Genome of a middle Holocene hunter-gatherer from Wallacea

Much remains unknown about the population history of early modern humans in southeast Asia, where the archaeological record is sparse and the tropical climate is inimical to the preservation of ancient human DNA1. So far, only two low-coverage pre-Neolithic human genomes have been sequenced from this region. Both are from mainland Hòabìnhian hunter-gatherer sites: Pha Faen in Laos, dated to 7939–7751 calibrated years before present (yr cal bp; present taken as ad 1950), and Gua Cha in Malaysia (4.4–4.2 kyr cal bp)1. Here we report, to our knowledge, the first ancient human genome from Wallacea, the oceanic island zone between the Sunda Shelf (comprising mainland southeast Asia and the continental islands of western Indonesia) and Pleistocene Sahul (Australia–New Guinea). We extracted DNA from the petrous bone of a young female hunter-gatherer buried 7.3–7.2 kyr cal bp at the limestone cave of Leang Panninge2 in South Sulawesi, Indonesia. Genetic analyses show that this pre-Neolithic forager, who is associated with the ‘Toalean’ technocomplex3,4, shares most genetic drift and morphological similarities with present-day Papuan and Indigenous Australian groups, yet represents a previously unknown divergent human lineage that branched off around the time of the split between these populations approximately 37,000 years ago5. We also describe Denisovan and deep Asian-related ancestries in the Leang Panninge genome, and infer their large-scale displacement from the region today.


5
Toalean stone artefacts (i.e., backed microliths). Layer 2 comprises a silty clay that is very similar to layer 1 in terms of sedimentary composition and colour, although it is noticeably drier. This stratum contains a rich aceramic Toalean stone artefact assemblage. The most diagnostic Toalean tool type consists of geometric backed microliths. Maros points were not recovered from this archaeological horizon. Layer 3 (silty clay) is a low-intensity occupational horizon that contains relatively few cultural remains compared with the immediately overlying and underlying strata (layers 3 and 4, respectively). Layer 4 is a thick silty clay unit containing a dense concentration of cultural and faunal remains, with the rich Toalean lithic assemblage including Maros points but no backed microliths (characteristic Toalean osseous points, often as not made on lower incisor roots of Suidae, were found throughout layers 2 and 4). Layer 4 yielded the deepest clear evidence for Toalean occupation in stratified deposits at the site. Layer 5 is a silty clay with a similar sedimentary composition to layer 4, but it contains fewer stone artefacts and faunal remains generally and no Maros points or other diagnostic Toalean tool types. Layers 6 and 7 consist of silty clay and sand layers, respectively, with abundant gravel and few cultural findings. Layer 8 comprises clay with fossils and lithic artefacts. Exposed clearly for the first time in 2019, layers 6-8 represent a transition to larger artefacts including limestone and volcanic materials. No macroscopically observable plant carbon remains or other materials amenable to 14 C dating were recovered from these lower strata, which are probably Late Pleistocene but are as-yet undated.

Recovered skeletal remains
The Toalean human skeleton from Leang Panninge was uncovered in 2015 within squares S16T6 and S17T6, and removed in a recovery excavation in 2018 [23][24] . The 2015 excavations were conducted in arbitrary 10-cm-thick spits and wet-sieved through a 3 mm mesh, to a depth of ~190 cm, at which point the human skeletal remains were encountered. The 2019 excavations took place between 20 July and 9 September. The backfilled squares, S16T6 and S17T6, were reopened and the deposit excavated in 5 cm spits to a depth of 200 cm (spit 19).
Square S16T6 was extended vertically an additional 130 cm to a maximum depth of 330 cm and S17T6 extended an additional 60 cm to a maximum depth of 260 cm. These trenches continued to be excavated in 5 cm spits and all spit labels were continued from 2015. Squares S16T7 and S17T7 were opened to the east contiguous with S16T6 and S17T6. Both squares were excavated to a maximum depth of 235 cm (spit 45). Artefact locations were recorded using a Leica total station. All excavated deposits were wet sieved through a 3 mm screen. 6 Descriptions of the human skeletal remains are based on original field documentation by the 2015 UNHAS/USM/Balar Sulsel team 23 , observations by BB who removed the sediment blocks with AMS in 2018, and a laboratory excavation of these blocks by DM. During the 2015 excavation of squares S16T6 and S17T6 a clear burial cut could not be identified in the largely homogenous sediment profile. However, the semi-articulated condition of the remains indicates a primary inhumation. It was also observed that the skeletal remains were partially covered over by five large water-worn volcanic cobbles and numerous smaller stones (Extended Data Figs. 2f,4). Much of the skull was represented, but in a fragmentary condition. Only the mandible, the four podials, and to a lesser degree the pelves were semicomplete. The left and right pelves were articulated with the fragmented sacrum and included well preserved left and right greater sciatic notches 23 . The skull and pelvic region were both extremely fragile; hence, it was necessary to remove these parts of the skeleton from the deposit en bloc for later processing in a laboratory. The remaining post-crania were poorly As observed during the 2015 and 2018 fieldwork 23 , the Leang Panninge skeleton was orientated in a north-south direction and lying in a tightly flexed or crouched position, with the skull facing left (towards the east) in the southern end of S16T6 (Extended Data Figs. 2f, 4). Most of the post-cranial remains were located in S17T6. The cranium was crushed postmortem, although the position and orientation of the cranial elements recovered suggest it was intact when interred. The associated cobbles appear to have been deliberately arranged over the human remains (Extended Data Figs. 2f, 4). Two of the largest cobbles, with a combined mass of 12 kg, flanked the skull 23 . A hand comprised of five fractured metapodials and a single phalanx was located below the largest of these (#1). A smaller cobble (#2) was placed immediately to the right (to the east) of the skull. Another of the largest cobbles (#4) 7 was positioned over the left foot. The southernmost cobble (#4) was placed over the pelvic region. The three cobbles in the pelvic region (#3, #4, and #5) had a combined mass of 13 kg.
A number of skeletal remains from non-human fauna were also excavated from spits 19 and 20 of squares S16T6 and S17T6 in the same context as the human skeleton. These included several phalanges, long bones, vertebrae, carpals and metacarpals of unidentified animal taxa as well as Bovidae teeth, Suidae astragalus, Siluriformes barbs, one Macaca sp. molar, various Chiroptera teeth, and 64 mollusc shells, including the freshwater gastropod Tylomelania (= Brotia) perfecta and several bivalves 23 . Tools made on bone or tooth, and over 2,000 stone artefacts -including around 13 modified flakes displaying clear attributes of Maros point technology (Extended Data Fig. 5i-k) -were also recovered 23 .

Toalean context of the human skeletal remains
The Leang Panninge individual was interred in a secure Toalean context, with distinctively Toalean artefacts recovered from the grave fill and surrounding and overlying deposits. As already noted, Toalean assemblages are commonly marked by the presence of distinctive artefact types, namely small points made on osseous materials, small stone Maros points, and/or backed microliths, occurring alongside more generic flakes and cores of generally small dimensions [1][2][3][4][5][6][7][8][9] . Maros points are distinctive and apparently unique in the world, consisting of pressure-flaked stone points made on a small flake blank with an indented base and/or regular denticulations around the remaining margins. None of these classic Toalean artefact types have ever been recorded in pre-Toalean sites, and hence they are used as a diagnostic cultural marker for this South Sulawesi technocomplex of ~8-1.5 kya 7 . A specialist analysis of the large lithic assemblage excavated from Leang Panninge is ongoing; however, tens of thousands of largely chert artefacts have been recovered from S16T6 and S17T6, including dozens of Maros points (e.g., Extended Data Fig. 5d-p), backed microliths, and bone/tooth points 23,25 . Among 4079 artefacts (total weight, 16.3 kg) recovered during excavation from spits 19 and 20 of S16T6 and S16T7, a total of 11 display diagnostic Maros point features (Extended Data Fig. 5i-k). While only two of these artefacts are classic Maros points, with an indented base and edge denticulations, the remaining specimens also have varying degrees of denticulations and represent incomplete or damaged Maros points.
Within the area of sediment removed en bloc around the skull of the individual in 2018, a total of 118 stone artefacts was recovered in association with the human remains, and these 8 were subjected to more detailed analysis. The majority of these stone artefacts (N = 71, 60%) were made from non-descript honey-to grey-coloured cherts common at the site. However, six were formed on a dark, fine-grained volcanic material with fine quartz veining that is exotic to the site. The distinctive appearance of the raw material of these particular artefacts strongly suggests that they were all struck from the same core during a single reduction event and likely deposited simultaneously, with minimal post-depositional disturbance.
In addition to this, two Maros points were recovered in direct association with the human skeleton during the laboratory excavation of the sediment blocks (Extended Data Figs. 2h-i, 5a-c). A chert Maros point (spit 19-20, square S17T6) was found directly associated with the pelvic area (~40 mm from the greater sciatic notch) of the Leang Panninge individual (Extended Data Figs. 2i, 5a-c). This artefact is a classic example of the tool type 8 , with a deep, well-defined basal notch and neat denticulations around the remaining margin. The point is 20 mm long, 17 mm wide and 3.5 mm thick. It should be noted, however, that the point has been broken by excessive heat and hence the first measurement probably only represents three-quarters of the original length. The stone contains embedded heat-induced potlids and a small number of potlid scars, further demonstrating that this Maros point was exposed to extreme heat. Differential gloss only occurs on the heat-damaged surfaces, indicating it was not deliberately heat-treated during manufacture and that the burning occurred after initial production and breakage. The artefact is pinkish red in colour, possibly the result of chemical change from the burning event. The point was produced using the classic Maros point reduction sequence, with pressure-flaked denticulations on the lateral margins converging to form the tip, and a basal notch at the proximal end of the flake blank. with minimal edge denticulations, also broken, was recovered from the skull block.

Faunal remains
The archaeological deposits at Leang Panninge are rich in non-human faunal remains, some of which may be attributed to human activities. Across the 11 excavation units, faunal assemblages are largely dominated by Chiroptera and Rodentia bones, sometimes articulated.
The Toalean deposits (layers 2-4), include the two endemic Suidae species Sus celebensis (Sulawesi warty pig) and Babyrousa sp. (babirusa), as well as Bubalus sp. (anoa), Ailurops ursinus, Strigocuscus sp., Varanus sp., Macaca sp., Tarsius sp., one possible Canis sp. bone 9 (S8T6, layer 2) 23 and members of the Anura, Aves, Osteichthyes, Serpentes, Viverridae, and Cervidae taxa 18-24 . The 2019 excavations revealed a near-complete Suidae skeleton in S16T7 spit 19 and 20, and extending into S17T7. Both species of Suidae, S. celebensis and babirusa, show signs of being butchered on site during the Toalean phase of occupation, in the form of processing marks on the bones and an assemblage dominated by sub-adult individuals 24 . High linear enamel hypoplasia levels were identified on Suidae teeth 24,26 . Modified faunal remains at Leang Panninge include osseous points, sometimes manufactured using Suidae incisors and canines 23 , a drilled shark tooth, and a small fragment of shell with a possible partial perforation 22 . Mollusc findings are dominated by specimens of the locally common freshwater gastropod T. perfecta and a small number of shells from the Bivalvia class 23 .

Dating evidence
In total, 13 AMS 14 C dates have been obtained on charcoal (N = 8), seeds (N = 3), and shells (N = 2) excavated from Leang Panninge, including three samples associated directly with the human burial context. These are reported in SI Table 1 (for additional contextual information, see Extended Data Figs. 2e, 3). All of the dated samples were hand-picked from the sediments under excavation -no samples were collected from sieve residues. The two dated shells were hand-picked from the layer 4 Toalean human burial feature under excavation in 2015. While exposing the skull area, several stone flakes and various other artefacts, including shells and non-human bones, were uncovered. An unidentified freshwater gastropod shell exposed near the human cranium (S17T6; spit 19, 181 cm depth) was collected for radiocarbon dating (Sample Wk-42781) using clean metal tweezers and aluminium foil to avoid any possible contamination. Another unidentified freshwater gastropod shell was collected using the same methods (Sample Wk-42782) during excavation of the burial area in S17T6 at spit 18 (170-180 cm). These two shells yielded AMS 14 C dates of 7423 ± 22 BP (8325-8176 cal BP) and 7008 ± 22 BP (7926-7748 cal BP), respectively (SI Table 1). Environmental offsets are unavailable to correct for the freshwater reservoir effect 30 , the magnitude of which is unknown in this region. Notably, the 14 C-dates on gastropod shells associated with the Toalean human skull are earlier than the 14 C age of a charred Canarium sp. seed (7264-7165 cal BP [Wk-48639]) recovered from the excavated sediment block containing the cranial bones, mandible, and dental remains (SI Table 1). It is therefore likely that the 14 C dates on freshwater shell are erroneously old -in the order of at least several centuries in this particular instance -owing to absorption of "dead" carbon from the ancient river or lake environment 30 . In this study, we inferred the age of the human burial 10 using the more reliable chronology obtained from 14 C-dating of charcoal and charred seeds (SI Table 1). We attempted to directly date the Leang Panninge burial by conducting AMS 14 C dating on the human petrous bone that yielded the ancient DNA. This effort was unsuccessful, however, owing to the lack of a sufficient amount of collagen in the petrous bone.

Morphological description of the Leang Panninge human skeleton
The Leang Panninge human remains (SI Table 2) are stored at the Archaeology Laboratory of the Archaeology Department (Departemen Arkeologi Fakultas Ilmu Budaya) at UNHAS, Makassar, South Sulawesi.   Table 3. The individual is mature enough to be treated as an adult and on that basis its sex may be determined as female, from the shape of the greater sciatic of the pelvis, the gracile status of the cranium, and the small and gracile status of the mandible. The right sciatic notch angle is ~80º, which is high within the female range (

15
Concerning the mandible, the weight of observations is consistent with a female status. These include bilaterally inverted gonia, with a narrow estimated bigonial breadth <91 mm, a small symphysis height <30 mm, a bilaterally slight anterior marginal tubercle, and a sulcus intertoralis that is slight on the right side and medium on the left side. One masculine feature, however, is a broad left ramus with a minimum ramus breadth >33 mm, producing an overall score of 8.5/18, which, if extrapolated to all 11 sexually dimorphic features in ref. 36    Overall tooth size can be summarised by multiplying the average lengths and breadths of each of the 32 teeth and summing the products to calculate overall tooth area 38 , applying this method to Asia-Pacific populations of determined sex (SI Table 6). Unsurprisingly, in any given population, males consistently have larger tooth area than females. Also, as expected, inconsistency would be better assigned to localised effects at the specific tooth sites observed.
Based on this approach (SI Tables 7 to 11), the strongest evidence is for a "D" event.
Although the evidence is not entirely consistent (SI Table 9), the right lower canine could also be included if we allowed for the defect recorded at its occlusal part of the crown, as could the right upper second molar and left lower second molar if we allowed for the defects recorded at their intermediate part of the crown. SI Table 9: Evidence for a D event as reflected in macroscopic enamel hypoplasia, following the system of ref. 31 (Table 6.3).    Table 9. A "D" event corresponds to a developmental age of three to four years old, and may have been caused by weaning at that age when the Leang Panninge individual was no longer protected by antibodies from breast milk.

SI
Numerous observations on the dental morphology of the Leang Panninge individual were possible, and these are presented in Additional observations on the teeth (see below).
Leang Panninge's expressions for focus tooth traits 45 are presented in SI Similarities to the Leang Panninge individual can be shown with a simple concordance statistic, which is the average of the percentage similarities for the traits in SI Table 12 where Leang Panninge does not have an asymmetric expression. In descending order to whole numbers, the results are: New Guinea 84%; Australia 81%; Melanesia 80%; Micronesia 78%; Jōmon 77%; prehistoric Southeast Asia, Polynesia 75%; recent Southeast Asia 74%; South Siberia 72%; recent Japan, China-Mongolia 67%; Northeast Siberia 65%; and American populations the least similar (e.g., "North and South America" 60%). These results confirm the inference that the Leang Panninge dental morphology is most similar to that of Sahul-Pacific populations, with the further indication of a greater similarity to Sunda-Pacific/Jōmon than to Sino-American populations. 23

Cranium
Reconstruction of the cranium proceeded as far as the right/median frontal, left zygomatic region, right and left temporal regions, right parietal, median occipital and left occipital condyle. Further joins could not be found and the general size and shape of the cranium is unknown. The recorded total weight of 78 grams of cranial bone and maxillary teeth (SI Table 2) excludes the right petrous portion of the tympanic, which had been collected for ancient DNA analysis. Italicised character states in the description below are those recommended 46 for distinguishing Aboriginal Australians from other populations.
In addition to the slight (right) zygomatic trigone, noted above, the frontal bone has a slight frontal crest on the right side, strong frontal bosses and absence of the median frontal ridge.
As reconstructed, the frontal bone is broad anteriorly, with fronto-malar breadth of ref. Panninge, the following procedure was followed: 1. Find all of the female crania scored for all of the characters scored for Leang Panninge 2. Focus on the female crania with a similar aggregate score to Leang Panninge (in practice, with a total difference of either 3 or 4 from Leang Panninge) 24 3. Focus on the characters for which the focus female crania either score identically with Leang Panninge (zygomatic trigone, orbital border rounding) or differ by at most one grade (median frontal ridge, parietal bossing, external occipital protuberance, supramastoid crest) 4. Accept a match with Leang Panninge for all female crania scored for these six characters whose expressions fall within the bounds described in (3) above 5. To extend the comparisons to include male crania, drop the requirement on zygomatic trigone, which is one of the characters used for sexing between males and females 35 .

Mandible
The observations recorded on the Leang Panninge mandible are detailed in Detailed observations on the mandible (see below). Detailed biometric analysis lies beyond the scope of this report. The mandible is U-shaped rather than V-shaped from above and below.
The mandible's small size can be shown by comparing its measurements with the means published 49 for two English cemetery series, Punjabi Indians (males only) and Aboriginal Australians. For three of the seven comparative measurements, Leang Panninge is smaller than the female range of means, and for another two measurements, it falls within the female (as well as the male) range of means (SI Table 14). 4. Accept a match with Leang Panninge for all mandibles scored for these six characters whose expressions fall within the bounds described in (3) above (SI Table 15). Around one-third of specimens classified as Aboriginal Australians and one-fifth of

SI
Melanesians have a mandibular morphology that matches Leang Panninge. This, however, is the case for less than one-tenth of Indo-Malaysian mandibles, again indicating that Leang Panninge does not lie outside of the recent Indo-Malaysian range of variation, but is more compatible with a southwest Pacific affinity.

Vertebral column
The vertebral column (above the pelvic girdle) is represented by a fragmentary axis vertebra and seven other fragments from the third to seventh cervical vertebrae, a seventh thoracic vertebra fragment and a process fragment to another thoracic vertebra, and a lumbar vertebra 27 fragment. None of these fragments are complete enough to provide information on the size or shape of the original vertebrae. In summary, the entire vertebral column (above the pelvic girdle) is sparsely represented, better so for the neck than the back.

Ribs
The ribcage is represented by fragments of the left second to fourth ribs and sixth to seventh ribs, and the right fourth to seventh ribs. None of these fragments are complete enough to provide information on the size or morphology of the original ribs. In summary, much of the ribcage is represented but in a very fragmentary manner.

Upper appendicular skeleton
The right clavicle is represented by two joining fragments that extend medially from the No measurements are available.
The right hand is represented by a cluster of bones that included the trapezium, the second metacarpal with the head recently fused to the distal shaft, the third metacarpal with a recently fused head, the fourth metacarpal with an unfused head, a shaft fragment to the fifth metacarpal, the third proximal manual phalanx with an unfused base, the fourth proximal manual phalanx distal shaft and head, the fifth proximal manual phalanx base and head, a shaft fragment to the second medial manual phalanx, a complete fifth medial manual phalanx with a fused base, and a complete distal manual phalanx with a fused base. A fragmentary capitate was also found with this cluster of bones, and although its morphology strongly 28 suggests it was a left capitate, it is more likely to be the right capitate based on its association.
The length of the fifth medial manual phalanx is 19.7 mm, which is shorter than the average of 21.1 mm recorded for Australian Aboriginal males 51  29

Lower appendicular skeleton
The right leg is represented by two joining fragments tentatively assigned to the distal tibia, and the left leg by a compressed fragment of the distal fibula. The left foot is represented by the cuboid and an articulating navicular fragment, all of the metatarsals and proximal pedal phalanges, and the first distal phalanx. Where measurable, the lengths of the Leang Panninge metatarsals and pedal phalanges generally approximate the Australian Aboriginal male averages but in some cases are about 3 mm shorter (SI Table 16). The right foot is represented by the right second metatarsal missing its unfused head, the right fourth metatarsal which is complete after being joined to its unfused head, the right third proximal phalanx missing its unfused base, and three fragments to the shaft of another proximal pedal phalanx. The length of the right fourth metatarsal (60.3 mm) is short compared to the average length of 65.8 mm recorded for Australian Aboriginal males 51 .

Sampling and extraction
The petrous bone of the Leang Panninge individual was sampled by sawing along the margo superior partis petrosae (crista pyramidis) and drilling 56.

Mitochondrial analysis
All unique mitochondrial reads (16.72x coverage) were used to reconstruct the mtDNA consensus sequence and estimate mitochondrial estimation using schmutzi 70 with quality 39 filters 0, 10, 20 and 30. Mitochondrial contamination was estimated to be 2 ± 1%. Since the mitochondrial/nuclear ratio of the shotgun sequence was low (61.78), the mitochondrial contamination can serve as a proxy for nuclear DNA contamination 71 . The mitochondrial haplogroup was ascertained as M1 using Haplofind 72 but with a low confidence score (SI Table 17).

PMD filtering
To confirm that quality control results were not affected by present-day DNA contamination, the merged 1240K-captured sequences were filtered for degraded DNA sequences with PMDtools v.0.6 restricting to a PMD score of 3 (https://github.com/pontussk/PMDtools 82 ).
Using the --UDGhalf parameter for the double-stranded library, this reduced the number of mapped reads from 735,985 to 81,525. For the single-stranded data the number reduced from 536,455 to 159,286 mapped reads.

Genetic sex determination
Based on the misincorporation pattern (SI Fig. 1b) of the double-stranded library, two base pairs were trimmed off the double-stranded 1240K-captured sequence on both ends of all reads to reduce damage-induced bias. From the resulting reads, the relative coverage of Xand Y-chromosomes was compared to that of the autosome, calculating the X-and Y-rate respectively. Due to the high X-rate (0.789) and low Y-rate (0.042), we concluded that the sample most likely possessed two X-chromosomes. Results after PMD-filtering were comparable with an X-rate of 0.740 and Y-rate of 0.024.

Genotyping
The double-stranded 1240K-captured sequence was genotyped for the aforementioned In both PCAs calculated on only East and Southeast Asian groups (Extended Data Fig. 6a), and including Near Oceanian groups (Extended Data Fig. 6b), the Leang Panninge genome falls into PCA space not occupied by any present-day populations. However, they are broadly located between Indigenous Australians and Onge from the Andaman Islands, which overlap with the Hòabìnhian-associated individuals 78 . The PMD-filtered genomes behave very similarly. The observed slight shift could also be due to the low coverage on the HO panel used for these calculations. Panninge showed stronger affinity to the Denisovan genome than Onge and other East Asian groups, but less than Papuan and Indigenous Australian individuals (Extended Data Fig. 7c).
We confirmed these observations when Papuans were kept in the third position in the f4statistics (Mbuti, Denisova; Papuan, X) where X are ancient and present-day individuals from the Asia-Pacific region (SI Fig. 3).  Fig. 7d).
A note of caution is warranted, however, since these statistics might in part be affected by ancient versus modern samples attraction.
Present-day groups in Wallacea, and particularly Sulawesi, speak Austronesian languages.
However, as published previously and indicated by the PCAs (Extended Data Fig. 6), there might also be Papuan ancestry present in these groups, similar to what has been observed in Vanuatu 88 . Therefore, we first tested if Papuan-like ancestry was present in modern-day groups from around the region in comparison to Ami individuals from Taiwan (HO dataset) 45 as representatives of Austronesian ancestry (SI Fig. 4a). These statistics were significantly positive only for the Lebbo and Mamanwa (traditionally, hunter-gatherer populations in Sarawak and the Philippines, respectively), confirming the dominance of Austronesian-like ancestry in the region. Targeting the minor Papuan-like component, a second statistic was performed to test potential differential affinities of present-day groups in the region towards Leang Panninge or Papuans. This showed that any detectable Papuan-like ancestry seemed to be more directly Papuan-than Leang Panninge-related (SI Fig. 4b).
We calculated the percentage of Denisova-related ancestry directly as in ref. 85:  To increase the SNP overlap of the test groups and the Leang Panninge genome, we also calculated the same statistics using the Vindija Neanderthal, Yoruba and Han from the 1000 Genomes project and present-day groups from the SGDP panel. Using YRI.SG and CHB.SG with other groups from SGDP achieved slightly higher values for all groups than the statistics using HO groups, while using all groups from SGDP panel resulted in lower values, notably 2.5% and 2.1% for Leang Panninge, respectively (SI Table 19).

D-statistics
Based on the archaic ancestry array (panel 4) 67 , we calculated D-statistics to gain insight into archaic ancestry in the Leang Panninge individual. Since the archaic ancestry array capture probes were applied to a single-stranded sequencing library, the orientation of deamination induced substitutions as C to T exchanges in forward oriented sequence alignments and G to A exchanges in reverse alignments was preserved. To reduce the effect of this deamination, we did not consider forward aligning sequences when the human reference genome, the two archaic human genomes or the inferred ancestor from the four ape genomes carried a C allele, 50 and did not consider reverse aligning sequences when any of these genomes carried a G allele  Table 21). No significant differences were found in the sharing of derived alleles with the Altai Neanderthal between the Han Chinese and Leang Panninge. Furthermore, the differences in allele sharing with the Altai Neanderthal were comparatively small between all tested genomes, except the French. Table 21: Allele sharing of the Leang Panninge genome and present-day individuals with the Denisova, Altai and Vindija genome, using the inferred great ape ancestor as outgroup. We also calculated D-statistics using data from the Tianyuan individual 90 . For this, we randomly sampled a single read at captured archaic ancestry sites from the Tianyuan data.

SI
Similar to the Leang Panninge data, higher sharing with Denisovans is observed (SI Table   22).   Table 23). For comparison we used the SGDP Sub-Saharan Africans 92 , three high-coverage Neanderthals and the high-coverage Denisovan as putative reference populations, and the chimpanzee reference genome (panTro4) as the putative ancestral allele. As no fine-scale recombination maps for ancient populations from the region under investigation were available, we ran the program using bins of size 10kb without incorporating a recombination map. Furthermore, as contamination has been estimated to be very low, we did not estimate contamination and error rates, but rather assumed a constant error rate of 0.01 for all reads.

SI
The exact command run was: In order to evaluate the robustness of some key parameters, we also used all East Asians from the SGDP data set as an alternative reference and varied the bin size to 20kb and 50kb. As results are consistent, we will focus the discussion on the primary parameter combination. For comparison, we also used admixfrog on all SGDP individuals using similar parameters, but calling the ancestry from the available genotypes by using the --gt-mode flag.
Admixfrog, as run above, results in a posterior decoding of the archaic ancestry; that is, for Neanderthal and Denisovan fragments are given in light blue and orange, respectively, homozygous fragments in darker colors. Bar height is proportional to the posterior probability of ancestry. Grey bars correspond to African ancestry. 56 We then investigated whether the 31 Denisovan fragments overlapped with known Denisova fragments from 109 modern human populations from Simons Genome Diversity Project (SGDP 90 ) using the same approach described in a previous study 106 . We summarised the overlap in terms of overlapping sequence and calculated p-values and 95% confidence intervals using 500 reshufflings of the archaic fragments in the Leang Panninge individual.
We found that there is a significant correlation between the Denisova fragments in Leang Panninge and those in Papuans (n=15 individuals), Indigenous Australians (n=2 individuals) and people from Bougainville (n=2 individuals), suggesting their Denisovan ancestry is shared (p-value < 1/500, Extended Data Fig. 8).
Overall the plot shows great similarities to the PCAs (Extended Data Fig. 6 Table 24). While the basic reference set was not able to differentiate between Leang Panninge and Papuan, these two groups are always consistent as deriving from at least two distinct waves of ancestry when using Denisova and/or ancient Asian individuals as additional reference groups. Using the qpWave results, we also attempted to find the best fitting source using the "rotating" approach described in ref. 108, keeping Mbuti as the calculation baseline and cycling in turn Kostenki14, Onge, Denisova, and the Asians groups listed above from the references to the source alongside Papuan. In this instance, Onge and Tianyuan are the only two sources that provide a working model, resulting in around 50% Asian-and 50% Papuanrelated contributions (Fig. 3a and SI Table 26).  Fig. 7b), but with relatively large residuals (Extended Data Fig. 10a). The first migration edge added by the algorithm 15, hires: YES, lambdascale: 1, initmix: 1000, inbreed: YES. We added one group after the other, moving from archaic humans to modern groups and ancient samples according to previously established branching patterns, testing all possible one-and two-way mixtures using a custom-made script. 67 The first scaffold was composed of five Mbuti individuals 92 and the Denisovan genome 94 . As expected, this two-group model did not result in any "outlier" -that is, f4-statistics where the calculated value using the proposed tree compared to the observed values differ with a zscore larger than 3 or smaller than -3 -and achieved a worst z-score of 0.013 (SI Fig. 7a).

SI
Then, we added the genomes of four Upper Paleolithic individuals from Sunghir (Russia) 111 as representatives of ancient non-African European-related lineages. The lowest worst z-score was achieved by fitting Sunghir as a sister group to Mbuti (worst z-score 0.024, SI Fig. 7b).

SI Figure 7:
The qpGraph scaffolds wibth the lowest worst z-scores modelling the relationships between: a, the Denisova and Mbuti (worst z-score 0.013), plus; b, ancient individuals from Sunghir, Russia (worst z-score 0.024).
Next, we added the genomes of two Onge individuals from the Andaman Islands 92 to introduce eastern Eurasian ancestry. The best fitting tree models Onge as a sister group to ancient Sunghir individuals (worst z-score 1.873, SI Fig. 8a). To represent more specifically mainland East Asian ancestry, the Neolithic Qihe individual 91 from present-day China was then added, forming a sister group to Onge (worst z-score -2.447, SI Fig. 8b).
The next added group is formed by 16 individuals from Papua New Guinea 93 Fig. 10a). The lowest worst z-scores are achieved by modelling Leang Panninge as a two-way admixture between Near Oceanian and Qihe-related ancestries (SI Fig. 10b and c), which is congruent with previous analyses (Fig. 3a, SI Table   26). Any other branching position from the Asian-related portion of the admixture graph results in branch lengths of zero between Qihe and the edge leading to Leang Panninge.The branching point on the Near Oceanian portion of the admixture graph is instead more uncertain. The overall lowest worst z-score of -2.190 is observed when Leang Panninge is 70 closer to Indigenous Australian than Papuan groups. However, this is only marginally different from the graph where Leang Panninge branches before the Near Oceanian populations split from each other (worst z-score -2.194), consistent with the f4-statistic results (Extended Data Fig. 7b). Both scaffolds contain branch length of 1 or 0, respectively, within the Leang Panninge-Oceanian clade; therefore, we present these groups as deriving from a trifurcation (Fig. 3c). Figure 10: Fitting Leang Panninge on the qpGraph scaffold: a, as a one-way admixture (worst z-score -4.818); b, as a two-way admixture branching closer to Indigenous Australian than Papuan (worst z-score -2.190); and c, as a two-way admixture branching before the Australo-Papuan population split (worst z-score -2.194).

SI
To rule out the possibility that the observed reduced amount of Denisovan ancestry in Leang Panninge compared with present-day Near Oceanians is due to repeated interactions with different Denisovan groups, we also modelled this scenario in four additional admixture graphs (SI Fig. 11). While these fit better than the one-way Denisovan adxmiture model 71 presented in SI Fig 10a -likely due to the different amounts of Denisovan-related ancestry present in Leang Panninge and Near Oceanians -the worst z-scores remain close to |4|. This is opposed to the graphs with one common Denisovan-related admixture into Oceanians Leang Panninge followed by an Asian-related admixed into Leang Panninge which provide worst z-scores below |3| (SI Fig. 10b-c Table 27).