Main

The metabolic rate of an organism is the sum of the energy necessary to acquire, convert and allocate energy to growth, reproduction and maintenance, which sets energetic limits on biological activities and establishes the pace and pattern of life8,9,10. The temperature- and mass-dependence of metabolic rate, from the smallest unicellular organisms to the largest plants and animals9, enables one to reconstruct the metabolic rates of extinct organisms and retrodict fundamental large-scale features of their palaeoenvironments11. Otherwise, efforts to obtain knowledge of internal and external influences on fossil vertebrate metabolism have been limited to bone microanatomical correlates of thermophysiology12 and palaeoenvironmental reconstructions, which typically depend on studies of stable isotopes, faunal composition and community ecology, dental wear analyses, and/or palynology or palaeovegetation in geological context3,13,14.

Here we explore molecular ecological strategies for obtaining ultrafine-scale details of an ancient organism’s metabolism that: (1) reflect its internal physiological responses to environmental conditions; and (2) reflect the diet that fuelled its metabolic rate. Metabolomic profiling of animals is typically performed on whole blood or its components, such as plasma, and on its transudates, such as urine and saliva1. However, we have learned that preservation niches that contain metabolites that were once in the free circulation exist within the mineralized tissue ultrastructure, being not too ‘loose’ (where they may be shielded) and not too ‘tight’ (where they may be permeable), as we describe in the ‘Analysis of bone mineral niches’ section of the Methods, as well as potentially within the crystalline matrices of palaeosols. We posit that metabolites in hard tissue mineral niches are a serum transudate that has become entombed in the mineralizing extracellular matrix. Metabolites putatively survive in small compartments of interstitial water that are present in these hierarchically structured organic–inorganic interfaces7.

“Palaeo-metabolomics” has been advanced to examine metabolite identities of scent from organic residues15 and tobacco use as obtained from drilled bone samples16 in the historic archaeological record. Here we provide genuine palaeometabolomes from the prehistoric fossil record using metabolomics as a systems biological approach and document that metabolites preserved in Plio-Pleistocene fossils may be used to study physiological health and disease of past life and research the palaeoecology.

The palaeometabolome recovered from a bone or tooth differs from the instantaneous metabolome of an organism obtained from biofluid, such as blood, urine or saliva. Instead, metabolites will have become stochastically trapped in mineral niches during the entire period of hard tissue formation. However, because size and shape changes remove bone (known as remodelling) formed during periods of early development17, peak bone mass at skeletal maturity of any one bone will not encompass the entire period of its growth. Thus, by a rough approximation to periods of maturity, adult mouse cortical bone will harbour a metabolite accumulation of several months, bone of a ground squirrel will contain metabolites accumulated over less than one year, and so on for larger mammals, reaching several years if not decadal for a modern human, depending on the bone. Animals that secondarily remodel their cortical bone and trabecular bone surfaces potentially enable the recovery of metabolites for the remainder of their lifetime.

Samples analysed in this study are listed in Table 1. Fossil samples were from early human localities in eastern, central and southern Africa. Five fossil rodent bones derive from Bed I at Olduvai Gorge, Tanzania (from 1.8–1.7 million years ago (Ma)), and one suid fossil derives from Bed II (from 1.3 Ma) at this locality. We advance an ecogeographic perspective by analysing an elephantid dentine and cementum sample (from 2.4 Ma) from the Chiwondo Beds, Malawi, and a presumed bovid bone sample (from 3.0 Ma) from the site at Makapansgat, South Africa. In addition to the fossils, palaeosols from the localities and extant representatives of the fossil taxa have been analysed for endogeneity and quality control (a detailed list of all samples is presented in Supplementary Table 1.

We chose to particularly represent the Olduvai Gorge fossiliferous locality for two reasons: (1) the original semi-arid environmental reconstruction has recently been re-examined to include dense woodland and wetland3, to which we predicted exogenous metabolites of the palaeometabolomes might correspond; and (2) fossil owl pellets at Olduvai Gorge contain identifiable rodent remains in abundance.

We focus on metabolites traceable to two major categories: (1) endogenous byproducts of metabolism that reflect internal metabolic processes, some of which present themselves in response to environmental factors, such as an infection or poor nutrition; and (2) exogenous metabolites derived from organisms they consume. Because the tropics and subtropics have been the focus of intense drug and cosmetic discovery efforts for factors derived from medicinal plants18,19, there is an increasing amount of plant metabolomics data available to draw on20. Recovering metabolites from these two categories enables us to address directly the questions of what metabolites are found in hard tissues that enable an interpretation of physiological signals and responses to the environment, and what metabolites derive from consumption patterns that fuelled their metabolisms.

We also performed metabolomics on fresh laboratory mouse bone samples—secondarily acquired from research investigations having no relationship to this study—and the diets of these mice to validate our retrieval and detection of metabolites. We also undertook an endogeneity assessment by performing metabolomics on palaeosols of the sites from which the fossils were excavated, because metabolites are present in soil that relate mainly to plant and microbial biological functions; palaeosol metabolites detected in common with fossil samples must putatively be removed from consideration of interpretations of exogenous metabolomes. Our assessment also includes an evaluation of the potential influence of the digestive system on the metabolomes obtained from the bones of rodents derived from regurgitated owl pellets.

Sample integrity

Prior to metabolite recovery, a portion of each fossil sample was subjected to histology and imaged by polarized light to illustrate putative collagen birefringence (Fig. 1). We present images for the five Olduvia Gorge rodent samples: specimen M-D (Extended Data Fig. 1), specimen Sm (Extended Data Fig. 2), specimen Gg (Extended Data Fig. 3), specimen Gi (Extended Data Fig. 4) and specimen Xi (Extended Data Fig. 5). Polarized light images are also presented for the Chiwondo Beds specimen HCRP-RC-11 883 (Supplementary Fig. 1) and the Makapansgat specimen TF12 Lime Dumps (Supplementary Fig. 2). Backscattered electron imaging in the scanning electron microscope (BSE-SEM) was also performed on the five Olduvai Gorge rodent specimens (Extended Data Figs. 6–10) and those of the Chiwondo Beds and Makapansgat specimens to confirm recognizable hard tissue microstructure (Supplementary Figs. 3 and 4).

Fig. 1: Representative images of fossil specimens.

Images are of fossils from Olduvai Gorge (left; Xerus cf. inauris bone; field width (FW) = 4.434 mm), Chiwondo Beds (middle; Elephas recki shungurensis dentine; FW = 6.5 mm) and Makapansgat (right; Bovidae bone; FW = 80 µm). Brightness in the images is taken to be birefringence of collagen (the brown hue on the right is diagenetic alteration).

To establish protein biomolecular integrity, proteomics analyses were performed on a representative of each Olduvai Gorge fossil rodent genus sampled (M-D, Sm, Gg and Xi), which revealed peptides related to collagen matrices (Supplementary Figs. 5 and 6), and other potentially exogenous peptides, including some from a parasitic infection (Supplementary Fig. 7).

Histology sections from the extant, fossil and palaeosol samples were also imaged by BSE-SEM for energy dispersive spectroscopy (EDS) evaluation to determine whether mineral concentrations would reflect those of bone and palaeosol matrices. Normalized mass concentrations of calcium (Ca) and phosphorous (P) from the extant, fossil and palaeosol samples are given in Supplementary Table 2. Ca/P ratios of extant bone ranged from 2.68–2.8, whereas those of fossil bone ranged from 3.02–3.9. This increase in ratios reflects two processes, as evident from the mean changes in Ca and P concentrations of the five Olduvai Gorge rodents compared with modern mouse bone (Supplementary Table 2). Phosphate is depleted by 13% in the fossils, which may be owing to hydrolysis of organo-phosphate and possibly a reduction in pH from microbial activity21. The fossils also exhibit a 10% increase in Ca from the soil. No palaeosol had an EDS X-ray peak for P, and in all cases Ca, carbon (C) and oxygen (O) dominated their compositions (Supplementary Table 2), which corresponds to the calcium carbonate (CaCO3)-based palaeosol matrices of each fossil site.

Metabolite recovery from study samples

In the first of 2 experiments relating diet to the laboratory mouse bone metabolome, from the long bone shafts of 10 genetically heterogeneous laboratory mice (UM-HET3) (Supplementary Tables 3.1a–c), we detected 25,676 features, which, after removing duplicates and compounds with no names from the pooled sample, was reduced to 2,193 metabolites for analysis. We detected 3,132 features from their diet, which was reduced to 577 metabolites for analysis. Of metabolites in the diet, 281 (48.7%) are shared with the total mouse metabolome. In the second experiment, using long bone shafts from six highly inbred mice (C57BL-6J) (Supplementary Tables 3.1c–f), we evaluated the exogenous metabolome against the ingredients of their diet (5053–PicoLab Rodent Diet 20), from which five ingredients could be related to specific metabolites (Supplementary Table 3f). There is no rationale for producing ecological profiling from exogenous metabolites derived from the artificial laboratory mouse diets. Yet, these results highlight the coupling of the metabolic profile to consumption patterns during bone formation.

From the bone fragments of 5 Olduvai Gorge Bed I rodents and 1 Bed II pig, 19,918 features were cumulatively rendered from specimens ranging in mass from 11–473 mg, which following the removal of duplicates and compounds with no names, was reduced to 2,507 metabolites for analysis (Supplementary Tables 4a–9a). The five Olduvai Gorge rodents represent four species that are still present in the vicinity of the site today. Bones from each species-matched extant representative were also analysed, rendering a cumulative 24,862 features detected and 2,600 metabolite IDs (Supplementary Tables 4a–8a).

Site metabolite comparisons derived from data in Supplementary Tables 4–11 are summarized in Supplementary Table 12. Endogenous metabolites shared between the four Olduvai Gorge fossils and extant species samples ranged from 50% to 100%, whereas those of exogenous metabolites shared ranged from 14.5% to 52.2%. From the dentine and cementum of the elephant sample from the Chiwondo Beds, 1,281 features were detected, 586 of which yielded metabolite IDs; and from putative bovid bone from Makapansgat, 1,489 features were detected, 694 of which yielded metabolite IDs. Endogenous and exogenous metabolites were shared with extant representatives at 66.25% and 56.6%, respectively, by the Chiwondo Beds samples, and at 53.3% and 37.8%, respectively, by the Makapansgat samples.

The metabolomes acquired from 2 dryland and 4 wetland palaeosol carbonates from Olduvai Gorge Bed I returned 39,210 features, which after removing duplicates yielded 897 metabolite IDs for comparing to the 5 rodent specimens (Supplementary Tables 4d–8d and 13). The Chiwondo Beds HCRP-RC-11 palaeosol yielded 6,580 features, resulting in 233 metabolite IDs for comparison (Supplementary Table 10d). The Makapansgat TF12 palaeosol returned 2,682 features, resulting in 1,293 metabolite IDs (Supplementary Table 11d). Supplementary Table 12 summarizes the comparisons between percentages of metabolites occurring in palaeosols against the fossils collected from these palaeosols. Olduvai Gorge rodent fossils shared 24.9–53.2% of metabolites with the palaeosols. Fossils recovered from the Chiwondo Beds and Makapansgat shared 16.0% and 61.3% of their metabolites with palaeosols obtained from their respective sites.

None of the representative extant species’ bone samples were exposed to a modern soil or palaeosol. Thus, we assessed the per cent sharing of their metabolomes with the palaeosols in anticipation of null results (Supplementary Table 12). The extant Olduvai Gorge rodents shared a range of 25.7–50.5% metabolites with the palaeosols. Extant bones of taxa similar to those of the Chiwondo Beds and Makapansgat shared 14.3% and 43.8% of their metabolites with the palaeosols, respectively.

Variance in the metabolite data

Multivariate and univariate analyses consistently reveal that the major sources of variance in metabolomic profiles across samples and sites are driven by a combination of endogenous degradation products, environmental influences and metabolic residues. Principal components analysis (PCA) clearly distinguishes fossil, soil and modern samples into distinct clusters (Supplementary Figs. 8 and 9), with substantial contributions to their separation from key metabolites such as 2,4-diaminopteridine, kojic acid and hypoxanthine. Hierarchical clustering heat maps of the 100 most variable metabolites show structured groupings, further confirming category-specific biochemical signatures (Supplementary Fig. 10). Statistical tests (ANOVA and Kruskal–Wallis; Supplementary Tables 14b and 14c, respectively) identify many metabolites with significantly different abundances across categories (for example, methyl acetoacetate and myristohydroxamic acid), many of which also rank highly in the supervised partial least squares discriminant analysis (PLS-DA) model. PLS-DA of the 100 most variable metabolites reveals excellent class separation, confirming results of PCA (Fig. 2). Variable importance in projection (VIP) scores highlight a subset of metabolites that overlap with ANOVA hits that strongly drive group differences. Examples include 2,4-diaminopteridine (which also contributes in the PCA; Supplementary Table 14b) and kojic acid (significant in both ANOVA and Kruskal–Wallis; Supplementary Tables 14b and 14c, respectively). Metabolites such as kojic acid and ethyl 2,4,6-trimethoxycinnamate are key drivers of group discrimination (Supplementary Table 14d). Together, the consistent convergence of PCA, clustering, statistical testing and supervised modelling demonstrates that metabolic differences across fossil, soil and modern bone samples reflect distinct biochemical pathways shaped by degradation, environmental incorporation and species physiology.

Fig. 2: PLS-DA of the top 100 most variable metabolites.

PLS-DA dimensionality reduction discriminates class labels along partial least squares (PLS) components.

Owl digestion and metabolite recovery

We considered that enzymes of the owl digestion system may affect the metabolomes of the bones that they consume. We evaluated this possibility in two ways. First, we compared the metabolite recovery between fossils and extant bones of the same species that were obtained by live capture (Saccostomus cf. mearnsi and Xerus sp.) versus extant bones of the same species that were obtained from owl pellets (Gerbilliscus sp. and Mus sp. or Dendromus sp.) (Table 1). The highest percentage of shared metabolites between a fossil and its extant counterpart obtained by live capture was 37.9% (Saccostomus cf. mearnsi) (Supplementary Table 5a). By contrast, the highest percentage of shared metabolites between a fossil and its extant counterpart obtained from owl pellets was 20.4% (Gerbilliscus gentryi) (Supplementary Table 6a).

Second, we experimented on the efficacy of metabolite recovery from the long bones of four C57BL-6J mice, in which the bones from two mice were treated with an enzyme detergent before metabolite extraction. The total number of metabolites with IDs recovered from the untreated samples was 70, whereas only 9 metabolites with IDs were recovered from the enzyme-treated samples (Supplementary Table 3g).

Endogenous mammalian metabolites

The extent to which endogenous metabolites are shared between the fossil and extant species samples is reported in Supplementary Table 12; it represents the proportions of shared canonical pathways, biological functions and gene networks presented in Supplementary Tables 4b–11b identified in ingenuity pathway analysis (IPA). These depict a variety of normal mammalian endogenous functions and disease states that describe the biology of each animal. The most common IPA-defined biological functions relate to amino acid metabolism, cell death and survival, carbohydrate metabolism, cellular growth and proliferation, organismal development, molecular transport, energy production, lipid metabolism, vitamin and mineral metabolism, the cell cycle, and developmental disorders, some of which are shared among the fossil specimens.

Exogenous non-mammalian metabolites

Although IPA maps IDs to endogenous and exogenous categories, it only provides pathway, biological function and network analyses for endogenous metabolites. There is currently no database where exogenous metabolites may be queried for ecological inferences. We present initial entries for such a database in Supplementary Tables 4–11.

The reconstructed environmental conditions derived from each fossil according to palaeontological site are summarized in Table 2. Acknowledging variability and some outliers in Supplementary Tables 4k–8k, 10k and 11k, in the main Early Pleistocene conditions at eastern African Olduvai Gorge Bed I and II sites and at the southern African Makapansgat site were wetter, minimum temperatures were higher, and the landscape contained more forest shade than in the present day, consistent with a mixed seasonally dry and wet tropical biome. The reconstructed conditions of the central African Chiwondo Beds site indicate a wetter environment consistent with a mixed seasonally dry and wet tropical biome.

Discussion

Metabolite preservation

Provided that the depositional environment is not diagenetically extreme, fossils more than 150,000 years old have been shown to harbour endogenous and exogenous amino acids and lipids22. Haem-containing compounds detected from dinosaur bone23 and eggshell24 have been found to enhance the resistance to protein degradation by haem-derived iron-catalysed non-enzymatic cross-linking of structural molecules25. Fourier transform infrared spectroscopy, transmission electron microscopy and time-of-flight secondary ion mass spectrometry (ToF-SIMS) of collagen in dinosaur bone is conclusive (in ref. 26, for example), as its preservation potential is fostered by resistance to hydrolysis27. Peptide sequence analyses and immunological data similarly demonstrate that a variety of bone matrix and blood vessel proteins survive in 80 million year-old dinosaur samples. Raman microscopy of fossils from the Phanerozoic eon has revealed that oxidizing diagenetic environments permit specific protein cross-linking, rendering intact endogenous peptide bonds that persist in fossil mineralized tissues28.

In all studies of metabolite stability, the matrix is either aqueous at room temperature, cooled, frozen or dried and exposed to air. These conditions differ from those used in the recovery of metabolites from hard tissues, particularly of the fossils we present here, which therefore requires an inferential explanation.

Metabolites located in the mineralized extracellular matrix may derive from two sources. The biological processes of overlying osteoblasts and incorporating osteocytes at forming and mineralizing fronts is one source of metabolites released into the extracellular space. A second, more abundant source of metabolites must be driven to the extracellular space from the vasculature during bone formation, which at forming and mineralizing fronts contains a rich capillary network29. By osmotic diffusion and transport mechanisms, capillaries provide the cellular milieu with metabolic requirements and metabolites30, thus metabolite infusion into this space appears to be derived from the free circulation, where the osteoblast cell-surface layer does not provide a barrier function. Tight junctions in extensive, belt-like cell-circumscribing arrangements (such as zonula occludens) similar to those that occur in epithelial cell layers do not form in osteoblasts, but whether they form a complete barrier is moot. Cultured primary osteoblasts form a monolayer with tight junctions, and the formation of a barrier has been suggested31. However, an investigation of osteoblasts on natural bone substrates found that tight junction contact density is relatively low32. Another cell culture experiment concluded that tight junctions were situated at vesicular trafficking points, and that when cells were otherwise confluent in a monolayer, the tight junctions were distributed only in a punctate manner that would not render a barrier function33.

Bone is formed of a carbonate-substituted hydroxyapatite mineral intermingling with the major, large and fibrillar structural protein collagen, which are known to be preserved intact in fossils of Plio-Pleistocene age as examined here34. Polarized light microscopy of the fossil specimens investigated in this study demonstrated positive birefringence typical of preserved bone microstructure in these Plio-Pleistocene fossils. We theorize that without evidence of permineralization effects, such polarized light signals derive from reasonably intact collagen because apatitic mineral has a very low birefringence, which is overwhelmed by that of the collagen. To confirm this result, we performed proteomics on the precipitate from the metabolomics preparations of four rodent fossils from Olduvai Gorge (M-D, Sm, Gg and Xi). Collagen peptide and related IDs that describe collagen matrices were detected from M-D and Sm (Supplementary Table 15 and Supplementary Figs. 5 and 6).

If polarized light microscopy of fossil specimens indicated brightness attributable to collagen, then we hypothesize that the preservation of other biomolecules as reported here is expected. Although this hypothesis remains unconfirmed, and although collagen peptides were detected in only two of four specimens examined, the presence of metabolites in the specimens is irrefutable. From first principles, we detail how we believe these metabolites are incorporated into protective nanoscopic niches of bone (Methods). We regard the nanocompartments as having metabolite preservation potential if two conditions are met: (1) circulating metabolites can enter the spatially constrained niche occupied by interstitial water; and (2) these compartments are not readily connected to an open system in bone subject to diagenesis (that is, they are not exposed to air or mobile water) and thus are shielded from diagenesis (Fig. 3).

Fig. 3: Bone ultrastructure and metabolite niches.

Within the hierarchical porosities of mineralized tissues, the niche that is suitable for palaeometabolome preservation is both permeable for transudate and shielded from diagenesis.

There are several non-exclusive pathways for the degradation of organic compounds in bone, but we suggest that if a bone’s protective mineral has maintained the fibrous collagenous extracellular matrix, then associated nanocompartments and metabolites are also likely to remain intact. This is consistent with biomolecular preservation and staining typical of oxidative diagenetic environments that render brown hues in histology28, as was often observed among the fossils in this study, in addition to some collagen that was presumed to be transmitting birefringent white light.

Although the conditions are different, metabolite preservation from plant resin35 and tobacco36 residues in archaeological contexts demonstrates a specific resistance to their degradation. In our study a t-test finds no significant difference between the average molecular mass in the fossil specimens and their extant skeletonized counterparts (Fig. 4 and Supplementary Table 16a). Yet, the average molecular mass in the laboratory mouse metabolomes is about 75 Da higher. It is likely that metabolites from bone cells included in the metabolite extraction are contributing to this larger molecular mass fraction; for example, the average molecular mass of 113 metabolites detected from a yeast cell37 is 269 Da, which is within the narrow range given for the laboratory mouse metabolomes (Fig. 4).

Fig. 4: Average metabolite molecular masses of extant and fossil hard tissues.

Molecular masses of metabolites extracted from fossils from five Olduvai Gorge rodents, a Chiwondo Beds elephant and a Makapansgat bovid. Fossil specimens are plotted together with their corresponding extant representatives and two laboratory mouse strains. Sample grand means ((\bar{x})) are given for each category.

The rat bone has a cellular density of approximately 60,000 mm−3 (ref. 38), and given the negative body mass dependence of the cell density, we expect the value for mouse to be around 75,000 mm−3. The 20 mg bone samples from the extant laboratory mouse are estimated to occupy approximately 1 mm3. Thus, metabolites from about 75,000 cells will be represented in each bone’s extract. Metabolomics39 analyses of powdered extant mouse femurs demonstrate that cells contribute a metabolite solute to the extraction solvent. Further research is required to more fully understand the extent to which degradation occurs in fossil and extant skeletonized bone devoid of cells.

Insights on the endogenous metabolome

Aside from the summary of normal mammalian endogenous functions and disease states described above, IPA analyses of the molecular physiological machinery responsible for the metabolome of each animal are detailed in Supplementary Tables 4b–11b, with a list of canonical pathways, functions, diseases and gene networks. Associations with genes related to oestrogen biosynthesis and signalling suggest that both Olduvai Gorge gerbils (specimens Gg and Gi; Supplementary Tables 6b and 7b, respectively) as well as the ground squirrel (specimen Xi; Supplementary Table 8b) were female, as was the Makapansgat bovid (sample TF12; Supplementary Table 11b). On the basis of oestrone detected in specimens Gg and Xi, we are less confident, relying only on the higher concentration of this form of oestrogen in females than in males. However, our confidence is higher regarding specimens Gi and TF12, in which placental oestriol and luteinizing hormone are indicated. Notes regarding sex hormones are shown adjacent to the gene network in each supplementary table referenced above.

The results from the Olduvai Gorge ground squirrel, the Chiwondo Beds elephant and the Makapansgat bovid also present an opportunity to connect the endogenous and exogenous metabolomes. Associations with genes related to proinflammatory cytokines in data from these individuals may be the result of the Trypanosoma brucei infection identified in each of their exogenous metabolomes (Supplementary Tables 8k, 10k and 11k), which is also revealed in proteomics analysis of the ground squirrel specimen Xi (Supplementary Fig. 7).

Endogeneity assessment

We considered that taphonomic research demonstrating the owl digestion of rodent bone might contribute to a metabolite signal. Rodents excavated from FLK N1 layer M3 and M2 were deposited as pellets by Verreaux’s eagle-owl (Bubo lacteus), whereas those from FLK N1 layer M1 were deposited by the spotted eagle-owl (Bubo africanus)5 (Table 1). Unlike other birds, owls lack a ‘crop’ to store and digest food slowly. Instead, non-digestible contents (such as bone and fur) are formed as a pellet in the gizzard and regurgitated after several hours, and thus not excreted through the digestive tract (owls cannot eat again until they eject this pellet). Owls have a higher pH in their ‘true’ stomach (proventriculus) than other birds, but experiments have confirmed that its acids may mildly corrode bone surfaces. However, experimental taphonomic research indicates no more surface damage than might occur during normal bone weathering40,41. A two-proportion z-test of the percentage of shared metabolites between the fossil of Saccostomus cf. mearnsi and its extant counterpart obtained by live capture versus the fossil of Gerbilliscus gentryi and its extant counterpart obtained in an owl pellet reveals a highly significant loss of shared metabolites as a result of it being subject to the owl’s digestion (P < 0.01; Supplementary Table 16b). The possibility of metabolites intruding into the bone matrix attributable to digestion is thus insignificant. This assessment was mirrored by our enzymatic laboratory test of fresh mouse bone in which the metabolite recovery was diminished over that of bones that were not subject to enzymatic digestion.

The potential incorporation of metabolites excreted by the gut microbiome may also be considered. Given our supposition that metabolites are incorporated into mineralizing surfaces, we expect only metabolite loss, and no metabolite gain. Further, the avian proventriculus has a reduced bacterial diversity compared with the gut42, the former being mainly responsible for acid and enzymatic break down of food, which the bones do not experience. Among the Olduvai Gorge rodent fossils, only the bone of Saccostomus cf. mearnsi, which was obtained from an owl pellet, contained a bacterial xenobiotic metabolite (6-hydroxyhexanoic acid; Supplementary Table 5k). However, species obtained by live capture (Saccostomus cf. mearnsi, Gerbilliscus sp. and Xerus sp. (Supplementary Tables 5a, 7a and 8a, respectively)) also contain this metabolite. Thus, it is more likely that the presence of xenobiotic metabolites derives from their incorporation into the mineralizing tissues of the animals independent of their passage through owls.

An endogeneity assessment of palaeosols’ metabolite contribution to the fossils was also considered. In testing the assumption of contamination by palaeosols from the three palaeontological sites, we discovered them to be independent repositories of metabolites detected from the fossils (Supplementary Table 12). The five rodent fossils from Olduvai Gorge share an average of 41.1% of total metabolites with the palaeosol. Yet, their extant counterparts, which had no contact with soil, share 40.4% of total metabolites. At the Chiwondo Beds, the fossil/extant proportions are 16.0/14.3. At Makapansgat, they are 61.3/43.8, respectively, the latter offset perhaps due to the sampled extant African bovid, while having a South African distribution, nevertheless originating in Malawi. The palaeosols at each palaeontological site are calcium carbonate-based (Supplementary Table 2). During natural carcass degradation, decomposition fluids containing metabolites (analogous to those that could have been sequestered in bone crystallite niches antemortem) seep into the soil. Loamy soils feature a broad size spectrum of porosities some of which can sequester and shield organic compounds originating from numberless decomposition events over time. We suspect that the crystalline matrices of these soils harbour metabolites from the environment, which are taken up and protected during their mineralization: it is not the soil compounds that contaminate the fossils, but rather decomposition compounds that contaminate the soils.

Insights on the exogenous metabolome

The various taphonomic, palynological and geological approaches taken to reconstruct the environment at Olduvai Gorge present a reasonably vivid macro-scale rendering of the environment in the Early Pleistocene3,4,5,43. This depiction ends where interpretations of the palaeometabolome begin, by detailing the exogenous metabolites derived from fossil bones, and from these the inferred attributes of soil, atmosphere, sun exposure, potential species names of organisms and the habitat and food preferences of the animals that lived there. These metabolites and their sources are listed in Supplementary Tables 4k–8k, 10k and 11k, which we summarize in Table 2 and generally remark upon here.

The exogenous metabolites recovered from the Olduvai Gorge fossils are mainly those of soil bacteria Escherichia and Streptomyces. Both are widely distributed in soils of the world.