The mammal fauna that we analyzed for this study originates from the Pondaung Fm. in the Central Basin of Myanmar. During annual fieldwork campaigns of the French-Myanmar expedition, the fossils were collected from different localities that can be attributed to one of three areas—Bahin, Pangan, and Mogaung (Figure 1).

The Middle Eocene Pondaung Fm. is framed by the underlying Early Eocene Tabyin and the overlying Late Eocene Yaw Fm., both of which are marine deposits (Bender and Bannert, 1983). The Pondaung Fm. itself is up to 2 km thick (Bender and Bannert, 1983) with a succession of marine to terrestrial sediments (Stamp, 1922), but only the upper part (Upper Member) with a thickness of around 500 m (Oo et al., 2015) yielded significant amount of fossils representing a terrestrial mammal fauna. The stratigraphic positions of the different fossil localities to one another has not been studied in depth as of today. Nevertheless, the fossiliferous sediments of some localities from the Bahin area (PK1, PK2, PK3, PK4, PK5, and PK8) belong to the same traceable claystone named Ayoedawpon Taung Claystone ranging in thickness from 8 to 20 m. It is underlain directly by another widely traceable sediment, the Ayoedawpon Taung Sandstone. For two other fossil localities from the same region, the stratigraphic relationship is well constrained. The fossil outcrops at PK9 are part of the Nyaungpinle Claystone, which is stratigraphically lower than the Ayoedawpon Taung Claystone. The claystones at the PK12 locality on the other hand overlay directly the sandstones above the Ayoedawpon Taung Claystone (Maung et al., 2005; Suzuki et al., 2006). It has been dated to 40.31 ± 0.65 Ma and 40.22 ± 0.86 Ma using LA-ICP-MS, U–Pb zircons on tuffaceous sandstones from the Bahin area (PK4 and PK8; Zaw et al., 2014). This means that the Pondaung Fm. dates to around 40 Ma, which corresponds to the late Middle Eocene or Bartonian.

Its depositional environment was characterized by fluvial sediments of meandering river channels and is mostly comprised of sandstone with interspersed silt/clay bands. Fossils are mostly associated with lithofacies indicative of small fluvial or crevasse channels, swale fills and point bar deposits. Some evidence of peat layers and coal seams might indicate peat swamp formation of the inter-channel areas (Soe et al., 2002).

In their magnetostratigraphic study on a 319 m thick section of the Upper Member of the Pondaung Fm., Benammi et al. (2002) detected a single normal polarity. Although this homogenous paleomagnetism makes correlation with a specific Bartonian chron difficult, it indicates a high sedimentation rate of the floodplain (>0.3 mm yr.⁻¹; Licht et al., 2014b) as the maximum length of the possible Bartonian polar chron is smaller than 1 Myr (Benammi et al., 2002). This inference can be further substantiated by the results of another study on sedimentation rates in Asia (Métivier et al., 1999). Given this high sedimentation rate and the observed homogeneity of the fauna documented notably by rodents, amynodontids, anthracotheres and primates, we will consider that the Pondaung fossil remains correspond to a contemporaneous mammal fauna assemblage (Ducrocq et al., 2019). We give a detailed evaluation of sympatry of the different taxa in our sample per locality and area in Table SI 1.

The fossil mammal fauna preserved in the Pondaung Fm. is famous for its unique and diverse primate assemblage. Two Sivaladapid adapiformes (Kyitchaungia takaii, and Paukkaungia parva), one Eosimiiforme (Afrasia djijidae), one Eosimiidae (Bahinia pondaungensis), a yet undescribed tarsid and most notably three different Amphipithecidae (Ganlea megacanina, Pondaungia cotteri, and Myanmarpithecus yarshensis) were found there (Marivaux et al., 2008b; Chaimanee et al., 2012; Fleagle, 2013; Jaeger et al., 2020). The phylogenetic position of the amphipithecids has been highly debated in the past century, but these primates are now firmly established as stem anthropoids by cranial, postcranial, and phylogenetic evidence (Jaeger et al., 2020). The high diversity in species underlines the importance of the Pondaung Fm. and Southeast Asia as the biogeographic region of early diversification and origin of this clade. This fact, together with the co-occurrence of three different stem anthropoid species together with eosimiids, which are basal anthropoids, and other primate taxa raises the interest in the paleoenvironment in which they all existed.

The diet of the Pondaung primates has been inferred from both dental morphology as well as dental microwear analysis. Pondaungia was likely predominantly frugivorous and also included some hard items in its diet. Myanmarpithecus subsisted on a mix of fruit and leaves (Kay et al., 2004; Ramdarshan et al., 2010). Although the unique tooth morphology of Ganlea complicates the dietary interpretations, this amphipithecid likely also ingested a mixture of leaves and fruit, but also engaged in seed predation (Beard et al., 2009; Jaeger et al., 2020). For the smaller primates Bahinia and Paukkaungia an insectivorous diet has been suggested (Kay et al., 2004; Ramdarshan et al., 2010). Only few primate postcranial remains have been found from the Pondaung Fm. Both Pondaungia and Ganlea have been described as active arboreal quadrupeds (Marivaux et al., 2003, 2010; Jaeger et al., 2020). For the Sivaladapidae a generalized locomotor profile as an arboreal quadruped has been suggested as well (Kay et al., 2004; Beard et al., 2007), with some evidence for leaping capabilities due to hip and pelvis morphology in K. takaii (Beard et al., 2007; Marivaux et al., 2008a,b). Given the small size of the teeth and their importance and rarity it was not possible to sample any of the Eocene anthropoid primates for stable isotope analysis.

Four large herbivore groups make up the majority of our sample of the Pondaung mammal fauna. The numbers in brackets correspond to the number of specimens sampled. There are three groups of perissodactyls, Brontotheriidae (n = 24), Amynodontidae (n = 16), Rhinocerotoidea indet. (n = 27), and the artiodactyl Anthracotheriidae (n = 56). Of these tooth specimens, 38 were sampled serially (9 Brontotheriidae, 9 Amynodontidae, 8 Rhinocerotoidea indet., and 12 Anthracotheriidae), hence providing a time series across the mineralization time of the fossil enamel allowing the investigation of seasonal variations. Brontotheriidae are described as being obligate browsers due to their brachydont teeth with bunoselenodont morphology and therefore preferring forest to woodland habitats in warm temperate to subtropical environments (Mader, 1998). Some of the brontotherids in our data set have been identified as Bunobrontops a more primitive genus in comparison to two other known genera, Metatelmatherium and Sivatitanops, from the Pondaung Fm. (Holroyd and Ciochon, 2000). Most of the Rhinocerotoidea are probably Amynodontidae, the most abundant family in the Pondaung Fm., but the fragmentary nature of the sampled fossils did not allow for an assignment of this taxonomic label with enough certainty. Previous work on fossil Rhinocerotoidea from Late Eocene sites in Vietnam characterized them as forest dwellers with their brachydont teeth and as being obligate browsers as well (Böhme et al., 2013). These brief descriptions of two of the main herbivorous taxonomic groups at the Pondaung Fm. reveal once more the lack of systematic studies especially on the Perissodactyl fauna. In addition, the tapiromorph Bahinolophus (n = 2), Eomoropidae (n = 1), and Ruminantia (n = 2) are also part of our data set. In four cases two teeth were sampled per individual (Siamotherium PNG-20 M2/M3 and PNG-141 M2/M2, Anthracokeryx tenuis PNG-22/23, Anthracotherium crassum PNG-56/57, and Amynodontidae PNG-92/93). In general, the ungulate fauna in the Pondaung mammal fauna is considered to be representative of the original living community, while a sampling bias is likely in the case of rodents and other small mammals including primates. This is due to the fieldwork method applied by the different teams of researchers, which mainly conduct surface prospecting in the area (Tsubamoto et al., 2005).

The Pondaung Fm. has also yielded some creodont (Egi and Tsubamoto, 2000; de Bonis et al., 2018), fish, reptile (crocodilian, lizard, and turtle; Hutchison et al., 2004; Tsubamoto et al., 2006a; Head et al., 2013) and bird fossils (Tsubamoto et al., 2006b). However, no extensive systematic studies on this material exist as of now and it has therefore not been included in our data set for isotopic analysis.

In the Pondaung Fm., we find a high diversity of anthracothere taxa. This comes as no surprise, as this taxonomic group likely originated in Southeast Asia and experienced a radiation there in the late Middle Eocene. Later on, anthracotheres spread from Asia to Europe, Africa, and North America. Three different anthracothere genera are known from the Pondaung Fm.—Anthracotherium, Anthracokeryx, and Siamotherium. In contrast to the perissodactyls from the Pondaung mammal fauna that were strictly herbivorous according to their tooth morphology, anthracotheres were more opportunistic feeders. They likely had a much more diversified diet that included leaves, fruits, roots, and invertebrates.

Anthracotherium (n = 20) was a medium to large sized anthracothere and the largest of the three genera found at the Pondaung Fm. It has bunodont teeth suitable for crushing tougher food like roots or even scavenge carcasses from time to time. Hence, its cranio-dental morphology indicates a more generalist diet not unlike the one of extant pigs (Ducrocq, 1999). Therefore, we hypothesize that it occupied more open forests than the other two and probably lived closer to the river system, meaning that it was more water dependent than the other anthracothere genera known from the Pondaung Fm. Two species could be identified in our sample, the smaller A. crassum (n = 5) and the larger A. pangan (n = 11).

The second genus, Anthracokeryx (n = 13) with the two species, the larger Anthracokeryx birmanicum (n = 4) and the smaller A. tenuis (n = 4), represents a more typical forest dweller with bunoselenodont teeth. These are typical for a more specialized herbivorous foraging ecology (Ducrocq, 1999). It has a body mass, of about 20–25 kg (Lihoreau and Ducrocq, 2007). This and its cranio-dental morphology are consistent with dwelling in denser forests and having a more specialized folivorous diet.

Siamotherium (n = 4) is the smallest and most primitive of the three genera with an estimated body mass of 7.5 kg for S. pondaungensis. It was a small, slender animal with a short snout showing first signs of selenodonty on its molars (Ducrocq et al., 2001; Soe et al., 2017; Ducrocq et al., 2021). Its subsistence strategy has been described as more omnivorous including occasional scavenging, and it is thought to have lived in open forested areas (Soe et al., 2017). For this study, we could sample four teeth from two individuals.

We sampled the specimens during a museum visit to the National Museum in Nay Pyi Taw in February 2020 and the fieldwork campaign in the following weeks. The specimens were cleaned from dust and sediment using ethanol and the outermost layer of fossil enamel was removed on the sampling site as these would be most susceptible to diagenetic alteration. During sampling 6–12 mg of fossil enamel powder were drilled using a portable diamond coated micro drill station. Dentine samples from these specimens were also sampled using the same sampling methodology. Sediment attached to the specimens was also collected in some instances, depending on availability. Both of these tissue types were used to assess diagenesis (see SI). They are reported in Supplementary Table SI 3.

The pretreatment and analysis in the IRMS (isotopic ratio mass spectrometer) was conducted in the laboratories of the Biogeology working group (Department of Geosciences) at the University of Tübingen using a modified Koch method (Bocherens et al., 1994; Koch et al., 1997; Wright and Schwarcz, 1999). We added 1.35 ml of NaOCl at a concentration of 2.5% to each of the powdered samples and left it react for 24 h to remove organic matter that might be present in the sample. Then the liquid was removed and after rinsing the samples three times with distilled water, 1.35 ml of 1 M acetic acid buffer solution (CH₃COOH) was added to the enamel and dentine samples and let to react with the samples for 24 h. Given difference in crystalline structure between enamel and other exogenous sources of carbonates, the latter ones were removed by this step. The samples were once again rinsed three times with distilled water and then dried at 35°C for 72 h. With each set of samples two internal standards of elephant and hippo enamel were treated following the same protocol. Prior to analysis one internal (LM = Laaser Marmor) and two international (IAEA-603, NBS-18) pure carbonate standards were added to the data set after every 15 samples. These were not subjected to any pretreatment prior to analysis.

For the analysis with the Elementar IsoPrime 100 IRMS 2.5–3 mg of enamel or 0.1 mg of pure carbonate were reacted with 99% phosphoric acid (H₃PO₄) at 70°C for 4 h. The isotopic ratios of the resulting CO₂ gas were measured four times for each sample over a time span of 15 min. This allowed us to monitor measurement precision by calculating the mean and standard deviation of these repeated measurements for each sample (Szpak et al., 2017). In addition, measurement uncertainty was assessed using the standards. It is reported for each sample in Supplementary Table SI 2. Isotopic ratios are reported using the δ-notation (in per mill). The calculation is based on formula (1) (Coplen, 2011; Bond and Hobson, 2012) where j marks the heavier and i the lighter isotope.

With the internal enamel standards the isotopic ratios were calibrated relative to VPDB (Vienna Pee Dee Belemnite). Oxygen isotopic ratios have been subsequently converted to δ¹⁸O_VSMOW values (Vienna Mean Ocean Water) using the formula (2) by (Coplen, 1988).

In Supplementary Tables SI 2, SI 3 we also report the CaCO₃ content of each sample. It was calculated with the Ion Vantage software based on an estimated elemental composition with sample weight, peak area and the internal LM standard as variables. The resulting CaCO₃ values are subsequently scaled up until one of the international pure carbonate standards reaches 100%. CaCO₃ content of sediment samples from various Pondaung localities was too low when we used the regular amount of 0.1 mg of powdered sediment sample in the IRMS. Even when we raised the sample size to 2.5–3 mg of powdered sample, CaCO₃ content remained very low for the Pondaung sediments (Supplementary Table SI 3). We discuss the results of the sediment and dentine samples in the Supplementary Information to evaluate the risk of diagenetic alteration of the original biogenic isotopic signal in our data set.

Prior to data analysis and isotopic niche modelling two types of data corrections were applied to account for differences in enrichment of δ¹³C values between diet and bioapatite as well as changing δ¹³C baselines, due to shifts of the isotopic composition of CO₂ in the atmosphere. All the data before and after the corrections were applied as well as the specific enrichment factor used to calculate δ¹³C_diet values are reported in Table 1.

In the past one uniform enrichment factor of 14‰ for the carbon isotope ratio from diet to apatite in dental enamel has been used for all medium to large bodied herbivores (Cerling and Harris, 1999). Recent studies however showed, that many different factors such as body mass and digestive physiology influence this enrichment factor (Passey et al., 2005; Tejada-Lara et al., 2018). Conducting an experimental study to establish a specific ε* for a species is only possible for extant animals. For our study we therefore apply the regression formula (3) proposed by Tejada-Lara et al. (2018) to account for differences of ε* due to variations in body mass. This is the conservative formula, which is not specific to foregut or hindgut fermenters.

Here, BM stands for the log transformed (ln) body mass in kg. The resulting enrichment factor is in per mill (‰), which can then be applied to the data prior to isotopic modelling (Table 1) to calculate δ¹³C_diet values (4).

Instead of correcting the individual δ¹³C values for changes in δ¹³C_CO2 of the atmosphere, which is the pool from which plants extract the carbon in their tissue during photosynthesis, we chose to apply the corrections to the thresholds used for the interpretation of vegetation type and density. Post-1,930 the δ¹³C_CO2 values were −8‰ (Zachos et al., 2001; Cerling et al., 2010 but see Keeling et al., 2005 for fluctuations within this time period). Tipple et al. (2010) reconstructed Cenozoic δ¹³C_CO2 values based on stable isotope analysis of benthic foraminifera. For the middle Eocene at 40 Ma their model suggest higher δ¹³C_CO2 values than today with an estimate of −5.5‰ (ranging from −6‰ to −5.2‰ at a confidence interval of 90%). Thus, we corrected the modern thresholds by +2.5‰ before interpreting the Eocene data.

Following the Hutchinsonian niche concept (Hutchinson, 1957, 1978), we calculated core ecological niches for each taxonomic group using the R package SIBER (version 2.1.6; Jackson et al., 2011). For this study, we quantify niches using standard ellipse area corrected for small sample sizes (SEA_C). They correspond to a confidence interval (CI) of 40%. As this modelling approach is based on a maximum likelihood estimation in a Bayesian framework, the limiting effect of small sample sizes on statistical power of our analysis is counteracted to a certain extent. Should the sample size of a taxonomic group be below five individuals we mention it explicitly during the interpretation of the results.

For the evaluation of niche packing and directional competition potential, we calculated three metrics to quantify niche overlap between two taxonomic groups. These are the percentage of the total niche space of the two taxa that is shared between them (% overlap), which is calculated with formula (5; Ogloff et al., 2019), the percentage of niche area of taxon A (A_A) that overlaps the niche space of taxon B (A_B), and vice versa. The latter two are reported as A_A–A_B and A_B–A_A, respectively. A_O which is the area shared between taxon A and taxon B was acquired by using the maxLikOverlap() function of the SIBER package. A_A and A_B are calculated using A_O and the unitary method.

Based on the modelled ecological niches we also calculated the relative individual niche (RINI; Sheppard et al., 2018) for the individuals from which serial samples were taken. RINI is a metric that relates the individual niche area to the population niche area. Here, it expresses the proportion the SEA_Cs (CI = 40%) of an individual that is shared with the union of all the SEA_Cs of the serially sampled individuals from a taxonomic group. For this we used the siberKapow() function that is part of the SIBER package. Code examples can be found in the explanatory vignettes at the SIBER Github repository.

Given the small sample sizes we are working with and the fact that normal distribution of our data cannot be attested in all cases non-parametric Wilcox rank sum tests are used to test for statistically significant differences in this study. To screen for relationships between δ¹³C and δ¹⁸O values in the specimen where intra-individual serial samples were taken, we performed Spearman correlation analyses. Habinger et al. (2022) identified one statistical outlier in the modern Bos data we also use in our study. This outlier was not included in the statistical analysis, but is shown in the plots. Statistical significance is ascribed to test results where p = <0.05. In cases where several statistical tests were run on the same data set, value of ps have been adjusted with the Bonferroni correction, a conservative method to correct for multiple comparisons (Bonferroni, 1936). All statistical tests and models were run in R version 4.1.2.

The δ¹⁸O values including the minimum values, which represent the seasonal precipitation maxima in tropical environments (Dansgaard, 1964; Sun et al., 2021), of the modern bovid are higher than the Eocene (Figure 2) ones even though the temperatures nowadays are around 5°C lower than they were back then. The temperature effect leads to higher δ¹⁸O values under warmer conditions (Dansgaard, 1964). However, the pattern in the comparison of the modern bovid with the Eocene mammals is reversed. This would be consistent with a generally wetter, more humid climate in the Eocene than in Myanmar today. Consistently lower minimum δ¹⁸O might even indicate an increased monsoon intensity in the Eocene, although they could have also been influenced by the possible closer paleogeographic proximity of the fossil localities to the coast at 40 Ma (Westerweel et al., 2019, 2020).

We screened for a correlation between the δ¹³C and δ¹⁸O values of the serially sampled specimen; as such a relationship would be consistent with an influence of seasonality on variation in the diet (e.g., the type of plant or plant part ingested) or habitat. We do see that there is a statistically significant correlation between seasonal precipitation and diet/habitat in the Anthracotheriidae, Amynodontidae, and Rhinocerotoidea as opposed to the Brontotheriidae. This relation however is not uniform across the taxonomic groups. This pattern is consistent with an interpretation of different strategies of the taxonomic groups to react to seasonal habitat changes such as for example adapting their subsistence strategy or migration. As there is no correlation of the δ¹³C or δ¹⁸O values diagenesis is likely not the cause for these correlations (see discussion in the Supplementary Information). The correlation in Anthracotheriidae and Amynodontidae is positive, whereas it is negative in Rhinocerotoidea. This difference between Amynodontidae and Rhinocerotoidea is surprising, since as stated earlier, most specimens which are classified as Rhinocerotoidea are likely amynodonts as no other genus from this superfamily has been described yet. Of course, this difference could be introduced by a sampling bias. If not, it could indicate that either there are two groups of amynodonts that have different ecological strategies to cope with seasonality or that these different ecological strategies also coincide with a taxonomic distinction on a generic level that has not been described as of yet. As there are no experimental studies that focus on this relationship, we cannot precisely characterize the difference between a positive versus a negative correlation of δ¹³C and δ¹⁸O values in plants or herbivorous animals.

Nevertheless, the occurrence of significant correlation of δ¹³C and δ¹⁸O values can be interpreted as an increased influence, e.g., due to the reconstructed monsoon intensity in the Eocene, of seasonality on the diet/habitat on some of the mammal groups from the Pondaung Fm when compared with the modern cow. Two important limitations of this interpretation are the small sample size for modern Bos, which is only represented by one individual and the fact that we are comparing wild Eocene mammals with a modern domestic Bos, whose precise diet is unknown. But even the statistically significant correlations of δ¹³C and δ¹⁸O values in the Eocene alone are another indication of an intense Eocene monsoon.

The Pondaung Fm. was a habitat with only C₃ plants present. When we consider the changes in δ¹³C_CO2 values of the atmosphere discussed in the Methods section, modern C₃ plants would lead to δ¹³C_diet values of −17.5 to −34.5‰ (modern thresholds from Kohn, 2010). Hence, δ¹³C_diet values higher than −17.5‰ would be evidence for the presence of C₄ plants in the Eocene. The δ¹³C values of all the mammals in the data set used in this study range from −20.7 to −29.2‰, and thus falling within the range of Eocene C₃ plants. As C₄ plants with their photosynthetic pathway adapted to dry and hot climatic conditions did not evolve and spread to Asia until the Late Miocene (Cerling et al., 1993, 1997) we did not expect to find any values that high in our data set. In Myanmar, the first evidence of the presence of C₄ plants based on isotopic data is from 6 Ma (Thein et al., 2011; Habinger et al., 2022).

The variation of δ¹³C_diet values that we observed within and between the two study areas is caused by differences in vegetation structure. Due to the canopy effect, including factors such as low irradiance and carbon recycling, δ¹³C values decrease with an increased canopy density (van der Merwe and Medina, 1989; Jackson et al., 1993; Bonal et al., 2000; Leffler and Enquist, 2002; Bonafini et al., 2013). This effect is less pronounced high up in the canopy and more open patches of a habitat. In these microhabitats C₃ plants yield delta-values of −18.5 to −24.5‰, whereas δ¹³C_diet values of −29‰ and below are indicative for the understory of a closed canopy forest (modern thresholds from Kohn, 2010). δ¹³C values from both the Bahin/Pangan (−20.7 to −27.2‰) and the Mogaung area (−21.2 to −29.2‰) are consistent with open-forested seasonal woodlands, like the ones reconstructed for the upper deltaic plain by Licht et al. (2015). These woodlands were denser in the Mogaung area, given the significantly lower δ¹³C values (Figure 3), and with specimen with the lowest δ¹³C_diet values falling just below the threshold for closed canopy vegetation. This possibly indicates proximity to the more closed dipterocarp forests in the upstream areas from the same vegetation model (Licht et al., 2015). An interpretation of the microhabitat differences as being caused by such spatial factors is also consistent with the paleoenvironmental similarities of the geographically close Bahin and Pangan areas with one another. However, given the currently available dates and correlations of the stratigraphic sequences of the different fossil localities of the Pondaung Fm. we cannot rule out the possibility of the two microhabitats representing a changing environment at two different late Middle Eocene time intervals completely.

The significant difference in the δ¹⁸O values between the Bahin/Pangan and the Mogaung area is consistent with a more humid climate in the latter. As the Mogaung area was probably more densely forested as well, the data are also consistent with a consumption of drinking water from water sources that are more protected from direct sunlight. Water sources that are subjected to evaporation would have higher δ¹⁸O values as the lighter oxygen isotope would evaporate more readily. However, there is no significant difference in the ranges of δ¹⁸O values of the intra-individually sampled individuals between the Bahin/Pangan and the Mogaung area (W = 104, value of p = 0.07336). Range here is defined as the minimum δ¹⁸O value subtracted from the maximum one. This indicates that there is no difference in the intensity of the seasonality.

Considering the occurrences of fossil primates (Supplementary Table SI4) in relation with these two different habitat types, we see that so far for only two of the four genera of the anthropoid Amphipithecidae have been found at all three localities, Aseanpithecus and Pondaungia. So at least for these two amphipithecid genera we can propose a certain degree of ecological flexibility. A more detailed discussion of the fossil primate occurrences and how they relate to the localities and the other mammal taxa in our data set can be found in the Supplementary Information.

An evaluation of differences in size of combined ecological niches on taxonomic family level, which potentially consist of multiple species, and their % overlap showed no significant differences between the two areas. In addition to these similarities the relative positions of the taxonomic groups to one another are very similar as well (Figure 4). Anthracotheriidae were occupying the more open patches of the habitat, which is most apparent in the Bahin/Pangan area. However, the high δ¹³C values that would indicate a more open vegetation in the anthracothere habitat can also be partly influenced by their more omnivorous subsistence strategy. As the Anthracotheriidae are more omnivorous than the other taxonomic group they had an ecological advantage there. They were able to reduce competition potential as evidenced by the indication of their use of more open habitats in the Bahin/Pangan area than in the more closed Mogaung habitat. Brontotheriidae, Rhinocerotoidea, and in the Bahin/Pangan area also Amynodontidae shared their aspects of their ecological niches as modelled by the carbon and oxygen isotopes in parts of the paleolandscape. The differentiation between the Brontotheriidae and Rhinocerotoidea niche space is mostly due to the δ¹⁸O, which would be consistent with the former drinking from more evaporated water sources than the latter. Given all these similarities in the relative positioning of the different taxonomic groups to one another, the differences that we saw between the paleoenvironments between the Bahin/Pangan and the Mogaung area, apparently did not translate to a different organization of the habitat use of the mammal communities there.

We took our analysis of the niche modelling of the Pondaung mammal fauna one step further and looked at the serially sampled individuals (Figure 5) separately. This revealed different patterns of the distribution of modelled ecological niches especially in the Anthracotheriidae. While the Anthracotheriidae appear to share the modelled aspects of their ecological niches, there is more diversity in the Brontotheriidae, Amynodontidae, and Rhinocerotoidea. This makes sense as the taxonomic resolution in the Anthracotheriidae is better (Figure 5), where all but two individuals could be identified as to at least belonging to the genus Anthracotherium, whereas potentially multiple genera or species with different ecologies could be included in the sample for the other taxonomic groups. Several individuals of the Brontotheriidae, Amynodontidae, and Rhinocerotoidea seem to have occupied different microhabitat niches than the other specimen of these groups. Given that this approach requires multiple samples per individual, it becomes apparent why the species diversity in Anthracotheriidae is not reflected here in its entirety as this sampling technique could only be applied to larger teeth, such as the two Anthracotherium species and some undetermined Anthracotheriidae.

Based on the ecological niche modelling on the individual level for the four main taxonomic groups we calculated a metric used in ecology to differentiate generalist from specialist individuals. Large modelled niche areas for a taxon or a taxonomic group are consistent with an overall generalist behavior. It can however, be caused by two different reasons. It can result either from many specialist individuals with different ecologies that would have a low RINI, or from generalist individuals that all more or less use the entire niche space of their taxonomic group and thus having a high RINI. As the total modelled niche area of the taxonomic groups is based on higher level taxa (families) there is the possibility of an overestimation of the isotopically modelled niche aspect, if there are big differences in habitat preferences and use in the lower level taxa. Hence, the RINIs could be lower in these cases. Both SEA_C and RINI were not significantly different between Anthracotheriidae and the rest of the taxonomic groups. The fact the same pattern is present in both metrics is an indication that no particular generalist or specialist individuals were present, and that the bias towards an underestimation of the RINIs has no significant effect on this comparison. The clustering pattern in Anthracotheriidae seems to be different than in the other three taxonomic groups. The lack of individuals plotting apart from the main cluster of modelled ecological core niches without overlapping them is consistent with a more continuous variation of ecological requirements in this group whereas the pattern in Brontotheriidae, Amynodontidae, and Rhinocerotoidea indicates more discrete ecological niches reflecting aspects of habitat use and dietary ecology.

Although the localities are located along a paleo river system, none of the taxonomic groups has δ¹⁸O values low enough to suggest semi-aquatic behavior. Previous studies showed that depending on the humidity of the terrestrial habitat, this offset can be 2–4‰ lower in freshwater living semi-aquatic species than in terrestrial ones (Clementz et al., 2008). An additional indication of semi-aquatic behavior in a taxonomic group is a reduced standard deviation (SD) of ≤0.5‰ (Yoshida and Miyazaki, 1991; Clementz and Koch, 2001). In our dataset such low SDs are only present in S. pondaungensis, Bahinolophus, and Amynodontidae from Mogaung (Table 2). These are all taxonomic groups with very low sample sizes (MNI = minimum number of individuals = 2). Therefore, we do not consider the low SD as an indication for semi-aquatic behavior in any of these cases. Differences in δ¹⁸O values can however result from the use of different water sources. The lower δ¹⁸O values in Rhinocerotoidea in comparison to Brontotheriidae would be consistent with them drinking from less evaporated water sources. This could mean shaded water or flowing and not stagnant water.

Between the three different anthracothere genera found in the Pondaung Fm. there is a lot of niche overlap (Figure 6), which was already suggested by the niche modelling of the serially sampled anthracotheres (Figure 5), although none of the smaller genera (Anthracokeryx and Siamotherium) were in this sample group. Apparently, the three genera did share habitats with similar ecological characteristics. Hence, the different feeding ecologies and habitat preferences in regard to forest density inferred from tooth and bone morphology (Ducrocq, 1999) do not translate to different habitats occupied by them based on the modelled core ecological niches. No clear patterns are visible that would differentiate the anthracothere genera based on canopy closure or isotopically different water sources. We described the hypothetical ecology of Anthracotherium as being both more water dependent and more reliant on open spaces in the habitat for their subsistence. Anthracotherium however does not have significantly higher δ¹³C values or lower δ¹⁸O values than the other genera, thus the isotopic data are not consistent with this hypothesis.

Anthracotherium shares almost all of its modelled ecological core niche space with Anthracokeryx (Figure 6B). This can indicate that Anthracotherium experienced a higher competition potential for its ecological niche than Anthracokeryx (Ogloff et al., 2019). However, the larger body mass of Anthracotherium counteracts this effect. The higher %overlap of the Anthracotherium with the Anthracokeryx niche could also mean that the advantage of being larger and more generalist in its diet enabled Anthracotherium to use more of the Anthracokeryx niche space (Schoener, 1983; Persson, 1985; Law et al., 1997). A similar pattern is also visible in the comparison of Anthracokeryx and Siamotherium, which would be consistent with similar interpretations. However, it is important to keep in mind that the Siamotherium niche is based on only two individuals, with two sampled teeth each, which differ vastly in δ¹³C values, and further interpretations will depend on a larger sample size for Siamotherium.

The limitation of sample size of course is even more prevalent when we are looking at niche partitioning between the anthracothere species (Figure 7A). Nevertheless, we wanted to investigate possible interesting patterns of niche partitioning and directional competition potential detectable to open new directions for future research. Given the severe problems with sample size in S. pondaungensis discussed in the previous section, we cannot discuss its niche overlap in comparison with the other species. It is apparent that there is a lot of niche overlap between the anthracothere species, leading to more diffuse competition than niche partitioning in this context (Figure 7). The Pondaung habitat must have provided sufficient area to fulfil the ecological requirements of the anthracotheres and sustain all of the species sampled in this study.

When we generalize the results formed based on the comparison Anthracotherium and Anthracokeryx in the previous section we propose the following hypothesis: Species with higher body mass have a competition advantage; hence, we expect them to have a higher percentage of shared niche space with the smaller species (Figure 7B). Anthracokeryx tenuis, the smallest of these anthracotheres, follows this prediction in every instance and always shares a smaller percentage of its niche area than the bigger species with which it is compared. For the other species this relationship is not as simple. Anthracotherium pangan is the largest anthracothere from Pondaung. However, the second largest A. crassum shares more of its core niche with it than vice versa. This might be because A. crassum also has a much narrower core ecological niche than A. pangan. This fact would be consistent with the latter having more generalist ecological requirements than the former. Anthracokeryx birmanicum mostly fulfils the expectations based on our hypothesis. The shared niche space between it and A. pangan however, are the same. They only share around 5% of their ecological core niches with one another. Given this small overlap area, distinct ecological requirements of these two anthracothere species probably led to a reduced competition potential where the body mass advantage of A. pangan was irrelevant.

To sum up, we could further prove the existence of a proto-monsoon already in the Eocene even at the low paleolatitude of 5°N of the equator. The δ¹³C and δ¹⁸O values also showed significant differences in both overall humidity and vegetation structure between the Bahin/Pangan and Mogaung areas. Although our results are consistent with the hypothesis that favor a spatial pattern of the vegetation structure, temporal differences cannot be ruled out and require more detailed investigation of the dating and stratigraphic relationship of the different fossil localities of the Pondaung Fm. Although the Mogaung area was more humid and more densely forested than the other two, the organization of the mammal fauna was not different, which is consistent with ecological flexibility of this Middle Eocene greenhouse mammal fauna. A high degree of niche overlap was observed on genus and species level resolution for the anthracotheres. Interestingly, there is no evidence for semi-aquatic adaptations in any of the taxonomic groups, when we test for significant offsets in δ¹⁸O values and a reduction of their variability. This means that at the same time when ancestral whales were already returning to aquatic life, anthracotheres that are ancestral to hippos still were fully terrestrial even though the paleo river channel associated with the Pondaung Fm. would have offered enough suitable habitats for this lifestyle. That the primate fossil record from the Pondaung Fm. also does not show any significant differences in the occurrences of the different taxa is also consistent with this and also indicates ecological flexibility in this taxonomic group. Therefore, the Pondaung Fm. represents diverse microhabitats that were used by a dynamically organized mammal fauna. This environment proved to be conducive for the diversification and radiation of many taxonomic groups such as primates, anthracotheres and rodents.