Main

The Middle to Upper Palaeolithic transition marks an important period in human evolutionary history, with the dispersal of Homo sapiens across Eurasia and the disappearance of other hominins such as Homo neanderthalensis from the fossil record. Archaeological and genetic evidence increasingly demonstrates that this transition involved a complex patchwork of archaeological and biological turnovers including many dispersals of H. sapiens1,2,3,4,5 (but see ref. 6). To interpret the evolutionary significance of these events, it is crucial to determine the environments and climatic conditions that H. sapiens groups encountered during these dispersals.

Prominent models have suggested that early range expansions of H. sapiens during the Late Pleistocene were linked to warm climatic phases that facilitated adaptation to higher latitudes7,8. Recent palaeoclimatic data generated directly from archaeological sites have challenged this idea9,10 but the scarcity of data from a few sites and archaeological technocomplexes means that climatic scenarios of dispersals still lack the complexity that is emerging from the genetic and archaeological records.

Here we add local palaeoclimatic data pertinent to a newly documented early incursion of Late Pleistocene H. sapiens into central Europe. We use multiple stable isotope analysis of faunal remains from Ilsenhöhle in Ranis (hereafter, Ranis), Germany, to document the climatic and environmental conditions H. sapiens faced during this dispersal, associated with the Lincombian–Ranisian–Jerzmanowician (LRJ) transitional technocomplex11,12,13. Ranis is located in the Orla valley (50° 39.7563′ N, 11° 33.9139′ E, Thuringia, Germany; Extended Data Fig. 1) and is a type site of the LRJ, an archaeological phenomenon of the Middle to Upper Palaeolithic transition, that extends across northern and central Europe. New genetic, proteomic and chronological evidence now links the LRJ with directly dated H. sapiens remains at Ranis and documents one of the earliest dispersals of our species into the European continent13. Using the isotopic data generated here we demonstrate the climatic and environmental conditions that pioneering H. sapiens groups exploited during their initial spread across central and northwestern Europe.

Originally extensively excavated by W. Hülle from 1932 to 193814, recent re-excavations from 2016 to 2022 (Thuringian State Office for Preservation of Historical Monuments and Archaeology (TLDA)/ Weimar and the Department of Human Evolution at the Max Planck Institute for Evolutionary Anthropology, Leipzig (MPI-EVA) excavations) have allowed a state-of-the-art reassessment of the stratigraphy and chronology of Ranis (Extended Data Fig. 1 and Supplementary Text 1). In the TLDA/MPI-EVA excavations, the LRJ occupations are associated with layers 9 and 8, dating to 47,500–45,800 cal bp and 46,800–43,300 cal bp, respectively, with the main occupation in layer 8 (ref. 13). H. sapiens fossil remains were identified in layers 9 (n = 1) and 8 (n = 3) and the Hülle collection (n = 9), with direct 14C dates of the Hülle specimens matching those of layers 9 and 8 (ref. 13). Zooarchaeological, archaeological and sediment DNA data suggest that the LRJ occupations were ephemeral and low-intensity, with most faunal remains accumulated by carnivores15. The LRJ layers are bracketed by Upper Palaeolithic deposits above and potentially Middle Palaeolithic deposits with low artefact density at the base of the sedimentary profile (Extended Data Fig. 1).

Here, we apply oxygen, carbon, nitrogen, strontium and zinc stable isotope analyses to directly 14C-dated Equus sp. teeth (enamel bioapatite and dentine and mandible bone collagen) from the transitional LRJ and Upper Palaeolithic occupations (layers 9–6) to generate evidence of seasonal palaeotemperatures, water availability and changes in vegetation cover experienced by the humans that produced the LRJ record. Fossil teeth were obtained from the Hülle collection (n = 14) and the TLDA/MPI-EVA excavation (n = 2; Supplementary Table 1). Furthermore, 24 tooth specimens representing a variety of herbivore, omnivore and carnivore taxa were chosen from the Hülle collection for further δ66Zn and 87Sr/86Sr analyses to explore their feeding ecology and mobility (Supplementary Table 1). The stratigraphic layers of the two excavations are clearly correlated and archaeologically equivalent. However, challenges with the documentation of the Hülle faunal collection often prevent clear stratigraphic assignments of faunal specimens (Supplementary Text 1). For this reason, all equid remains studied here were directly 14C-dated and correlated to the LRJ and H. sapiens fossils through the obtained ages.

Results

Chronology

Direct radiocarbon dating of 16 equid specimens yielded calibrated 14C ages ranging from 48,800 to 36,300 cal bp, covering ~12,500 years (Supplementary Table 2). The radiocarbon dates from the Hülle faunal collection show a large spread of ages for the layers labelled ‘brown’ (Supplementary Fig. 1). The LRJ grey layer (X) has more consistent ages, ranging from 45,900 to 42,100 cal bp, which overlaps with the date range of the LRJ layer 8 but also with layer 7 of the TLDA/MPI-EVA excavation13. Quality control indicators demonstrate exceptional collagen preservation and purity (Supplementary Table 2) and thus this overlap should be attributed to the excavation methodology of the 1930s and the documentation quality of the Hülle faunal collection. To avoid stratigraphic attribution errors, all analyses in the following are made only on the basis of the direct radiocarbon dates of the equid specimens.

Stable isotope analyses

Oxygen isotope measurements of sequentially sampled tooth enamel phosphate (δ18Ophos) show sinusoidal seasonal cycles in all specimens, with high δ18Ophos peaks representing summers and low δ18Ophos troughs representing winter inputs (Supplementary Fig. 2, Supplementary Text 2 and Supplementary Table 3). The 87Sr/86Sr values of equids (0.7090–0.7120), undertaken to confirm that δ18O values are representative of local conditions without bias from long-distance migrations, fall into the range of bioavailable values of Thuringian lithological units and match those observed in hyenas and ursids with typically small to modest home range sizes (Supplementary Text 3, Supplementary Figs. 3 and 4 and Supplementary Table 4). Seasonal 87Sr/86Sr intratooth differences are mostly very small (<0.0005), with no systematic changes through time (Supplementary Fig. 3). One individual, R10131, shows a slightly larger seasonal change (0.0008) and also shows a larger seasonal difference in δ66Zn (Supplementary Figs. 5 and 6), an isotopic tracer that reflects some impacts of underlying bedrock type—in addition to the more prominent dietary effects (Supplementary Text 4). Overlap of 87Sr/86Sr values of this specimen with regionally expected values means that long-distance movement remains unlikely but cannot be ruled out entirely. Thus, excluding this specimen from climatic interpretations is the most cautious approach. Furthermore, the correlation between 87Sr/86Sr and δ66Zn seems driven by two outliers and has a very shallow slope, suggesting that δ66Zn values are predominantly driven by diet (Supplementary Figs. 7and 8, Supplementary Text 3 and Supplementary Text 4).

Seasonal and mean annual δ18Ophos values show distinct changes over time (Fig. 1 and Supplementary Table 5), starting with mean annual values of ~12–13.5‰ at 48,000–45,000 cal bp, then dipping by ~3‰ to ~9–10‰ at ~45–43 ka cal BP. After this, δ18Ophos values rise back up to a similar level at around 42,500 cal bp, followed by a gap in the data and a phase of high δ18Ophos variability at ~39–36.5 ka cal bp. During the low δ18Ophos excursion at ~45–43 ka cal bp, winter δ18Ophos values fall as low as 9.0‰, while summers only reach 13.1‰ at their highest value. Seasonal amplitudes of δ18Ophos (summer–winter differences) range between 0.9‰ and 4.1‰ overall and are highest at ~45–43 ka cal bp with a mean seasonal amplitude of 3.3 ± 0.8‰. Seasonal δ18Ophos amplitudes are lower in periods of comparatively higher δ18O and correlate negatively with winter δ18O (P = 0.0049, R = −0.39, n = 50, Pearson correlation) but not with summer δ18O (Supplementary Fig. 9). Hence, changes in seasonality are driven predominantly by changes in winter δ18O. The lowest point in δ18Ophos at ~45–43 ka cal BP coincides with the highest dentine and mandible bone collagen δ15N values of 8.7–6.8‰ (Fig. 1) and the two proxies show a statistically significant correlation, particularly during this time (Supplementary Figs. 10 and 11, for all data R = −0.36, P = 0.045, n = 32; for >42,000 cal bp, R = −0.59, P = 0.012, n = 17, Pearson correlation). After the ~45–43 ka cal BP high point, δ15N values decline steadily by ~4‰ until they fluctuate between 3.4‰ and 4.5‰.

Fig. 1: Oxygen, nitrogen, carbon and zinc stable isotope analyses of directly dated equid teeth show changes in climate and environment through the LRJ and Upper Palaeolithic sequence of Ranis.
figure 1

Summer peak, mean annual and winter trough oxygen isotope values show low values throughout the sequence and a temperature decline from ~48 ka cal bp to a temperature minimum at ~45–43 ka cal bp. This oxygen isotope minimum coincides with high δ15N (dentine and mandible bone collagen) and δ66Zn values, suggesting a hypergrazer niche of equids in open steppe environments or very dry soil conditions similarly indicative of an open environment. This is supported by high δ13C (dentine and mandible bone collagen) values consistent with a steppe or tundra biome. One individual has been marked with an asterisk as it has been excluded from climatic interpretations because 87Sr/86Sr and δ66Zn seasonal amplitudes are high enough that a seasonal movement cannot be completely excluded. Oxygen isotope data points represent δ18O summer peak, winter trough and annual means of individual annual cycles represented in sinusoidal δ18O time series obtained from sequentially sampled tooth enamel (marked in Supplementary Fig. 2). Stable isotope data are presented as the mean ± measurement uncertainty based on sample replicates (1 s.d., nreplicates = 3 for δ18O and nreplicates = 2 for all other proxies where error bars are present; replicate measurements represent repeated isotopic measurements of aliquots of each single prepared sample). Measurement uncertainty for δ66Zn is smaller than the symbol size. Horizontal error bars indicate the 95% calibrated age range of direct radiocarbon dates (n = 1 tooth sample for each data point). Symbol shapes indicate the excavation origin from either the Hülle (1932–1938, circles) or TLDA/MPI-EVA (2016–2022, triangles) campaigns. Collagen analysed for δ13C and δ15N was obtained from tooth dentine for all 1932–1938 samples and from adhering mandible bone for the two 2016–2022 samples marked by triangle shapes. Stable isotope delta values are reported in relation to the relevant scale-defining reference materials Vienna Mean Ocean Water (VSMOW), atmospheric N2 (AIR), Vienna Pee Dee Belemnite (VPDB), and Johnson Mattey zinc metal (JMC Lyon).

Full size image

Carbon stable isotope values of dentine and mandible bone collagen change little over time, with most individuals falling into a range of less than 1‰ (−21.3‰ to −20.6‰; Fig. 1). The δ13C values do not show any statistically significant correlations with the other isotope systems (Supplementary Fig. 10, Pearson correlation). Across a variety of taxa in the food web, δ66Zn values of the Ranis fauna follow expected dietary and trophic patterns, with the lowest values observed in carnivores and highest values in herbivores (Supplementary Text 4 and Extended Data Fig. 2). Within herbivores, woolly rhinoceros (Coelodonta antiquitatis) show the lowest values, followed by typically browsing or mixed feeding cervid taxa, while equids show the highest δ66Zn values (0.79‰ to 1.51‰). Across time, δ66Zn values of equids are highest at ~45–43 ka cal bp, coinciding with the lowest δ18O values and highest δ15N values (excluding R10131 ; Fig. 1), resulting in a positive correlation with δ15N (R = 0.43, P = 0.0013, n = 15, Pearson correlation) and a negative one with δ18O in the time before 42,000 cal bp (R = −0.58, P = 0.015, n = 17, Pearson correlation; Supplementary Figs. 11 and 12). However, diachronic δ66Zn change is relatively small and equid δ66Zn values never overlap with non-herbivore taxa or even some of the lower-δ66Zn herbivores such as woolly rhinoceros or cervids (Extended Data Fig. 2).

Water oxygen isotopes and palaeotemperatures

Reconstructed δ18O values of drinking water (δ18Odw), which enable comparisons with modern meteoric water sources and those of other fauna, fall systematically below modern day δ18O of local precipitation (δ18Oprecip), as well as Thuringian rivers and springs (Extended Data Fig. 3 and Supplementary Table 6). Summer δ18Odw partially overlap with modern spring and river water, as these seasonally more buffered water bodies represent amount-weighted annual averages of δ18Oprecip. The lowest δ18Odw values at ~45–43 ka cal bp fall more than 5‰ below the mean annual δ18O of the modern water sources and more than 8‰ below winter water source values.

This is mirrored by air temperature estimates (Supplementary Table 6), which fall substantially below modern-day conditions with the largest difference in winter (Fig. 2). During the ~45–43 ka cal bp cold phase, air temperature estimates are lower than modern day by 7.3 ± 3.6 °C in summer, 11.3 ± 3 °C for mean annual conditions and 15.3 ± 4.5 °C in winter. In the oldest data (~48–45 ka cal bp), temperature estimates, while less extreme, still fall 3.0–8.3 °C below modern-day temperatures across all seasons. Temperature seasonality ranges from 22.2 ± 2.6 °C at ~38 ka cal bp to 27.4 ± 3.8 °C during the temperature minimum at ~45–43 ka cal bp, compared to the modern-day temperature seasonality of 19 ± 2.9 °C.

Fig. 2: Air temperature estimates derived from δ18O measurements generally fall below modern-day conditions.
figure 2

Lowest temperatures are observed in the ~45–43 ka cal bp interval, where they fall ~7–15 °C below modern day and mean annual temperatures below freezing. Oxygen isotope data from several individuals were grouped into time bins according to clusters of radiocarbon dates and δ18O measurements (Supplementary Table 6). Plotted points represent temperature estimates for each time bin. Error bars represent combined uncertainty for each temperature estimate, taking into account the uncertainty of each temperature calibration step (Supplementary Text 5). Ndatapoints for each error bar varies by season and time bin and can be found in detail in Supplementary Table 6. In the time bins δ18Odw estimates are based on a variable number of tooth specimens with n36–39 ka = 7, n42–43 ka = 3, n43–45 ka = 2 and n45–48 ka = 3. Summer and winter temperatures were estimated from inverse modelled δ18O time series, while annual means were derived directly from unmodelled δ18O (Supplementary Text 5). Lines and shaded ribbons of modern comparative data represent means and one standard deviation of modern climate observations (MAT, Tcoldest month, Twarmest month) for 1961–2009 from the ClimateEU model59).

Full size image

Relationship with human presence

To test the temporal overlap between the equid specimens yielding the isotopic climate data and the presence of H. sapiens we used χ2 tests and agreement indices of the OxCal Combine function for groups of direct dated equids, H. sapiens remains and anthropogenically modified faunal bone fragments. A table of all test results can be found in the associated online supplementary material at https://osf.io/wunfd/.

The direct dates of all equid individuals overlapping with the age ranges of the LRJ deposits (~48–43 ka cal bp) are statistically indistuinguishable from at least one directly dated H. sapiens fragment and, in many cases, also from at least one anthropogenically modified faunal bone fragment. Importantly, this includes the equids that yielded the lowest δ18O values (R10124, ETH-111922 and R10126, ETH-111920), which show a calibrated date range of 45,000–43,100 cal bp (Fig. 3). The direct date of R10126 is statistically indistinguishable (χ2 = 0.170, d.f. = 2, (5% 5.991), Acomb = 117.7) to those of a H. sapiens fragment (R10875, ETH-127625) and a cut-marked bone from layer 8 (16/116-159091, ETH-118367), while the date of R10124 is statistically indistinguishable (χ2 = 0.079, d.f. = 3, (5% 7.815), Acomb = 194.5) from two H. sapiens fragments (R10879, ETH-127628 and R10396 ETH-115246) and an equid fragment with percussion notches (16/116-159318, ETH-111935). Other H. sapiens specimens and anthropogenically modified bone fragments from the LRJ deposits date to a slightly earlier period ~48–45 ka cal bp coinciding with equid specimens yielding temperatures that fall ~3–8 °C lower than today but are less extreme than those from the ~45–43 ka cal bp interval (Fig. 3).

Fig. 3: H. sapiens presence coincides with the coldest temperatures documented by equid δ18O data.
figure 3

Comparison of equid δ18O data (top) with directly dated H. sapiens remains 13 (bottom, turquoise symbols) demonstrates extensive overlap of H. sapiens presence with the coldest temperatures documented between ~45 and 43 ka cal bp (marked by blue shading). This coldest, low δ18O phase overlaps with the age ranges of both the LRJ layer 8 and the beginning of undiagnostic layer 7 (ref. 13) (modelled 95% probability layer age ranges of the MPI-EVA/TLDA excavation in purple) but the direct dates of H. sapiens remain and faunal bone fragments with anthropogenic surface modifications clearly show that they endured the cold subarctic steppe conditions evidenced by the stable isotope data at this time. This also holds true independent of the calibration curve, as seen in the uncalibrated dates (Supplementary Fig. 13). Top panels show the relevant Greenland stadials (GS) and interstadials (GI) recorded in the NGRIP ice cores60 and the proportions of total arboreal pollen (dark brown) and Betula pollen (ochre) in the Füramoos pollen record from southwestern Germany29,61. Data are presented as mean ± 95% calibrated age ranges (n = 1 bone or tooth enamel sample for each data point), while point shape indicates whether specimens were found in the Hülle (1932–1938, circles) or the TLDA/MPI-EVA (2016–2022, triangles) excavation collection. We argue that H. sapiens fragments from the Hülle collection (labelled IX–XI here) all originate from the LRJ deposits (layer X) and were sometimes assigned to a mixture of layer X and adjacent strata by the original excavators due to rough excavation methods (details in ref. 13). We have pooled all these samples here to reflect this. Credits: equid silhouette by Mercedes Yrayzoz, vectorized by T. Michael Keesey (PhyloPic); human silhouette from NASA Pioneer plaque.

Full size image

Comparison of the uncalibrated dates confirms that the chronological overlap between H. sapiens specimens and the coldest phase at ~45–43 ka cal bp is independent of the calibration curve used (Supplementary Fig. 13). The two equids yielding lowest temperature data (ETH-111922 41,490 ± 360 14C bp and ETH-111920 40,740 ± 330 14C bp) overlap with the dates of seven of ten human specimens (Supplementary Fig. 13) and are almost identical to the ages of R10879 (41,429 ± 765 14C bp) and R10396 (41,570 ± 420 14C bp).

Taxonomic identifications of bone fragments with anthropogenic surface modifications from layers 9 and 8 show that they predominantly originate from reindeer but also include equid fragments. While the total number of fragments (n = 12) is too low to make robust inferences about taxonomic representation of hunted herbivores, this does indicate that some equids did overlap with H. sapiens occupations and that there is no immediate indication of a systematic difference in the prey taxa targeted by H. sapiens compared to carnivores15.

Discussion

Using multiple isotope analyses of directly 14C-dated equid teeth compared to directly dated H. sapiens fossils and an updated site chronology, we provide evidence that H. sapiens associated with the LRJ occupations of Ranis were present in central Europe during subarctic climatic conditions in a cold open environment, probably including a severe cold episode at ~45–43 ka cal bp. This shows that H. sapiens successfully operated in harsh environmental conditions during an early northward range expansion into central Europe.

Equid δ18Ophos data reported here are among the lowest ever reported in Europe for MIS 5 to MIS 3 (δ18Omean annual = 9–13.5‰ for the LRJ). Direct comparisons are limited to data from other equids, where most MIS 3 data from Germany do not fall below 14‰ (ref. 16). Examples with low values from Late Pleistocene stadial contexts in Germany (Bocksteinhöhle, Vogelherdhöhle and Villa Senckendorff17,18) and in Switzerland (Boncourt Grand Combe, MIS 3, Courtedoux-Va Tche Tcha, MIS 5a, ref. 19) range from 12.1‰ to 13.2‰ for mean annual data. This is similar to the higher values from Ranis observed for ~48–45 ka cal bp and ~43–39 ka cal bp. Data from Ranis for ~45–43 ka cal bp, however, descend 1.5–2‰ lower than this (Fig. 1). The data from ~45–43 ka cal bp fall ~0.5–1.5‰ lower than even the lowest equid δ18Ophos data reported from the Initial Upper Palaeolithic occupation of Bacho Kiro Cave (minimum δ18Omean annual = 11.2‰), Bulgaria, which has been used to reconstruct subarctic climatic conditions for an early presence of H. sapiens in southeastern Europe9. Reconstructed drinking water δ18Odw comparisons with other species suggest closest matches with Late Pleistocene data in Scandinavia and Russia, including data from the Last Glacial Maximum20,21. On the basis of characteristics of the study area and on the presence of a sinusoidal signal, we argue that δ18Ophos data reflect palaeotemperatures (Supplementary Text 2), thus our results demonstrate low temperatures, some of them remarkably so. Palaeotemperature estimates show a variability of conditions with a cooling trend but even the comparatively warmer episode at ~48–45 ka cal bp shows mean annual temperatures of <5 °C and based on δ18Oprecip most closely matches current climatic conditions in subarctic climate zones in northern Finland (for example, GNIP station Rovaniemi22). Mean annual temperatures then descend even further to below freezing for the coldest interval at ~45–43 ka cal bp. For this interval, our results correspond to temperature anomalies of ~7–15 °C below modern-day conditions with largest anomalies in winter (temperature estimation assumptions in Supplementary Text 2). This is paired with a very strong temperature seasonality of up to 27 ± 5 °C, indicating a continental cold subarctic to tundra climate with closest modern-day matches in northwestern Russia (for example, GNIP stations in Amderma and Pechora22). Importantly, summer temperature estimates in some cases fall below the 12 °C warmest month isotherm that dictates the Eurasian northern tree line23, indicating that tree growth may have been impossible in some of the climatic phases captured here. Air temperatures of ~10–15 °C below modern day are consistent with full stadial conditions and have been reconstructed for particularly severe Greenland stadials (GS) in central Europe24,25,26,27.

The cooling trend from ~48 to 44 ka cal bp into full stadial conditions followed by a rapid temperature increase after ~44 ka cal bp matches well with the documented slow cooling and rapid warming of Dansgaard–Oeschger (DO) events and based on tentative correlations with long-term climatic records may capture a Greenland interstadial (GI) to GS transition culminating in a pronounced cold phase such as GS12 or GS13 (refs. 27,28,29,30). While an assignment to a specific DO event may not be possible because of the chronometric dating uncertainties involved, the comparison suggests that the equid δ18O record captures millenial-scale climatic variability that occurred during the time of the LRJ.

Remarkably high δ15N values of equid dentine and mandible bone collagen in the ~45–43 ka cal bp interval (~7–9‰) compared to earlier and later equid data suggest either a hypergrazer feeding ecology or dry soil conditions during this interval. Nitrogen isotope variability in arctic biomes shows the highest values in grasses and herbaceous plants over shrubs or trees and this transfers to high δ15N values in specialized grazers31. Glacial phases are often accompanied by low δ15N values in fauna due to limited nitrogen availability and reduced bacterial activity in cold-wet soils, while high δ15N values are observed in phases with higher temperatures or low moisture availability32,33. As δ18O values demonstrate low temperatures in this interval, high δ15N values could indicate dry soil conditions and/or strong grazing specialization of equids, which both imply the presence of an open steppe environment. This matches with reconstructions of grass steppe environments in central Europe during MIS 3 stadials30,34. Equid dentine and mandible bone collagen δ13C values are consistent with feeding in an open grassland environment35 and the lack of diachronic change is in line with relatively small climatic impacts on δ13C of C3 plants common to Pleistocene European biomes.

Zinc stable isotope values of the food web at Ranis follow expected trophic level relationships with low values in carnivores and high values in herbivores36,37. Within herbivores, δ66Zn seems to reproduce a pattern of higher values in taxa commonly consuming more grass (equids) and lower values in typical browsers to mixed feeders such as Cervus elaphus (Extended Data Fig. 2). A similar pattern has been observed in a few European and African food webs and agrees well with higher δ66Zn values observed in low-growing plants over higher-standing tree or shrub leaves37,38,39 but its robustness is still debated36 (Supplementary Text 4). While the Ranis data cannot be used to definitively confirm this idea, they are tentatively consistent with it. If true, particularly high and seasonally invariant δ66Zn observed in equids in the ~45–43 ka cal bp cold interval would support an interpretation of a hypergrazer feeding niche of equids in an open steppe environment. The statistically significant correlation of δ66Zn with δ15N (Supplementary Fig. 12) is also consistent with both proxies being driven by grass consumption but due to the effects of soil nutrient cycling on both isotopic systems40, a relationship with dry soils and consequent changes in soil biochemical cycles is also possible.

Cold temperature conditions and an open grassland or tundra environment for the LRJ at Ranis match the faunal spectrum which includes cold-adapted fauna such as wolverine (Gulo gulo), reindeer (Rangifer tarandus), woolly mammoth (Mammuthus primigenius) and woolly rhinoceros (Coelodonta antiquitatis), with reindeer being the predominant herbivore taxon15. Furthermore, sedimentological analyses suggest a drop in temperature from layer 9 to the start of layer 7 based on a pronounced decrease in organic carbon and total nitrogen content13, lending support to the decreasing temperatures from ~48 to 43 ka cal bp reported here.

Our results show that climatic conditions throughout the LRJ occupations, even during the earliest phase ~48–45 ka cal bp, were characterized by temperatures substantially below modern-day conditions. Although a direct contextual connection through anthropogenic modification cannot be established, the chronological overlap between the direct dates of H. sapiens remains and anthropogenically modified bone fragments with those of the equid individuals that yielded low temperature results indicates that H. sapiens faced subarctic to tundra climatic conditions, probably even those of the severe cold climatic phase 45,000–43,000 cal bp. Zooarchaeological analysis suggests that this presence was characterized by ephemeral occupations, either due to short occupation, task-specific site use or small group sizes15, although the direct radiocarbon dates of H. sapiens remains suggest intermittent site visits across at least a thousand years. Such a site-use pattern, perhaps in the context of frequent movements between sites, may have been a response to the subarctic steppe environment reconstructed for the LRJ. Micromorphological evidence for increased fire use in layer 8 compared to layers 9 or 7 (ref. 13) could also be indicative of a behavioural adaptation to the cooling climate. Owing to the few human modifications on faunal remains, we cannot determine the seasonality of site occupations by LRJ H. sapiens groups15. However, more long-term palaeoclimatic records and limited evidence from other LRJ sites (see below) indicate that a subarctic steppe or tundra landscape would have extended over large areas in central Europe, where human groups would have faced similar conditions during potential seasonal movements. The ephemeral H. sapiens presence at the site implies that the climatic data probably also cover phases of site formation where humans were absent but we argue that the temporal overlap between directly dated equids, H. sapiens remains and anthropogenically modified bones and the reflection of millenial-scale climatic variability in our record does suggest that we can broadly characterize the climates faced by LRJ H. sapiens as cold to very cold during most of the LRJ formation period.

The association of H. sapiens with the LRJ suggests a rapid range expansion across the northern European plain as far as the British Isles13, which may have been enabled by the resilience to cold conditions and success in steppe environments documented here. Direct environmental evidence from other LRJ sites is sparse. Nonetheless, the association of LRJ material with cold-adapted fauna at Grange Farm, United Kingdom, and Schmähingen, Germany, as well as biomarker and pollen evidence for climate cooling and open landscapes during the Jerzmanowician occupation of Koziarnia Cave, Poland, suggests that association with cold climatic conditions may be a more common feature of the LRJ than previously noted41,42,43. Genetic data and technological analyses allude to a potential connection between the Ranis LRJ to populations and technocomplexes further east13, including potentially the Initial Upper Palaeolithic44, which is associated with a cold-climate H. sapiens presence at Bacho Kiro Cave, Bulgaria9. Cold-steppe environments that provided open landscapes and supported large herds of prey fauna may have actively supported a rapid dispersal of these connected populations across the northern and eastern European Plain45,46. At the same time, our study joins increasing recent evidence for a more complex patchwork of early dispersals of our species in different periods and in more diverse ecological settings than previously appreciated1,2,3,5,9,10, raising the question of whether pioneering groups of H. sapiens may not be more accurately described as climatically resilient generalists.

Methods

Study design, materials and sampling

A multi-isotope study design was chosen to reconstruct a variety of climatic, environmental and ecological aspects of the LRJ and Upper Palaeolithic deposits of Ilsenhöhle in Ranis. A multi-isotope approach also has substantial benefits in reducing equifinality in the interpretation of each isotopic proxy. Oxygen stable isotope analysis was chosen as the main palaeoclimatic proxy, while strontium isotope analysis serves to confirm that sampled animals did not undergo long-distance migrations that could affect the δ18O signal (Supplementary Texts 2 and 3). Carbon and nitrogen stable isotope analyses were conducted to reconstruct dietary ecology, the structure of the plant biome and water availability in the past. We add to this aspect using zinc stable isotope analysis, a non-traditional stable isotope proxy with potential to elucidate herbivore dietary ecology (Supplementary Text 4).

A total of 16 equid teeth were selected for stable isotope analysis and radiocarbon dating (Supplementary Table 1). Only fully formed and mineralized teeth were considered and first molars (M1) were excluded to prevent the influence of mother’s milk consumption on oxygen isotope ratios. Identification of tooth position was achieved with the help of an experienced equid tooth specialist and teeth where an M1 identification could not be confidently excluded were not sampled. Specimens were obtained both from the collections of the 1930s excavation by W. Hülle (sample numbers starting with ‘R’, n = 14) and the 2016–2022 excavation by the TLDA and the MPI-EVA (sample numbers starting with ‘16/116’, n = 2). Teeth were chosen to obtain data on the lower part of the depositional sequence from the black layer (TLDA/MPI-EVA: 6 black, Hülle VIII) downward including the LRJ occupations and adjacent layers (Extended Data Fig. 1) and sequentially sampled to yield subannually resolved stable isotope data (Supplementary Text 5). In the Hülle excavation, this encompasses layers VIII (Schwarze Schicht), IX (Mittlere Braune Schicht), X (Graue Schicht) and XI (Untere Braune Schicht), where layer X is associated with the LRJ technocomplex. These layers correspond to the depositional sequence from layer 6 black to layer 14 in the TLDA/MPI-EVA excavation, where the LRJ is associated with layers 8 and 9 (ref. 13). Layer information in the Hülle faunal collection is recorded using deposit colour (for example, ‘Graue Schicht’, meaning grey layer) and approximate depth. For the two brown layers identified by Hülle in the lower depositional sequence (IX and XI), these labels can include colour variations (for example, Braune Schicht, Schokobraune Schicht and Rotbraune Schicht), which probably reflect stratigraphic information but also colour differences between site areas, while layer positions (for example, ‘mittlere’ and ‘untere’, meaning middle and lower) are almost always omitted. Owing to sloping terrain and compression of deposits by rockfall in some areas of the site13, depth information is often of limited use in assigning layer designations. Colour and depth descriptions were used to assign layer association as best as possible but all equid specimens were also directly radiocarbon dated using dentine or mandible bone collagen samples to confirm their chronological position.

Because of the predominant role of carnivores in accumulating the faunal remains found in the Ranis LRJ and Upper Palaeolithic deposits and the palimpsest nature of the deposits, the link between faunal stable isotope data and H. sapiens activity at the site is less direct than for sites where the faunal assemblage is predominantly anthropogenically accumulated. Moreover, we rely on a comparison of direct radiocarbon dates of the equids analysed for stable isotopes with those of the archaeological layer boundaries, direct dates of H. sapiens skeletal remains and anthropogenically modified faunal fragments to establish the archaeological context for the equid remains sourced from the Hülle collection. This approach, while unavoidable due to the characteristics of the site, carries some uncertainty in relating the isotopic climate evidence to periods of H. sapiens site occupation. Indeed, it is most likely that the climatic data generated in this study cover both periods of H. sapiens presence at Ranis and periods where humans were absent. We use χ2 tests and agreement indices of the OxCal Combine function to test the chronological agreement between the direct dates of the equid specimens yielding the climatic data on the direct dates of H. sapiens remains and anthropogenically modified bones from LRJ contexts to test the probability of a link with H. sapiens presence as best as possible.

Samples of ~300–600 mg of dentine or mandible bone were obtained for collagen extraction from tooth roots when available. If roots were not preserved, pieces were cut from lower sections of the tooth crown and tooth enamel was mechanically removed before demineralization. For two specimens (16/116-123510 and 16/116-124286), adhering mandibular bone was available and sampled instead of tooth dentine. Sample pieces were removed using a diamond-coated rotary disk after cleaning of surfaces using air abrasion (Supplementary Text 5).

In addition to the equid specimens chosen for δ18O, δ13C, δ15N, δ66Zn and 87Sr/86Sr analysis, a total of 24 tooth enamel specimens representing a variety of herbivore, omnivore and carnivore taxa were chosen for further δ66Zn and 87Sr/86Sr analyses to explore patterns across the food web (Supplementary Table 1). These specimens were obtained from the Hülle collection from contexts thought to correspond to the brown layer IX. Given the documentation of the finds from the brown layers in the collection described above, this sample probably represents to some degree a mix of specimens of different LRJ and Upper Palaeolithic stratigraphic units (Supplementary Text 1 and Supplementary Fig. 1) dating between ~48 and 36 ka cal bp. While less than ideal, analysis of the directly dated equid specimens shows that diachronic changes in δ66Zn are too small to affect broader dietary patterns and trophic relationships in the food web (Supplementary Text 4). Tooth enamel samples from these specimens were obtained either as powder or piece samples in a positive pressure Flowbox following methods described in Supplementary Text 5. In some cases, several teeth from the same mandible were sampled to obtain sufficient tooth enamel.

Oxygen stable isotope analysis

Tooth enamel powder samples were converted to silver phosphate for oxygen isotope analysis of bioapatite phosphate using digestion with hydrofluoric acid, followed by crash precipitation of silver phosphate47,48 (Supplementary Text 5). Following recommendations in ref. 49 we did not use an oxidative pretreatment before silver phosphate preparation. Oxygen isotope delta measurements of Ag3PO4 were conducted in triplicate using a high-temperature elemental analyser (TC/EA) coupled to a Delta V isotope ratio mass spectrometer via a Conflo IV interface (Thermo Fisher Scientific) (Supplementary Text 5). Oxygen isotope delta values were two-point scale normalized to the VSMOW scale using matrix-matched standards calibrated to international reference materials and scale normalization was checked using three separate quality control standards. Details of normalization standards and quality control outcomes can be found in Supplementary Text 5. Average reproducibility of sample replicate measurements was 0.25‰.

Inverse modelling and palaeotemperature estimation

Before seasonal palaeotemperature estimation an inverse model following ref. 50 was applied to sinusoidal δ18Ophos time series to remove the time averaging and amplitude damping effects caused by the extended nature of tooth enamel mineralization and the sampling procedure (Supplementary Text 5). It should be noted that this inverse model does not account for the successive decrease in tooth growth and mineralization speed that is known to occur in horses towards the completion of tooth formation. Seasonal amplitudes reconstructed here, therefore, probably represent minimum amplitudes (Supplementary Text 5). Following recommendations developed in ref. 9, summer peak and winter trough values were obtained from the inverse modelled δ18O curves and processed for palaeotemperature estimation, while mean annual temperatures were estimated on the basis of unmodelled annual means. Summer peak and winter trough values were first identified by visual inspection in the original unmodelled δ18O time series (Supplementary Fig. 2) and corresponding areas on the inverse model outcome were used to yield corrected summer δ18O and winter δ18O values. Unmodelled annual means were calculated as the mean of unmodelled summer peak and winter trough values following ref. 9. Detailed information on the modelling procedure is provided in Supplementary Text 5 and we provide all associated code and data, including the parameters used for each model run and the model outcomes for all specimens of this study in the associated online repository at https://osf.io/wunfd/.

Following methods in ref. 51, air temperature estimates were derived via two regression steps using the empirically determined relationships between (1) tooth enamel δ18Ophos and drinking water δ18O (δ18Odw) and (2) δ18O of precipitation (δ18Oprecip) and air temperature. Regression relationships were established using modern calibration datasets of equid tooth enamel δ18O and drinking water δ18O as well as temperature and δ18Oprecip data from meteorological measurement stations. Details on modern calibration datasets and the conversion procedure are described in Supplementary Text 5. All calculations with the specific conversion equations can be reproduced using the data and code provided in the associated online repository at https://osf.io/wunfd/. Printed conversion equations based on the same data and excel files to conduct equivalent conversions are also available in ref. 9.

To confirm that herbivore δ18O values reflect climatic influences without ecological or behavioural biases, some studies recommend the use of several taxa from the same archaeological units. Owing to a lack of teeth from other taxa suitable for sequential sampling (for example, large bovids) this was unfortunately not possible in this study (Supplementary Text 2). However, equid δ18Odw values have been shown to be in excellent agreement with those of other sympatric taxa (Supplementary Text 2).

Collagen extraction and radiocarbon dating

Collagen was extracted from the equid teeth in the Department of Human Evolution at the MPI-EVA using HCl demineralization, NaOH humic acid removal, gelatinization and ultrafiltration steps, following the protocol in refs. 52,53. (Supplementary Text 5). The suitability of the extracts for dating was assessed on the basis of collagen yield (minimum requirement ~1%) and the elemental values54, as reported in Supplementary Text 5 and Supplementary Table 2. All extracts were characteristic of well-preserved collagen (Supplementary Table 2) so were submitted for 14C dating via accelerator mass spectrometry. Three of 16 collagen extracts were graphitized and dated at the Curt Engelhorn Center for Archaeometry gGmbH (CEZA, laboratory code: MAMS), while the remaining 13 extracts were dated at the Laboratory for Ion Beam Physics at ETH Zurich, Switzerland (laboratory code: ETH; Supplementary Text 5). Aliquots of a background bone (>50,000 bp) were pretreated and dated alongside the equid samples to monitor laboratory-based contamination and were used in the age calculation of the samples. The 14C dates were calibrated in OxCal 4.4 (ref. 55) using the IntCal20 calibration curve56. Uncalibrated 14C dates (14C bp) are reported with their 1σ error and, in the text, calibrated ranges (cal bp) are reported at the 2σ range (95% probability). Calibrated ages at the 1σ range (68% probability) can be found in Supplementary Table 2. All dates have been rounded to the nearest 10 years.

Carbon and nitrogen stable isotope analysis

Subsamples of collagen extracts were analysed for their elemental composition (%C, %N) and carbon and nitrogen stable isotope composition using a Flash 2000 Organic Elemental Analyser coupled to a Delta XP isotope ratio mass spectrometer via a Conflo III interface (Thermo Fisher Scientific; Supplementary Text 5). Samples were analysed in duplicate and stable isotope delta values were two-point scale normalized using international reference material IAEA-CH-6, IAEA-CH-7, IAEA-N-1 and IAEA-N-2 for δ13C and δ15N, respectively. Two inhouse quality control standards were used to check scale normalization and evaluate analytical precision. Replicate sample measurements and measurements of the in-house standards indicate a measurement precision of 0.1‰ or better for both δ13C and δ15N. Details of calibration methods and quality control indicators can be found in Supplementary Text 5. Elemental composition and C/N ratios used for quality checks of collagen integrity are reported in Supplementary Text 5 and Supplementary Table 2.

Strontium and zinc stable isotope analysis

Zinc and strontium extraction were both conducted on ~10 mg of tooth enamel powder or pieces. In the case of sequentially sampled equid teeth, two subsamples per specimen, representing the summer and winter seasons (based on δ18O sinusoid peaks and troughs), were selected for δ66Zn and 87Sr/86Sr analysis. Samples were processed using standard acid digestion and column chromatography purification protocols following refs. 37,57,58. (Supplementary Text 5). Isotopic measurements were conducted using a Neptune Multi-Collector Inductively Coupled Plasma Mass Spectrometer (MC-ICPMS, Thermo Fisher Scientific). Procedural blanks and aliquots of quality control standards were processed alongside each batch of samples to quality check wet chemistry and isotope measurement quality. Details of the instrumental setup, scale normalization and quality control indicators can be found in Supplementary Text 5. A subset of samples was analysed in duplicate with an average reproducibility of 0.000008 for 87Sr/86Sr and 0.01‰ for δ66Zn.

It has been shown that dental enamel reliably preserves biogenic strontium and zinc isotope values (see Supplementary Text 4 for details on the preservation of δ66Zn in fossil tooth enamel). As an additional diagenetic check we evaluated isotope measurements against elemental concentration data to confirm the preservation of biogenic isotopic ratios. A lack of relationship between the two in our results indicates that incorporation of diagenetic Zn or Sr is unlikely (Supplementary Figs. 14 and 15).

Reporting summary

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