Deep-time preservation of amino acids in mammalian fossil tooth enamel

TL;DR

Tooth enamel preserves amino acids for over 40 million years, offering a valuable resource for studying ancient life and evolution.

Contribution

Demonstrates the long-term preservation of amino acids in fossil enamel, up to 48 million years.

Findings

Amino acids persist in enamel at least to the Eocene period.

The intra-crystalline organic fraction stabilizes after an initial rapid decline.

Preservation is independent of taphonomic context and shows similar variability in modern and fossil samples.

Abstract

Tooth enamel, primarily composed of bioapatite, is a promising archive of endogenous organic matter for studying ancient fauna. Despite its low organic content (~1%), protein residues have been identified in teeth up to 24 million years old. This study investigates the preservation of total hydrolysable amino acids (THAAs) in fossil enamel dating back as far as 48 million years. Modern and fossil enamel from large herbivorous mammals (Equidae, Rhinocerotidae, Proboscidea) across various taphonomic settings and Cenozoic periods reveal that AAs persist at least to the Eocene. The “intra-crystalline” organic fraction stabilizes after an initial rapid decline within the first 0.10 million years. Preservation appears independent of taphonomic context, and the relative abundance of amino acids is similarly variable in both modern and fossil samples. These findings demonstrate that enamel is a diagenetically robust substrate for long-term organic preservation, with significant potential for phylogenetic and ecological reconstructions in the fossil record.

Tooth enamel preserves endogenous amino acids for over 40 million years, representing a robust and underexplored archive for deep-time biomolecular research and enabling future paleoproteomic and isotopic studies of ancient ecologies and evolution.

Figures

Amino acid loss in tooth enamel over geological time.Total absolute AAs abundances (pM/mg) of modern (open points) and fossil (filled points) tooth enamel from Equidae (yellow), Proboscidea (pink), and Rhinocerotidae (blue) plotted against geological age in millions of years (Ma) of the respective fossil sites (Supplementary Data 3). The dark gray area represents 1 SD, and the light gray area represents 2 SD for averaged modern samples abundances across all taxa. The scatter plot (A) displays the fast decrease in AAs abundance with time. The panel B focuses on the 0–0.15 Ma time interval and h

Temporal variability of amino acid concentrations in fossil tooth enamel.Scatter plot of the variability of AA concentrations in enamel from fossil mammal teeth from a range of different aquatic and terrestrial taphonomic settings (Supplementary Data 1 and 3) ranging in age from the Late Pleistocene to older periods (up to the middle Eocene). Variability is related more to the age of the tooth sample rather than taxonomy or taphonomic setting. The size of the circles is related to the age: smaller circles are younger than 0.125 Ma, while the larger circles are older than 0.125 Ma.

Absolute and relative amino acid abundances in fossil mammal enamel across geological time.A Absolute and B relative abundance of abundance of AAs in modern (open points) and fossil (filled points) enamel according to geological age for each of the three large mammal taxonomic groups Equidae (yellow), Rhinocerotidae (blue), Proboscidea (pink). The gray shaded area represents 1 SD of modern enamel sample’s abundances of these mammal taxa.

B), Arg, Phe, and Ile are the most important predictors of age, with Tyr also showing substantial importance, whereas His and Ser show negligible importance. This outcome is consistent with their tendency to degrade rapidly during the early stages of diagenesis, meaning that over the long timescales captured in our dataset they do not correlate well with age. However, it should be emphasized that in shorter temporal frameworks or different preservation contexts, Ser and His are known to perform as good indicators of age^[45](#CR45)^. For the proboscidean model (Supplementary Fig. 4C), Tyr stan

Full text

Introduction

Well preserved organic biomolecules in fossils represent key archives of taxonomic, phylogenetic, ecological and behavioral information^1–3^. Biominerals are produced by living organisms via extra- or intra-cellular biomineralization and are made up of an organic (commonly proteinaceous phase) and an inorganic (mineralized phase) fraction. The way in which the two phases are spatially organized is complex and still not well characterized for many biominerals. The organic components are commonly classified in terms of their spatial relationship with the biomineral crystals as inter- and intra-crystalline fractions^4,5^. The former are located between the crystals and can be exposed post-mortem to external contamination, alteration or loss by degradation and leaching, while the latter, entrapped within the crystals, may behave as a closed system with respect to taphonomic processes during diagenesis, potentially preserving original biomolecules for thousands or even up to millions of years in fossils^4–7^.

Recent studies have demonstrated that the “intra-crystalline” organic fraction of fossil biominerals (including enamel bioapatite) preserves the original pristine nitrogen isotope composition of both invertebrates and vertebrates^8–10^ with clear trophic enrichment between herbivores and carnivores in fossil tooth enamel, and permit reconstruction of ancient food chains^11–15^. Furthermore, alteration experiments with enamel showed little to no change in organic N isotope composition or concentrations when partially dissolved, subjected to oxidative attack and heated up to 500 °C^8^. Similarly, only thin enamel alteration rims (≈200 µm) of element uptake were observed after in vitro alteration experiments of dental cubes placed in aqueous solutions at sub-boiling temperatures^16^. These studies suggest that tooth enamel is a diagenetically robust tissue in which endogenous organic nitrogen (N) is preserved as long as enamel bioapatite does not recrystallize. However, despite preservation of original N content and isotopic composition, the structural integrity of the original biomolecules (e.g., as proteins, peptides or amino acids) remains unclear.

Endogenous intra-crystalline organic matter has been extracted from eggshells^6,17–22^, foraminifera^23,24^, brachiopods^25,26^ and corals^27^ demonstrating that peptide-bound amino acids (AAs) can survive in ancient calcareous biominerals as old as 66 Ma^18,20^. These studies highlight the potential for tooth enamel as a reservoir of pristine peptides (i.e., protein fragments) and AAs in extant and extinct vertebrate taxa^28–30^.

Enamel is comprised predominantly of highly mineralized carbonate-bearing hydroxylapatite (≈96%), water (≈3%) and organic matter (≈1%)^31,32^. Enamel formation occurs through a multi-phase process comprising an initial appositional phase, in which an organic-rich matrix is secreted, followed by a maturation phase marked by progressive crystal growth and matrix protein removal. As enamel matures, its mineral content increases dramatically, from ≈15–25% to over 95% by weight, while porosity decreases. This densification, particularly in inner and mid-enamel regions that mineralize earlier than the outer layer, results in a densely packed hydroxylapatite structure that isolates intra-crystalline organic residues^33,34^. The organic fraction includes mainly proteins and proteinases, such as amelogenin, ameloblastin, enamelin and metalloproteins^31^. Despite the low abundance of organic matter (≈1%) in enamel, peptides have been found in fossil teeth that are several million years old (≈24 Ma)^35,36^, protected by the hydroxylapatite crystals^32,34^. The characteristic prismatic microstructure of enamel plays a crucial role in protecting organic constituents from chemical alteration during the process of fossilization or diagenesis. Indeed, the packed hydroxylapatite structure results in extremely low porosity and limited fluid permeability^34^. These characteristics restrict the movement of water and ions within this hard tissue, creating a physically stable and chemically isolated microenvironment. Organic molecules that become entrapped during crystal growth, especially those located within the intra-crystalline compartment, are effectively shielded from microbial degradation, oxidative agents, and leaching. As a result, the preservation potential of enamel bound peptides and AAs is significantly enhanced compared to other biominerals such as dentine or bone, where greater porosity and vascularization facilitate post-mortem molecular alteration^32,34^. However, there are still different open questions: how and where proteinaceous intra-crystalline organic matter is bound, how long this matter can persist within the enamel matrix, and what form it takes (e.g., peptides, AAs^37^). Indeed, protein preservation in geological settings depends on multiple factors including temperature, pH, humidity, and sediment type^4,18^. Foundational studies on different types of fossil materials have long demonstrated the differential stability of AAs over geological timescales^24,30,38^, emphasizing the impact of the diagenetic processes on proteinaceous residues in mineralized tissues^25^.

While AA preservation over time has been widely studied in some archeological materials, such as mollusk shells and corals^18,19,27,39–43^, AAs in enamel have not been as-well-studied. The limited studies that have been published focus primarily on proteome detection through mass spectrometry^1,36,44^ or on amino acid racemization (AAR) as a technique for dating^4,28,37^. To date, the oldest partial peptide sequences identified from fossil enamel are 21‒24 Ma^36^. However, the stability of AA composition in fossil enamel over long geological time scales and across different taphonomic settings has not been systematically assessed. Understanding the extent and limitations of AA preservation in deep time is a first necessary step for understanding diagenetic processes that impact and may potentially bias paleoproteomics and isotopic analyses.

The investigation of total hydrolysable amino acids (THAAs) could provide the estimation of the total amino acid pool, capturing both free and peptide-bound AAs without requiring extensive sample amounts or destructive processing. For this reason, we systematically explore preservation of THAAs in fossil enamel through time as proxy for assessing the overall preservation of intra-crystalline organic matter in fossil enamel. To do so, we characterized the composition and variation of 11 AAs in mammalian enamel of specimens of different geological ages (0.04‒48 Ma) and distinct taphonomic settings (fluvial, limnic, peat bog, coal seam, karst fissures and others) from central Europe. Enamel from 72 fossil teeth belonging to three different large mammal clades (Proboscidea, Rhinocerotidae, and Equidae), was measured to assess whether taxon-specific enamel characteristics influence AA preservation in different taphonomic settings. Enamel from modern individuals belonging to the same taxonomic clades (i.e., elephants, rhinoceroses, and horses; n = 12 samples) was analyzed for comparison to the fossil dataset. Additionally, taxa (n = 13 different families) with different feeding strategies (i.e., carnivores, herbivores, omnivores) were represented in the modern samples, which permitted us to evaluate potential taxon-specific variations in enamel AA composition.

Results

To minimize phylogenetic effects, fossil teeth of Proboscidea, Rhinocerotidae, and Equidae specimens recovered in different taphonomic settings (Supplementary Data 1) have been compared with modern enamel samples from the same taxonomic groups (Supplementary Data 2). The absolute AAs abundances in fossil enamel (Supplementary Data 3) of the three taxonomic clades have been evaluated in relation to their geological age (Fig. 1). The data revealed that AA concentrations in fossil enamel of the three mammal clades consistently diminished compared to modern samples. Notably, in all three mammal clades analyzed, the overall AA concentration strongly decreased during the first ≈0.10 Ma (by about 55–96% compared to modern levels, depending on the specific AA), but it stabilized after this initial decline.Fig. 1. Amino acid loss in tooth enamel over geological time.Total absolute AAs abundances (pM/mg) of modern (open points) and fossil (filled points) tooth enamel from Equidae (yellow), Proboscidea (pink), and Rhinocerotidae (blue) plotted against geological age in millions of years (Ma) of the respective fossil sites (Supplementary Data 3). The dark gray area represents 1 SD, and the light gray area represents 2 SD for averaged modern samples abundances across all taxa. The scatter plot (A) displays the fast decrease in AAs abundance with time. The panel B focuses on the 0–0.15 Ma time interval and highlights the rapid decrease in AA abundance observed during the Holocene and Late Pleistocene (from 55 up to 96% according to the AA).

Specifically, Equidae exhibit greater stability in AA concentrations over time, likely as the result of lower starting concentrations and reduced inter-individual variability in modern specimens. Proboscidea and Rhinocerotidae showed a more pronounced initial decrease in AA abundance than Equidae, with a 70–75% reduction in AAs abundance relative to their modern counterparts, probably due to the higher inter-individual variability observed in modern teeth. Regardless, the majority of the AA abundances for fossil enamel are within 2 SD of the modern samples.

Additionally, we examined the impact of eight different taphonomic settings on the degradation of enamel AAs (Fig. 2). When subjected to the same taphonomic setting, teeth display comparable temporal patterns of variation in AA concentrations, irrespective of taxon. However, samples younger than 0.125 Ma (i.e., Late Pleistocene) exhibit higher AAs concentrations, while those older than 0.125 Ma revealed markedly lower concentrations (Fig. 2) highlighting that AA concentrations change predominantly in the early stages of the fossilization process. Only the samples from a limnic taphonomic setting are an exception. Indeed, there is a substantial overlap between specimens younger and older than 0.125 Ma in a limnic environment. This overlap reduces the distinction between the two age groups, indicating that in this depositional environment the concentrations of AAs are similar, regardless of age. Overall, these findings demonstrate that for fossil samples, while burial conditions may affect AA preservation in certain cases, they are not the primary factor in determining AA stability over time.Fig. 2. Temporal variability of amino acid concentrations in fossil tooth enamel.Scatter plot of the variability of AA concentrations in enamel from fossil mammal teeth from a range of different aquatic and terrestrial taphonomic settings (Supplementary Data 1 and 3) ranging in age from the Late Pleistocene to older periods (up to the middle Eocene). Variability is related more to the age of the tooth sample rather than taxonomy or taphonomic setting. The size of the circles is related to the age: smaller circles are younger than 0.125 Ma, while the larger circles are older than 0.125 Ma.

Total AA content and normalized AA amounts decreased across all taxa (Supplementary Data 4, Fig. 3A). Gly, Asx, and Glx were the most unstable AAs in our dataset and thus particularly prone to poor preservation. For example, in proboscideans, Asx declined from ≈182.8–715.8 pM/mg in modern samples to ≈121.61–409.09 pM/mg within the first 0.125 Ma (a reduction of ≈35–43%), and further to a minimum of 16.1 pM/mg in samples older than 0.125 Ma (>90% decrease). In Rhinocerotidae, Asx dropped from ≈71.68–528.9 pM/mg in modern samples to ≈40.84 pM/mg in fossils younger than 0.125 Ma (≈40–45% decrease), and then to ≈18.2 pM/mg in older specimens (>70% decrease). Similarly, Equidae values decreased from ≈57.6–300.8 pM/mg in modern samples to 161.52 pM/mg in younger than 0.125 Ma fossils (≈40% decrease), and to 15.10 pM/mg in older ones (>90% decrease). Across all three clades, Asx, Glx, and Gly showed comparable proportions of decline. Importantly, most fossil concentrations still fell within 1 SD of the variability observed in modern samples. To better illustrate these temporal changes, absolute and relative abundances of each AA were compared between modern and fossil taxa (Fig. 3).Fig. 3. Absolute and relative amino acid abundances in fossil mammal enamel across geological time.A Absolute and B relative abundance of abundance of AAs in modern (open points) and fossil (filled points) enamel according to geological age for each of the three large mammal taxonomic groups Equidae (yellow), Rhinocerotidae (blue), Proboscidea (pink). The gray shaded area represents 1 SD of modern enamel sample’s abundances of these mammal taxa.

Relative abundance (sum to 100%) (Supplementary Data 5, Supplementary Fig. 1) was also considered in evaluating AA preservation over time due to the importance of the overall ratio(s) of AAs to one another for proteomic and isotopic analyses. We observe high variability in AA relative abundances in both modern and fossil samples (Fig. 3B). Despite the variability, the data obtained are close to the ones obtained by Dickinson et al.^28^, confirming that the relative ratios between different AAs was the expected one. When Asx, Glx, and Gly are excluded (on the grounds of their aforementioned instability), variability in modern samples is comparable to that observed in fossil samples (Supplementary Fig. 2).

In order to better understand the significance of the AAs changes over time, the variance in AA relative abundance of the three modern taxa was evaluated in comparison with the relative values obtained from the fossils (Supplementary Fig. 3). Fossil proboscideans have the highest secular variation, Rhinocerotidae intermediate, and Equidae showed again the lowest one in AAs abundance, likely linked to a better preservation of the enamel-bound organic matter and thus AAs. Considering the variance for each AA in the different taxa, proboscideans showed the highest intra-taxon variance within modern samples. On the contrary, Equidae showed the lowest intra-taxon variance both in modern and fossil samples for the majority of AAs. Rhinocerotidae intra- and inter-taxon data revealed a behavior in the middle, closer to the Equidae ones than to the Proboscidea ones.

To assess the role of AAs concentrations in predicting sample age, independent of racemization rates, we built a Random Forest (RF) model using all the AA data of Equidae, Rhinocerotidae and Proboscidea (Supplementary Fig. 4). Random trees were built with n = 4 random variables at a time (i.e., optimized mtry parameter = 4). The model performances were evaluated with a 10-fold cross-validation (resulting R^2^ = 0.67, RMSE = 6.8). Three additional sub-models were also run, one for each taxonomic family. The importance of the predictors in the random forest model was computed based on the %IncMSE which is the increase of the cross-validation mean squared error, randomly permuting values of that specific variable; this value is then scaled to its standard deviation (Supplementary Data 6).

In the global model (Supplementary Fig. 4A), Phe (62.8% ± 4.0), Tyr (24.8% ± 1.5), Arg (24.7% ± 2.1), and Ile (29.5% ± 2.9) emerged as the most influential predictors of sample age, with the highest increases in prediction error when permuted. These amino acids show the most consistent concentration shifts with increasing geological age across all three large mammal families. Conversely, Gly (3.2% ± 0.9) and Leu (2.9% ± 0.7) contributed only marginally to model performance.

When examining the taxon-specific models, distinct patterns of AAs importance become apparent. In the Equidae model (Supplementary Fig. 4B), Arg, Phe, and Ile are the most important predictors of age, with Tyr also showing substantial importance, whereas His and Ser show negligible importance. This outcome is consistent with their tendency to degrade rapidly during the early stages of diagenesis, meaning that over the long timescales captured in our dataset they do not correlate well with age. However, it should be emphasized that in shorter temporal frameworks or different preservation contexts, Ser and His are known to perform as good indicators of age^45^. For the proboscidean model (Supplementary Fig. 4C), Tyr stands out as by far the most important amino acid for age prediction, followed by Thr, His, Glx, and Ser. This highlights a differential diagenetic pathway or preservation characteristic in proboscidean enamel where Tyr’s degradation or alteration is a strong indicator of age. The Rhinocerotidae model (Supplementary Fig. 4D) shows a less pronounced hierarchy of importance compared to the other taxa. Here, Ile, Gly and Arg show highest positive importance, while several other AAs exhibit very low or near-zero importance.

This suggests that, within the scope of our dataset, age-related changes in individual amino acids are less distinct for Rhinocerotidae than for the other clades.

Discussion

Comparing modern samples, it can be noted that among the three large mammal clades (Rhinocerotidae, Proboscidea, and Equidae), Proboscidea exhibited the highest intra-taxon variability in total concentration of AAs. This is likely related to the unique mineralization pattern of Proboscidea teeth, which combine hypsodont and brachydont features^46^, display daily incremental growth, and undergo horizontal tooth replacement throughout an individual’s lifetime^47–49^. In contrast, Rhinocerotidae and Equidae possess hypsodont teeth with diphyodont vertical replacement (i.e., one set of deciduous teeth replaced by one set of permanent teeth), which generally exhibit more uniform prismatic enamel^50,51^. Proboscidea enamel includes elephant-type plate enamel (Mammuthus primigenius), composed of extensive vertical prismatic plates, and other Proboscidea taxa (e.g., Mastodon, Anancus, Stegodon, Gomphotherium), which have mixed prismatic and thicker prismless bands^47–49^. These structural differences appear to influence AA preservation: elephant-type enamel tends to be more stable and similar to the Rhinocerotidae and Equidae model, while other Proboscidea teeth show higher intra-tooth variability. This suggests that the amelogenesis of Proboscidea enamel results in a lower overall degree of mineralization and thus higher susceptibility to protein degradation. To minimize potential biases, we analyzed only mature enamel from fully mineralized permanent teeth of adult individuals, allowing a more accurate comparison across taxa.

AAs were detected in all fossil teeth analyzed and the results obtained in this study indicate that while absolute AA concentrations decreased initially, AAs are nevertheless preserved in fossil tooth enamel well back into the Paleogene (at least up to 48 Ma). These results suggest that in the first 0.10 Ma an organic fraction less protected by the hydroxylapatite crystals is lost from the enamel during the fossilization process, but when this more labile fraction is lost the remaining organic matter is well encapsulated and protected from diagenetic alteration within the enamel mineral structure, which seems to be the case independent of diagenetic setting.

Moreover, the taphonomic setting does not appear to markedly affect the total AAs amount, instead, the geological age is more important (see Fig. 2). Our observations suggest that the Late Pleistocene (≈0.125 Ma) is the turning point when initial AAs deterioration stabilizes, regardless of external factors such as taphonomic conditions and intrinsic factors such as different enamel microstructure/mineralization^46,50,51^. Although taphonomic setting does not drive preservation as shown in Fig. 2, the fossil equids from the UNESCO heritage site of Messel (Germany, ≈48 Ma)^52^, nevertheless demonstrate that the right convergence of factors may facilitate better preservation of AA in certain circumstances. Indeed, the site of Messel is a world famous conservation Lagerstätte known particularly for its exceptional fossil preservation (including e.g., soft tissue and pigmentation due to the anoxic maar lake setting^53^). Previously published geochemical data from Messel (including enamel from the same specimens included in this study) demonstrate that the enamel preserves pristine isotope and element compositions suggesting minimal diagenetic alteration^54^. This preservation is also reflected in the AA abundances measured in this study which are strikingly similar to those of modern horses. Messel, however, is likely an exceptional case reflecting a unique convergence of factors, including characteristics of Equidae enamel and specific, favorable taphonomic conditions at Messel, which facilitated the preservation of amino acids despite their considerable age.

These findings confirmed that enamel of these three clades of large mammals could preserve and provide valuable geochemical information about the past. Indeed, Hypsodont equid teeth are commonly used for past environmental and paleoclimatic reconstruction due to their sequential and fast mineralization over multiple years^33,55^, but hypsodont molars of Rhinocerotidae and Proboscidea have also been used as archives of information about species evolution, mobility and feeding ecology, due to their survival in fossil context thanks to the thick enamel and large size of teeth^50,56^. Despite the analysis of ancient teeth for reconstructing the species evolution, previous works did not consider the investigation of the organic fraction of samples older than 24 Ma, thus the impact of diagenesis on the organic matter content of enamel over several million-year timescales has not been characterized.

We also show that some AA, are more prone to degradation than others. For example, Asx and Glx, are hydrophilic AAs and thus prone to decarboxylation (see Fig. 3). Glx is known to be a highly degradable AA through processes such as γ-decarboxylation, leading to γ-aminobutyric acid formation, lactamization, production of pyroglutamic acid, and deamidation^4,25,38^. This high susceptibility to degradation is consistently reflected in their high contribution to age prediction in the Random Forest model. On the other hand, there are some more stable AAs, as Leu, which is reported as one of the AAs being most resistant to decomposition^38^. However, our Random Forest analysis indicates that while Leu may be chemically stable, its predictive power for age is relatively low in the global model, suggesting its changes are less diagnostic of age compared to more reactive AAs.

Despite the overall similar behavior for the different taxa considering the relative AAs abundance, the fossil enamel of different taxa reacted differently to alteration processes for some AAs, as further emphasized by the taxon-specific Random Forest models: for instance, Gly in Proboscidea and Equidae, Ser in Equidae and Rhinocerotidae show lower contents in fossils and are thus less well preserved. In detail, Ser is prone to a fast decomposition, previously associated mainly with three chemical reactions (dehydration, reversible aldol cleavage and decarboxylation)^25,57,58^, which likely cause its decrease over time resulting in formation of Gly and formaldehyde^40^. The Random Forest model for proboscideans, however, shows Ser as a moderately important predictor of age, indicating its degradation pattern is more diagnostic within this clade. Moreover, Gly should be a quite stable AA, but it showed a significant decrease in both Proboscidea and Equidae, which could suggest degradation by a decarboxylation reaction. While temperature is likely the primary driver, we cannot fully exclude the possibility that other burial conditions (e.g., sedimentary context or microenvironmental pH) may also play a role under certain circumstances^24,25,38,59^. Moreover, Gly may also arise as a secondary product from the breakdown of less stable residues such as Ser, through dehydration or aldol cleavage, complicating the interpretation of its abundance in fossil materials^59^. In this study, we observe significant reductions in Gly content in older samples, suggesting that its preservation is more sensitive to diagenetic conditions than previously assumed, and cannot be used uncritically as a marker of molecular stability. However, according to the Random Forest models on the relative data, only in Equidae its change is really affecting the prediction of the age and thus is a significant feature of degradation.

The ability of the Random Forest model to predict sample age with reasonable accuracy (R^2^ = 0.67, RMSE = 6.8), and the varying importance of specific amino acids, highlight their potential as geochronological markers. For example, Tyr, Arg, Glx, and Phe consistently emerge as key predictors across the global model, while Tyr is exceptionally important for proboscideans, suggesting that there is also a taxon-specific AA degradation kinetics that can be harnessed for more refined dating.

The persistence of the detected AAs in the analyzed samples is considered consistent and not related to contaminations. Indeed, enamel powders were oxidatively cleaned with NaOCl as previously reported^28^ and some tests with longer times of oxidative cleaning were performed in order to evaluate the possibility to further reduce possible exogenous contaminations, with no impact on the final data.

These findings reveal that tooth enamel effectively protects and preserves intra-crystalline bound organic residues (i.e. AAs) over geological time (at least up to 48 Ma) and represents an untapped archive of information for future investigations into paleoproteomic and especially compound-specific isotopic approaches.

Conclusion

Amino acids in modern enamel from different large mammals did not show significant inter- and intra-taxon differences in their AAs distribution, demonstrating that phylogeny does not strongly influence tooth enamel AAs composition. The results of analyses of fossil enamel from Proboscidea, Rhinocerotidae and Equidae, demonstrate that AAs are characterized by an initial rapid decrease in concentration during the first 0.10 Ma, after which concentrations stabilize. From the Late Pleistocene to the Eocene fossil mammal enamel AA content and relative abundance remain similar at least as far back as 48 Ma. AAs preservation in enamel of Equidae teeth may be better than that of other taxa, potentially related to differences in tooth biomineralization or to higher relative content of intra-crystalline organic matter.

These findings suggest that fossil tooth enamel is a diagenetically robust, highly mineralized phosphatic biomineral archive in which endogenous organic matter such as AAs are frequently preserved back in deep time.

Future investigations are needed to clarify if these AAs are still bound in peptides or represent free AAs, to further our understanding of protein degradation phenomena in biominerals. The minimal sample requirement (~1 mg) of this method makes it ideal for non-destructive triage prior to targeted paleoproteomic analysis. This approach holds significant promise for expanding the analytical potential of enamel in paleoproteomics, by enabling the identification of protein fragments in samples where ancient DNA is no longer preserved.

Moreover, enamel-bound amino acids, once characterized at the compound-specific level, offer a powerful avenue for stable isotope analysis, enabling dietary and ecological reconstructions in extinct taxa. Applying such techniques to stratified enamel sequences could also provide temporally resolved insights into seasonal dietary changes, life history traits, or diet and habitat shifts through deep time. These results therefore pave the way for enamel to be developed not only as a molecular archive for paleoprotein survival, but also as a geochemical proxy of unprecedented resolution for ancient vertebrate ecology and evolution.

Methods

Samples

Powder from mature (i.e., fully mineralized) enamel from different modern and fossil tooth samples has been sampled from different museum collections. All modern samples are from different individuals and three replicates were measured per sample to characterize inter- and intra-population variability. Modern tooth enamel powder from and African elephant Loxodonta africana (AG-Lox^60^) was used as an in-house reference material for protocol development.

Modern samples comprised enamel powders from three different clades (Rhinocerotidae, proboscidean, and Equidae) (Supplementary Data 2) and were analyzed as reference for fossil samples. The same method was then applied to 76 fossil enamel samples from different large mammal species belonging to the same three clades from fossil sites covering a broad range of ages from 0.04 Ma to 48 Ma and different taphonomic settings (Supplementary Data 1). In addition to these, modern samples from 13 different herbivorous and carnivorous species of mammals, reptiles and fish were also analyzed, to assess potential variation in AA composition across taxa (covering a broader phylogenetic range) that could arise from differences in diet, due to the different types of AAs (in particular essential AAs) processed and incorporated in the enamel organic matter (Supplementary note for the results, Supplementary Fig. 5, Supplementary Data 2, 7, 8). The sampling procedure involved the initial removal of the outermost enamel surface to eliminate potential contaminants. Subsequently, the inner enamel layer was drilled under controlled conditions to minimize the risk of external contamination. Finally, the enamel layer adjacent to the dentin was isolated and stored separately not for these analyses.

Pure synthetic hydroxylapatite was used as a white reference to check possible contaminants from the procedure. The white samples were prepared and run following the same procedure of the enamel samples and no AAs were detected proving purity of the procedure (Supplementary Fig. 6).

Optimized protocol for sample preparation

The analyses were performed in the laboratories of the Organic Isotope Geochemistry Group of the Department of Climate Geochemistry at the MPIC in Mainz, Germany. Prior to AA analysis, enamel powder was cleaned following an oxidative cleaning step able to isolate the intra-crystalline biomolecules but avoiding heating^6^. Powders were weighed into sterile vials (2 mg) and 4.25 mL of sodium hypochlorite (NaOCl, Sigma Aldrich) 10‒15% was added. The vials were shaken at 110 rpm for 24 h. Then the supernatant NaOCl solution was removed using a pipette connected to a vacuum pump and the powder was washed 5 times with MilliQ water and once with methanol. Afterwards, the samples were completely dried, and 1 mg of this cleaned and dried enamel powder was weighed into microbore vials for AA extraction.

For the determination of total hydrolysable amino acids (THAA) we used the methods proposed by Dickinson et al.^28^, with minor modifications as described below. Samples (1 mg) were subjected to complete dissolution and subsequent hydrolysis by adding 20 μL/mg of 7 M HCl and heating to 110 °C in an oven for 24 h. Afterwards, excess acid was removed by centrifugal evaporation for 40 min. The sample was then redissolved in 20 μL/mg of 1 M HCl and sonicated for 10 min. To reduce the high concentration of phosphate ions, the previously developed protocol by Dickinson et al.^28^ was adjusted in order to achieve the best reproducibility and maximize the AA extraction. In brief, potassium hydroxide (1 M KOH, Sigma Aldrich) was added to the samples (35 μL/mg), leading to the formation of a cloudy liquid. Centrifugation at 13,500 rpm for 10 min resulted in a clear supernatant above a gel. The supernatant was collected and dried by centrifugal evaporation for 50 min. Before the HPLC analysis, 35 μL of spiked HCl (known concentration of L-Homoarginine in 0.1 N HCl) was added as an internal standard.

HPLC analyses

The analyses were performed using an Agilent AdvanceBio AAA LC Column 4.6 ×100 mm, 2.7 m 655950-802 and a Diode Array Detector (DAD). For the pre-column derivatization o-Phthalaldehyde (OPA) (10 mg/mL each of o-phthalaldehyde and 3-mercaptopropionic acid in 0.4 M borate buffer) and Fluorenylmethyloxycarbonylchloride (FMOC) (2.5 mg/mL in acetonitrile, 9-fluorenylmethylchloroformate) reagents were used, both purchased from Agilent Technologies.

Purchased reagents were delivered in sealed glass ampoules. Reagent vials were opened, distributed into daily portions and stored at 4 °C for up to 7 days. For distributing reagents, the used syringe was thoroughly rinsed with MilliQ water (OPA) or acetonitrile (FMOC) before use. The injection diluent was prepared by transferring 100 mL of mobile phase C and 0.4 mL of phosphoric acid (H_3_PO_4_) into a bottle. Two mobile phases were used: mobile phases B and C. The mobile phase B was prepared in a DURAN YOUTILITY® brown glass bottle, mixing acetonitrile, methanol, and MilliQ water in the proportions 45:45:10 by volume. The bottle was then ultrasonicated for 30 min to degas. For mobile phase C, disodium hydrogen phosphate (Na_2_HPO_4_, 1.4 g) and sodium tetraborate decahydrate (Na_2_B_4_O_7_ ∙ 10H_2_O, 3.8 g) were dissolved in 1 L of MilliQ water while stirring. After dissolution, the pH was adjusted to 8.2 by slow addition of 32% HCl. The resulting solution was filtered through a regenerated cellulose filter with 0.45 µm pore size by means of a vacuum filtration apparatus and subsequently transferred to a DURAN YOUTILITY® brown glass bottle.

Standard in-house solutions of mixed amino acids (AAs) were prepared in known concentrations. The solutions included 11 AAs: aspartic acid/asparagine (Asx), glutamic acid/glutamine (Glx), serine (Ser), histidine (His), glycine (Gly), threonine (Thr), arginine (Arg), tyrosine (Tyr), phenylalanine (Phe), isoleucine (Ile), leucine (Leu). For each AA a calibration curve was obtained and the LOD was determined (around 1 pM/mL for all the AAs).

Hydroxyproline (Hyp) and Proline (Pro) were included at the beginning of the analyses, but then they were excluded from this analysis due to several analytical and methodological considerations. As imino acids, Hyp and Pro often necessitate specific derivatization conditions and chromatographic optimization that can differ significantly from those applied to other proteinogenic amino acids. Furthermore, we observed that their quantification demonstrated reduced reproducibility and increased instability within our established analytical protocol for ancient enamel samples. Given the potential uncertainty in their accurate quantification and their primary association with collagen, we focused our analysis on the remaining amino acids, which were considered more robust and representative for this investigation into diagenetic changes in enamel.

At the beginning of each run a blank with just 0.1 N HCl was run to check the absence of nonspecific signal from analytical impurities. No signals were detected in the blank of the reported data.

Statistics and reproducibility

The integrated areas of the AAs in the chromatograms were converted to substance amounts in pM using AA-specific calibration curves obtained from analysis of in-house standard solutions at various concentrations. Concentrations per mg of cleaned enamel powder were determined via division by sample weight. The concentrations (pM/mg) obtained were then normalized using an internal standard (L-Homoarginine). A calibration curve for the internal standard (L-Homoarginine) was obtained by adding 35 μL of three different concentrations of pure L-Homoarginine solution (100 pM/μL, 32.25 pM/μL, 1 pM/μL) to AG-Lox samples. This additional correction was applied to correct for variations in peak areas observed in samples, likely due to the presence of small residues of phosphate ions. Principal component analysis (PCA) was then performed to the data (e.g., ref. ^6^). This being a compositional dataset, a centered-log ratio transformation (CLR) was performed prior to the actual PCA analysis. This approach is useful to evaluate the relative abundance of each AA in the different samples, considering the ratio between different AAs. In addition, this approach also limits possible variability in terms of total AAs content across different specimens, for example those belonging to different taxa. All the data analysis was performed using R. To evaluate the importance of AA in predicting the age of a sample, we built a Random Forest (RF) model using the caret package in R (version 4.0.5). Random trees were built with n = 4 random variables at a time (i.e., optimized mtry parameter = 4). The model performances were evaluated with a 10-fold cross-validation (resulting R^2^ = 0.67, RMSE = 6.8). The importance of the predictors in the random forest model was computed based on the %IncMSE which is the increase of the cross-validation mean squared error, randomly permuting values of that specific variable; this value is then scaled to its standard deviation.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Supplementary information

Supplementary Information Description of Additional Supplementary Files Supplementary Data1-8 Reporting summary