Evidence from Tinshemet Cave in Israel suggests behavioural uniformity across Homo groups in the Levantine mid-Middle Palaeolithic circa 130,000

Main

Studies on the mid-Middle Palaeolithic (mid-MP; ~130–80 thousand years ago (ka)) document an increased pace of out-of-Africa hominin dispersals and provide a growing body of evidence that southwest Asia was the stepping stone for an Asia-wide expansion of Homo sapiens^1,2. The Levant during this period is marked by exceptional evidence of synchronous human morphological variability within the region, including fossil evidence of highly variable populations^3,4,5,6. This variability is expressed on the inter-site level with archaic Neanderthal-like fossils reported at Nesher Ramla⁵ and H. sapiens-like fossils at Skhul and Qafzeh caves^3,4,7, between the Skhul and Qafzeh fossils^8,9,10,11, and on the intra-site level within the Qafzeh human fossil assemblage^{6,7,8,11,12,13,14}. This variability has been a source of major debate about the anatomical affinities of Levantine mid-MP hominins. Skhul and Qafzeh hominins have been variously seen as variable populations of archaic H. sapiens^7,15,16,17, distinct human types or hybrids^{3,8,11,13,18,19}.

High human morphological diversity in the Levantine mid-MP is accompanied by abundant evidence of complex social behaviour such as intentional burial of the dead, the interment of grave goods and ochre use in burial sites^{3,4,20,21,22,23,24}, which predates all other such evidence in the global record by tens of thousands of years. At present, the Levantine mid-MP behavioural evidence is unevenly spread across sites, resulting in long-lingering uncertainties in interpreting regional variability in material culture and its behavioural implications.

The four key Levantine sites that include human-fossil-bearing deposits and are broadly assigned to the mid-MP—namely, Qafzeh, Skhul, Tabun and Nesher Ramla—are currently characterized by a lack of coherence. To date, the most complete package of material markers of technological and symbolic human behaviours derives from Qafzeh Cave. Excavations in this site have uncovered seven burials^4,21, associated with grave goods and abundant ochre pieces^22,23, and large lithic assemblages that demonstrate the technological dominance of the centripetal Levallois method²⁵. This knapping technology is considered a defining characteristic of mid-MP lithic technology in southwest Asia^{5,25,26,27,28,29,30}.

The assignment of two other mid-MP-affiliated sites, Skhul and Tabun Layer C, to a mid-MP stage of cultural development has long been a matter of debate^{25,26,28,31,32,33,34,35,36}, which has drawn on a range of considerations from material culture, chronology and human anatomical remains. The affiliation of Skhul and Tabun Layers B and C lithic assemblages with those of Qafzeh remains uncertain due to their small size, non-systematic early twentieth-century collection methods and inconsistency in the dating results. Various authors continue to argue for the absence of chronological overlap among these mid-MP-associated strata^{37,38,39,40,41,42}.

Additionally, the recently uncovered, large lithic assemblage of Nesher Ramla reveals that the Levallois centripetal reduction system was the primary technology employed at this site^28,29,43. While mid-MP Nesher Ramla may closely align with Skhul and Qafzeh chronologically²⁹, its human fossils belong to a palaeodeme associated with archaic Neanderthal-like hominins⁵. The south Levantine mid-MP thus represents a pivotal testing ground for issues involving the technological, symbolic, anatomical and genetic repercussions of local population admixture processes.

Here we report on the recent discoveries at Tinshemet Cave, Israel, which is located only 10 km from the open-air site of Nesher Ramla. We systematically evaluate the likelihood that these sites, together with Qafzeh, Skhul and Tabun C, belonged to a coherent regional cultural tradition widely shared among hominin groups of diverse anatomical affiliations. Tinshemet Cave has yielded large lithic assemblages dominated by centripetal Levallois technology associated with numerous Homo spp. remains, including fully articulated skeletons, and abundant ochre chunks. Excavations at the cave put into perspective the question of lithic regional technological variability, shedding light on the unifying role of the centripetal Levallois reduction system across the mid-MP southern Levant and revealing the widespread occurrence of a novel repertoire of complex social behaviours. This study reveals tight connections between mid-MP lithic technological behaviour and the local development and elaboration of social and symbolic behaviours within one coherent and uniform cultural complex.

Site formation, stratigraphy and chronology

Tinshemet Cave (Mugharet Al Watwat⁴⁴) is situated on the east bank of Nahal Beit Arif in central Israel (Fig. 1 and Extended Data Fig. 1)⁴⁵. The cave is composed of a Terrace and three chambers, with the deepest one now featuring an open chimney (Fig. 1 and Supplementary Section 1). Preliminary inspection of the cave by M. Stekelis in 1940 documented the presence of Middle and Upper Palaeolithic lithic artefacts⁴⁴. Our excavations revealed that MP deposits extend across much of the exterior Terrace and the First Chamber of the cave, whereas the second and third chambers of the cave lack evidence of human occupation.

**Fig. 1: Tinshemet Cave, geographical location, cave plan and stratigraphy.**

The main activity area of the site was the Terrace, which was probably partially roofed during the human occupation. The Terrace consists of 2–2.5-m-thick cemented sediments composed of varying amounts of calcite, clays and quartz, with the first originating mainly from wood ash, fragments of the cave wall and secondary calcite. The clays originate primarily from reworked terra rossa soil above the cave; the quartz grains are of aeolian origin, most likely from the active dunes along the coastal plain to the west^46,47 (Supplementary Sections 2 and 3). The sediments also contain carbonated hydroxylapatite that primarily originates from bones, carnivore coprolites and meagre authigenic phosphate nodules. The deposits were divided into three units—A, B and C (Figs. 1 and 2), with Unit A further subdivided into five layers (A1–A5) and Unit B into two layers (B1a and B1b) on the basis of differences in the density of the archaeological remains and the presence/absence of rock fragments that have collapsed from the cave ceiling and walls (Supplementary Section 2).

**Fig. 2: Tinshemet Cave: major characteristics.**

The sedimentological sequence begins with the archaeologically poor Unit C. This unit includes a small number of lithic artefacts, carnivore coprolites and no traces of wood ash, and it is likely to represent either ephemeral or no human occupation. Unit C rests on a phosphatized limestone bedrock, indicating the degradation of bat guano and carnivore coprolites—that is, a roofed cave environment. According to absolute dates, the sediments of Unit C accumulated at a slow rate (Fig. 2). Well-preserved pollen uncovered in Unit C indicates a Mediterranean open forest with fairly wide-spaced trees, small shrubs and herbs dominated by evergreen oak (Supplementary Section 4). In contrast, the overlying Units A and B of the Terrace deposits show evidence of intensive human occupation, with the sediments consisting of recrystallized wood ash and an accumulation of other fire residues such as microcharcoal, carbonized and calcined bone fragments, and ash pseudomorphs (Extended Data Fig. 2 and Supplementary Section 3). Unit A includes more secondary calcite and is more indurated than Unit B. Cementation of the remains of combustion activities probably occurred by water percolation downwards and along cracks, where secondary calcite can be detected. The sediments are bioturbated (primarily insect and plant root turbation). Therefore, clear combustion features (such as hearths) were not detected. The intensive use of fire at the site is also evident from the high frequencies of burnt lithics.

Within the cave, the First Chamber is divided into inner and outer parts. The outer part of this chamber, located near the present-day dripline, is composed of anthropogenic sediments similar to those found on the Terrace, albeit less cemented. The sedimentological sequence of the First Chamber is about 1 m thick down to the bedrock (Fig. 1 and Supplementary Section 2), and the bulk of the deposits are assigned to Layer III (Fig. 1). Layer III shows mineralogical composition similar to that of Unit A on the Terrace and is likely to represent the same stratigraphic unit⁴⁶ (Supplementary Section 3).

The inner part of the chamber contains a soft, dark brown, non-anthropogenic accumulation composed primarily of clay and quartz (Dark Brown (DB) Layer). This layer consists of reworked terra rossa soil that infiltrated the cave from its interior chimney and entrance (Figs. 1 and 2 and Supplementary Sections 2 and 3). The DB Layer contains some artefacts, which were probably redeposited from in situ deposits in the outer part of the First Chamber.

Thermoluminescence dating

The thermoluminescence (TL) dating method applied to burnt flints from Layer III in the First Chamber provides ages with relatively low dispersion, ranging from 84 to 109 ka, with a mean age of 96.3 ± 7.2 ka (Fig. 2, Extended Data Table 1 and Supplementary Section 5). Burnt flints from the Terrace demonstrate a much higher dispersion of ages than the cave deposits, mainly because of two samples from Unit B1b, which give ages of 173 ± 20 and 153.8 ± 12 ka, and two samples from Unit C, which give ages of 124 ± 14 and 117 ± 14 ka (mean age, 120.4 ± 14 ka). These four samples indicate that the bottom of Unit B1b and Unit C may represent an earlier human presence at the site. Apart from these four samples, the remaining six TL samples from the Terrace yielded a mean age of 98.3 ± 8.0 ka, which is indistinguishable from the mean value obtained for the flints from the First Chamber (Extended Data Fig. 3). The burnt flints from the non-anthropogenic DB Layer in the inner part of the First Chamber (1, 22 and 14-1) probably represent reworked material and provide a mean age of 87.9 ± 7.7 ka.

Optically stimulated luminescence dating

Optically stimulated luminescence (OSL) ages were obtained from quartz grains extracted from the sediments. The OSL ages for the Terrace Unit A layers yielded a mean age of 106.5 ± 16 ka (Fig. 2, Extended Data Table 2 and Supplementary Section 6). A single age from Unit B1a is 106 ± 7 ka, while the ages from Unit B1b show a high dispersion with ages of 106 ± 6, 129 ± 6 and 153 ± 14 ka. A single sample from Unit C yielded an age of 129 ± 11 ka. The OSL age for Layer III in the First Chamber is 114 ± 8 ka. The DB Layer provided an average age of 71 ± 8 ka, suggesting that these sediments accumulated after human use of the cave had ceased.

Uranium-series dating with laser-ablation inductively coupled plasma mass spectrometry

Images of radioisotopes relevant for uranium-series (U-series) dating were produced using a laser ablation (LA) module and a high-resolution inductively coupled plasma (ICP) mass spectrometry (MS) system. Two types of samples were dated: veins of secondary calcite and snail shells. Two samples were taken from thin calcite veins deposited in fissures formed in the cemented deposits of Unit A on the Terrace. The isochron ages computed using the Java code⁴⁸ are 15.6 ± 0.2 ka and 35.6 ± 1.1 ka. These ages suggest that the infilling of the fissures with calcite is a relatively recent Late Pleistocene phenomenon (Supplementary Section 7). Snail shells taken from Unit B revealed U values distributed according to the growth layers of the shells⁴⁹, with the variability ranging from a few ppb up to 200–300 ppb (Supplementary Section 7). The ²³²Th content is very low and homogeneous inside the shell, indicating that it was incorporated during growth. Three samples provide ages (113.3 + 3.3 / − 3.5 ka, 105.5 ± 4.1 ka and 138.9 + 6.4 / − 6.0 ka) that are in agreement with the TL and OSL age distributions (Fig. 2, Extended Data Fig. 3 and Extended Data Table 3). A fourth shell gives an older age (>272 ka), with higher uncertainties mainly related to the asymptotic nature of the growth curve of the ²³⁰Th/²³⁴U activity ratio.

Cryptotephra

The results of the cryptotephra (microscopic volcanic ash) analysis show low shard abundance both on the Terrace and in the First Chamber (Supplementary Section 8). Compositional data were successfully obtained from three glass shards, one representing Unit A on the Terrace and two representing Layer III in the First Chamber. The shard from Unit A and one of the shards from Layer III exhibit evolved rhyolitic signatures that are similar enough to suggest a correlation between Unit A on the Terrace and Layer III in the First Chamber. The measured samples also exhibit a glass chemical signature similar to volcanoes from the Aegean Arc, such as Kos, as well as eruption centres in Central Anatolia (for example, Acıgöl⁵⁰). The Tinshemet Cave shards show a good level of similarity to the recently published tephra from the Marine Isotope Stage (MIS) 5 Nahal Aqev site, Israel³⁴ (Supplementary Section 8).

The micromammals

Altogether, eight taxa were identified (Supplementary Section 9). Most of these taxa typically occur in mid–late Pleistocene fossil assemblages from southwest Asia (Supplementary Section 9). However, two of the taxa (Mastomys batei and Arvicanthis ectos) represent key index fossil species (that is, fossiles directeurs) of the regional biostratigraphy, which alternately occupied the region during different stages of the MP³³ and were last documented within this region during MIS 5 (ref. ⁵¹). The presence of these two taxa, together with their relative and rank abundance in the assemblage, identify the Tinshemet Cave deposits as an analogue of mid-MP Qafzeh Layers XV–XXV. Despite the rather small size of the Tinshemet Cave assemblage and attendant limitations for quantitative analysis, the observation that one tooth of either Mastomys or Arvicanthis was uncovered for every 18 specimens of all other rodent taxa (a 1:18 ratio) suggests temporal affiliation with Qafzeh. The equivalent ratio for all Qafzeh layers combined is 1:20 (ref. ⁵¹), while the early-MP deposits of Misliya and Hayonim Layer E show a remarkable difference with ratios of 1:70 and 1:600, respectively^33,51.

Summary of the chronostratigraphic results

Dosimetric (TL and OSL) and U/Th dating methods indicate that Tinshemet Cave was inhabited during the MIS 5 (130–80 ka; Fig. 2 and Extended Data Fig. 3). The major stages of occupation of the cave are represented by Units A and B on the Terrace and Layer III at the front of the First Chamber. The geoarchaeological, geochronological and biochronological evidence suggests that Unit A on the Terrace and Layer III in the outer part of the First Chamber belong to the same stratigraphic unit. They exhibit similar sedimentological characteristics that include a greyish deposit containing abundant clasts of flint, bones and rocks, variably cemented by partial dissolution and recrystallization of wood ash (Extended Data Fig. 2). The only difference is the presence of carnivore coprolites in Layer III, probably due to its position under the roofed part of the cave. According to TL ages, the main phase of human occupation on the Terrace and in the First Chamber of the cave is dated to 97 ± 9 ka (mean age) (Extended Data Table 1). This mean age is in the range of the U-series ages obtained for Unit B and the OSL ages, which are 106.5 ± 16 ka on average for Unit A, 106 ± 7 ka for Unit B1a and 114 ± 8 ka for Layer III in the First Chamber (Extended Data Tables 2 and 3). These results suggest that the Terrace (Units A and B) and the outer part of the First Chamber (Layers III) were inhabited at roughly the same time. The contemporaneity of the deposition is also supported by the volcanic glass shards that suggest a potential temporal correlation between Unit A on the Terrace and Layer III in the First Chamber.

The occurrence of the African-tropical rodent taxa Mastomys and Arvicanthis suggests that similarly to Qafzeh Cave, Tinshemet Cave is likely to represent mid-MP Biozone II of the regional biostratigraphy (130/120–80 ka; Qafzeh Layers XV–XXV³³). Both taxa at Tinshemet Cave occur within the First Chamber’s deposits (Layer III) and in Units A and B of the Terrace, supporting the contemporaneity of the deposits. The lower part of the Terrace sediment sequence, especially Unit C, probably represents an earlier phase of accumulation (late MIS 6/early MIS 5).

The behavioural repertoire of the Tinshemet Cave hominins

The lithic assemblages

The Tinshemet Cave excavation yielded abundant lithic material from both the First Chamber and the Terrace. In the framework of this study, assemblages from Layer III in the First Chamber (n = 809) and from Units B and C on the Terrace (n = 1971) were studied (Fig. 1 and Extended Data Fig. 4). The results show close technological similarity between these assemblages, suggesting that they are likely to represent the same cultural unit. Local Mishash Formation Campanian flint dominates the assemblages (53% on the Terrace and 70% in the First Chamber; Supplementary Section 10). Flint from indeterminate (non-local) sources forms 15% of the lithic assemblage of the Terrace and 14% of the lithic assemblage of the First Chamber. Thermal alterations were recorded on 44% of artefacts on the Terrace and 36% in the First Chamber. The technology in all excavated areas predominantly comprised flake production by the centripetal Levallois method (Fig. 3 and Extended Data Table 4). Cortical flakes and different types of core trimming elements were identified both on the Terrace and in the First Chamber, suggesting that at least part of the knapping activities took place on-site (Extended Data Fig. 4a).

**Fig. 3: Lithic artefacts from Tinshemet Cave.**

The centripetal Levallois reduction system is represented by cores and debitage, including both by- and end-products (Fig. 3). Levallois by-products are represented by débordant flakes, pseudo-Levallois flakes and points, and striking platform preparation flakes (Fig. 3). Typical Levantine MP lithic components such as Kombewa flakes and core-on-flakes were identified in the assemblages. While Levallois points are exceptionally rare on the Terrace (1%), their frequency is slightly higher in the First Chamber (5%). The frequency of retouched tools is very low: 1.3% on the Terrace and 3.1% in the First Chamber (Fig. 3). The assemblage from the First Chamber shows a minor dominance of single sidescrapers (Extended Data Fig. 4b). In addition, two ‘bulb retouchers’ were identified, one in each of the assemblages. Use-wear analysis identified evidence of use on 19 of the 44 tools examined (Supplementary Section 11). A wide variety of activities were documented, including whittling, scraping, cutting/sawing and perforating bones; whittling and scraping wood; and butchering (Supplementary Section 11).

Ochre

The excavations at Tinshemet Cave yielded more than 7,500 ochre fragments of different sizes, shapes, textures and colours (Fig. 4 and Supplementary Section 12). There is a clear dominance of red to orange hues (75.7%), followed by brown (16.7%), yellow (5%) and purple hues (2.6%). Some of the exploited pieces have a combination of lighter to darker red tonalities that suggest heating. The density of ochre pieces larger than 5 mm per excavated volume of sediment increases notably from Unit C to Unit A (Fig. 2), with the highest densities recorded in the layers with human burials (Units A and B). Ochre fragments were often found close to the burials, as in the case of a 4–5 cm large chunk of red ochre found between the legs of the Tinshemet 2 individual (Fig. 4b).

Six different classes of ochre (I to VI) have been identified on the basis of the grain size, grain morphology, and the presence/absence and size of quartz crystals (Fig. 4 and Supplementary Section 12). The characteristics of the ochre allow us to infer at least four different sources including sandstones (with closest available source as far as the Galilee in the north, at least 60–80 km from the site), limestone and iron-rich formations possibly as far as the central Negev (>100 km to the south). The exploitation of distant sources suggests that great efforts were invested to obtain these materials, hinting at the importance of ochre in the human activities at the site.

Faunal remains

The studied faunal sample (n = 191) is dominated by ungulate remains (88%; Extended Data Table 5 and Supplementary Section 13). Among the ungulates, four taxa are almost evenly represented: aurochs (Bos primigenius), equids (Equus spp.), Mesopotamian fallow deer (Dama mesopotamica) and mountain gazelle (Gazella gazella). The equid remains are variable in size and probably represent two coexisting forms. Wild boar (Sus scrofa), red deer (Cervus elaphus) and caprine remains were also found, the latter most likely belonging to the bezoar goat, Capra aegagrus. A rhinoceros phalanx completes the ungulate repertoire. Two large carnivore remains were found; one was identified as a calcaneus of a hyeanid. Isolated remains of porcupine (Hystrix sp.), small to medium-sized birds (yet unidentified), spur-thighed tortoise (Testudo graeca), and lizard and snake vertebrae complete the taxonomic spectrum. Taphonomic analysis has yet to be performed, but humans may be considered the primary (but probably not exclusive) accumulation agent; some hammerstone percussion notches were observed on limb bones, as well as a few cases of carnivore damage.

Differences were observed between faunal samples from three depositional contexts: the Terrace (Units A–B), the outer part of the First Chamber (Layers II–III) and the DB Layer. While our sample remains small, these contexts significantly differ in the relative abundance of the four most common taxa (χ² = 13.17, d.f. = 6, P = 0.04; weak to moderate effect size, Cramer’s V = 0.24). The variability is driven by the absence of equids in the DB Layer and the relatively low Dama proportions in the Terrace. Overall, the Terrace (all units combined) is dominated by gazelle, aurochs and equids, while Layer III of the First Chamber also includes Dama in a significant proportion (Extended Data Table 5 and Supplementary Section 13).

Human remains

Human remains, totalling five individuals, were discovered in different layers of Tinshemet Cave. Notable findings include two fully articulated skeletons, three isolated skulls in varying states of preservation, assorted appendicular bones and isolated teeth. Ongoing investigations focus on the remains recovered from Units A and B1a on the Terrace and in Layer III of the First Chamber. Tinshemet Cave fossils are still under study and are preliminarily characterized as Homo spp.

The comprehensive analysis of several of the burials at Tinshemet Cave, including a burial of a nearly complete adult skeleton (Tinshemet 2; Fig. 5) from Unit B1a, a burial of a nearly intact child skeleton (Tinshemet 1) from Layer III of the First Chamber and a partially exposed burial (Tinshemet 8) from Unit A2, has unravelled important findings about funerary practices and interment processes at Tinshemet Cave. These include formal inhumation of both adults and children, and the fetal or sleeping position of the dead (bodies lying on their sides with highly flexed legs, arms bent towards the chest and face, and the head facing down; Fig. 5). The dominance of primary burials (on the basis of the fully or partial articulated skeletons) suggests immediate burial after death. Additional insights into the burial practices at Tinshemet Cave suggest grave inclusions—that is, large pieces of ochre (Fig. 4b).

**Fig. 5: Human burials at Tinshemet, Qafzeh and Skhul Caves.**

Discussion

Lithic technologies as a measure of inter-population connectivity

The centripetal Levallois technology is the dominant lithic reduction method used at Tinshemet Cave, as at all other mid-MP sites (Fig. 6a), with unidirectional convergent Levallois point production representing an auxiliary knapping method (Supplementary Section 14)^{28,29,34,52,53,54}. The neighbouring northern part of the Arabian Peninsula shows similar technological features^27,55 and is probably part of the same techno-complex²⁹. Precursor early MP (~250–140 ka) lithic industries of the Levant exhibit an entirely different technological set than that of the Levantine mid-MP, with the dominance of blades produced by non-Levallois laminar methods and the production of Levallois blanks employing convergent unidirectional and bidirectional methods^56,57,58. The Levantine early MP industries display no evidence of the systematic use of the centripetal Levallois method, which became dominant in the Levant in the mid-MP (Fig. 6a and Supplementary Section 14). The centripetal Levallois method persisted into the late MP (80–45 ka), where it occurs alongside other Levallois as well as non-Levallois and laminar methods, but rarely dominates the lithic assemblages (but see refs. ^59,60,61,62). The Levantine mid-MP lithic technological homogeneity differs from the neighbouring regions, such as the technologically heterogeneous East Africa during the Middle Stone Age^{28,29,63,64,65}. The technological homogeneity that occurs across Homo groups in southwest Asia suggests high inter-population connectivity within this region during the mid-MP²⁹.

**Fig. 6: Centripetal Levallois method and large-game hunting in the Levantine MP.**

Ochre—a marker of socially mediated activities

During the mid-MP, ochre emerged as an important cultural substance whose exploitation is likely to have involved trips to distant sources and heat treatment (Supplementary Sections 12 and 15). The use of ochre in the Levantine MP was reported only in mid-MP sites roughly contemporaneous with Tinshemet Cave. Its use has been linked to funerary practices²² and has been taken as evidence of a human behavioural mode probably undergirded by the emergence of symbolic thought. For instance, at Qafzeh Cave, the quantity of ochre increases in Layer XVII, where most burials are located. Some particular associations between ochre and burials were also reported (that is, burial 8), whereas the concurrent disappearance of both burials and ochre above Layer XVII at Qafzeh Cave is striking²². The ochre chunks at Skhul Cave were retrieved from Layer B2, where human burials were found. Ochre also occurs in mid-MP sites that lack burials, including Nesher Ramla, Israel, where it is found throughout the stratigraphic sequence, and at Nahr Ibrahim, Lebanon⁶⁶.

Heating of the ochre to obtain red colour, reported at Skhul, Qafzeh and Tinshemet, is likely to be related to the importance of red in symbolic communication^{22,24,67,68,69,70,71}. Ochre stain was also found on perforated marine shells from Qafzeh, which are likely to have been used as personal ornaments²³. An additional aspect of ochre use is the long-distance trips involved in its acquisition²⁴.

Exploitation of large ungulates

From the perspective of game exploitation, the mid-MP displays a consistent pattern that differs from those of the early and late MP in the region (Supplementary Section 16). The data show a more even representation of species and increased reliance on large-bodied ungulates in the mid-MP compared with the early and late stages of the Levantine MP. The proportion of the largest ungulate prey taxa, aurochs and equids, at Qafzeh, Skhul, Tabun C, Tinshemet Cave and the open-air site of Nesher Ramla^72,73,74 is consistently greater than during the earlier and later stages of the MP, especially in respect to cave sites^{75,76,77,78,79} (Fig. 6b). A game exploitation system focused on the largest prey species in the landscape may reflect a shift in human prey choice or may be linked to a major behavioural change in carcass transport patterns associated with different settlement and site-use strategies. It is also possible that the specialized funerary activity that occurred in some of these sites impacted the archaeofaunal pattern—for example, by focusing on consuming the largest animals in the landscape (an issue discussed in a much later context⁸⁰).

Funerary practices

The Tinshemet burials present an exceptional opportunity for comparative analysis of burial customs across prominent mid-MP Levantine sites. All three major Levantine mid-MP sites with skeletal remains, Tinshemet Cave (MNI = 5), Skhul Cave (MNI = 10) and Qafzeh Cave (MNI = 25), fall within the multiple-burials site category^4,81, signifying that a major activity at these sites involved the deliberate treatment and deposition of the deceased. The cultural responses to death were similar in all three sites. All three sites show remarkable similarities in how people disposed of their dead. These characteristics include the highly flexed position of the deceased (Fig. 5) and the placement of various objects inside the grave, including animal remains and chunks of ochre^{4,7,20,21,52,82,83}, which demonstrate similarities in symbolically mediated behaviour across these sites and suggest that the intricacy of funerary rites in the mid-MP surpasses that of late MP burials of Levantine Neanderthals exposed at Amud and Kebara Caves, dating to 70–50 ka^84,85. The mid-MP burials at Tinshemet, Qafzeh and Skhul caves represent the earliest instances of intentional Homo burials, predating the formalized burial practices in Europe and Africa by tens of thousands of years^86,87.

Behavioural uniformity across Homo groups

The consolidation of a unified behavioural package during the ~50,000-year-long Levantine mid-MP deserves particular attention as it stands in stark contrast to the heterogenous morphology of the local hominin population. The MP Levant was inhabited by three hominin groups: archaic Neanderthal-like Homo, Neanderthals and H. sapiens^{3,5,6,88,89,90,91,92,93,94}. The evidence suggests that some of these hominins coexisted. The site of Nesher Ramla yielded Homo remains that bear a combination of Neanderthal and archaic Homo features, which chronologically overlap with the Qafzeh and Skhul fossils^{5,29,38,95,96}. The stratigraphic position and chronometric age of the Tabun C1 pre-Neanderthal/Neanderthal skeleton are controversial, although some authors suggest a mid-MP age for the fossil^{39,97,98,99,100} (but see ref. ¹⁰¹). The identification of the Skhul/Qafzeh group as a single population of archaic H. sapiens is also debated. The identification of the Skhul specimens as modern humans was supported by some scholars^15,17,102, while others identified them as a separate morphological group or as hybrids^{3,18,19,103,104,105}, emphasizing their peculiar characteristics in the morphology of the chin and the brow ridges^{3,8,9,10,106,107}, and the morphological differences between the Skhul and Qafzeh fossils^8,9,10. The morphological variability of the Qafzeh hominins is also notable^6,12, including fossils that show characteristics that do not align with H. sapiens^8,10,13,14. Noticeably, all Qafzeh fossils occur in the same archaeological context, close to each other within the site’s stratigraphy.

The morphological variability of the Levantine MP hominins has become a source of heated debate on their taxonomic identity and on the question of H. sapiens and Neanderthal interactions in the region. The hypotheses raised include coexistence with no interactions¹⁰⁸, alternating occupations^17,109, brief episodes of engagement followed by extinction³², and interactions that facilitated cultural transmission and assimilation across species^29,106,110. The final scenario is supported by recent studies of the facial and dental morphology of the Qafzeh and Skhul fossils that show that the Qafzeh and Skhul hominins fall well within the expected morphology of hybrid populations^{11,18,111,112}, supporting previous similar views^{3,13,19,104,106,107}. A recent study based on the lithic technological behaviour of the Nesher Ramla hominins demonstrated that Nesher Ramla, Skhul and Qafzeh shared core reduction technologies, suggesting that cultural diffusion and interaction across Homo populations is the most likely reason for the close cultural similarity between these sites²⁹. The behavioural evidence from Tinshemet Cave further supports a high level of inter-population connectivity in the Levantine mid-MP. The summary of the MP human behavioural features in the Levant presented in Fig. 7 reveals a well-defined mid-MP entity with a widely shared behavioural repertoire found across the four major Levantine sites (Qafzeh, Skhul, Nesher Ramla and Tinshemet) with chronological overlaps (Supplementary Sections 14–16). These strata broadly share a uniform lithic technology, the use of ochre, a large-ungulate hunting pattern, the presence of articulated human remains and the presence of grave goods or non-utilitarian artefacts^113,114. We suggest that the association between behavioural uniformity and high human biological variability could be a result of intensifying social interactions and admixture among African H. sapiens and Eurasian Neanderthal-like hominins in the mid-MP Levant. It is now increasingly understood that several taxa provided population sources for the Levantine MP and that population influxes from different sources chronologically overlapped in a way that created opportunities for genetic and social admixtures^{5,11,18,29,115}.

**Fig. 7: Cultural and environmental characteristics of the Levantine MP.**

Methods

Excavation

The fieldwork consisted of the excavation of 5-cm-thick spits in 0.5 m × 0.5 m squares. Each flint artefact longer than 2 cm, bones longer than 3 cm, ochre larger than 0.5 cm, ground stone tools, charcoals, shells and all human remains were recorded with a total station. The excavation of the cemented deposits was conducted using pneumatic tools (PaleoTool), die grinders, electric hammers and hand chisels. The deposits were softened with water. This combination of techniques allowed us to extract complete artefacts and bones and perform 3D measurements.

Geoarchaeology

Mineralogical identification was conducted on 307 sediment samples using Fourier transform infrared spectroscopy with the well-established KBr method (ref. ¹¹⁶, Ch. 12). Phytoliths and wood ash pseudomorphs were extracted and quantified from 68 sediment samples (~22% of the total samples) following the procedures of Katz et al.¹¹⁷ and Gur-Arieh et al.¹¹⁸. Phytoliths and wood ash pseudomorphs were sought to obtain information on plant use and combustion activities by hominins. Small (<5 mm) fragments of bones (n = 141) found in sediment samples were assigned to burn colour codes, which were verified using Fourier transform infrared spectroscopy following the guidelines outlined in Stiner et al.¹¹⁹.

Monolithic blocks for micromorphological examination were sampled along the sedimentary sequence, impregnated by polyester resin, cut and ground to 30 μm thickness according to established procedures. A total of 43 thin sections were produced and studied using a Nikon petrographic microscope.

The particle size distribution of the detrital grains (nine carbonate-free samples) was analysed using a Malvern 3000 laser diffraction analyser following Crouvi et al.¹²⁰. The samples were then examined using a binocular, scanning electron microscopy, and energy dispersive spectroscopy.

TL dating

Among the flint samples showing evidence of past heating and therefore selected for TL dating¹²¹, 20 have signals that had been reset in the past according to preliminary TL tests. Ten samples come from the Terrace, from squares AD23, AE23, AF23, AF24 and AE25; they are associated with Units B1a, B1b and C. The other ten samples were unearthed from Layer III in the First Chamber, at the entrance of the cavity, in squares AO19, AO20, AM20, AM21 and AN21, and from the DB Layer (squares AO18 and AN18).

Current gamma and cosmic dose rates have been measured with dosimeters¹²² buried for months at different locations and at different depths in the Terrace and the First Chamber. Their analysis following Kreutzer et al.¹²³ lead to gamma dose rates ranging from 357 to 454 µGy/a for the Terrace, and from 361 µGy/a at the entrance of the First Chamber to 551 µGy/a deeper in the cavity (Supplementary Table 2). For each flint, the gamma dose rate used for the TL age calculation was then obtained by averaging the values recorded by the nearest dosimeters, and considering a past moisture content of 10 ± 3% on average for the Terrace sediments and of 15 ± 4% for the sediments of the First Chamber. Cosmic dose rates were computed following Prescott and Hutton¹²⁴ considering, for the Terrace, the thickness of sediments at the time of the first excavation, and the protection induced by the shelter for samples coming from the First Chamber.

For each flint, the alpha and beta dose rates were deduced from its radioactive content in U, Th and K, which was determined via ICP-MS on a fraction (100 mg) of the powder prepared for the TL measurements.

Overall, the total dose rate ranges from about 700 to 3,800 µGy/a, including an external fraction that varies greatly from sample to sample and represents 22–81% of the total (Extended Data Table 1).

After extraction by sawing with a diamond saw, the core of each flint, which was thus externally irradiated only by gamma and cosmic rays, was prepared following the recommendations of Valladas¹²⁵. TL measurements were done using coarse (100–160 µm) grains dispersed on stainless-steel cups and performed with a TL reader to which a calibrated ⁹⁰Sr beta source was fitted and used for delivering artificial doses¹²⁶. The additive dose technique was applied to describe the growth of the TL signal, integrated between 330 and 430 °C, as a function of the total dose received (natural + artificial). The same technique was also applied to a fraction of the natural powder reset by heating at 350 °C for 90 min: the measurement of regenerated TL signals thus allowed the description of the growth curve in the low dose range. The palaeodose values deduced from the analysis of the individual growth curves range from 56 to 477 Gy (Extended Data Table 1) and increase as a function of the internal (alpha and beta) dose rate (Supplementary Fig. 1).

OSL

OSL uses the dosimetric properties of quartz to date the most recent exposure of the grains to sunlight. When buried, quartz accumulates a signal due to the natural radiation emitted from radioactive elements in the sediment. When exposed to sunlight, this signal is reset (‘zeroed’), and it accumulates once more after burial. This signal is measured in the lab, and the OSL intensity is compared to laboratory-induced OSL created by artificial beta doses (the equivalent dose, or D_e). The burial age is determined by the ratio between the D_e and the environmental dose rate measured in the surrounding rocks.

The limestone bedrock at Tinshemet Cave does not contain any quartz. The source of all the quartz found in the deposits in the cave and on the Terrace (which had also been part of the cave) is aeolian, blown in from the coast or the southern desert by strong winds. The grain size of this quartz is silt to very fine sand, and the latter is sufficiently coarse to be used for OSL coarse-grain dating¹²⁷.

Twenty-four samples for OSL dating were collected at the site during three excavation seasons from all areas and layers (Extended Data Table 2). Due to the indurated character of the sediments on the Terrace, samples were collected as blocks cut from the breccia using a diamond saw. The blocks were further cleaned in the laboratory under safe amber light. In the inner chamber, the softer sediments were sampled either by drilling with a hand-held auger or, when rocks interfered, with a gardening trowel. All samples were placed immediately in light-tight black bags to avoid any exposure to sunlight. One or two additional samples were collected from each location for dose-rate evaluations.

The surfaces of the breccia blocks were first cleaned of grains that might have been exposed to the sun by soaking them in diluted hydrochloric acid (8% HCl), which dissolved the cementing calcite, followed by scrubbing with a metal brush. The cleaned blocks were then placed in light-tight bags and crushed using a pneumatic press. The diminuited material was then placed in diluted HCl for several days until a sufficient amount was dissolved and enough quartz was released from the breccia. The samples were washed and sieved to 75–125 µm, the most common sand-size fraction in the sample.

The sediment samples from the chamber were processed using routine laboratory procedures. Wet sieving was used to isolate the 75–125 µm fraction, followed by soaking in HCl to remove carbonates, followed by rinsing and drying. This fraction from all the samples was passed through a Frantz magnetic separator to remove heavy minerals and some feldspars. The grains were etched in 40% HF for 40 min to dissolve the remaining feldspars and remove the outer rim of the grains, followed by soaking overnight in 16% HCl to remove any fluorides that may have precipitated. After thorough rinsing and drying, the samples were ready for measurements.

The additional samples for dose-rate evaluations were crushed, homogenized and split, and a representative ~50 g subsample was powdered. The concentrations of uranium (U) and thorium (Th) were measured using ICP-MS, whereas potassium (K) was measured by ICP–optical emission spectrometry. For each sample, two or more analyses were carried out, either from duplicate samples taken at the site or from duplicated measurements of one powdered fraction. Alpha, beta and gamma dose rates were evaluated from the concentrations of the radioactive elements using attenuation factors of Nambi and Aitken¹²⁸. Gamma dose rates were also evaluated from aluminium oxide dosimeters placed in the locations of the flint samples (see ‘TL dating’) in cases where they were within 30 cm of the OSL sample. The gamma dose rates to samples were weighted from the dosimeters and the values calculated from the chemical analyses, according to the distance between the OSL sample and the dosimeter. Moisture contents of 10 ± 3% and 15 ± 4% were estimated for the Terrace and chamber sediments, respectively. Cosmic dose rates were evaluated following Prescott and Hutton¹²⁴ and considering the additional 0.6 m of topsoil removed during excavations on the Terrace, and the rock overburden for the samples from the chambers.

The quartz D_e values were measured on 17–20 2-mm aliquots for each sample. Five dose points were used to construct the dose response curve, including a zero point and a repeated (recycling) point. The sample average D_e was calculated using the central age model, which assumes that all grains in a sample have the same depositional age and absorbed the same dose¹²⁹.

A dose recovery test over a range of preheat and cutheat temperatures showed that a recovery ratio of 0.99 can be obtained under a preheat of 10 s at 260 °C, a test dose of 8 Gy and a test dose cutheat of 5 s at 220 °C (Supplementary Fig. 1a). These measurement conditions were used for all samples.

The OSL signal of all the samples is bright and decays to background level within 2–3 s; the dose response curves can be well fitted with an exponential + linear fit, and recycling ratios are mostly within 5% of unity. The distribution of the measured aliquots is usually normal with low overdispersion values, justifying the use of the central age model (Extended Data Table 2 and Supplementary Fig. 2).

U-series dating on flowstones and snail shells

The U-series dating method was applied to two flowstones from the top of Unit A and to six fragments of snail shells unearthed from the Terrace: two from Unit B1a (Tin19SSh1 and 2) and four from Unit B1b (Tin19SSh6 to 9). The samples were prepared following the protocol described in Martin et al.⁴⁸. All the samples were first embedded in resin (Araldite 2020), cut into slices in a horizontal plane and dry polished to obtain a flat surface suitable for LA sampling.

The measurement of isotopes relevant for U-series dating (²³⁸U, ²³⁵U, ²³⁴U, ²³⁰Th and ²³²Th) was made by coupling a femtosecond LA system (Lambda 3, Nexeya SA/Amplitude System) with a sector field (SF) ICP mass spectrometer (SF-ICPMS Element XR, Thermo Scientific). LA sampling was carried out using adjacent sub-parallel transects of 30 µm height, with a horizontal displacement speed of 30 µm s⁻¹, resulting in 2D imaging of the isotopes’ distribution within each analysed section with a resolution of 30 × 30 µm² per pixel (one measure of each isotope per second). Two to five replicas were made for each sample, for a total ablated mass ranging from 2.6 to 4.9 mg for speleothems and 0.5 to 3.2 mg for snail shells. The FsLA SF-ICP-MS coupling sensitivity was optimized daily by measuring a certified glass (NIST612) while maintaining a U/Th ratio of 1.00 ± 0.05 to ensure efficient particle atomization into the SF-ICP-MS plasma. Oxide corrections (<0.23% for both ThO/Th and UO/O), mass bias correction and background correction were considered. The images were processed with ImageJ (v. 1.53)¹³⁰. Their analysis aimed to identify areas affected by recrystallization or polluted by exogenous materials (then defined as the region of exclusion) and to calculate isotope ratios and ages only in the preserved areas (regions of interest). For more details about the protocol used, equipment optimization and data processing, refer to A. Galy et al. (manuscript in preparation).

Cryptotephra

Samples for cryptotephra analyses were collected at 5 cm intervals from existing and freshly cleaned profiles under active excavation at the site. A total of 43 bulk sediment samples were taken from 5 sampling columns (Supplementary Section 8 and Supplementary Figs. 3 and 4). On the Terrace, within the cemented sediments (Supplementary Fig. 3), samples from Columns 1–4 were collected using a rotary power saw with a 5-cm-diameter blade. Columns 1 and 2 were located in Square AI24 (east profile), with Column 1 extending from the site surface to a depth of 40 cm (88.04–87.64 m datum) and Column 2 stepped immediately below, extending a further 40 cm (87.61–87.21 m datum) and reaching the contact between the very hard breccia and Unit B1. Column 3 was located in Square AD24 (west profile) in a transitional position between Column 2 and Column 4, spanning a 25 cm sequence of sub-units B1a and B1b (86.92–86.67 m datum). Column 4 included two overlapping sections, designated as Columns 4A and 4B, within Square AG25 (east profile). Column 4A extended 30 cm (86.78–86.48 m datum) from Unit B1a to the B1b/C contact. The adjacent Column 4B measured 20 cm (86.44–86.24 m datum), extending from the Unit B1/C contact a further 15 cm into Unit C. Within the cave’s First Chamber, Column 5 also comprised two overlapping sections, including Columns 5A and 5B (Square AM21, south profile), each spanning 30 cm (87.11–86.81 m and 87.26–86.96 m datums, respectively) of the unconsolidated sediment sequence designated as the ‘light brown’ Layer III (Supplementary Fig. 4).

In the laboratory, the samples were immersed in 10% HCL to dissolve the carbonate fraction. Residual material was passed through disposable nylon sieve meshes, retaining the intermediary fraction between 125 and 15 μm. This material was subjected to a standard density separation procedure based on Blockley et al.¹³¹, which preferentially removes organic detritus and concentrates volcanic glass shards of a more felsic chemical composition. The remaining material following the density separation phase was mounted onto glass slides using Canada balsam and examined under a polarizing light microscope. Glass shard identification was undertaken manually at ×100 and ×400 magnification. Glass shards for major and minor elemental analysis were extracted from a water mount using a mechanical ‘picking’ device and embedded in an epoxy resin. Major and minor elemental analyses were conducted using a Cameca Electron Probe Microanalyser housed at the Tephra Analysis Unit, University of Edinburgh. Probe conditions and procedures followed the guidelines of Hayward¹³².

Lithic technology and use-wear studies

The number of lithic artefacts retrieved so far at Tinshemet Cave is approximately 10,000 (>2 cm). The sample of lithic material studied here is about 30% of the total assemblage. The sample comprises ~800 artefacts from the squares most dense in lithic artefacts in the First Chamber (AM 21-22 and AN 21) and almost 2,000 artefacts from the densest squares on the Terrace (AF 23-24-25 and AE 23-24-25). Both samples represent complete assemblages excavated from about 1-m-thick archaeological deposits. For both assemblages, flint artefacts found within the topsoil layer were excluded from the analyses. The lithic study encompassed attribute analyses and general observations related to the chaîne opératoire concept, such as raw material identification, state of preservation, blank types and scar patterns. The typology follows the traditional Bordes list¹³³.

Forty-four flint artefacts were inspected to check the preservation of use-wear. Sample selection was done including bone retouched and unretouched items. All artefacts were analysed at the TraCEr laboratory (Neuwied, Germany) using a stereo-microscope (ZEISS SteREO Discovery V8), a digital microscope (Zeiss Smartzoom 5) and a metallographic microscope (Zeiss-Axio ScopeA1). The magnification ranged from ×10 to ×50 with the stereomicroscope and digital microscope and from ×50 to ×500 with the metallographic microscope. Extended depth-of-focus images were created using the post-processing software Helicon Focus (v. 8.0.2).

The analysis under low magnification focused mainly on the observation of macro-wear (scars and edge rounding) and on the possible identification of ancient residues. The analysis with a metallurgical microscope allowed a more detailed analysis of wear patterns, such as polish and striations¹³⁴. Interpretations relied on comparisons with published images found in the literature^134,135,136.

Micromammal remains

Remains of micromammals were retrieved for analysis from soft-sediment deposits; further analysis of material extracted from well-brecciated contexts will require considerably more time and effort. A fair sample size of 229 molar teeth of micromammals (rodents and insectivores) was obtained for a preliminary taxonomic description of the assemblage in the present study. This sample represents a relatively limited spatial coverage compared with the entire area of exposed archaeological deposits (retrieved from 9 out of the 150 m²). It is, however, representative of the different excavation contexts, including material from within and outside the present extent of the cave and from some of the excavation squares in which human remains were uncovered.

Palynology study

Eleven samples were taken during seasons 2021 and 2022 using sterile equipment to prevent outside contamination (Supplementary Table 4). Eight samples were recovered from archaeological layers (sample nos. 1–8), and one sample was collected from a compact layer located at the lowest archaeological stratum (sample no. 9). This sample was characterized by a yellowish-greyish colour. Two additional samples were collected to serve as controls (sample nos. 10 and 11): one sample from the surface sediment, 20 m northeast of the cave, and one sample of recent guano collected from the inner part of the cave.

Laboratory procedure

The pollen extraction process began with soaking the samples in 32% HCl for ten days, combined with 1 min of sonication per day. This physical–chemical treatment removes the calcium carbonates within the samples, loosens the different compact debris and dissolves the Lycopodium spores (a tracer added at the beginning of the process that enables us to calculate pollen concentrations¹³⁷). The samples were then rinsed with distilled water several times until pH 7 was achieved. Next, a density separation was performed using ZnBr₂ solution with a specific gravity of 1.95. After mixing and vortex, the samples were placed in an ultrasonic water bath for 2 min. After sonication, the samples were centrifuged for 20 min at 3,500 rpm (all other steps were followed by only 5 min of centrifuging at the same rpm). The floated suspension was sieved through a 150 µm mesh screen and rinsed with distilled water. After sieving (150 µm mesh screen), the unstained residues were homogenized and mounted onto microscope slides using glycerin. The control sample of the recent fruit bat guano (sample no. 11) passed through a short acetolysis process after sieving due to its high organic content.

Pollen identification

Pollen grains were identified under a light microscope at magnifications of ×200, ×400 and ×1,000 (oil immersion) to the most detailed possible systematic level. For pollen identification, a comparative reference collection at Tel Aviv University (Steinhardt Museum of Natural History) was used, in addition to pollen atlases^{138,139,140,141}. When possible, all the extracted pollen grains were counted and identified.

Reporting summary

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

Evidence from Tinshemet Cave in Israel suggests behavioural uniformity across Homo groups in the Levantine mid-Middle Palaeolithic circa 130,000–80,000 years ago

Main