Since then the list of secure remains has grown slowly: a mandible and a rib from a cave on the Tibetan Plateau, a mandible dredged from the Taiwan Strait, a nearly complete cranium from Harbin in north-eastern China. Of the body below the neck almost nothing was known: one phalanx and one rib.
Two papers published in Nature on 9 September add seven specimens, and among them the first long bone of the arm ever attributed to this group.
They come from Bianfu Cave, in Heqing County, western Yunnan, a couple of kilometres from a secondary tributary of the Yangtze. Excavation began in 2019. The cultural deposit spans, according to a Bayesian model combining optically stimulated luminescence and uranium-series dating, roughly 190,000 to 70,000 years ago. It is the oldest Palaeolithic site known in western Yunnan.
The layers yielded 1,366 stone artefacts and more than 64,000 mammal fossils, of which over 60,000 are bone fragments unidentifiable by eye. Four human teeth had been found in the middle of layer 7.
The question was whether that mountain of splinters held any more.

Testing them one by one with mass spectrometry would have cost too much. The technique is called ZooMS, zooarchaeology by mass spectrometry: collagen is extracted from a fragment and its peptide fingerprint read, which identifies the animal family it came from. It works well, but on sixty thousand pieces it becomes impractical. And the hit rate is low: at Denisova Cave and Baishiya, human remains recovered this way amounted to just 0.1 per cent.

The team led by Qiaomei Fu, of the Institute of Vertebrate Paleontology and Paleoanthropology at the Chinese Academy of Sciences, added a step before the laboratory. They inspected every fragment by hand, assessing size, cortical thickness, articular surfaces, muscle attachments and the vascular impressions on the inner face of cranial bones. Twenty-two candidates emerged: five from the upper layers, seventeen from the lower ones.
Only then the spectrometry. Three fragments showed collagen markers pointing to the genus Homo: two pieces of parietal bone, labelled BFD767 and BFD769, and a proximal fragment of radius, BFD771.
The attempt to recover ancient DNA from those three specimens failed. In Yunnan the warm, humid climate destroys nucleic acids.
The proteins held. This is the methodological point that makes the work important: where DNA cannot reach, palaeoproteomics can still say who that individual was.
Collagen is the most abundant protein in bone and resists degradation longer than almost anything else. Its sequences differ between species at single positions, called single amino acid polymorphisms. One of these, in the alpha-2 chain of type I collagen, works as a marker: the variant COL1A2 R996K appears in every Denisovan identified so far, from Harbin to Penghu to Siberia, and in no Neanderthal or modern human.

The screening path for human remains at Bianfu Cave. More than sixty thousand bone fragments were inspected by hand; twenty-two were judged possible human remains; three were confirmed as such by protein analysis. Mass spectrometry was applied to seventy-six fragments in all, candidates and controls. Chart produced by Scienzaonline from data in Rao et al., Nature 2026, doi 10.1038/s41586-026-10976-9.

All five Bianfu specimens analysed carry it. The two teeth, the two parietals and the radius.
For the teeth there is more, because enamel preserves different proteins from bone. A variant of ameloblastin, AMBN M273V, occurs in East Asian Homo erectus, at Harbin, at Penghu, in the Siberian Denisovans and in some modern humans: almost all of them from Southeast Asian and Oceanian populations, those with the highest Denisovan ancestry. A second ameloblastin variant distinguishes the Bianfu individuals from Asian Homo erectus.
And the enamel of both teeth contained peptides from the Y isoform of amelogenin, the protein encoded on the male sex chromosome. They were two males.
Phylogenetic trees built on the protein sequences place each Bianfu specimen in a group with the reference Siberian Denisovan, with a posterior probability of one hundred per cent, and place modern humans as sister group to the branch joining Neanderthals and Denisovans. It is the same topology obtained from nuclear genomes.
Now the specimen that made the headlines, and the real numbers.
BFD771 preserves about forty-five millimetres of the proximal portion of the radius, the forearm bone running from the thumb side of the wrist to the elbow. It includes a partially intact radial head and extends to the distal part of the radial tuberosity, the ridge where the biceps tendon attaches. Computed tomography shows a robust cortex and a fully fused proximal epiphysis: the individual was adult or near-adult. It is most probably a right radius.

It is not an elbow bone, as many have written. The elbow is a joint. This is a piece of forearm bone, the part near the elbow.
Now the point on which the coverage split, some speaking of strength and some of precision. The paper says both, and calls them mosaic features.
On the Neanderthal side: the radial tuberosity faces inwards, a condition seen in 76.5 per cent of late Neanderthals, whereas in modern humans the prevailing orientation is anteromedial in 88.8 per cent of cases. And the radial head is large, with a margin of 23.2 millimetres, closer to Neanderthals than to us.
On our side: the overall external shape and the cross-sectional geometry at mid-neck align with modern humans. Surface-matching analysis brings BFD771 closer to modern humans because of its relatively short neck in proportion to the radial head, whereas in Neanderthals the neck is more elongated.
And a third figure that sits on neither side: the mid-neck cross-section shows greater total area and torsional rigidity than in the Neanderthals, australopithecines and Homo erectus specimens compared, while remaining within the range of modern humans and living great apes.

Why these measurements matter. Tuberosity orientation, neck length, head size and cross-sectional geometry determine how efficiently the biceps rotates the forearm upwards and how much load the joint can bear. A mosaic of traits indicates biomechanical demands different from both Neanderthals and ourselves. Which ones, we do not know: one fragment is not enough.
The second paper, led by Qijun Ruan of the Yunnan Provincial Institute of Cultural Relics and Archaeology, describes how they lived.
Faunal and pollen records indicate a conifer-dominated forest or forest-steppe environment, sustained across three marine isotope stages. The inhabitants practised specialised hunting of medium- to large-bodied prey: more than a hundred individuals of deer and bovids, with smaller numbers of rhinoceros and hyena.
The technology is the most interesting part. The authors describe it as expedient core reduction and tool production, alongside pervasive use of unmodified bone. Where a bone or a stone already served, it was used as it was. The authors speak of an adaptive system prioritising the exploitation of what an object already affords over intensive manufacture.
This is not crudeness. It is economics: less labour invested in a tool that will be used once.

Two Italian researchers appear on the second paper, Marco Peresani and Davide Delpiano of the University of Ferrara, alongside Francesco d’Errico, Jean-Jacques Hublin, Clément Zanolli and Fabrice Demeter.
The limitations remain and should be stated. One radius does not make a skeleton: Denisovan stature, body proportions, posture and gait remain unknown. There is no DNA, so this population cannot be placed precisely against the different Denisovan lineages that left traces in Asian and Oceanian genomes. The authors state plainly that the genomic and proteomic data currently available are too scarce for that.
And the direct uranium-series dates on the human remains came out far younger than the layers: around 102,000 years for two samples and 143,000 for the older parietal. The authors explain that bone in the ground exchanges uranium with its surroundings, so those figures should be read as minimum ages. The stratigraphic dates remain the more reliable.
There is, finally, something the work teaches anyone who excavates. For fifteen years Denisovans were a kind of people without a body. The method that gave them one is not a new technology: it is looking by hand at sixty thousand fragments before switching on the spectrometer. The authors propose it as a repeatable procedure for other Palaeolithic sites full of broken bone, which is most of them.

References
Rao H., Xing S., Ruan Q., Bai F., Zhao Y., Wu Z., Tang S., Bennett E.A., Feng X., Lin S., Shao Q., Yang X., Li Z., Zhou K., Liu F., Fu Q., "Ancient proteins identify various Denisovan remains from Southwest China", Nature, received 16 January 2026, accepted 31 July 2026, published online 9 September 2026, open access under CC BY-NC-ND 4.0. doi: 10.1038/s41586-026-10976-9
Ruan Q., Li H., Xing S., Zhao K., Liu J., Zanolli C., Doyon L., Delpiano D., Peresani M., d’Errico F., Demeter F., Hublin J.-J. et al., "Denisovans from southwestern China and their subsistence strategies", Nature, published online 9 September 2026. doi: 10.1038/s41586-026-10997-4
Chinese Academy of Sciences, "Denisovan Radius Discovered for First Time in Southwest China", press release, September 2026.
Mass spectrometry data deposited with the ProteomeXchange Consortium, identifier PXD073112. Three-dimensional models of the fossils available on MorphoSource.

*Board Member, SRSN (Roman Society of Natural Science)
Past Editor-in-Chief Italian Journal of Dermosurgery