Introduction

A relationship between climate and the mean body size of endotherms was first proposed by Bergmann1 on the observation that mean body size seems to correspond with latitude2. Today, dwarfing detected in numerous modern endotherms has regularly been attributed to anthropogenic pressures, including climate change3,4. The fossil record allows to test the impact of climate on body size from a deep-time perspective. The Paleogene period is of particular interest because it not only witnessed the abrupt appearance, radiation and global dispersal of most modern mammal orders but it also spanned the transition from greenhouse to icehouse Earth5.

Several arguments based on the fossil record have been advanced in support of a negative correlation between endotherm body size and and environmental temperature. For example, ungulates during the much hotter Eocene were on average one magnitude less massive than their extant descendants6,7. On shorter timescales, abrupt warming events that punctuated the early Paleogene are associated with transient dwarfing of mammals8,9,10. Here we analyze the middle Eocene site of Geiseltal, whose record is considered to span ca. 5 Myrs of gradual global cooling, and generate one of the most extensive mammalian Paleogene body size datasets outside North America.

Modern-day Germany hosts a number of internationally renowned Eocene fossil vertebrate lagerstätten, such as the Eckfelder Maar and the Messel pit (Fig. 1). However many of those sites are stratigraphically restricted and commonly span shorter time intervals of less than 1 million years11,12. By contrast, biostratigraphy of the coal succession at the former brown coal surface mine in Geiseltal (51°18′49″N, 11°52′08″E; Fig. 1) indicates a broadly Middle Eocene age and a temporal range of several million years of around 47.5 to 42.5 Ma for the fossiliferous sections of Geiseltal13,14. This makes Geiseltal a suitable target for the study of early Cenozoic environmental and ecological dynamics as this age corresponds to a period of gradual global cooling5. Unfortunately, despite a large vertebrate fossil volume15 and comparable quality of preservation16,17, the east German site of Geiseltal has not yet received as much scientific attention as its west German counterparts like Messel (Fig. 1).

Figure 1
figure 1

Global paleogeographic configuration (a) and Northern Europe (b) during the middle Eocene. Locations of Geiseltal, Grube Messel and Eckfeld Maar are indicated by a golden star, green circle and blue circle, respectively. The global reconstruction was created using the GPlates software (http://www.gplates.org) and the regional reconstruction was adapted from ref. 76.

The Geiseltal fossil lagerstätte is composed of several large coal seams that are each subdivided into between one and four fossiliferous stratigraphic horizons18. Coal seams are intercalated with silty and sandy grey clastic sediments in which vertebrate fossil material is largely absent. Fossil abundance is highest in the lowest section of Geiseltal and gradually declines towards the top of the profile18, restricting the analysis presented here to the lower coal (LC), lower middle coal (LMC), upper middle coal (UMC) and upper coal (UC). The stratigraphic interval from LC to UMC is biostratigraphically interpreted to span approximately 5 Myrs13,14, the ages of horizons within individual coal seams as well as the stratigraphic position of UC remain unconstrained. Fossil vertebrates in the Geiseltal strata hence provide a number of distinct geologic windows into the Eocene rather than a single continuous record.

We constructed a comprehensive analysis of changes in ungulate body mass and taxonomy across the Geiseltal profile for the tapiromorph and hippomorph taxa Lophiodon (Fig. 2a) and Propalaeotherium (Fig. 2b) - the two most abundant vertebrates at this locality. To investigate a potential coupling between dynamical change in the climate state and body size in Geiseltal, we also extracted in-situ oxygen and carbon isotopic measurements from the teeth of 26 individual of these two taxa.

Figure 2
figure 2

Examples of well-preserved Lophiodon remensis [GMH VII-1-1952] (a) and Propalaeotherium isselanum [GMH Ce IV-7011-1933] (b) skeletons from Geiseltal. Although fossil remains are common, complete and articulated mammal body fossils like those pictured are still rare in Geiseltal. The reconstruction of body mass therefore relies on a tooth crown proxy instead (Methods). Photographs taken by O.W.

Results

Taxonomic revision

Species-level distinctions in Geiseltal Propalaeotherium and Lophiodon are largely based on relative size differences in dental, cranial and autopodial characteristics19,20,21,22,23,24. We find that proposed diagnostic traits are non-discrete or otherwise impossible to replicate due to intraspecific variation and insufficiently well-preserved fossil material (see Supplementary Information). This indicates an artificial inflation of local mammal diversity and, as we are unable to distinguish more than a single consistent morphotype of each taxa, we here recognize only a single hippomorph and tapiromorph species at Geiseltal which we tentatively refer to Propalaeotherium isselanum and Lophiodon remensis, respectively (see Supplementary Information).

Body mass analysis

Results for Propalaeotherium and Lophiodon are given in Fig. 3 (see Supplementary Table 1). Measurements were normalized for the lower first molar (m1) because it has the closest correlation between crown area and body mass10,25 and results in the lowest number of total corrected data points. Estimated Propalaeotherium mass varies between 10.4 and 65.3 kg (Fig. 3a), excluding one outlier of >80 kg, and between 68.0 and 404.5 kg for Lophiodon, in broad agreement with estimates inferred from cranial dimensions26,27 and biomechanics28. The box-and-whiskers plot in Fig. 3 shows the mass distribution when our results are binned into coal seam datasets.

Figure 3
figure 3

Estimated body mass of Propalaeotherium (a) and Lophiodon (b) in Geiseltal. Left column, all 257 data points, color-coded according to their position within the jaw. Right column, box-and-whiskers plot of body mass when data points are pooled together in their respective coal seams (data from Code H was included in the UC bin).

Propalaeotherium mass ranges from 19.4 to 52.8 kg (median: 39.2 kg) in LC, 15.7 to 44.0 kg (median: 25.8 kg) in the LMC and 10.4 to 47.2 kg (median: 26.0 kg) in UMC, after which the record terminates. While median body masses in the two middle coal seams are indistinguishable (p = 0.74 Wilcoxon-Mann-Whitney test; p = 0.27 two-tailed t-test), lower coal Propalaeotherium are significantly more massive (p = <0.01 Wilcoxon-Mann-Whitney test; p = <0.01 two-tailed t-test). Despite lower absolute abundance, the vertebrate record of Lophiodon is longer than that of Propalaeotherium, extending into UC (Fig. 3b). Interestingly, Lophiodon sample size in the LC (n = 16) is comparable to the relatively fossil-poor UC seam (n = 11). This is in contrast to the LC-centered general vertebrate fossil distribution in Geiseltal18 and may reflect a low population size of tapiromorphs in early Geiseltal. Body mass of Lophiodon is lowest in LC, occupying a range from 84.3 to 153.6 kg (median: 123.9 kg), compared to ranges between 80.6 and 300.0 kg (median: 152.4 kg) in the LMC, 68.0 to 278.0 kg (median: 143.0 kg) in the UMC and 82.5 to 307.2 kg (median: 223.0 kg) in the UC (including the inter-coal horizon H; Fig. 3). Statistically significant steps occur from LC to LMC (p = <0.01 Wilcoxon-Mann-Whitney test; p = <0.01 two-tailed t-test) and from UMC to UC (p = <0.01 Wilcoxon-Mann-Whitney test; p = <0.01 two-tailed t-test).

Stable isotopic (O, C) analysis

Oxygen and carbon isotopic composition of carbonate is reported as δ18Oc and δ13Cc (where δ18O = ((18O/16O)sample/(18O/16O)standard − 1) · 1.000‰ and δ13C = ((13C/12C)sample/(13C/12C)standard − 1) · 1.000‰). Carbonate content in most samples falls within or slightly below the range of modern mammals29 and reveal no discernible correlation to the isotopic signature (Supplementary Fig. S3), indicating that exogenic incorporation of carbonate is negligible. Isotopic data are presented in Supplementary Table 2.

Analysis yielded a narrow δ18Oc range from 22.1 to 24.8‰ (mean: 23.4‰; Fig. 4a). Because the third molar is last to form in the ungulate tooth eruption sequence30 it has been traditionally considered to be a more reliable proxy for paleoclimate reconstruction due to the reduced potential influence of weaning. However, intervals in Geiseltal where measurements from both the M3 and other tooth positions are available show very good agreement in general, supporting our contention that tooth position does not impose an important measurement bias (see Methods). Using the enamel carbonate-precipitation transfer function31, we compute a mean isotopic composition of precipitation (δ18Op) of between −10.8 and −7.6‰ (mean: −9.2‰). Carbon isotopic ratios (Supplementary Table 2, Fig. 4b) scatter between −13.8 and −9.7‰, with an overall mean of 12.0‰, well within the range expected for C3 plant-feeders32. When measurements are binned into their respective coal seams, neither isotopic system reveals a discernible change in composition with stratigraphic age (p = >0.05). Analytical standard deviation (1σ) is generally below 0.2‰ for the majority of samples. Within paired enamel-dentine samples (Supplementary Table 2), we find small mean offsets of only −0.4‰ and −1.5‰ for δ18O and δ13C, respectively. Considering the high susceptibility of dentine for isotopic exchange, this signals at most only very limited post-depositional diagenetic alteration33. Nevertheless, we find an unexpected pattern when δ13C paired samples are separated according to stratigraphic age. Enamel-dentine samples from the LC exhibit very minor positive offset of +1.1‰, while samples from the two middle coals have a slightly stronger negative mean offset of −2.7‰ (Supplementary Fig. S4).

Figure 4
figure 4

Oxygen (a) and carbon (b) isotopic results of the carbonate component in the enamel of ungulates in Geiseltal against relative stratigraphic age. Brown and green arrows represent the carbon isotopic trends for Propalaeotherium and Lophiodon, respectively.

Discussion

Middle Eocene local climate

When inferred δ18Op values are substituted into an accepted Eocene transfer function34, mean annual surface temperature across all samples in Geiseltal is estimated to be 14.9 °C (±1.3 °C; 1σ). This is well below previous estimates of Central European temperature during the hot Eocene35,36,37,38,39 (Fig. 5a). Significant non-climatic overprint of tooth geochemistry is unlikely because most potential isotopic reservoirs in Geiseltal have an isotopic signature that differs markedly from our results. For example, our methodology implicitly assumes that ingested water is sourced from local precipitation. For Eocene ungulates in Geiseltal, who had a predominantly foliage-based diet and may not have been obligate drinkers, a considerable fraction of water intake might be derived from leaf water, thereby introducing a systematic bias into our data. However, because lighter oxygen isotopes are preferentially lost during evapotranspiration40 leaf water is characterized by a high 18O/16O ratio. Similarly, substantial post-depositional alteration is inconsistent with the close overlap between enamel and dentine and fails to account for our results because plausible alteration pathways, such as influx of dissolved carbonate ions considered to be crucial for fossil formation in Geiseltal41, involve δ18O signatures well above our results42. We therefore consider our results to approximate in-vivo conditions and attribute the anomalous oxygen isotopic composition to the ‘amount effect’ of precipitation.

Figure 5
figure 5

Evolution of European surface temperature and precipitation during the Eocene. When taken at face value, enamel δ18O-calculated surface temperature (a) yields anomalous values when compared to previously published proxy records (see Supplementary Information for reference list), which we interpret as an amount effect. Our δ13C-based MAP estimate (b), by contrast, is in good agreement with existing reconstructions and is comparable to modern tropical rainforests. Error bars reflect the 1σ interval.

The amount effect describes a decoupling between ambient temperature and isotopic rainfall chemistry due to deep moist convection, droplet re-evaporation and moisture convergence43 and manifests itself in an anti-correlation with precipitation intensity (i.e. δ18Op decreases as rainfall amount increases). The amount effect is today observed at tropical latitudes in the tracks of regional summer monsoonal systems43. For past greenhouse intervals like the Eocene that were characterized by a expanded tropics and a shallow equator-to-pole temperature gradient44 it would stand to reason that this isotopic insensitivity to temperature may have extended into mid-latitudes45. Quantifying the relative importance of temperature and rainfall amount on isotopic chemistry during the Eocene is difficult, however, we note that our δ18Op results are in good agreement with the composition of modelled Eocene precipitation and those inferred from organic n-alkane record of the nearby sites of Lillebaelt and Possagno46, indicating that the amount effect is the dominant hydrological mechanism in Geiseltal.

Carbon isotopic ratios provide a second independent way to investigate middle Eocene environment in Geiseltal. When accounted for isotopic enrichment between enamel and diet (ϵ = 14.6‰; ref. 32) and corrected for the different isotopic signature of middle Eocene CO2 (−6‰ compared to a modern value of −8‰; ref. 47), the isotopic composition inferred from our measurements is entirely within the range expected from C3 vegetation and signals a warm-temperate to subtropical open-canopy forest habitat48. For C3 plants, the fractionation between 13C and 12C is dominated by the biophysical response of plants to the availability of light and water49, such that δ13C increases with aridity. Intensity of past mean annual precipitation (MAP) can hence be constrained if the relationships between isotopic partitioning and its forcers are sufficiently well resolved. Using the quantitative relationship50 (assuming an absolute paleolatitude of 43° and an arbitrary low elevation of 50 m), we reconstruct Geiseltal paleo-rainfall variations of 2068 mm/yr (±319 mm/yr; 1σ) which compares favorably with Eocene general circulation model spread51 and proxies35,39 from Geiseltal coals, which indicate variability in precipitation from approx. 600 to 2500 mm/yr (Fig. 5b). Paleoclimate models agree that precipitation is generally linearly dependent on surface temperature on a global level51 and our isotopic measurements (both oxygen and carbon) would hence suggest no detectable long-term trend in temperature in Geiseltal, in contrast to the pronounced global cooling during the inferred time of coal deposition5. However, relating both variables at a single location is difficult because on a regional scale, precipitation is also sensitive to moisture convergence modulated by regional atmospheric circulation.

Potential body size drivers

To our knowledge, the results presented here represent the most extensive investigation of Paleogene mammal physiology in Europe. Previous works that include body mass-related characteristics in Geiseltal concluded relative changes through time that are either opposite20 or congruent23 to our results - though our analysis is based on sample sizes around ten times larger.

The invariant isotope geochemistry and opposite sign of body mass change through the LC-LMC transition in both genera precludes a straightforward climatic effect analogous to early Paleogene hyperthermals, as a direct climate forcing would elicit a physiological response in the same direction. A broad similarity in body mass behavior (i.e. marked change from LC to LMC and identical mean body masses between LMC and UMC) may nevertheless indicate size-specific responses to a common driver.

Additionally, ecology places as an equally important constraint on mammal body size that is independent of climate. Because body size directly scales with many life history traits (such as reproductive rate or home range size)52 that determine the efficiency of resource acquisition, different dietary niches are likely associated with distinct optimal body size ranges53. Interaction with other organisms can additionally influence the resource use and therefore the available niche space that is open for a given taxon. For example, Smith and Lyons54 note that the body size of African hyraxes dropped by several orders of magnitude following the immigration of competitive herbivores during the Oligocene. Under this interpretation, the substantial increase in mammalian body mass during the Cenozoic reflects an ecological release and progressive filling of empty niche space that favors large-bodied organisms rather than the effect of an abiotic forcing like temperature55.

Sample sizes of Lophiodon are lowest when Propalaeotherium abundance is highest and vice versa, consistent with a changing competitive balance56 and conventional theory predicts that increasing competitive pressure will result in character displacement (CD) in which both competitors will tend towards minimizing their trait similarities57. Nevertheless, unambiguous examples of CD are rare58 and trade-offs in population sizes may also arise from independent dynamics of species ranges instead of CD. It is further unclear to which extent large tapiromorphs and small horses were truly subject to resource competition as we observe an offset in enamel δ13C values, suggestive of a body-size related difference in resource utilization, similar to other extant and fossil mammal communities48,52. Furthermore, a strong competitive pressure would be expected to lead a decrease in trait variance. We, by contrast, document a sharp increase in body size variance for Lophiodon and no pervasive variance change in Propalaeotherium (Fig. 3).

Alternatively, different feeding strategies and resource partitioning between larger, less forage-selective mammals and smaller, more specialized mammals could lead to divergent responses under selective pressures. According to the fast-slow life history concept59, small animals have a higher mortality but more rapid reproduction compared to larger animals. Therefore, a divergent body size evolution could be the result of a maximization of selective advantages as both taxa evolve in opposite directions along the fast-slow life history continuum. For example, a larger body mass could confer a larger migratory range, reduced predation pressure and more effective metabolization on Lophiodon, the largest herbivores of the European Eocene. On the other hand, the dwarfing of Propalaeotherium would result in a faster transfer of resources to reproduction per unit of body mass60. A progressive differentiation into their respective fast and slow lifestyles explains, in contrast to CD, why the tapiromorph body size increase persists after Propalaeotherium becomes rare in Geiseltal.

Implications for the formation and placement of Geiseltal

Krutzsch et al.39 proposed that the Geiseltal coal succession resulted from a cyclical transition between episodes of persistently elevated humidity that promoted deposition of organic matter and a semi-arid climate with diminished vegetation and reduced coal formation. Though vertebrate fossils are overwhelmingly concentrated in coal-bearing sequences, few remains are preserved in the grey, unconsolidated sands between (Code H in Fig. 4), allowing for an independent evaluation. We find that the isotopic composition of enamel preserved in coal-poor strata are not substantially different from samples of other stratigraphic horizons, refuting a strong climatic control on local stratigraphy. Furthermore, because coal-forming peats are normally unconducive towards the preservation of biogenic hard tissues due to the build-up of humic acids, buffering by dissolved bicarbonate from the Muschelkalk limestone group is considered as a necessary condition for the deposition of vertebrate fossils in Geiseltal41. Enamel in the LC is on average depleted in 13C relative to dentine (above the 1:1 line where enamel δ13C is equal to dentine δ13C) while enamel from the LMC and UMC are relatively enriched in δ13C (below the 1:1 line in Supplementary Fig. S4). One possible, but non-restrictive, explanation is that this represents the receding influence of percolating dissolved bicarbonate which has a δ13C signature well above Geiseltal mammals42, and the increasing role of compositionally low-δ13C lignites61, potentially explaining why fossil abundance is highly concentrated at the base of Geiseltal.

Due to an absence of radiometrically datable igneous rocks, the principal means by which the age of Geiseltal is constrained is biostratigraphy. For individual coal-bearing sequences, age is inferred by comparison of its mammal composition with more well dated European localities. A similarity in reptilian62,63, avian64 and mammal faunal assemblages18,20,65,66 has lead to the conclusion that the Geiseltal LC is contemporaneous to the early Lutetian site of Messel, while LMC, UMC and UC have been correlated with the sites of Issel (France), Bouxwiller (France) and Egerkingen (Switzerland), respectively14,18,20. However, numerous traits considered to be diagnostic of Geiseltal Propalaeotherium species that are predominantly used in Eocene mammal biochronology are poorly defined, not reproducible or potentially within intra-specific variability (see Supplementary Information). This indicates that multiple Geiseltal palaeotheriids are subject to taxonomic over-splitting which would strongly distort biostratigraphic analyses. The deposition of lignite beds with the same minimal vertical thickness as in Geiseltal requires just over 1 Myrs, assuming the maximum present-day coal accumulation rate67.

In our view, this highlights the need for a reevaluation of regional biostratigraphic mammal zonation (i.e., the age of Geiseltal relative to European Eocene vertebrate localities) and indicates a temporally more restricted Geiseltal record although still likely at the Myr scale.

Conclusions

Middle Eocene lignites in Geiseltal document a discontinuous succession of floral and faunal assemblages. Oxygen and carbon compositional in-situ measurement of molars support near-in vivo preservation and suggest that both form complementary records of intensified palaeo-precipitation, consistent with an expansion of the (sub)tropics during the Paleogene greenhouse interval. We further find little evidence of long-term change in climate, despite supposedly coeval progressive global-scale cooling. Extensive measurement of m1 crown area of the two most abundant vertebrates (Propalaeotherium and Lophiodon) indicates that both varied significantly in body size through the Geiseltal profile and exhibit an opposite directionality. We explore multiple interpretations and conclude that instead of climatic adaptation, the exploitation of relative selective advantages resulting from different life histories (‘slow’ vs. ‘fast’ living) of both species is the most plausible explanation. As large herbivorous mammals appear in Europe relatively late in the Paleogene, a promising direction of further research is to better constrain when and how these body size-related life histories first evolved compared to larger continents. We were unable to replicate most of the diagnostic traits outlined for Geiseltal hippomorphs and tapiromorphs that otherwise largely underpin the regional biochronology, which suggests that previous alignment of Geiseltal with other localities is subject to arbitrary inflation of species diversity. This highlights the need for a reexamination of current biostratigraphic correlations of both Geiseltal and the Geiseltalian European Land Mammal mega-zone with the European terrestrial mammal chronology.

Methods

Taxonomic revision

Mammal taxonomy is a crucial tool for determining the biostratigraphic provenance of Geiseltal and has importance for our body mass analysis. We performed an assessment of the most recent published systematic diagnoses presented in refs.19,20,23 and tested for re-producibility by examining holotypes and referred specimens of the species Propalaeotherium hassiacum, P. isselanum, P. voigti, Lophiodon cuvieri, L. remensis, L. tapirotherium, Rhinocerolophiodon buchsovillanum and Eurohippus parvulus, kept at the Geiseltal Collection in Halle an der Saale, Germany (see Supplementary Information).

Body mass analysis

Complete fossil skeletons are exceedingly rare compared to the number of incomplete vertebrate remains from Geiseltal. Therefore, to largely overcome this taphonomic effect, rather than measuring anatomy directly, a variable that approximates body size is used instead. The proxy analyzed in this study exploits an ungulate-specific observational relationship between the crown area of the first molar (M1; defined as the buccal-lingual length multiplied by distal-mesial length; see Supplementary Fig. S1) and mean body mass (Eq. (1))25:

$$\mathrm{ln}({\rm{a}})=1.564\cdot \,\mathrm{ln}({\rm{b}})+3.267$$
(1)

where a is the mean body mass (in g) and b is the tooth crown area (in mm2). This approach has proven to be a reliable tool for body size reconstruction25 and has recently been used to successfully demonstrate climate-driven mammalian dwarfing during the early Eocene10. All measurements were taken with a digital caliper in the Geiseltal Collection in Halle an der Saale (Germany), which hosts the largest collection of Geiseltal fossil material. Because body mass can differ markedly from one species to another20 it is necessary to restrict analysis to fewer taxa that have a large stratigraphic range and are found in sufficient number to produce meaningful results. Therefore our analysis was restricted to 157 specimens of Propalaeotherium, and 100 specimens of Lophiodon where the m1 could be firmly identified. To reduce measurement inaccuracy, every tooth dimension was measured three times (which translates to more than 1500 individual measurements) and the respective arithmetic means were used for calculation. Furthermore, owing to differing tooth morphology between upper and lower first molar, the crown area proxy can produce discrepant results for the same individual. To avoid bias by inhomogeneous sampling of upper and lower first molars, we constructed an intrageneric correction factor for Propalaeotherium and Lophiodon using a least-squares tooth regression of fossil specimen from individuals that preserve both their maxilla and mandible. We find that tooth dimensions between upper and lower M1 are broadly correlated (see Supplementary Fig. S2) for Propalaeotherium (n = 9; r2 = 0.69) and Lophiodon (n = 9; r2 = 0.66) through the following equations:

$$\mathrm{ln}({{\rm{l}}}_{{\rm{P}}})=0.9144\cdot \,\mathrm{ln}({{\rm{u}}}_{{\rm{P}}})-0.0327$$
(2)
$$\mathrm{ln}({{\rm{l}}}_{{\rm{L}}})=0.7689\cdot {\mathrm{ln}({\rm{u}}}_{{\rm{L}}})+0.8549$$
(3)

where u and l are the respective crown areas (in mm2) of upper and lower M1. Subscripts P and L denote the genera Propalaeotherium and Lophiodon, respectively.

Stable isotopic (O, C) analysis

Vertebrate skeletal hard tissues such as enamel and dentine are principally composed of a calcium-phosphate lattice (Ca5(PO4)3) that can contain significant substitution of carbonate (CO32−). The isotopic composition of these fossilized biominerals is commonly considered to be a reliable proxy for paleoenvironmental reconstruction because of their refractory nature and ability for primordial geochemical preservation at geologic timescales68,69. While carbon isotopic values can be directly interpreted in terms of dietary and habitat preferences58, oxygen stable isotope values are tied to a step-wise reasoning process: because homeothermic animals have a constant body temperature, oxygen isotopic composition in bioapatite and carbonate reflect the composition of ingested water70,71, of which meteoric surface water is the dominant reservoir for obligate drinkers. Fractionation of meteoric water molecules moving through the hydrological cycle, in turn, is climate-dependent72, and can therefore be used for paleoclimate reconstructions. We sampled the isotopic composition of the carbonate component in enamel and dentine from 26 isolated Propalaeotherium and Lophiodon molars from Geiseltal. Bulk powder-samples were extracted using a micromotor drill and subsequently subjected to a pre-treatment process to remove the potentially polluting effect of secondary carbonate (see Supplementary Information). Measurements were made with an isotope ratio mass spectrometer (IRMS) at the University of Tübingen.

We do not anticipate significant intra-jaw variability of δ18O because of the likely accelerated enamel biomineralization of smaller, brachydont equids73. To obtain the isotopic composition of local paleo-surface water, we invoke the calibration between ingested water and the CO32–component of bioapatite presented by Zanazzi et al.31:

$${\delta }^{18}{{\rm{O}}}_{{\rm{p}}}=1.16\cdot {\delta }^{18}{{\rm{O}}}_{{\rm{c}}}-36.4\textperthousand $$
(4)

We prefer this approach because it reduces the propagated error compared to first relating carbonate to phosphate isotopic composition before constraining paleo-precipitation chemistry. We are unaware of a carbonate-meteoric calibration for tapirs and therefore also apply Eq. (4) to our measurements of molars from Lophiodon, implicitly assuming that basal sections of both branches of perissodactyls (tapiromorphs and hippomorphs) had similar isotopic fractionation.

Though there is evidence for limited ice volume in the middle Eocene74,75, we nevertheless apply the −1‰ correction for the ice-free ocean on δ18Op to provide consistency with previous studies.