Inside the Denisova Cave in Siberia

Ancient hominin mtDNA
We detected ancient hominin mtDNA in 175 samples (24%), covering nearly all layers in all three chambers (Fig. 1a–c, Extended Data Figs. 3a, b, 5b). Four samples showed evidence for the presence of predominantly one haplotype and yielded sufficient mtDNA fragments to reconstruct mtDNA consensus sequences that are more than 80% complete (Supplementary Information sections 5, 6). Three of the sequences (samples E202 and E213 from East Chamber (layers 11.4 and 11.4/12.1) and sample M65 from Main Chamber (layer 19)) group with Neanderthals in phylogenetic trees built with previously published hominin mtDNAs (Supplementary Information section 7), specifically with Denisova 5, Denisova 15, Mezmaiskaya 1 and Scladina I-4A (Fig. 1d, Extended Data Fig. 6a). The fourth sequence (sample M71 from Main Chamber (layer 20)) is of the Denisovan type and falls basal to Denisova 2 and Denisova 8, albeit with low bootstrap support (Extended Data Fig. 6b). The most complete mtDNA sequence (over 99% of the genome reconstructed) for a Neanderthal from Main Chamber (M65) has a genetic age estimate of 140 ka (95.4% highest posterior density interval of 181–98 ka) (Supplementary Information section 7), consistent with the time of deposition of layer 19 (151 ± 17 to 128 ± 13 ka; here and below, uncertainties on optical ages are given at 95.4% probability)8.

For the remaining 171 samples, we assigned mtDNA fragments to specific hominin groups by counting the number of fragments that support lineage-specific states at diagnostic sites that distinguish between modern human, Neanderthal and Denisovan mtDNA genomes. We distinguished three Neanderthal lineages: the Sima de los Huesos lineage (representing Neanderthals who lived approximately 430 ka in Spain and whose mtDNA is most closely related to that of Denisovans)18,19; the Hohlenstein–Stadel (HST) lineage, which falls basal to all other Neanderthal mtDNAs20; and the ‘typical’ Neanderthal mtDNA, known from all other Neanderthals. The presence of ancient modern human mtDNA was evaluated by restricting the analysis to deaminated fragments to mitigate the effect of present-day human DNA contamination. We identified Denisovan and typical Neanderthal mtDNA in 79 and 47 samples, respectively (based on 54–9,093 unique hominin mtDNA fragments), and modern human mtDNA in 35 samples (based on 55–2,200 deaminated fragments) (Fig. 1a–c, Extended Data Fig. 3a, b). We detected DNA from two hominin groups in ten samples, either within a single library or across libraries that were prepared from independent subsamples in some cases (Extended Data Fig. 6c, Supplementary Information section 4). In addition, we identified one sample (M76 from Main Chamber (layer 20)) containing hominin mtDNA fragments that support the branch shared by HST and typical Neanderthal mtDNA, but neither of the branches defining those lineages. This signal cannot be created by mixing mtDNA fragments from Neanderthals, Denisovans and ancient or present-day modern humans. On the basis of simulations with ancestralized Neanderthal mtDNA, the mtDNA in this sample is compatible with the presence of a previously unknown Neanderthal mtDNA lineage that diverged from typical Neanderthal mtDNA between 255 and 230 ka, 20 to 45 thousand years after the split of the HST and typical Neanderthal mtDNA lineages (Supplementary Information section 9).

The oldest hominin mtDNA recovered—identified as Denisovan—originates from a sample in Main Chamber layer 21, which began to accumulate 250 ± 44 ka. This provides the earliest genetic evidence for hominin occupation in Denisova Cave; Denisova 2 was found in layer 22.1, but is probably intrusive from an overlying layer and has an estimated age of 194–123 ka5. Among all 223 samples from the early Middle Palaeolithic layers in Main and East Chambers, 50 contained evidence for Denisovan mtDNA and only three (all from layer 20 in Main Chamber) for Neanderthal mtDNA. Two of these (M174 and M235) contain typical Neanderthal mtDNA and are from areas in which small-scale mixing with overlying sediments may have occurred8; the third (M76) is from the middle of the layer, and carries the previously unknown Neanderthal mtDNA lineage. These results point to Denisovans as the first and principal makers of the early Middle Palaeolithic assemblages, which are older than 170 ± 19 ka. Consistent with this interpretation, the detection of Neanderthal mtDNA in a sediment sample from early Middle Palaeolithic layer 14 in East Chamber in the pilot study13 was due to an incorrect assignment, which was later corrected to middle Middle Palaeolithic layer 11.4 in this chamber8 (Supplementary Information section 2). Our results also suggest that Neanderthals first occupied Denisova Cave towards the end of the early Middle Palaeolithic and may therefore have contributed to the production of these assemblages in their later stages.

Forty out of 173 samples from the middle Middle Palaeolithic layers in Main and East Chambers (deposited approximately 160–60 ka) yielded Neanderthal and/or Denisovan mtDNA, with both present in six samples (Fig. 1a–c). DNA from both groups also occurs in the deformed Middle Palaeolithic layers in South Chamber (Extended Data Fig. 3b). Notably, sediments deposited between 120 ± 11 and 97 ± 11 ka in Main and East Chambers produced no traces of Denisovan mtDNA, whereas 12 samples contained Neanderthal mtDNA. This suggests that only Neanderthals may have occupied the cave during that period, and possibly for most of Marine Isotope Stage (MIS) 5 (Fig. 2).

Fig. 2: Timeline of hominin and faunal mtDNA presence, archaeological phases and environmental records at Denisova Cave.



a, Baikal Drilling Project 1996 composite biogenic silica record of diatom productivity in Lake Baikal, a proxy for regional annual temperature28. b, c, Summary data for Main Chamber (b) and East Chamber (c). Start and end ages for layers and corresponding data are based on the common time scale in Extended Data Fig. 1. Time intervals in white represent gaps in the stratigraphic sequence, or were not sampled. Archaeological phases (early Middle Palaeolithic, middle Middle Palaeolithic, Initial Upper Palaeolithic and Upper Palaeolithic) follow the colour scheme in Fig. 1 and Extended Data Fig. 1. Genetic data for hominins are from Fig. 1 (excluding samples M174 and M235 from Main Chamber, which are thought to be out of context), and for dominant hyaenid and ursid populations from Extended Data Fig. 10. Symbols for specific hominin mtDNA lineages (Fig. 1d) are inset in white. Environmental conditions are inferred from pollen records and skeletal remains of vertebrate fauna8. Relative proportions of bovid, canid, equid, hyaenid and ursid mtDNA are from Extended Data Figs. 8, 9.

Only ancient modern human mtDNA was detected in the Initial Upper Palaeolithic and Upper Palaeolithic layers in Main Chamber (layers 11.4 and above, deposited 44 ± 5 to 21 ± 8 ka), except for one sample from Initial Upper Palaeolithic layer 11.2 that yielded Neanderthal mtDNA (Fig. 1c). The association between Initial Upper Palaeolithic assemblages and the appearance of modern humans is further supported by the recovery of modern human mtDNA from a sample from Initial Upper Palaeolithic layer 11 in South Chamber, which was deposited after 47 ± 8 ka (Extended Data Fig. 3a). The situation in East Chamber is more complex: Denisovan, Neanderthal and ancient modern human mtDNAs were recovered from Initial Upper Palaeolithic layer 11.2, and Neanderthal and ancient modern human mtDNA from Initial Upper Palaeolithic layer 11.1 (Fig. 1a). Given these results and the recovery of two Denisovan fossils (Denisova 3 and Denisova 4) from layers associated with Initial Upper Palaeolithic assemblages, we cannot discount the possibility that—in addition to modern humans—Denisovans and Neanderthals may have been present during the period of Initial Upper Palaeolithic production9,10,11.

For 34 out of 37 samples that yielded 100 or more deaminated hominin mtDNA fragments, we identified similarities with specific Neanderthal and Denisovan mtDNA genomes using a k-mer-based approach21,22 (denoted by symbols other than circles in Fig. 1a–c, Extended Data Fig. 3b, Supplementary Information section 8). This analysis revealed that the early Middle Palaeolithic layers in Main and East Chambers and the earliest middle Middle Palaeolithic layer in East Chamber, which span the period between 250 ± 44 and 146 ± 11 ka, contain Denisova-2- and Denisova-8-like mtDNA fragments. This contrasts with the Denisovan mtDNA recovered from middle Middle Palaeolithic layers deposited after 80 ± 9 ka, which yielded assignments to Denisova-3- and Denisova-4-like sequences, as did a sample from South Chamber. These results suggest a transition in mtDNA sequences sometime in the 146–80-ka interval, possibly reflecting different Denisovan populations. Our results also align well with the modelled ages of Denisova 2, Denisova 3 and Denisova 4, and with the relative age of Denisova 8 inferred from molecular dating5. Sediments deposited between about 130 and 100 ka (and possibly longer, given the subsequent time gap of 20 millennia)—during MIS 5—contain mtDNA and fossil evidence only of Neanderthals, with Denisova-11-like mtDNA sequences only appearing in sediments deposited after 80 ka.