were sourced from the Global Ocean Physics Analysis and Forecast from E.U. Copernicus Marine Service Information 44 . All monthly temperature and salinity data from 2023 were imported into ArcGIS Pro using the Make NetCDF Raster Layer tool. Data from the shallowest available depth (0.494025 m) of the raster layers were averaged using the Cell Statistics tool in ArcGIS Pro to generate the annual mean surface temperature and salinity maps. Raman Spectroscopy Raman spectra were acquired from pearl samples using two Raman spectrometers: a Renishaw inVia spectrometer and a Qontor Renishaw inVia spectrometer, both integrated with an optical microscope. Calibration for the Renishaw inVia spectrometer was conducted with the 1331.8 cm -1 diamond Raman line. A diode-pumped solid-state laser with an excitation wavelength of 514 nm was used for the measurements. The pearls were exposed to an average laser power of 4 mW. Spectra were acquired from 100 cm -1 to 2000 cm -1 with a resolution of ca. 2 cm -1 , using a grating of 1800 grooves/mm and a 40 μm slit. A 50×/0.75 short-distance objective lens was used, with a laser acquisition duration of 30 seconds and 7 accumulations. The remainder of the pearl Raman spectra acquisition was performed with the Qontor Renishaw inVia spectrometer. Calibration was conducted with a silicon wafer exhibiting a peak at 520.6 cm -1 . A diode-pumped solid-state laser with an excitation wavelength of 457 nm was used, and the laser power was 40 mW. The spectral range was 100 cm -1 to 2000 cm -1 with a detector resolution of ca. 1 cm -1 . A grating of 2400 grooves/mm, a notch filter, and a 40 μm slit were used. A 50× long-distance objective lens was used with an exposure duration of 1 second and 10 accumulations. Algorithm Development Pearl Measurements: Carbon and oxygen isotope data of saltwater pearls from Australia, Bahrain, Japan, and Qatar were used with an 80:20 train:test split. Raman data were collected and pre-processed using Python glob module (glob.glob) for a linear fit. ML Algorithms: All algorithms were developed and optimized using the SciKit Learn (sklearn) module in Python. For the distinction of saltwater pearls based on isotope data, a logistic regression model, a k-nearest neighbors (k-NN) model, and a support vector machine (SVM) for four classes (Australia, Bahrain, Japan, Qatar), and a logistic regression model for three classes (Australia, Bahrain and Qatar, Japan) were developed. Logistic regression models had the following parameters: solver = ‘lbfgs’, max_iter = 1000, and multi_class = ‘multinomial’. The k-NN model was developed with n_neighbors = 5. For the C-Support Vector Classification (SVC) model, a grid search algorithm (GridSearchCV) was used to determine the best parameters. Regularization term (C) values of 0.1, 1, 10, and 100, and kernel coefficient (gamma) values of 1, 0.1, 0.01, and 0.001 for ‘rbf’, ‘linear’, and ‘poly’ kernels were investigated. The best model had parameter values of C = 100 and gamma = 1 for the ‘poly’ kernel, which was then used to plot decision boundaries between classes. A similar SVC model was developed for predicting pearl species based on Raman spectra. A linear kernel with C = 1.0 and gamma = 0.5 was applied. Accuracy scores and confusion matrices were imported from sklearn.metrics for evaluating model performance. Selected plots were generated using matplotlib.plot and seaborn.heatmap. Statistics Two-tailed comparisons between pairings of pearl, shell aragonite, and shell calcite δ 18 O values for samples from Bahrain (n = 101) were performed using a linear mixed effects model (mixedlm) from the SciKit Learn module in Python with a = 0.05. For each pairing the calculated p-values were < 0.001: for shell aragonite vs. shell calcite, p = 0.00050; for shell aragonite vs. pearl, p = 0.000025; and for shell calcite vs. pearl, p = 8.7 ´ 10 -15 . A linear mixed effects model was used given the hierarchical structuring of the data and expected variations in mineral δ 18 O values by specimen and by location. Location was set as a cluster group and a random slope was allowed for unique specimens. A histogram and a quantile-quantile plot of the residuals are shown in Supplementary Fig. 7a-b, and the Shapiro-Wilk test (shapiro) was conducted using the SciPy module in Python to generate W = 0.983 and p = 0.21 to support the assumption of normality of the residuals. Additionally, a plot of the residuals vs. fitted values and a boxplot of the distribution of residuals by location are shown in Supplementary Fig. 7c-d, and White’s Lagrange Multiplier Test for Heteroscedasticity (het_white) was conducted using the Stastmodels module in Python to generate a Lagrange Multiplier statistic of 0.18218 with a corresponding p-value of 0.91294 and an F-statistic of 0.08854 with a corresponding p-value of 0.91534, to support the assumption of homoscedasticity of variance. Declarations Data Availability All relevant data are available from the corresponding author upon reasonable request, subject to evaluation by the authors. Acknowledgments We would like to thank the team at the Bahrain Institute for Pearls and Gemstones (DANAT) for acquiring all samples analyzed in this work. We would also like to thank Claire Hayhow for assistance in collecting pearl and shell isotope data. We would like to thank Dr. Nicu-Viorel Atudorei and Dr. Abdul-Mehdi Ali from the University of New Mexico for the coordination and collection of seawater isotope and trace element measurements. Author Contributions Conceptualization: NJ, JCW, KDB, VB, AM Secured Funding: NJ, VB, AM Project Supervision: NJ, JCW, KDB, VB, AM Provided Instrumentation: NJ, KDB, VB, AM Experimental Design: All authors Data Collection: DK, RZ, ABJ, AZ, AA, JCW Data Analysis: DK, RZ, ABJ, AZ Manuscript Writing: DK, RZ, JCW, AM All authors reviewed and edited the manuscript. Competing Interests The authors declare no competing interests. References Weiner, S. & Traub, W. Macromolecules in mollusc shells and their functions in biomineralization. Philos. Trans. R. Soc. Lond. B Biol. Sci. 304 , 425–434 (1984). Addadi, L., Joester, D., Nudelman, F. & Weiner, S. Mollusk Shell Formation: A Source of New Concepts for Understanding Biomineralization Processes. Chem. Eur. J. 12 , 980–987 (2006). Marin, F. The formation and mineralization of mollusk shell. Front. Biosci. S4 , 1099–1125 (2012). Strack, E. Pearls . (Rühle-Diebener-Verlag, Stuttgart, 2006). Sturman, N. et al. A Pearl Identification Challenge. Gems Gemol. 55 , 229–243 (2019). The Pearl Oyster . (Elsevier Science, 2011). Simkiss, K. & Wada, K. Cultured pearls—commercialised biomineralisation. Endeavour 4 , 32–37 (1980). Miyoshi, T., Matsuda, Y. & Komatsu, H. Fluorescence from Pearls to Distinguish Mother Oysters Used in Pearl Culture. Jpn. J. Appl. Phys. 26 , 578 (1987). Wehrmeister, U. et al. Visualization of the internal structures of cultured pearls by computerized X-ray microtomography. J. Gemmol. 31 , 15–21 (2008). Vigorelli, L. et al. X-ray Micro-Tomography as a Method to Distinguish and Characterize Natural and Cultivated Pearls. Condens. Matter 6 , 51 (2021). Krzemnicki, M. S., Friess, S. D., Chalus, P., Hänni, H. A. & Karampelas, S. X-Ray Computed Microtomography: Distinguishing Natural Pearls from Beaded and Non-Beaded Cultured Pearls. Gems Gemol. 46 , 128–134 (2010). Hänni, H. A., Kiefert, L. & Giese, P. X-ray luminescence, a valuable test in pearl identification. J. Gemmol. 29 , 325–329 (2005). Karampelas, S. Spectral Characteristics of Natural-Color Saltwater Cultured Pearls from Pinctada Maxima . Gems Gemol. 48 , 193–197 (2012). Tsai, T.-H. & Zhou, C. Rapid detection of color-treated pearls and separation of pearl types using fluorescence analysis. in Novel Optical Systems, Methods, and Applications XXIII (eds. Hahlweg, C. F. & Mulley, J. R.) 6 (SPIE, Online Only, United States, 2020). doi:10.1117/12.2566590. Shi, L., Wang, Y., Liu, X. & Mao, J. Component Analysis and Identification of Black Tahitian Cultured Pearls From the Oyster Pinctada margaritifera Using Spectroscopic Techniques. J. Appl. Spectrosc. 85 , 98–102 (2018). Yan, J. et al. Origin of the common UV absorption feature in cultured pearls and shells. J. Mater. Sci. 52 , 8362–8369 (2017). Agatonovic-Kustrin, S. & Morton, D. W. The Use of UV-Visible Reflectance Spectroscopy as an Objective Tool to Evaluate Pearl Quality. Mar. Drugs 10 , 1459–1475 (2012). Karampelas, S., Fritsch, E., Gauthier, J.-P. & Hainschwang, T. UV-Vis-NIR Reflectance Spectroscopy of Natural-Color Saltwater Cultured Pearls from Pinctada Margaritifera . Gems Gemol. 47 , 31–35 (2011). Soldati, A. L., Jacob, D. E., Wehrmeister, U., Häger, T. & Hofmeister, W. Micro-Raman spectroscopy of pigments contained in different calcium carbonate polymorphs from freshwater cultured pearls. J. Raman Spectrosc. 39 , 525–536 (2008). Karampelas, S., Fritsch, E., Makhlooq, F., Mohamed, F. & Al‐Alawi, A. Raman spectroscopy of natural and cultured pearls and pearl producing mollusc shells. J. Raman Spectrosc. 51 , 1813–1821 (2020). Hardman, M. F. et al. Classification of Gem Materials Using Machine Learning. Gems Gemol. 60 , 306–329 (2024). Homkrajae, A., Sun, Z., Blodgett, T. & Zhou, C. Provenance Discrimination of Freshwater Pearls by LA-ICP-MS and Linear Discriminant Analysis (LDA). Gems Gemol. 55 , 47–60 (2019). McCrea, J. M. On the Isotopic Chemistry of Carbonates and a Paleotemperature Scale. J. Chem. Phys. 18 , 849–857 (1950). Epstein, S., Buchsbaum, R., Lowenstam, H. A. & Urey, H. C. REVISED CARBONATE-WATER ISOTOPIC TEMPERATURE SCALE. Geol. Soc. Am. Bull. 64 , 1315 (1953). Immenhauser, A., Schöne, B. R., Hoffmann, R. & Niedermayr, A. Mollusc and brachiopod skeletal hard parts: Intricate archives of their marine environment. Sedimentology 63 , 1–59 (2016). Clark, I. D. & Fritz, P. Environmental Isotopes in Hydrogeology . (CRC Press/Lewis Publishers, Boca Raton, FL, 1997). Epstein, S. & Mayeda, T. Variation of O 18 content of waters from natural sources. Geochim. Cosmochim. Acta 4 , 213–224 (1953). Bigg, G. R. & Rohling, E. J. An oxygen isotope data set for marine waters. J. Geophys. Res. Oceans 105 , 8527–8535 (2000). Locarnini, R. A. et al. World Ocean Atlas 2023, Volume 1: Temperature. (2024) doi:10.25923/54BH-1613. LeGrande, A. N. & Schmidt, G. A. Global gridded data set of the oxygen isotopic composition in seawater. Geophys. Res. Lett. 33 , 2006GL026011 (2006). Spero, H. J., Bijma, J., Lea, D. W. & Bemis, B. E. Effect of seawater carbonate concentration on foraminiferal carbon and oxygen isotopes. Nature 390 , 497–500 (1997). McConnaughey, T. A. & Gillikin, D. P. Carbon isotopes in mollusk shell carbonates. Geo-Mar. Lett. 28 , 287–299 (2008). Clayton, R. N. & Degens, E. T. Use of Carbon Isotope Analyses of Carbonates for Differentiating Fresh-Water and Marine Sediments: GEOLOGICAL NOTES. AAPG Bull. 43 , 890–897 (1959). Silverman, S. R. & Epstein, S. Carbon Isotopic Compositions of Petroleums and Other Sedimentary Organic Materials. AAPG Bull. 42 , 998–1102 (1958). Keith, M. L., Anderson, G. M. & Eichler, R. Carbon and oxygen isotopic composition of mollusk shells from marine and fresh-water environments. Geochim. Cosmochim. Acta 28 , 1757–1786 (1964). Boutton, T. W. Stable carbon isotope ratios of natural materials: 2. Atmospheric, terrestrial, marine, and freshwater environments. in Carbon isotope techniques (1991). Quay, P. D., Emerson, S. R., Quay, B. M. & Devol, A. H. The carbon cycle for Lake Washington—A stable isotope study. Limnol. Oceanogr. 31 , 596–611 (1986). Weber, J. N. & Woodhead, P. M. J. Temperature dependence of oxygen-18 concentration in reef coral carbonates. J. Geophys. Res. 77 , 463–473 (1972). Gilbert, P. U. P. A. et al. Biomineralization: Integrating mechanism and evolutionary history. Sci. Adv. 8 , eabl9653 (2022). Yoshimura, T., Nakashima, R., Suzuki, A., Tomioka, N. & Kawahata, H. Oxygen and carbon isotope records of cultured freshwater pearl mussel Hyriopsis sp . shell from Lake Kasumigaura, Japan. J. Paleolimnol. 43 , 437–448 (2010). Farfan, G. A., Zhou, C., Valley, J. W. & Orland, I. J. Coupling Mineralogy and Oxygen Isotopes to Seasonal Environmental Shifts Recorded in Modern Freshwater Pearl Nacre From Kentucky Lake. Geochem. Geophys. Geosystems 22 , e2021GC009995 (2021). Kawahata, H., Inoue, M., Nohara, M. & Suzuki, A. Stable isotope and chemical composition of pearls: Biomineralization in cultured pearl oysters in Ago Bay, Japan. J. Oceanogr. 62 , 405–412 (2006). Woo, K. S. Textural, Isotopic, and Chemical Investigation of Cultured Pearls. J. Oceanol. Soc. Korea 24 , 69–78 (1989). European Union-Copernicus Marine Service. Global Ocean 1/12° Physics Analysis and Forecast updated Daily. Mercator Ocean International https://doi.org/10.48670/MOI-00016 (2016). Reynolds, M. R. Physical oceanography of the Gulf, Strait of Hormuz, and the Gulf of Oman—Results from the Mt Mitchell expedition. Mar. Pollut. Bull. 27 , 35–59 (1993). Vaughan, G. O., Al-Mansoori, N. & Burt, J. A. The Arabian Gulf. in World Seas: an Environmental Evaluation 1–23 (Elsevier, 2019). doi:10.1016/B978-0-08-100853-9.00001-4. The Gulf Ecosystem: Health and Sustainability . (Backhuys, Leiden, 2002). Rivers, J. M. et al. The Geochemistry of Qatar Coastal Waters and its Impact on Carbonate Sediment Chemistry and Early Marine Diagenesis. J. Sediment. Res. 89 , 293–309 (2019). Conroy, J. L. et al. Spatiotemporal variability in the δ 18 O‐salinity relationship of seawater across the tropical Pacific Ocean. Paleoceanography 32 , 484–497 (2017). Zeebe, R. & Wolf-Gladrow, D. CO 2 in Seawater: Equilibrium, Kinetics, Isotopes . (Elsevier, Amsterdam, 2007). Rubinson, M. & Clayton, R. N. Carbon-13 fractionation between aragonite and calcite. Geochim. Cosmochim. Acta 33 , 997–1002 (1969). Lécuyer, C. et al. Carbon and oxygen isotope fractionations between aragonite and calcite of shells from modern molluscs. Chem. Geol. 332–333 , 92–101 (2012). Romanek, C. S., Grossman, E. L. & Morse, J. W. Carbon isotopic fractionation in synthetic aragonite and calcite: Effects of temperature and precipitation rate. Geochim. Cosmochim. Acta 56 , 419–430 (1992). Ghosh, P. et al. 13 C– 18 O bonds in carbonate minerals: A new kind of paleothermometer. Geochim. Cosmochim. Acta 70 , 1439–1456 (2006). Eiler, J. M. “Clumped-isotope” geochemistry—The study of naturally-occurring, multiply-substituted isotopologues. Earth Planet. Sci. Lett. 262 , 309–327 (2007). McConnaughey, T. 13 C and 18 O isotopic disequilibrium in biological carbonates: I. Patterns. Geochim. Cosmochim. Acta 53 , 151–162 (1989). Daëron, M. & Vermeesch, P. Omnivariant Generalized Least Squares Regression: Theory, Geochronological Applications, and Making the Case for Reconciled Δ47 calibrations. Chem. Geol. 647 , 121881 (2024). Kim, S.-T., Mucci, A. & Taylor, B. E. Phosphoric acid fractionation factors for calcite and aragonite between 25 and 75 °C: Revisited. Chem. Geol. 246 , 135–146 (2007). Daëron, M. Full Propagation of Analytical Uncertainties in Δ 47 Measurements. Geochem. Geophys. Geosystems 22 , e2020GC009592 (2021). Bernasconi, S. M. et al. InterCarb: A Community Effort to Improve Interlaboratory Standardization of the Carbonate Clumped Isotope Thermometer Using Carbonate Standards. Geochem. Geophys. Geosystems 22 , e2020GC009588 (2021). Bernasconi, S. M. et al. Reducing Uncertainties in Carbonate Clumped Isotope Analysis Through Consistent Carbonate‐Based Standardization. Geochem. Geophys. Geosystems 19 , 2895–2914 (2018). Anderson, N. T., Bergmann, K. D., Braun, M. G., Griffith, E. M. & Saltzman, M. R. High-resolution record of global cooling during a large Mississippian positive carbon isotope excursion. Earth Planet. Sci. Lett. 668 , 119557 (2025). Additional Declarations There is NO Competing Interest. Supplementary Files SupplementaryInformation.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6874927","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":593983261,"identity":"03042d6f-038a-4086-9806-1724a1427422","order_by":0,"name":"Admir Masic","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAnklEQVRIiWNgGAWjYJCCAwwVDDwghgQJWs4w8PCQpIWBsY2BgXgtuu3HHx74OW+bjD0D88HbPMRoMTuTY3Cwd9ttoMPYkq2J03Igh+EwI1gLj5k0cVrOP39wmHEOSAv/NyK13EgwOMzYALaFjVgtbwwO9hwDajnMZmw5hziHpT/+8KPmtj17e/PDG2+I0YIAzKQpHwWjYBSMglGADwAAVdcvkCx0wqAAAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0002-8791-175X","institution":"Massachusetts Institute of Technology","correspondingAuthor":true,"prefix":"","firstName":"Admir","middleName":"","lastName":"Masic","suffix":""},{"id":593983262,"identity":"9afff5cc-8413-47f8-ba01-c95bd37da8e7","order_by":1,"name":"Dahyun Kyung","email":"","orcid":"https://orcid.org/0009-0000-6399-9876","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Dahyun","middleName":"","lastName":"Kyung","suffix":""},{"id":593983263,"identity":"b9595de8-9cdb-44d1-b840-b407ca8e683f","order_by":2,"name":"Ruiqi Zhang","email":"","orcid":"https://orcid.org/0000-0003-3901-8737","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Ruiqi","middleName":"","lastName":"Zhang","suffix":""},{"id":593983264,"identity":"78b576a6-e317-44d0-a62f-8c24d21e73c2","order_by":3,"name":"Alice Zehner","email":"","orcid":"https://orcid.org/0009-0000-3265-7806","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Alice","middleName":"","lastName":"Zehner","suffix":""},{"id":593983265,"identity":"dbf8bb69-5a90-45d1-a362-4bbeea0bac2d","order_by":4,"name":"Ali Alatawi","email":"","orcid":"","institution":"Bahrain Institute for Pearls and Gemstones (DANAT)","correspondingAuthor":false,"prefix":"","firstName":"Ali","middleName":"","lastName":"Alatawi","suffix":""},{"id":593983266,"identity":"6b787870-30e7-4ac9-a8b9-9cfd3058770e","order_by":5,"name":"Noora Jamsheer","email":"","orcid":"","institution":"Bahrain Institute for Pearls and Gemstones (DANAT)","correspondingAuthor":false,"prefix":"","firstName":"Noora","middleName":"","lastName":"Jamsheer","suffix":""},{"id":593983267,"identity":"2950ba66-d533-4760-b140-7c009f664e02","order_by":6,"name":"James Weaver","email":"","orcid":"","institution":"Harvard University","correspondingAuthor":false,"prefix":"","firstName":"James","middleName":"","lastName":"Weaver","suffix":""},{"id":593983268,"identity":"2c288e36-f908-4d1c-846c-609f32df2075","order_by":7,"name":"Kristin Bergmann","email":"","orcid":"https://orcid.org/0000-0002-6106-2059","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Kristin","middleName":"","lastName":"Bergmann","suffix":""},{"id":593983269,"identity":"e24a12be-4ebb-49b1-a332-f98fe30a1b16","order_by":8,"name":"Vladimir Bulovic","email":"","orcid":"https://orcid.org/0000-0002-0960-2580","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Vladimir","middleName":"","lastName":"Bulovic","suffix":""},{"id":593983270,"identity":"414be576-c00f-4217-b32e-51101ec2e2d3","order_by":9,"name":"Adam Jost","email":"","orcid":"","institution":"Massachusetts Institute of Technology","correspondingAuthor":false,"prefix":"","firstName":"Adam","middleName":"","lastName":"Jost","suffix":""}],"badges":[],"createdAt":"2025-06-11 21:30:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6874927/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6874927/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":103517515,"identity":"199cd6a8-ec24-483e-b7b6-184e8710364a","added_by":"auto","created_at":"2026-02-26 14:33:01","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":1811717,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eData analysis pipeline for the identification of pearl provenance.\u003c/strong\u003e Left: The outer surface of pearls is primarily composed of nacre, which exhibits a characteristic brick and mortar-like organization of aragonite tablets (CaCO\u003csub\u003e3\u003c/sub\u003e; Ca: green, C: gray, O: red) separated by thin organic inter-layers. Using a wide range of materials characterization tools (center) in conjunction with data clustering and machine learning (ML) algorithms, the present study focuses on the development of robust methods for the identification of pearl geographical origin and host species (right).\u003c/p\u003e","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-6874927/v1/849d5c90382636dd75e57d70.png"},{"id":103517516,"identity":"9cfb58c0-16bb-427d-890d-23f60f77fd32","added_by":"auto","created_at":"2026-02-26 14:33:02","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":542049,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMajor drivers of mineral δ\u003c/strong\u003e\u003csup\u003e\u003cstrong\u003e18\u003c/strong\u003e\u003c/sup\u003e\u003cstrong\u003eO and isotopic analysis of pearl carbonate minerals\u003c/strong\u003e.\u003cstrong\u003e \u003c/strong\u003ea) Hydrological processes such as evaporation and precipitation are among the main mechanisms driving the differences in freshwater and saltwater δ\u003csup\u003e18\u003c/sup\u003eO. Evaporation leads to the preferential uptake of \u003csup\u003e16\u003c/sup\u003eO into the vapor phase; as such, regions marked by high evaporation, such as the Arabian Gulf, can show pronounced enrichment in \u003csup\u003e18\u003c/sup\u003eO. Freshwater bodies are supplied by meteoric water, which becomes isotopically lighter than the evaporative source (i.e., seawater) through repeated precipitation events. b) Mean annual sea surface temperature\u003csup\u003e29\u003c/sup\u003e (top left) and seawater δ\u003csup\u003e18\u003c/sup\u003eO\u003csup\u003e30\u003c/sup\u003e (bottom left) maps illustrate the compounding effects of these processes on a global scale. A map of the sampled saltwater pearl source locations is provided on the right. c) A plot of the stable carbon and oxygen isotope ratios (i.e., δ\u003csup\u003e13\u003c/sup\u003eC vs. δ\u003csup\u003e18\u003c/sup\u003eO) of pearls from freshwater and saltwater sources. Plot marker colors correspond to source location. Separation between freshwater (green region) and saltwater (pink region) samples is highlighted in the middle panel. Furthermore, there is clustering of the saltwater pearl data, primarily along the δ\u003csup\u003e18\u003c/sup\u003eO axis, based on geography. The rightmost panel shows\u003cstrong\u003e \u003c/strong\u003elogistic regression decision boundaries, demonstrating the separation of the saltwater pearl carbonate isotopes by geography. Error bars show standard deviations for samples with multiple measurements, for which plot markers show the mean δ\u003csup\u003e13\u003c/sup\u003eC and δ\u003csup\u003e18\u003c/sup\u003eO values.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-6874927/v1/6d7e8c83f1de9ece60f82ca8.png"},{"id":103517159,"identity":"d21f5b3a-c8b6-40ca-971c-23191f8d055b","added_by":"auto","created_at":"2026-02-26 14:31:36","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":730424,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eδ\u003c/strong\u003e\u003csup\u003e\u003cstrong\u003e13\u003c/strong\u003e\u003c/sup\u003e\u003cstrong\u003eC and δ\u003c/strong\u003e\u003csup\u003e\u003cstrong\u003e18\u003c/strong\u003e\u003c/sup\u003e\u003cstrong\u003eO analysis of shells and pearls from the Arabian Gulf.\u003c/strong\u003e a) Map showing the Arabian Gulf (the source of the samples analyzed in this figure) in a broader geographic context. b) Annual mean surface temperature (left) and salinity (right) maps of the Arabian Gulf\u003csup\u003e44\u003c/sup\u003e, with markers denoting the specific locations from which samples were collected. Plot marker colors correspond to source location, with triangles representing Bahrain locations, crosses representing Kuwait locations, and diamonds representing Qatar locations. c) δ\u003csup\u003e13\u003c/sup\u003eC vs. δ\u003csup\u003e18\u003c/sup\u003eO plot of shell aragonite from specific locations in Bahrain, Kuwait, and Qatar. Differences in shading in plot markers of the same shape correspond to different source locations within a single country. d) δ\u003csup\u003e13\u003c/sup\u003eC vs. δ\u003csup\u003e18\u003c/sup\u003eO plots showing separability of samples from different locations in Qatar (left; Q1: Al Ghariyah, Q2: Al Aaliya Island) and between samples from Bahrain and Kuwait (right; B: Bahrain, K: Kuwait). e) δ\u003csup\u003e13\u003c/sup\u003eC vs. δ\u003csup\u003e18\u003c/sup\u003eO plot of shell aragonite, shell calcite, and pearls from \u003cem\u003eP. radiata\u003c/em\u003e oysters from Bahrain. Differences in plot marker shading correspond to different source locations in Bahrain, with triangles corresponding to shell aragonite, squares corresponding to shell calcite, and circles corresponding to pearl aragonite. Error bars in c)-e) show standard deviations for samples with multiple measurements, for which plot markers show the mean δ\u003csup\u003e13\u003c/sup\u003eC and δ\u003csup\u003e18\u003c/sup\u003eO values. f) Violin plots comparing δ\u003csup\u003e18\u003c/sup\u003eO values of shell aragonite, shell calcite, and pearl aragonite samples from pearl oysters collected in Bahrain. Boxes in box plots show the interquartile range and whiskers extend to the most extreme values within 1.5 times the interquartile range beyond the first and third quartiles. The white bands in the boxes represent the mean values. A mixed linear model was used to determine significant differences between pairings of δ\u003csup\u003e18\u003c/sup\u003eO values of shell aragonite, shell calcite, and pearl aragonite, with specimen and source location considered as random effects. Three asterisks above bars connecting two groups represent a significant difference between the two groups with p \u0026lt; 0.001.\u003c/p\u003e","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-6874927/v1/3efff488a408df3b11ed1b9a.png"},{"id":103517317,"identity":"430f8abd-3981-44f1-b5a4-84251f3307bb","added_by":"auto","created_at":"2026-02-26 14:32:22","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":762910,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eRaman spectra and SVM results of pearls from different oyster species.\u003c/strong\u003e a) Data distribution of the training and testing datasets. Green, cyan, and blue represent the pearls from \u003cem\u003eP. maxima\u003c/em\u003e, \u003cem\u003eP. radiata\u003c/em\u003e, and \u003cem\u003eP. fucata\u003c/em\u003e, respectively. The larger pie chart beneath represents the training dataset, while the upper pie chart represents the testing dataset. b) Averaged SVM training weights corresponding to each data point (250 points in total) on the input Raman spectra. The training weight of each Raman data point is denoted by a green dot corresponding to its position on the processed Raman spectrum. Chemical signatures of the highly weighted parameter areas are listed in purple. c) Confusion matrix of 3000 SVM prediction results. The x-axis represents the SVM prediction category, and the y-axis represents the ground truth category. The green confusion matrix (left) indicates the number counts of each prediction result, whereas the blue confusion matrix (right) indicates the percent error of the predictions. Matrices are interpreted from both their horizontal and vertical entries. The diagonal entries in the matrix represent accurate predictions, whereas the off-diagonal entries indicate incorrect predictions.\u003c/p\u003e","description":"","filename":"floatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-6874927/v1/cd278c4221dffa854f8eb8b2.png"},{"id":103517467,"identity":"ebd1b50d-c8a0-4569-b137-2dc72dbbfa25","added_by":"auto","created_at":"2026-02-26 14:32:54","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":1024498,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eEffects of Raman spectral features on SVM prediction accuracy.\u003c/strong\u003e a) Confusion matrices of 3000 SVM prediction results based on the first (left, red) and second (right, yellow) halves of the Raman spectra (125 points each). b) Prediction accuracy comparison over trials based on the original full spectrum (black), first-half (red), and second-half spectrum (yellow). c) 100 SVM training weights based on selected spectral regions from Fig. 4b. A processed Raman spectrum with selected wavenumber values is illustrated as a reference, shown in the black dotted-line graph. d) Accuracy scores for a prediction of pearl oyster species using four different input spectral features. The original full spectrum has the overall best prediction performance compared to the other three selected-feature Raman spectrum inputs.\u003c/p\u003e","description":"","filename":"floatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-6874927/v1/9e9896ecec0a7eacb6efea95.png"},{"id":104400067,"identity":"774e4849-6d72-4d3f-a64d-0db9c7b9845b","added_by":"auto","created_at":"2026-03-11 12:08:43","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5719339,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6874927/v1/c0d7c216-6cbf-4997-95dd-3d57b7ea35c9.pdf"},{"id":103517419,"identity":"f090e428-b107-4aaa-91af-8b6cf7e23e9c","added_by":"auto","created_at":"2026-02-26 14:32:39","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2434424,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryInformation.docx","url":"https://assets-eu.researchsquare.com/files/rs-6874927/v1/a519ef0cb966cf3176df72ba.docx"}],"financialInterests":"There is \u003cb\u003eNO\u003c/b\u003e Competing Interest.","formattedTitle":"Decoding Nacre: Uncovering Unique Identifiers of Pearl Provenance","fulltext":[{"header":"Introduction","content":"\u003cp\u003ePearls are unique among precious gems in that they are produced through biological processes. The outer surfaces of pearls are primarily composed of nacre (Fig. 1), a hierarchically structured brick and mortar-like nanocomposite material consisting of aragonite (CaCO\u003csub\u003e3\u003c/sub\u003e) tablets separated by thin organic inter-layers\u003csup\u003e1\u0026ndash;3\u003c/sup\u003e. Within the broad range of commercially available pearl types\u003csup\u003e4,5\u003c/sup\u003e, one major classification distinction is based on whether a pearl is natural or cultured. Natural pearls are formed without any human intervention and rarely occur\u003csup\u003e4\u003c/sup\u003e, while in cultured pearls, pearl formation is artificially induced through the implantation of a piece of tissue from a donor mollusk, oftentimes along with an inorganic bead, into a host mollusk\u003csup\u003e6,7\u003c/sup\u003e. In addition, pearls can be further subdivided based on whether they were formed by saltwater or freshwater mollusks, and further distinctions are made between pearls produced in different geographical locations and by different mollusk species\u003csup\u003e4,6,8\u003c/sup\u003e. Despite this incredible diversity in the employed mollusk species, production method, and geographical origin, all of which have a profound impact on pearl economic value, the ability to accurately classify pearls based on visual inspection alone has been historically problematic since the unique identifying features of different pearl types can be extremely subtle or ambiguous.\u003c/p\u003e\n\u003cp\u003eTo help augment these visual inspection-based workflows, gemologists can also employ a diverse set of analytical tools to classify pearls by structure and chemistry, some examples of which are shown in Fig. 1\u003csup\u003e4,5,9\u003c/sup\u003e. For example, cultured and natural pearls can be differentiated using x-ray imaging techniques such as x-ray microradiography and micro-computed tomography (micro-CT), \u0026nbsp; which can non-destructively visualize internal structural features such as pearl nuclei, growth rings, and voids\u003csup\u003e9-11\u003c/sup\u003e. Freshwater and saltwater pearls can be routinely distinguished using x-ray fluorescence (XRF) techniques, as freshwater pearls have a higher manganese content than saltwater pearls and thus fluoresce under x-ray irradiation\u003csup\u003e12\u003c/sup\u003e. Additionally, spectroscopy methods including photoluminescence\u003csup\u003e13,14\u003c/sup\u003e, ultraviolet-visible-near infrared\u003csup\u003e13,15\u0026ndash;18\u003c/sup\u003e, Raman\u003csup\u003e15,19,20\u003c/sup\u003e, and fluorescence spectroscopy\u003csup\u003e8\u003c/sup\u003e have been used to identify organic compounds (e.g., pigments) in pearl nacre. The unique combinations and quantities of these organic compounds in nacre have been linked to different mollusk species, allowing for the identification of specific spectral features to be attributed to pearls produced by distinct species and enabling the identification of pearls that have been artificially enhanced through color treatments.\u003c/p\u003e\n\u003cp\u003eAs with visual inspection-based classification workflows, however, the utilization of specific spectroscopic features for pearl classification may not always present themselves clearly and can be difficult to interpret\u003csup\u003e5,14,21,22\u003c/sup\u003e. Furthermore, previous studies have been largely correlative and have not been widely employed for pearl identification purposes across the broad range of commercially available pearl types. There thus remains a need for the development of methods that can enhance the existing suite of characterization tools by providing useful information on the different aspects of a pearl, which would consequently allow for improved provenance tracing and economic valuation of pearls of uncertain origin. Toward achieving this goal, in the present study we present two techniques\u0026mdash;stable isotope analysis of carbonate minerals and Raman spectroscopy\u0026mdash;for the identification of pearl geographic provenance and species, respectively (Fig. 1). A summary of the analyzed samples can be found in Supplementary Table 1, with additional details in Supplementary Tables 2 and 3.\u003c/p\u003e"},{"header":"Results","content":"\u003ch2\u003eStable Isotope Analysis of Carbonates\u003c/h2\u003e\n\u003cp\u003eCarbonate stable isotope analysis is a well-documented technique which is commonly utilized for paleoclimate reconstructions, as the carbon and oxygen isotopic compositions of biogenic carbonate minerals are reflective of the conditions in which they were formed. Carbonate \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO are measures of the ratio of \u003csup\u003e13\u003c/sup\u003eC to \u003csup\u003e12\u003c/sup\u003eC and \u003csup\u003e18\u003c/sup\u003eO to \u003csup\u003e16\u003c/sup\u003eO, respectively, relative to a standard (Vienna Pee Dee Belemnite or VPDB) and expressed in per mil (\u0026permil;). Correlations between mineral \u0026delta;\u003csup\u003e18\u003c/sup\u003eO, temperature, and water \u0026delta;\u003csup\u003e18\u003c/sup\u003eO (standardized relative to Vienna Standard Mean Ocean Water or VSMOW), have long been studied and form the basis of a commonly applied palaeothermometry method\u003csup\u003e23\u0026ndash;25\u003c/sup\u003e. Carbonate \u0026delta;\u003csup\u003e18\u003c/sup\u003eO is linked to both mineralization temperature and hydrological processes such as evaporation and precipitation. Temperature exerts a direct control on mineral \u0026delta;\u003csup\u003e18\u003c/sup\u003eO through the temperature-dependent equilibrium fractionation between water and dissolved carbonate: at lower temperatures, an increased proportion of \u003csup\u003e18\u003c/sup\u003eO is incorporated into dissolved carbonate, which is subsequently captured by the carbonate mineral\u003csup\u003e23,24\u003c/sup\u003e. Hydrological processes are a primary driver of water \u0026delta;\u003csup\u003e18\u003c/sup\u003eO, which is in turn recorded in the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO of biogenic carbonates\u003csup\u003e24,25\u003c/sup\u003e. As illustrated in Fig. 2a, the evaporation of seawater leads to enrichment of the heavier \u003csup\u003e18\u003c/sup\u003eO isotope in the source water as the lighter \u003csup\u003e16\u003c/sup\u003eO isotope preferentially enters the vapor phase, which then becomes the water source for subsequent precipitation. Since freshwater bodies are primarily supplied and replenished by meteoric water, they are consequently more abundant in the lighter \u003csup\u003e16\u003c/sup\u003eO isotope than the evaporative source (i.e., seawater)\u003csup\u003e26\u003c/sup\u003e. Environmental factors such as water depth, circulation, and salinity can additionally influence water \u0026delta;\u003csup\u003e18\u003c/sup\u003eO\u003csup\u003e26\u0026ndash;28\u003c/sup\u003e and, by consequence, mineral \u0026delta;\u003csup\u003e18\u003c/sup\u003eO. To illustrate the compounding effects of these processes on a global scale, maps of temperature\u003csup\u003e29\u003c/sup\u003e and seawater \u0026delta;\u003csup\u003e18\u003c/sup\u003eO\u003csup\u003e30\u003c/sup\u003e are shown in Fig. 2b. Beyond the value of measuring biomineral \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values, complementary information can also be obtained from C isotopic analysis, since the \u0026delta;\u003csup\u003e13\u003c/sup\u003eC of biogenic carbonates has been linked to metabolic processes such as respiration and feeding, and the \u0026delta;\u003csup\u003e13\u003c/sup\u003eC of dissolved inorganic carbon in the water\u003csup\u003e31,32\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eMotivated by these observed variabilities in global isotopic concentrations and their impact on precipitated biominerals, analyses of stable C and O isotopes in pearl nacre were conducted to determine whether carbonate isotopic composition sufficiently captures geographic differences for accurate identification of pearl origins. The plot of \u0026delta;\u003csup\u003e13\u003c/sup\u003eC against \u0026delta;\u003csup\u003e18\u003c/sup\u003eO in Fig. 2c shows a clear distinction between freshwater and saltwater pearl samples, with the saltwater pearls being more enriched in both \u003csup\u003e13\u003c/sup\u003eC and \u003csup\u003e18\u003c/sup\u003eO. There is also considerably wider spread in the freshwater data and no clear trends related to geographic origin. However, the saltwater pearls demonstrate separation, mainly along the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO axis, based on where they were formed. To assess the possibility of automating the task of identifying the geographic provenance of saltwater pearls, a logistic regression model was constructed to separate the saltwater pearl isotope data. The logistic regression model produced an accuracy score of 88%, demonstrating that the geographic origin of saltwater pearls can indeed be classified based on pearl carbonate stable isotopic composition. The logistic regression decision boundaries are shown in the rightmost panel of Fig. 2c. A confusion matrix for this model can be seen in Supplementary Fig. 1, along with decision boundaries and confusion matrices for support vector machine and k-nearest neighbors models, for both of which we report accuracy scores of 85%.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eIn agreement with these observed results, previous comparisons of marine and non-marine carbonates show similar trends in \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO in numerous types of geologic and biogenic samples\u003csup\u003e33\u0026ndash;35\u003c/sup\u003e. The relative enrichment of saltwater pearls in \u003csup\u003e18\u003c/sup\u003eO compared to freshwater pearls aligns with the differences in freshwater and saltwater \u0026delta;\u003csup\u003e18\u003c/sup\u003eO that are primarily driven by the hydrological processes referenced above (Fig. 2a)\u003csup\u003e26,28\u003c/sup\u003e. The difference in carbon isotope ratios may be accounted for by the sources of carbon in the environment where the pearl was grown. In freshwater systems, for example, \u0026delta;\u003csup\u003e13\u003c/sup\u003eC values can be strongly influenced by organic carbon cycling; the remineralization of abundant, isotopically light organic matter can dominate the isotopic composition of dissolved inorganic carbon utilized in pearl mineralization\u003csup\u003e36,37\u003c/sup\u003e. There may also be differences in pearl biomineralization processes between freshwater and saltwater species that affect isotopic fractionation, as variations in fractionation among carbonate minerals from different taxa have been reported\u003csup\u003e38,39\u003c/sup\u003e. Due to limited prior data, however, it remains unknown whether pearls exhibit similar species-specific controls on precipitation kinetics that might affect isotopic equilibrium.\u003c/p\u003e\n\u003cp\u003eThe considerable spread in the freshwater pearl data in Fig. 2c, which was collected from a limited number of locations, prevents any definitive conclusions on \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO sensitivity to geographic provenance. The wide range of freshwater pearl \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values may be attributed to local environment dependencies and larger seasonal changes in lakes and rivers, which lead to greater variability in freshwater chemistry compared to their marine counterparts\u003csup\u003e36,37\u003c/sup\u003e. For example, seasonal shifts have been previously observed by measuring \u0026delta;\u003csup\u003e18\u003c/sup\u003eO across transects of freshwater pearls\u003csup\u003e40,41\u003c/sup\u003e, which in one study were found to span a range upwards of 4 \u0026permil;\u003csup\u003e41\u003c/sup\u003e. Conversely, the saltwater data (as seen in Fig. 2c) broadly cluster by geographic region along the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO axis. This trend suggests that the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values are primarily influenced by the conditions in which the pearls were grown. While the literature on pearl carbonate isotopes is limited, \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of saltwater pearls from Japan\u003csup\u003e42\u003c/sup\u003e and South Korea\u003csup\u003e43\u003c/sup\u003e have been previously linked to seawater temperature. On a larger spatial scale, as shown by the global seawater temperature (Fig. 2b, top left) and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO (Fig. 2b, bottom left) maps, water \u0026delta;\u003csup\u003e18\u003c/sup\u003eO also plays an important role in determining saltwater pearl \u0026delta;\u003csup\u003e18\u003c/sup\u003eO. Of the source countries in our analysis of saltwater pearls, Japan has the lowest average sea surface temperature and seawater \u0026delta;\u003csup\u003e18\u003c/sup\u003eO. The \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of pearls from Japan are higher than the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of pearls from (northern) Australia. However, the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of pearls from Bahrain and Qatar are higher than the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of pearls from Japan. While Bahrain and Qatar have temperatures that fall between those of Australia and Japan, they have the highest seawater \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values due to high rates of evaporation in the Arabian Gulf region. These results suggest that there is a combined importance of water \u0026delta;\u003csup\u003e18\u003c/sup\u003eO and temperatures in determining pearl \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values. It should still be noted, however, that there remains some overlap in \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of pearls from different locations, particularly between Bahrain and Qatar, where the isotopic compositions of pearls are difficult to distinguish. Of the source locations, Bahrain and Qatar have the greatest proximity to one another, meaning that the water \u0026delta;\u003csup\u003e18\u003c/sup\u003eO and temperatures in these locations (and consequently pearl \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values) may be similar, especially relative to the other source locations. Pearls from Qatar were most commonly misclassified as pearls from Bahrain by the logistic regression, support vector machine, and k-nearest neighbors models, and we note that grouping pearls from Bahrain and Qatar into the same class noticeably increases accuracy scores (Supplementary Fig. 1). Furthermore, due to material availability limitations, the number of pearls analyzed in this study are not consistent across classes (see Supplementary Table 1), and as such, we expect that having a more balanced dataset may produce more reliable performance metrics for the models. Moreover, saltwater pearl \u0026delta;\u003csup\u003e13\u003c/sup\u003eC has been associated with feeding and organic matter degradation\u003csup\u003e42\u003c/sup\u003e and has not been concluded to significantly change with temperature. Saltwater pearl \u0026delta;\u003csup\u003e13\u003c/sup\u003eC data in Fig. 2c appear to remain within the same range of ca. \u0026ndash;1 to 2 \u0026permil; for all groups, with the pearls from Australia showing a more limited span. Given the available source locations for this study, these results suggest that saltwater pearl \u0026delta;\u003csup\u003e13\u003c/sup\u003eC may not be sensitive to location or temperature in the way \u0026delta;\u003csup\u003e18\u003c/sup\u003eO appears to be, and remains within a specific range regardless of growth environment, and possibly even species (pearls from Bahrain, Qatar, and Japan were formed by oysters from the (\u003cem\u003ePinctada\u003c/em\u003e) \u003cem\u003eP. fucata/martensii/radiata/imbricata\u003c/em\u003e species complex, while pearls from Australia were formed by \u003cem\u003eP. maxima\u003c/em\u003e oysters). It is possible, however, that pearls from other locations may exhibit distinct \u0026delta;\u003csup\u003e13\u003c/sup\u003eC values depending on regional circulation patterns and rates of organic matter remineralization and primary productivity.\u003c/p\u003e\n\u003cp\u003eTo investigate the link between environmental conditions and the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of carbonates formed by pearl oysters on a more local scale, isotopes from pearl oyster shells in the Arabian Gulf (Fig. 3a) were also examined. Aragonite from shells in Bahrain, Kuwait, and Qatar were analyzed and compared to annual mean surface temperature (Fig. 3b, left) and salinity (Fig. 3b, right) maps of the Arabian Gulf\u003csup\u003e44\u003c/sup\u003e. Salinity is controlled by much of the same evaporative processes as water \u0026delta;\u003csup\u003e18\u003c/sup\u003eO\u003csup\u003e28\u003c/sup\u003e, and the calculation of regional slopes correlating these two variables have been previously used to construct seawater \u0026delta;\u003csup\u003e18\u003c/sup\u003eO datasets\u003csup\u003e30\u003c/sup\u003e. As there is better availability and spatial resolution of salinity data in the Arabian Gulf compared to seawater \u0026delta;\u003csup\u003e18\u003c/sup\u003eO data, the salinity maps are shown in Figure 3. While overall there is considerable overlap in the \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of the shell samples from Qatar with the shell samples from Bahrain and Kuwait (Fig. 3c), some separation can be observed between the isotopic signatures of shell samples from different locations within Qatar, as well as between samples from Bahrain and Kuwait (see Fig. 3d). To better understand potential trends in the shell and pearl datasets, additional calculations showing the offset of these results with expected \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of inorganic aragonite and calcite precipitated at equilibrium\u003csup\u003e39\u003c/sup\u003e using available local temperature datasets are provided in Supplementary Fig. 2. The different biomineralization products of pearl oysters were additionally examined: comparisons between shell aragonite, shell calcite, and pearl aragonite in oysters collected in Bahrain are shown in Fig. 3e, revealing that shell calcite is depleted in \u003csup\u003e13\u003c/sup\u003eC compared to its aragonitic counterparts. In contrast, pearl and shell aragonite display more similar \u0026delta;\u003csup\u003e13\u003c/sup\u003eC values, with slightly lower \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values on average in pearl samples compared to shell samples as demonstrated in the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO violin plot in Fig. 3f (a \u0026delta;\u003csup\u003e13\u003c/sup\u003eC violin plot can be seen in Supplementary Fig. 3).\u003c/p\u003e\n\u003cp\u003eThe Arabian Gulf is a unique and relatively isolated water body characterized by high evaporation rates, elevated salinity, shallow bathymetry, and large seasonal temperature variability\u003csup\u003e45\u0026ndash;47\u003c/sup\u003e. The analysis of carbonates grown in these distinctive conditions could thus offer insights on the relationship between the isotopic composition of carbonates and the environments in which they were formed. Figure 3c shows that the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of shell aragonite from the Arabian Gulf span a range of ca. -1 to 2 \u0026permil;. More variability is seen in \u0026delta;\u003csup\u003e13\u003c/sup\u003eC values compared to \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values, and a wider overall range of shell \u0026delta;\u003csup\u003e13\u003c/sup\u003eC values are observed than in the pearl samples seen in Fig. 2c. The shells collected in Kuwait came from the coldest and lowest salinity regions in the Arabian Gulf, owing to the Shatt Al Arab river, which is the primary freshwater input for the Gulf\u003csup\u003e47\u003c/sup\u003e. As such, these \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values cluster close to the origin, while shells from Bahrain have high \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values and variable \u0026delta;\u003csup\u003e13\u003c/sup\u003eC values (Fig. 3c and 3d). Shells from Qatar, which come from a region of the Arabian Gulf with higher average temperatures and salinity, exhibit \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values spanning most of the range of \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values observed in the samples from both Bahrain and Kuwait. In particular, shells from Al Wakrah and Al Safliya Island, Qatar (light green and light blue diamonds, respectively, in Fig. 3c and 3d), which experience the highest average annual temperature and salinity of all the source locations, have lower \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values compared to other locations in Qatar and more closely resemble those from Kuwait. These observations suggest that elevated temperature, which drives the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO composition toward lower values, exerts a stronger influence than evaporation.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eOverall, correlations between \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values and the average temperature and salinity maps shown in Fig. 3b are not always straightforward, in part because the two main controls\u0026mdash;evaporation and elevated temperature\u0026mdash;exert opposing effects on \u0026delta;\u003csup\u003e18\u003c/sup\u003eO. The results in Fig. 3c generally suggest a greater relative role of local evaporation in shell aragonite \u0026delta;\u003csup\u003e18\u003c/sup\u003eO within the Arabian Gulf, although the influence of mineralization temperature appears to become stronger in locations with higher average sea surface temperatures. As suggested from the data presented in Fig. 2 that water \u0026delta;\u003csup\u003e18\u003c/sup\u003eO noticeably influences the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of pearls from Bahrain and Qatar, the observation that temperature could still play a dominant role in determining \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of shells within the Arabian Gulf signifies that the relative effects of temperature and water \u0026delta;\u003csup\u003e18\u003c/sup\u003eO on the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of minerals produced by pearl oysters may be sensitive to context and scale.\u003c/p\u003e\n\u003cp\u003eAs there is substantial overlap between the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of shells sourced from different countries in Fig. 3c (particularly with regard to Qatar), it may be important to additionally consider that the proximity of the source locations within Bahrain and Qatar may render a country-based separation of isotope data less effective. Previous mapping of water data in Qatar has shown considerable differences in salinity between the East and West coast\u003csup\u003e48\u003c/sup\u003e, which is not fully captured in the salinity map shown in Fig. 3b due to differences in resolution between the datasets. Given such variations that are highly sensitive to locality, trends in mineral \u0026delta;\u003csup\u003e18\u003c/sup\u003eO may be more readily seen based on the specific water conditions of the source locations and on their proximity to one another. Furthermore, while the salinity-\u0026delta;\u003csup\u003e18\u003c/sup\u003eO correlation has been well-studied, it has been shown to exhibit spatiotemporal sensitivity\u003csup\u003e30,49\u003c/sup\u003e, and given that the Arabian Gulf is isolated and evaporative, definitive conclusions about the primary influences on mineral \u0026delta;\u003csup\u003e18\u003c/sup\u003eO within this region using salinity data would require a nuanced understanding of the local relationship between salinity and water \u0026delta;\u003csup\u003e18\u003c/sup\u003eO. In addition, the maps do not account for the extent of seasonal variations; furthermore, variables such as circulation and water depth may also help to paint a more complete picture of the observed \u0026delta;\u003csup\u003e18\u003c/sup\u003eO\u003csup\u003e50\u003c/sup\u003e, and improved temporal and spatial resolution of seawater \u0026delta;\u003csup\u003e18\u003c/sup\u003eO measurements in the Arabian Gulf could help resolve the relative importance of these controls.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePearls are primarily composed of aragonite in the form of nacre, which is the same material that constitutes the innermost layer of pearl oyster shells\u003csup\u003e3\u003c/sup\u003e. Above this nacreous shell layer is a layer of prismatic calcite (another polymorph of CaCO\u003csub\u003e3\u003c/sub\u003e)\u003csup\u003e3\u003c/sup\u003e. In Fig. 3e it is observed that shell aragonite and pearl aragonite fall in similar regions of the plot, with shell aragonite being more variable in \u0026delta;\u003csup\u003e13\u003c/sup\u003eC, whereas shell calcite is visibly more depleted in \u003csup\u003e13\u003c/sup\u003eC (which can be more readily observed in Supplementary Fig. 3a). Additionally, \u0026delta;\u003csup\u003e18\u003c/sup\u003eO in shell calcite is slightly higher than in shell aragonite and pearl aragonite, with \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of pearls being slightly lower compared to \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of shells (Fig. 3f). The general trends in \u0026delta;\u003csup\u003e13\u003c/sup\u003eC values are in line with expectations of distinct carbon isotope equilibrium fractionation factors for calcite and aragonite\u003csup\u003e51\u0026ndash;53\u003c/sup\u003e, where shell aragonite was previously found to be enriched in \u003csup\u003e13\u003c/sup\u003eC relative to calcite\u003csup\u003e52\u003c/sup\u003e. \u0026delta;\u003csup\u003e13\u003c/sup\u003eC values for pearl and shell aragonite, while similar on average, appear more variable in shell aragonite (see Supplementary Figure 3), which may be a result of greater heterogeneity in the shell samples compared to the pearl samples. Regarding oxygen isotopes, our results show that pearl \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values appear slightly lower than \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values for both shell aragonite and shell calcite in samples from Bahrain, although the reason for this remains unclear. The process of pearl mineralization may indeed be distinct from shell nacre mineralization, resulting in differences in oxygen isotope fractionation, or there may be certain seasons or environmental conditions in which pearls are more likely to form, which may put pearl growth out of phase with shell nacre growth.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWith the large suite of analyzed materials from a wide range of locations, our results present isotope ratio mass spectrometry as a promising method for improving pearl classification as well as an advancement of insights on isotope fractionation and biomineralization processes in shell and pearl minerals produced by pearl oysters. However, there are some considerations to be made for the practical implementation of this tool. The collection of samples from cultured pearls with a large beaded nucleus and a thin layer of native nacre must be performed with caution\u0026mdash;drilling into the nucleus should be avoided as this would certainly affect the results\u003csup\u003e43\u003c/sup\u003e. Most of the saltwater pearls analyzed in this study did not have a beaded nucleus\u0026mdash;the pearls from Bahrain, Qatar, and Australia were determined to be natural, as these pearls were either collected firsthand by the authors and/or were verified with x-ray imaging. The pearls from Japan, however, included both non-beaded and beaded pearls. For these samples drilling was performed to a shallow depth (ca. 1 mm), and no notable outliers were observed in the isotope data\u0026mdash;from this, it is expected that there was no unintentional drilling of the nuclei, which are typically made from freshwater shells\u003csup\u003e7\u003c/sup\u003e and would therefore likely produce a distinct isotopic signature from saltwater nacre. Additionally, when considering a real-world use case for carbonate isotope analysis, it must be noted that the method requires some destruction of the sample. While a relatively small sample mass is required (around 50-100 \u0026micro;g), this may be difficult to acquire from smaller pearls. In these cases, a subsampling of entire pearls from a large batch may be required (since batches of small wild pearls routinely come from a single locality).\u003c/p\u003e\n\u003cp\u003eFurthermore, to improve the robustness of utilizing mass spectrometry data as a means of geographic provenance identification, a larger dataset of the isotopic composition of pearls from all known locations of natural and cultured pearls should be further developed. For example, the pearls from Qatar in Fig. 2c were only sourced from one specific location (Al Aaliya Island), which exhibit a different range of \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values compared to the shells from Qatar, which were sampled from more locations. If we assume that the isotopic composition of shell aragonite, as seen in Fig. 3c, is similar to that of the pearls formed in the same environment, the analysis of pearls from different locations in Qatar may affect the performance of the classification models in distinguishing between pearls from Bahrain and Qatar as presented in Fig. 2c. Although shell aragonite\u0026mdash;which would be more convenient to sample from than pearls for expanding the dataset\u0026mdash;could be a sufficiently suitable proxy for pearl aragonite, the observed differences in the isotopic composition of pearls and shell aragonite (Fig. 3e) may affect the accuracy of the dataset and should thus be subject to further analysis and consideration. From preliminary results, we show that machine learning algorithms such as logistic regression (Fig. 2c) could be applied to predict geographic origin from pearl isotope data. We expect that dataset expansion may be able to improve the accuracy of such algorithms and hopefully resolve ambiguities pertaining to some areas of overlap in the \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of pearls from different locations. Clumped isotope analysis, which measures the extent of \u0026lsquo;clumping\u0026rsquo; of heavy isotopes (e.g., \u003csup\u003e13\u003c/sup\u003eC-\u003csup\u003e18\u003c/sup\u003eO bonds) rather than measuring isotope ratios of individual elements, could be additionally investigated to reconstruct seawater temperature from pearl \u0026Delta;\u003csub\u003e47\u003c/sub\u003e\u003csup\u003e54,55\u003c/sup\u003e. This method requires a larger sample mass than the isotope ratio measurements used in this study, but may be useful in improving the dataset as it could help clarify the temperature-mineral \u0026delta;\u003csup\u003e18\u003c/sup\u003eO correlation and isolate vital effects (i.e., kinetic and metabolic contributions that result in an isotopic offset between biogenic and inorganic carbonate when the minerals are precipitated in the same conditions)\u003csup\u003e56\u003c/sup\u003e.\u003c/p\u003e\n\u003ch2\u003eRaman Spectroscopy and Support Vector Machine (SVM) Prediction\u003c/h2\u003e\n\u003cp\u003eSince species of origin is another important factor in pearl classification, and considering previous investigations that have utilized Raman spectroscopy for the identification of species-specific organic compound content in pearl oyster nacre\u003csup\u003e15,19,20\u003c/sup\u003e, we next explored the robustness of this technique for such purposes. In this study, we utilized Raman spectroscopy data from the surface of 771 pearls to train a machine learning (ML) algorithm to differentiate pearl species based on Raman-collected chemical features. While forming the training dataset for the ML model, it was essential to ensure consistency between independent Raman measurements from all analyzed pearls. We thus pre-processed each measured Raman spectrum so that it was constrained to a wavenumber range of 100 cm\u003csup\u003e-1\u003c/sup\u003e to 2000 cm\u003csup\u003e-1\u003c/sup\u003e, as shown in Fig. 4a. To maintain uniform data intervals and ensure an equal number of data points, each processed Raman spectrum was resampled to 250 equally spaced data points within this range. For the sampled data points that did not precisely align with the original measurement, linear interpolation was applied using the two nearest neighboring points in the raw spectrum (Supplementary Fig. 4). To retain information potentially contained within fluorescent signals and signals from high-energy organic components for the ML algorithm, no background subtraction was carried out on the raw Raman spectra. In total, 700, 170, and 51 Raman spectra of pearls from \u003cem\u003eP. maxima\u003c/em\u003e, \u003cem\u003eP. radiata\u003c/em\u003e, and \u003cem\u003eP. fucata\u003c/em\u003e saltwater oysters, respectively, were collected and analyzed. The collected samples were then randomly partitioned into two groups: a training dataset and testing dataset, followed by a train-to-test ratio of 90:10, as illustrated in Fig. 4a.\u003c/p\u003e\n\u003cp\u003eA Support Vector Machine (SVM) model was developed to classify the three pearl-forming oyster species based on their Raman spectral features. Hyperparameter optimizations were performed to enhance the SVM classification performance: a linear kernel matrix was applied with a kernel coefficient (gamma) value of 0.5 and a regularization parameter (C) of 1.0. The training weights assigned to each Raman feature are presented in Fig. 4b. Notably, data points around 1080, 700 and 200 cm\u003csup\u003e-1\u003c/sup\u003e wavenumbers exhibit strong contributions to specimen classification prediction. These highly weighted Raman bands correspond to the vibrational modes of aragonite, which is the primary crystalline component of the pearls. Specifically, bands at 1080 and 703 cm\u003csup\u003e-1\u003c/sup\u003e correspond to the v\u003csub\u003e1\u003c/sub\u003e symmetric and v\u003csub\u003e4\u003c/sub\u003e in-plane bending modes of the aragonite carbonate CO\u003csub\u003e3\u003c/sub\u003e\u003csup\u003e2-\u003c/sup\u003e, whereas the wavenumber at ca. 230 cm\u003csup\u003e-1\u003c/sup\u003e corresponds to the vibration of the aragonite lattice\u003csup\u003e15\u003c/sup\u003e. In addition, the highly weighted data points at wavenumbers above 1250 cm\u003csup\u003e-1\u003c/sup\u003e are attributed to different organic components within the pearls. In particular, Raman band peaks at ca. 1260, 1350, 1410, and 1560 cm\u003csup\u003e-1\u003c/sup\u003e are indicative of the in-plane deformation of C-H bonds, biological pigment vibrations, hexagonal carbon ring stretching, and C=C bond stretching, respectively\u003csup\u003e15\u003c/sup\u003e. The consistency between these dominantly weighted Raman features with known pearl spectral signatures validates the capacity of the SVM model to extract chemically relevant classification features when making a prediction. Moreover, the SVM model is capable of capturing subtle spectral variations that may not be directly attributed to typical Raman features of pearls. This model also accounts for background fluorescence, which can vary according to the structure and composition of individual pearls. Unlike conventional physical measurements or manual baseline correction of the spectra, the SVM model can autonomously learn and integrate these complex and overlooked spectral perturbations, thereby enhancing the correlations between the pearl Raman spectral data and the prediction outcomes, and ultimately improving the classification accuracy.\u003c/p\u003e\n\u003cp\u003eEach testing dataset was randomly selected according to the overall data distribution (55 \u003cem\u003eP. maxima\u003c/em\u003e, 15 \u003cem\u003eP. radiata\u003c/em\u003e, and 5 \u003cem\u003eP. fucata\u003c/em\u003e spectra per trial, for a total of 75 testing data, which is 10% of the overall collected data). Across the 3000 independent testing trials, this yields a total of 2200, 600, and 200 samples for \u003cem\u003eP. maxima\u003c/em\u003e, \u003cem\u003eP. radiata\u003c/em\u003e, and \u003cem\u003eP. fucata\u003c/em\u003e, respectively. Two statistical confusion matrices summarizing the prediction results are shown in Fig. 4c. Each entry in the % error confusion matrix was calculated by dividing the number of misclassifications of that species by its total number of test samples. For example, within 2200 \u003cem\u003eP. maxima\u003c/em\u003e test samples, 42 samples were misclassified as \u003cem\u003eP. radiata\u003c/em\u003e, giving an error of (42/2200)\u0026acute;100% = 1.91%. The three diagonal elements of each matrix specify an accurate pearl specimen prediction through SVM, whereas the off-diagonal elements in the matrix represent a misclassification.\u003c/p\u003e\n\u003cp\u003eAmong the three species, \u003cem\u003eP. maxima\u003c/em\u003e pearls exhibited the highest classification accuracy, whereas \u003cem\u003eP. radiata\u003c/em\u003e and \u003cem\u003eP. fucata\u003c/em\u003e pearls are more frequently misclassified as \u003cem\u003eP. maxima\u003c/em\u003e by the model. This predicted classification bias is likely due to the imbalanced distribution of data in the training dataset, wherein spectra from pearls formed by \u003cem\u003eP. maxima\u003c/em\u003e account for 73% of the total samples, compared to the 21% for \u003cem\u003eP. radiata\u003c/em\u003e pearls and 6% for \u003cem\u003eP. fucata\u003c/em\u003e pearls. However, despite this imbalance in data distribution, an overall high prediction accuracy of 96.4% was achieved over the 3000 testing samples across the three pearl species (Supplementary Fig. 5, Supplementary Table 4).\u003c/p\u003e\n\u003cp\u003eTo further investigate the influence of specific Raman spectral regions on the prediction accuracy, we divided the full Raman spectrum into halves, with the first half containing data from a wavenumber range of 100 cm\u003csup\u003e-1\u003c/sup\u003e to1050 cm\u003csup\u003e-1\u003c/sup\u003e and the second half containing data from a range of 1050 cm\u003csup\u003e-1\u003c/sup\u003e to 2000 cm\u003csup\u003e-1\u003c/sup\u003e. Each separated spectrum consisted of 125 data points (125 data points each from the first and second halves of the spectra from the original 250 Raman data points; this is detailed in Supplementary Fig. 6). The same hyperparameter-optimized SVM model was retrained using only one spectral region at a time for over 3000 randomly selected testing pearl spectra.\u003c/p\u003e\n\u003cp\u003eFig. 5a presents the accuracy confusion matrices of the prediction results for the halved spectral regions (same procedure as previous; see Supplementary Tables 5 and 6 for details). Compared to the original Raman spectrum input, using only half of the spectral data led to a decrease in the overall classification accuracy, particularly for \u003cem\u003eP. radiata\u003c/em\u003e and \u003cem\u003eP. fucata\u003c/em\u003e. This decline in accuracy can be attributed to the loss of data variations and spectral information, including critical vibrational features associated with the pearl organic matrix components and lattice vibrations, thus reducing the prediction power of the model. A statistical box plot overlaid with a scatter plot of accuracies from individual trials further illustrates the variations in classification accuracy across different trials, as shown in Fig. 5b. The predictions from the full-spectrum SVM model exhibits lower error variability and superior predictive performance compared to the models using data points from partial spectra.\u003c/p\u003e\n\u003cp\u003eGiven that the majority of the Raman features exhibit a minimal impact weight in making the SVM prediction, a refined feature selection approach was implemented. Indicated by the highly weighted Raman features obtained in Fig. 4b, we extracted a subset of 100 highly weighted Raman data points, ranging from the spectral regions of 100-400 (lattice vibration), 670-750 (CO\u003csub\u003e3\u003c/sub\u003e\u003csup\u003e2-\u003c/sup\u003e bending modes), 1050-1120 (CO\u003csub\u003e3\u003c/sub\u003e\u003csup\u003e2-\u003c/sup\u003e stretching), 1280-1400 (organic pigments), 1510-1580 (C=C bond vibrations), and 1850-1960 cm\u003csup\u003e-1\u003c/sup\u003e (triple-bonding organic components and overtones). After training the SVM algorithm with these 100 highly weighted data points with the same model criteria, we extracted the corresponding feature importance weights, shown in Fig. 5c. A similar result was obtained as with the previous extraction in Fig. 4b, which matches the characteristic band peak of the Raman spectrum. To evaluate model performance with these 100 highly weighted data points, the species for 3000 randomly selected pearl spectra were again predicted, as shown in Fig. 5d. Compared to the half-spectrum input dataset, the dataset using the 100 highly weighted data points achieves a comparable prediction accuracy with respect to the original full-spectrum model, except for a notable drop in classification accuracy for \u003cem\u003eP. fucata\u003c/em\u003e.\u003c/p\u003e\n\u003cp\u003eOverall, SVM-based classification models can predict pearl species from their chemical signatures obtained through Raman spectroscopy, with our models yielding a prediction accuracy of 96.4%. By analyzing the importance of Raman features, we identify the key Raman band wavenumbers that dominate pearl species classification, which consistently align with known pearl aragonite and organic spectral features. Furthermore, by optimizing feature selection, we can reduce the complexity of input data while maintaining a high prediction accuracy. These findings demonstrate that SVM can be an effective tool for pearl identification and classification, which leverages Raman spectroscopy as a non-destructive analytical method that fits in the as-developed numerical ML method. The demonstrated models provide a foundational proof-of-principle framework that can serve as a basis for future pearl identification studies.\u003c/p\u003e\n\u003cp\u003eIn summary, we demonstrate that stable isotope analysis and Raman spectroscopy, respectively, are capable of answering two major questions in identifying pearl provenance\u0026mdash;geography and species. We found that saltwater pearls were enriched in both \u003csup\u003e13\u003c/sup\u003eC and \u003csup\u003e18\u003c/sup\u003eO relative to freshwater pearls, and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of saltwater pearls were found to further separate by country of origin, while \u0026delta;\u003csup\u003e13\u003c/sup\u003eC data remained within a similar range of values for all saltwater pearl groups. Further isotopic analysis of shell aragonite samples from the Arabian Gulf revealed that both temperature and evaporation could play an important role in \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of carbonates produced by pearl oysters, with the influence that these controls exert being variable from location to location. A comparison of the isotopic composition of shell aragonite, shell calcite, and pearl in oysters from Bahrain revealed that pearls have slightly lower \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values than shell samples, while shell calcite has lower \u0026delta;\u003csup\u003e13\u003c/sup\u003eC values than pearl and shell aragonite. Our results show that the isotopic analysis of saltwater pearls could be utilized as a powerful tool for identifying geographic provenance of pearls with unknown origins, and there is a rich body of existing work on the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of minerals formed by bivalves and their link to temperature and hydrological processes in support of this proposal. However, there are uncertainties and limitations with this tool that remain to be examined and further addressed in future studies. Forming a clearer understanding of the factors influencing pearl \u0026delta;\u003csup\u003e18\u003c/sup\u003eO by acquiring more environmental data on source locations and/or conducting additional isotopic analysis will be key to addressing overlaps in isotopic composition between pearls from different locations and improving the use of mass spectrometry for pearl classification. For species identification, Raman spectra of pearls produced by \u003cem\u003eP. maxima\u003c/em\u003e, \u003cem\u003eP. radiata\u003c/em\u003e, and \u003cem\u003eP. fucata\u003c/em\u003e oysters were used to develop an SVM model that could predict species with an accuracy of 96.4%. SVM training weights were the highest in areas of the spectra corresponding to the characteristic Raman features of aragonite and organic compounds found in pearls. The best model performance was observed using full Raman spectra for training and prediction, with accuracy scores dropping when using a dataset containing only the first or second halves of the Raman spectra. A model using 100 points from the Raman spectra corresponding to the highest SVM training weights achieved similar results as the full-spectrum model but produced a considerably lower accuracy score for \u003cem\u003eP. fucata\u003c/em\u003e pearls. Together, these results show that stable isotope analysis and Raman spectroscopy reveal chemical information encoded into pearl nacre that can help uncover important insights on provenance. These techniques thus represent a robust strategy for improving aspects of the pearl classification process, while simultaneously highlighting their value in future studies aimed at addressing underlying questions on the mechanisms of biogenic carbonate formation.\u003c/p\u003e"},{"header":"Methods","content":"\u003cp\u003e\u003cstrong\u003eSample Acquisition\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSamples of pearls and shells were acquired from diverse geographic locations; specimens of \u003cem\u003ePinctada radiata\u003c/em\u003e were collected from the Arabian Gulf (Bahrain, Qatar, and Kuwait), whereas \u003cem\u003ePinctada fucata\u003c/em\u003e were sourced from pearl culturing farms and markets in Japan. \u003cem\u003ePinctada maxima\u003c/em\u003e specimens were obtained from the northern oceans of Australia, whereas the freshwater pearls were collected from Scotland and markets in Japan.\u003c/p\u003e\n\u003cp\u003eThe samples were categorized according to varying levels of acquisition confidence ranging from 1-5 in decreasing order of confidence (Supplementary Table 2). Most of the samples were obtained directly from the source by the Bahrain Institute of Pearls and Gemstones (DANAT) field research team (i.e., the team went into the sea to collect samples); this direct acquisition corresponds to a sample confidence level of 1. Samples from Scotland were gathered for DANAT by a trusted external entity without paper trail evidence (Confidence Level 3). The remainder of the samples were acquired from wholesale pearl merchants from the pearl market in Kobe, Japan (Confidence Level 5). Supplementary Table 3 displays samples from different species and their corresponding confidence levels. All natural samples were considered to be untreated raw materials, whereas cultured pearls were regarded as having potentially undergone various treatments, including bleaching.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMass Spectrometry\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFor carbon and oxygen isotopic analysis, powder samples were collected from both pearls and shells. The majority of the analyzed pearls were drilled using a jewelry drill press with a 0.6 mm carbide bur drill bit, whereas pearls that were too small for drilling (i.e., similar to or smaller than the size of the drill bit) were instead ground with an agate mortar and pestle. Powdered shell nacre was collected by passing a Dremel with a rough conical bit over an approximately 1.5 cm strip of nacre at the center of the inner surface of the shell until a 1-2 mm deep groove was formed. For calcite collection, the outer surface of the shell was blasted with compressed air and scraped with a metal spatula to remove larger pieces of debris. An approximately 0.5 cm\u0026nbsp;\u0026acute;\u0026nbsp;2 cm piece of prismatic calcite from the growth edge of the shells was then broken off using pliers and ground using an agate mortar and pestle. Isotope analysis was conducted using a Nu Perspective isotope ratio mass spectrometer. Carbonate samples weighing 50\u0026ndash;100 \u0026mu;g were digested in a Nu Carb automated sample preparation unit for 25 minutes in individual glass vials with 150 \u0026mu;L orthophosphoric acid (\u0026rho; = 1.93 g/cm\u0026sup3;), and the evolved CO₂ gas was purified cryogenically. Purified sample gas and reference gas of known composition were alternately measured on six Faraday collectors (m/z 44\u0026ndash;49) in 6 cycles, each with a 30-second integration time (3 minutes total integration time). Each session of approximately 50 individual analyses began with two ETH anchors, then alternated between blocks of six to eight unknowns and two ETH anchors, totaling twelve anchors per run. Data were processed using the \u0026ldquo;D47crunch\u0026rdquo; Python package\u003csup\u003e57\u003c/sup\u003e with IUPAC \u003csup\u003e17\u003c/sup\u003eO parameters and 70\u0026deg;C \u003csup\u003e18\u003c/sup\u003eO acid fractionation factors of 1.00871 for calcite and 1.009091 for aragonite\u003csup\u003e58\u003c/sup\u003e. Raw measurements were converted to Vienna Pee Dee Belemnite (VPDB) using a pooled regression approach\u003csup\u003e59\u003c/sup\u003e that used the ETH anchor values from Bernasconi \u003cem\u003eet al.\u003c/em\u003e\u003csup\u003e60\u003c/sup\u003e. Nominal anchor values for \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO (in \u0026permil; VPDB), respectively, are ETH-2: -10.17, -18.69; and ETH-3: 1.71, -1.78\u003csup\u003e61\u003c/sup\u003e. The \u0026delta;\u003csup\u003e13\u003c/sup\u003eC and \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values of a portion of the pearl samples were obtained with a set of run parameters optimized for clumped isotope spectrometry (see Supplementary Table 7), which are described in Anderson \u003cem\u003eet al\u003c/em\u003e.\u003csup\u003e62\u003c/sup\u003e. Standard errors on the anchor measurements can be seen in the Supplementary Information as a metric of accuracy (Supplementary Table 8).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eSeawater Temperature, \u0026delta;\u003csup\u003e18\u003c/sup\u003eO, and Salinity Maps\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eGlobal sea surface temperature data for Fig. 2b (top left) were sourced from the World Ocean Atlas 2023 using the annual statistical mean on 1/4\u0026deg; grid for all decades\u003csup\u003e29\u003c/sup\u003e. The data were imported into ArcGIS Pro and the Empirical Bayesian Kriging tool (Geostatistical Analyst Tools) was applied to the surface Z value field for interpolation. The resulting data were classified in equal intervals to generate the map. Global seawater \u0026delta;\u003csup\u003e18\u003c/sup\u003eO data for Fig. 2b (bottom left) were sourced from LeGrande and Schmidt\u003csup\u003e30\u003c/sup\u003e and imported into ArcGIS Pro, where it was classified in equal intervals to generate the map. Temperature and salinity data of the Arabian Gulf as presented in Fig. 3b were sourced from the Global Ocean Physics Analysis and Forecast from E.U. Copernicus Marine Service Information\u003csup\u003e44\u003c/sup\u003e. All monthly temperature and salinity data from 2023 were imported into ArcGIS Pro using the Make NetCDF Raster Layer tool. Data from the shallowest available depth (0.494025 m) of the raster layers were averaged using the Cell Statistics tool in ArcGIS Pro to generate the annual mean surface temperature and salinity maps.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eRaman Spectroscopy\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eRaman spectra were acquired from pearl samples using two Raman spectrometers: a Renishaw inVia spectrometer and a Qontor Renishaw inVia spectrometer, both integrated with an optical microscope. Calibration for the Renishaw inVia spectrometer was conducted with the 1331.8 cm\u003csup\u003e-1\u003c/sup\u003e diamond Raman line. A diode-pumped solid-state laser with an excitation wavelength of 514 nm was used for the measurements. The pearls were exposed to an average laser power of 4 mW. Spectra were acquired from 100 cm\u003csup\u003e-1\u003c/sup\u003e to 2000 cm\u003csup\u003e-1\u003c/sup\u003e with a resolution of ca. 2 cm\u003csup\u003e-1\u003c/sup\u003e, using a grating of 1800 grooves/mm and a 40 \u0026mu;m slit. A 50\u0026times;/0.75 short-distance objective lens was used, with a laser acquisition duration of 30 seconds and 7 accumulations.\u003c/p\u003e\n\u003cp\u003eThe remainder of the pearl Raman spectra acquisition was performed with the Qontor Renishaw inVia spectrometer. Calibration was conducted with a silicon wafer exhibiting a peak at 520.6 cm\u003csup\u003e-1\u003c/sup\u003e. A diode-pumped solid-state laser with an excitation wavelength of 457 nm was used, and the laser power was 40 mW. The spectral range was 100 cm\u003csup\u003e-1\u003c/sup\u003e to 2000 cm\u003csup\u003e-1\u003c/sup\u003e with a detector resolution of ca. 1 cm\u003csup\u003e-1\u003c/sup\u003e. A grating of 2400 grooves/mm, a notch filter, and a 40 \u0026mu;m slit were used. A 50\u0026times; long-distance objective lens was used with an exposure duration of 1 second and 10 accumulations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAlgorithm Development\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cem\u003ePearl Measurements:\u0026nbsp;\u003c/em\u003eCarbon and oxygen isotope data of saltwater pearls from Australia, Bahrain, Japan, and Qatar were used with an 80:20 train:test split. Raman data were collected and pre-processed using Python glob module (glob.glob) for a linear fit.\u003c/p\u003e\n\u003cp\u003e\u003cem\u003eML Algorithms:\u0026nbsp;\u003c/em\u003eAll algorithms were developed and optimized using the SciKit Learn (sklearn) module in Python. For the distinction of saltwater pearls based on isotope data, a logistic regression model, a k-nearest neighbors (k-NN) model, and a support vector machine (SVM) for four classes (Australia, Bahrain, Japan, Qatar), and a logistic regression model for three classes (Australia, Bahrain and Qatar, Japan) were developed. Logistic regression models had the following parameters: solver = \u0026lsquo;lbfgs\u0026rsquo;, max_iter = 1000, and multi_class = \u0026lsquo;multinomial\u0026rsquo;. The k-NN model was developed with n_neighbors = 5. For the C-Support Vector Classification (SVC) model, a grid search algorithm (GridSearchCV) was used to determine the best parameters. Regularization term (C) values of 0.1, 1, 10, and 100, and kernel coefficient (gamma) values of 1, 0.1, 0.01, and 0.001 for \u0026lsquo;rbf\u0026rsquo;, \u0026lsquo;linear\u0026rsquo;, and \u0026lsquo;poly\u0026rsquo; kernels were investigated. The best model had parameter values of C = 100 and gamma = 1 for the \u0026lsquo;poly\u0026rsquo; kernel, which was then used to plot decision boundaries between classes. A similar SVC model was developed for predicting pearl species based on Raman spectra. A linear kernel with C = 1.0 and gamma = 0.5 was applied. Accuracy scores and confusion matrices were imported from sklearn.metrics for evaluating model performance. Selected plots were generated using matplotlib.plot and seaborn.heatmap.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eStatistics\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTwo-tailed comparisons between pairings of pearl, shell aragonite, and shell calcite \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values for samples from Bahrain (n = 101) were performed using a linear mixed effects model (mixedlm) from the SciKit Learn module in Python with\u0026nbsp;a\u0026nbsp;= 0.05. For each pairing the calculated p-values were \u0026lt; 0.001: for shell aragonite vs. shell calcite, p = 0.00050; for shell aragonite vs. pearl, p = 0.000025; and for shell calcite vs. pearl, p = 8.7\u0026nbsp;\u0026acute;\u0026nbsp;10\u003csup\u003e-15\u003c/sup\u003e. A linear mixed effects model was used given the hierarchical structuring of the data and expected variations in mineral \u0026delta;\u003csup\u003e18\u003c/sup\u003eO values by specimen and by location. Location was set as a cluster group and a random slope was allowed for unique specimens. A histogram and a quantile-quantile plot of the residuals are shown in Supplementary Fig. 7a-b, and the Shapiro-Wilk test (shapiro) was conducted using the SciPy module in Python to generate W = 0.983 and p = 0.21 to support the assumption of normality of the residuals. Additionally, a plot of the residuals vs. fitted values and a boxplot of the distribution of residuals by location are shown in Supplementary Fig. 7c-d, and White\u0026rsquo;s Lagrange Multiplier Test for Heteroscedasticity (het_white) was conducted using the Stastmodels module in Python to generate a Lagrange Multiplier statistic of 0.18218 with a corresponding p-value of 0.91294 and an F-statistic of 0.08854 with a corresponding p-value of 0.91534, to support the assumption of homoscedasticity of variance.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003eData Availability\u003c/p\u003e\n\u003cp\u003eAll relevant data are available from the corresponding author upon reasonable request, subject to evaluation by the authors.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAcknowledgments\u003c/p\u003e\n\u003cp\u003eWe would like to thank the team at the Bahrain Institute for Pearls and Gemstones (DANAT) for acquiring all samples analyzed in this work. We would also like to thank Claire Hayhow for assistance in collecting pearl and shell isotope data. We would like to thank Dr. Nicu-Viorel Atudorei and Dr. Abdul-Mehdi Ali from the University of New Mexico for the coordination and collection of seawater isotope and trace element measurements.\u003c/p\u003e\n\u003cp\u003eAuthor Contributions\u003c/p\u003e\n\u003cp\u003eConceptualization: NJ, JCW, KDB, VB, AM\u003c/p\u003e\n\u003cp\u003eSecured Funding: NJ, VB, AM\u003c/p\u003e\n\u003cp\u003eProject Supervision: NJ, JCW, KDB, VB, AM\u003c/p\u003e\n\u003cp\u003eProvided Instrumentation: NJ, KDB, VB, AM\u003c/p\u003e\n\u003cp\u003eExperimental Design: All authors\u003c/p\u003e\n\u003cp\u003eData Collection: DK, RZ, ABJ, AZ, AA, JCW\u003c/p\u003e\n\u003cp\u003eData Analysis: DK, RZ, ABJ, AZ\u003c/p\u003e\n\u003cp\u003eManuscript Writing: DK, RZ, JCW, AM\u003c/p\u003e\n\u003cp\u003eAll authors reviewed and edited the manuscript.\u003c/p\u003e\n\u003cp\u003eCompeting Interests\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eWeiner, S. \u0026amp; Traub, W. Macromolecules in mollusc shells and their functions in biomineralization. \u003cem\u003ePhilos. Trans. R. Soc. Lond. B Biol. Sci.\u003c/em\u003e \u003cstrong\u003e304\u003c/strong\u003e, 425\u0026ndash;434 (1984).\u003c/li\u003e\n\u003cli\u003eAddadi, L., Joester, D., Nudelman, F. \u0026amp; Weiner, S. Mollusk Shell Formation: A Source of New Concepts for Understanding Biomineralization Processes. \u003cem\u003eChem. Eur. J.\u003c/em\u003e \u003cstrong\u003e12\u003c/strong\u003e, 980\u0026ndash;987 (2006).\u003c/li\u003e\n\u003cli\u003eMarin, F. The formation and mineralization of mollusk shell. \u003cem\u003eFront. Biosci.\u003c/em\u003e \u003cstrong\u003eS4\u003c/strong\u003e, 1099\u0026ndash;1125 (2012).\u003c/li\u003e\n\u003cli\u003eStrack, E. \u003cem\u003ePearls\u003c/em\u003e. (R\u0026uuml;hle-Diebener-Verlag, Stuttgart, 2006).\u003c/li\u003e\n\u003cli\u003eSturman, N. \u003cem\u003eet al.\u003c/em\u003e A Pearl Identification Challenge. \u003cem\u003eGems Gemol.\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 229\u0026ndash;243 (2019).\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eThe Pearl Oyster\u003c/em\u003e. (Elsevier Science, 2011).\u003c/li\u003e\n\u003cli\u003eSimkiss, K. \u0026amp; Wada, K. Cultured pearls\u0026mdash;commercialised biomineralisation. \u003cem\u003eEndeavour\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 32\u0026ndash;37 (1980).\u003c/li\u003e\n\u003cli\u003eMiyoshi, T., Matsuda, Y. \u0026amp; Komatsu, H. Fluorescence from Pearls to Distinguish Mother Oysters Used in Pearl Culture. \u003cem\u003eJpn. J. Appl. Phys.\u003c/em\u003e \u003cstrong\u003e26\u003c/strong\u003e, 578 (1987).\u003c/li\u003e\n\u003cli\u003eWehrmeister, U. \u003cem\u003eet al.\u003c/em\u003e Visualization of the internal structures of cultured pearls by computerized X-ray microtomography. \u003cem\u003eJ. Gemmol.\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 15\u0026ndash;21 (2008).\u003c/li\u003e\n\u003cli\u003eVigorelli, L. \u003cem\u003eet al.\u003c/em\u003e X-ray Micro-Tomography as a Method to Distinguish and Characterize Natural and Cultivated Pearls. \u003cem\u003eCondens. Matter\u003c/em\u003e \u003cstrong\u003e6\u003c/strong\u003e, 51 (2021).\u003c/li\u003e\n\u003cli\u003eKrzemnicki, M. S., Friess, S. D., Chalus, P., H\u0026auml;nni, H. A. \u0026amp; Karampelas, S. X-Ray Computed Microtomography: Distinguishing Natural Pearls from Beaded and Non-Beaded Cultured Pearls. \u003cem\u003eGems Gemol.\u003c/em\u003e \u003cstrong\u003e46\u003c/strong\u003e, 128\u0026ndash;134 (2010).\u003c/li\u003e\n\u003cli\u003eH\u0026auml;nni, H. A., Kiefert, L. \u0026amp; Giese, P. X-ray luminescence, a valuable test in pearl identification. \u003cem\u003eJ. Gemmol.\u003c/em\u003e \u003cstrong\u003e29\u003c/strong\u003e, 325\u0026ndash;329 (2005).\u003c/li\u003e\n\u003cli\u003eKarampelas, S. Spectral Characteristics of Natural-Color Saltwater Cultured Pearls from \u003cem\u003ePinctada Maxima\u003c/em\u003e. \u003cem\u003eGems Gemol.\u003c/em\u003e \u003cstrong\u003e48\u003c/strong\u003e, 193\u0026ndash;197 (2012).\u003c/li\u003e\n\u003cli\u003eTsai, T.-H. \u0026amp; Zhou, C. Rapid detection of color-treated pearls and separation of pearl types using fluorescence analysis. in \u003cem\u003eNovel Optical Systems, Methods, and Applications XXIII\u003c/em\u003e (eds. Hahlweg, C. F. \u0026amp; Mulley, J. R.) 6 (SPIE, Online Only, United States, 2020). doi:10.1117/12.2566590.\u003c/li\u003e\n\u003cli\u003eShi, L., Wang, Y., Liu, X. \u0026amp; Mao, J. Component Analysis and Identification of Black Tahitian Cultured Pearls From the Oyster \u003cem\u003ePinctada margaritifera\u003c/em\u003e Using Spectroscopic Techniques. \u003cem\u003eJ. Appl. Spectrosc.\u003c/em\u003e \u003cstrong\u003e85\u003c/strong\u003e, 98\u0026ndash;102 (2018).\u003c/li\u003e\n\u003cli\u003eYan, J. \u003cem\u003eet al.\u003c/em\u003e Origin of the common UV absorption feature in cultured pearls and shells. \u003cem\u003eJ. Mater. Sci.\u003c/em\u003e \u003cstrong\u003e52\u003c/strong\u003e, 8362\u0026ndash;8369 (2017).\u003c/li\u003e\n\u003cli\u003eAgatonovic-Kustrin, S. \u0026amp; Morton, D. W. The Use of UV-Visible Reflectance Spectroscopy as an Objective Tool to Evaluate Pearl Quality. \u003cem\u003eMar. Drugs\u003c/em\u003e \u003cstrong\u003e10\u003c/strong\u003e, 1459\u0026ndash;1475 (2012).\u003c/li\u003e\n\u003cli\u003eKarampelas, S., Fritsch, E., Gauthier, J.-P. \u0026amp; Hainschwang, T. UV-Vis-NIR Reflectance Spectroscopy of Natural-Color Saltwater Cultured Pearls from \u003cem\u003ePinctada Margaritifera\u003c/em\u003e. \u003cem\u003eGems Gemol.\u003c/em\u003e \u003cstrong\u003e47\u003c/strong\u003e, 31\u0026ndash;35 (2011).\u003c/li\u003e\n\u003cli\u003eSoldati, A. L., Jacob, D. E., Wehrmeister, U., H\u0026auml;ger, T. \u0026amp; Hofmeister, W. Micro-Raman spectroscopy of pigments contained in different calcium carbonate polymorphs from freshwater cultured pearls. \u003cem\u003eJ. Raman Spectrosc.\u003c/em\u003e \u003cstrong\u003e39\u003c/strong\u003e, 525\u0026ndash;536 (2008).\u003c/li\u003e\n\u003cli\u003eKarampelas, S., Fritsch, E., Makhlooq, F., Mohamed, F. \u0026amp; Al‐Alawi, A. Raman spectroscopy of natural and cultured pearls and pearl producing mollusc shells. \u003cem\u003eJ. Raman Spectrosc.\u003c/em\u003e \u003cstrong\u003e51\u003c/strong\u003e, 1813\u0026ndash;1821 (2020).\u003c/li\u003e\n\u003cli\u003eHardman, M. F. \u003cem\u003eet al.\u003c/em\u003e Classification of Gem Materials Using Machine Learning. \u003cem\u003eGems Gemol.\u003c/em\u003e \u003cstrong\u003e60\u003c/strong\u003e, 306\u0026ndash;329 (2024).\u003c/li\u003e\n\u003cli\u003eHomkrajae, A., Sun, Z., Blodgett, T. \u0026amp; Zhou, C. Provenance Discrimination of Freshwater Pearls by LA-ICP-MS and Linear Discriminant Analysis (LDA). \u003cem\u003eGems Gemol.\u003c/em\u003e \u003cstrong\u003e55\u003c/strong\u003e, 47\u0026ndash;60 (2019).\u003c/li\u003e\n\u003cli\u003eMcCrea, J. M. On the Isotopic Chemistry of Carbonates and a Paleotemperature Scale. \u003cem\u003eJ. Chem. Phys.\u003c/em\u003e \u003cstrong\u003e18\u003c/strong\u003e, 849\u0026ndash;857 (1950).\u003c/li\u003e\n\u003cli\u003eEpstein, S., Buchsbaum, R., Lowenstam, H. A. \u0026amp; Urey, H. C. REVISED CARBONATE-WATER ISOTOPIC TEMPERATURE SCALE. \u003cem\u003eGeol. Soc. Am. Bull.\u003c/em\u003e \u003cstrong\u003e64\u003c/strong\u003e, 1315 (1953).\u003c/li\u003e\n\u003cli\u003eImmenhauser, A., Sch\u0026ouml;ne, B. R., Hoffmann, R. \u0026amp; Niedermayr, A. Mollusc and brachiopod skeletal hard parts: Intricate archives of their marine environment. \u003cem\u003eSedimentology\u003c/em\u003e \u003cstrong\u003e63\u003c/strong\u003e, 1\u0026ndash;59 (2016).\u003c/li\u003e\n\u003cli\u003eClark, I. D. \u0026amp; Fritz, P. \u003cem\u003eEnvironmental Isotopes in Hydrogeology\u003c/em\u003e. (CRC Press/Lewis Publishers, Boca Raton, FL, 1997).\u003c/li\u003e\n\u003cli\u003eEpstein, S. \u0026amp; Mayeda, T. Variation of O\u003csup\u003e18\u003c/sup\u003e content of waters from natural sources. \u003cem\u003eGeochim. Cosmochim. Acta\u003c/em\u003e \u003cstrong\u003e4\u003c/strong\u003e, 213\u0026ndash;224 (1953).\u003c/li\u003e\n\u003cli\u003eBigg, G. R. \u0026amp; Rohling, E. J. An oxygen isotope data set for marine waters. \u003cem\u003eJ. Geophys. Res. Oceans\u003c/em\u003e \u003cstrong\u003e105\u003c/strong\u003e, 8527\u0026ndash;8535 (2000).\u003c/li\u003e\n\u003cli\u003eLocarnini, R. A. \u003cem\u003eet al.\u003c/em\u003e World Ocean Atlas 2023, Volume 1: Temperature. (2024) doi:10.25923/54BH-1613.\u003c/li\u003e\n\u003cli\u003eLeGrande, A. N. \u0026amp; Schmidt, G. A. Global gridded data set of the oxygen isotopic composition in seawater. \u003cem\u003eGeophys. Res. Lett.\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e, 2006GL026011 (2006).\u003c/li\u003e\n\u003cli\u003eSpero, H. J., Bijma, J., Lea, D. W. \u0026amp; Bemis, B. E. Effect of seawater carbonate concentration on foraminiferal carbon and oxygen isotopes. \u003cem\u003eNature\u003c/em\u003e \u003cstrong\u003e390\u003c/strong\u003e, 497\u0026ndash;500 (1997).\u003c/li\u003e\n\u003cli\u003eMcConnaughey, T. A. \u0026amp; Gillikin, D. P. Carbon isotopes in mollusk shell carbonates. \u003cem\u003eGeo-Mar. Lett.\u003c/em\u003e \u003cstrong\u003e28\u003c/strong\u003e, 287\u0026ndash;299 (2008).\u003c/li\u003e\n\u003cli\u003eClayton, R. N. \u0026amp; Degens, E. T. Use of Carbon Isotope Analyses of Carbonates for Differentiating Fresh-Water and Marine Sediments: GEOLOGICAL NOTES. \u003cem\u003eAAPG Bull.\u003c/em\u003e \u003cstrong\u003e43\u003c/strong\u003e, 890\u0026ndash;897 (1959).\u003c/li\u003e\n\u003cli\u003eSilverman, S. R. \u0026amp; Epstein, S. Carbon Isotopic Compositions of Petroleums and Other Sedimentary Organic Materials. \u003cem\u003eAAPG Bull.\u003c/em\u003e \u003cstrong\u003e42\u003c/strong\u003e, 998\u0026ndash;1102 (1958).\u003c/li\u003e\n\u003cli\u003eKeith, M. L., Anderson, G. M. \u0026amp; Eichler, R. Carbon and oxygen isotopic composition of mollusk shells from marine and fresh-water environments. \u003cem\u003eGeochim. Cosmochim. Acta\u003c/em\u003e \u003cstrong\u003e28\u003c/strong\u003e, 1757\u0026ndash;1786 (1964).\u003c/li\u003e\n\u003cli\u003eBoutton, T. W. Stable carbon isotope ratios of natural materials: 2. Atmospheric, terrestrial, marine, and freshwater environments. in \u003cem\u003eCarbon isotope techniques\u003c/em\u003e (1991).\u003c/li\u003e\n\u003cli\u003eQuay, P. D., Emerson, S. R., Quay, B. M. \u0026amp; Devol, A. H. The carbon cycle for Lake Washington\u0026mdash;A stable isotope study. \u003cem\u003eLimnol. Oceanogr.\u003c/em\u003e \u003cstrong\u003e31\u003c/strong\u003e, 596\u0026ndash;611 (1986).\u003c/li\u003e\n\u003cli\u003eWeber, J. N. \u0026amp; Woodhead, P. M. J. Temperature dependence of oxygen-18 concentration in reef coral carbonates. \u003cem\u003eJ. Geophys. Res.\u003c/em\u003e \u003cstrong\u003e77\u003c/strong\u003e, 463\u0026ndash;473 (1972).\u003c/li\u003e\n\u003cli\u003eGilbert, P. U. P. A. \u003cem\u003eet al.\u003c/em\u003e Biomineralization: Integrating mechanism and evolutionary history. \u003cem\u003eSci. Adv.\u003c/em\u003e \u003cstrong\u003e8\u003c/strong\u003e, eabl9653 (2022).\u003c/li\u003e\n\u003cli\u003eYoshimura, T., Nakashima, R., Suzuki, A., Tomioka, N. \u0026amp; Kawahata, H. Oxygen and carbon isotope records of cultured freshwater pearl mussel \u003cem\u003eHyriopsis sp\u003c/em\u003e. shell from Lake Kasumigaura, Japan. \u003cem\u003eJ. Paleolimnol.\u003c/em\u003e \u003cstrong\u003e43\u003c/strong\u003e, 437\u0026ndash;448 (2010).\u003c/li\u003e\n\u003cli\u003eFarfan, G. A., Zhou, C., Valley, J. W. \u0026amp; Orland, I. J. Coupling Mineralogy and Oxygen Isotopes to Seasonal Environmental Shifts Recorded in Modern Freshwater Pearl Nacre From Kentucky Lake. \u003cem\u003eGeochem. Geophys. Geosystems\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, e2021GC009995 (2021).\u003c/li\u003e\n\u003cli\u003eKawahata, H., Inoue, M., Nohara, M. \u0026amp; Suzuki, A. Stable isotope and chemical composition of pearls: Biomineralization in cultured pearl oysters in Ago Bay, Japan. \u003cem\u003eJ. Oceanogr.\u003c/em\u003e \u003cstrong\u003e62\u003c/strong\u003e, 405\u0026ndash;412 (2006).\u003c/li\u003e\n\u003cli\u003eWoo, K. S. Textural, Isotopic, and Chemical Investigation of Cultured Pearls. \u003cem\u003eJ. Oceanol. Soc. Korea\u003c/em\u003e \u003cstrong\u003e24\u003c/strong\u003e, 69\u0026ndash;78 (1989).\u003c/li\u003e\n\u003cli\u003eEuropean Union-Copernicus Marine Service. Global Ocean 1/12\u0026deg; Physics Analysis and Forecast updated Daily. Mercator Ocean International https://doi.org/10.48670/MOI-00016 (2016).\u003c/li\u003e\n\u003cli\u003eReynolds, M. R. Physical oceanography of the Gulf, Strait of Hormuz, and the Gulf of Oman\u0026mdash;Results from the Mt Mitchell expedition. \u003cem\u003eMar. Pollut. Bull.\u003c/em\u003e \u003cstrong\u003e27\u003c/strong\u003e, 35\u0026ndash;59 (1993).\u003c/li\u003e\n\u003cli\u003eVaughan, G. O., Al-Mansoori, N. \u0026amp; Burt, J. A. The Arabian Gulf. in \u003cem\u003eWorld Seas: an Environmental Evaluation\u003c/em\u003e 1\u0026ndash;23 (Elsevier, 2019). doi:10.1016/B978-0-08-100853-9.00001-4.\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eThe Gulf Ecosystem: Health and Sustainability\u003c/em\u003e. (Backhuys, Leiden, 2002).\u003c/li\u003e\n\u003cli\u003eRivers, J. M. \u003cem\u003eet al.\u003c/em\u003e The Geochemistry of Qatar Coastal Waters and its Impact on Carbonate Sediment Chemistry and Early Marine Diagenesis. \u003cem\u003eJ. Sediment. Res.\u003c/em\u003e \u003cstrong\u003e89\u003c/strong\u003e, 293\u0026ndash;309 (2019).\u003c/li\u003e\n\u003cli\u003eConroy, J. L. \u003cem\u003eet al.\u003c/em\u003e Spatiotemporal variability in the \u0026delta;\u003csup\u003e18\u003c/sup\u003eO‐salinity relationship of seawater across the tropical Pacific Ocean. \u003cem\u003ePaleoceanography\u003c/em\u003e \u003cstrong\u003e32\u003c/strong\u003e, 484\u0026ndash;497 (2017).\u003c/li\u003e\n\u003cli\u003eZeebe, R. \u0026amp; Wolf-Gladrow, D. \u003cem\u003eCO\u003csub\u003e2\u003c/sub\u003e in Seawater: Equilibrium, Kinetics, Isotopes\u003c/em\u003e. (Elsevier, Amsterdam, 2007).\u003c/li\u003e\n\u003cli\u003eRubinson, M. \u0026amp; Clayton, R. N. Carbon-13 fractionation between aragonite and calcite. \u003cem\u003eGeochim. Cosmochim. Acta\u003c/em\u003e \u003cstrong\u003e33\u003c/strong\u003e, 997\u0026ndash;1002 (1969).\u003c/li\u003e\n\u003cli\u003eL\u0026eacute;cuyer, C. \u003cem\u003eet al.\u003c/em\u003e Carbon and oxygen isotope fractionations between aragonite and calcite of shells from modern molluscs. \u003cem\u003eChem. Geol.\u003c/em\u003e \u003cstrong\u003e332\u0026ndash;333\u003c/strong\u003e, 92\u0026ndash;101 (2012).\u003c/li\u003e\n\u003cli\u003eRomanek, C. S., Grossman, E. L. \u0026amp; Morse, J. W. Carbon isotopic fractionation in synthetic aragonite and calcite: Effects of temperature and precipitation rate. \u003cem\u003eGeochim. Cosmochim. Acta\u003c/em\u003e \u003cstrong\u003e56\u003c/strong\u003e, 419\u0026ndash;430 (1992).\u003c/li\u003e\n\u003cli\u003eGhosh, P. \u003cem\u003eet al.\u003c/em\u003e \u003csup\u003e13\u003c/sup\u003eC\u0026ndash;\u003csup\u003e18\u003c/sup\u003eO bonds in carbonate minerals: A new kind of paleothermometer. \u003cem\u003eGeochim. Cosmochim. Acta\u003c/em\u003e \u003cstrong\u003e70\u003c/strong\u003e, 1439\u0026ndash;1456 (2006).\u003c/li\u003e\n\u003cli\u003eEiler, J. M. \u0026ldquo;Clumped-isotope\u0026rdquo; geochemistry\u0026mdash;The study of naturally-occurring, multiply-substituted isotopologues. \u003cem\u003eEarth Planet. Sci. Lett.\u003c/em\u003e \u003cstrong\u003e262\u003c/strong\u003e, 309\u0026ndash;327 (2007).\u003c/li\u003e\n\u003cli\u003eMcConnaughey, T. \u003csup\u003e13\u003c/sup\u003eC and \u003csup\u003e18\u003c/sup\u003eO isotopic disequilibrium in biological carbonates: I. Patterns. \u003cem\u003eGeochim. Cosmochim. Acta\u003c/em\u003e \u003cstrong\u003e53\u003c/strong\u003e, 151\u0026ndash;162 (1989).\u003c/li\u003e\n\u003cli\u003eDa\u0026euml;ron, M. \u0026amp; Vermeesch, P. Omnivariant Generalized Least Squares Regression: Theory, Geochronological Applications, and Making the Case for Reconciled \u0026Delta;47 calibrations. \u003cem\u003eChem. Geol.\u003c/em\u003e \u003cstrong\u003e647\u003c/strong\u003e, 121881 (2024).\u003c/li\u003e\n\u003cli\u003eKim, S.-T., Mucci, A. \u0026amp; Taylor, B. E. Phosphoric acid fractionation factors for calcite and aragonite between 25 and 75 \u0026deg;C: Revisited. \u003cem\u003eChem. Geol.\u003c/em\u003e \u003cstrong\u003e246\u003c/strong\u003e, 135\u0026ndash;146 (2007).\u003c/li\u003e\n\u003cli\u003eDa\u0026euml;ron, M. Full Propagation of Analytical Uncertainties in \u0026Delta;\u003csub\u003e47\u003c/sub\u003e Measurements. \u003cem\u003eGeochem. Geophys. Geosystems\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, e2020GC009592 (2021).\u003c/li\u003e\n\u003cli\u003eBernasconi, S. M. \u003cem\u003eet al.\u003c/em\u003e InterCarb: A Community Effort to Improve Interlaboratory Standardization of the Carbonate Clumped Isotope Thermometer Using Carbonate Standards. \u003cem\u003eGeochem. Geophys. Geosystems\u003c/em\u003e \u003cstrong\u003e22\u003c/strong\u003e, e2020GC009588 (2021).\u003c/li\u003e\n\u003cli\u003eBernasconi, S. M. \u003cem\u003eet al.\u003c/em\u003e Reducing Uncertainties in Carbonate Clumped Isotope Analysis Through Consistent Carbonate‐Based Standardization. \u003cem\u003eGeochem. Geophys. Geosystems\u003c/em\u003e \u003cstrong\u003e19\u003c/strong\u003e, 2895\u0026ndash;2914 (2018).\u003c/li\u003e\n\u003cli\u003eAnderson, N. T., Bergmann, K. D., Braun, M. G., Griffith, E. M. \u0026amp; Saltzman, M. R. High-resolution record of global cooling during a large Mississippian positive carbon isotope excursion. \u003cem\u003eEarth Planet. Sci. Lett. \u003c/em\u003e\u003cstrong\u003e668\u003c/strong\u003e, 119557 (2025).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-6874927/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6874927/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Pearls are nacreous biogenic products that can be classified by whether they are natural or cultured, what species of mollusk produced them, and what environment they were grown in. Due to the subtle compositional and morphological differences between pearl types, determining pearl provenance can be problematic. To address these challenges, in this study we introduce a pearl identification workflow combining stable isotope analysis of pearl carbonate minerals to identify their geographic origin, and a Raman spectroscopy-based machine learning model to determine pearl species. Stable isotope data reveal clustering in δ18O based on the geographic origin of saltwater pearls. Additionally, pearl oyster shells from the Arabian Gulf were used as a tractable model system to investigate links between seawater geochemistry and carbonate stable isotope signatures. Paralleling these studies, Raman spectra of pearls formed by P. radiata, P. maxima, and P. fucata oysters were utilized to train a support vector machine classifier to predict pearl species with 96.4% accuracy. The combined results from these investigations demonstrate their utility in tracing pearl provenance by identifying geography and species of origin, which could be employed at a larger scale to improve pearl classification and provide insights into the formation mechanisms of biogenic carbonates.","manuscriptTitle":"Decoding Nacre: Uncovering Unique Identifiers of Pearl Provenance","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-26 14:20:26","doi":"10.21203/rs.3.rs-6874927/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"a845e1ba-1b27-40e4-99d4-ec2a40857ec2","owner":[],"postedDate":"February 26th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":63209291,"name":"Physical sciences/Materials science"},{"id":63209292,"name":"Physical sciences/Materials science/Techniques and instrumentation/Characterization and analytical techniques"},{"id":63209293,"name":"Physical sciences/Materials science/Biomaterials"}],"tags":[],"updatedAt":"2026-03-02T16:27:30+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-26 14:20:26","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6874927","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6874927","identity":"rs-6874927","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}