In a limestone cave in the karst lowlands of South Sulawesi, a forager died sometime between 25,000 and 16,000 years ago, leaving behind only a fragment of upper jaw and three molars. What survived of those teeth is strange enough that when Adam Brumm, an archaeologist at Griffith University’s Australian Research Centre for Human Evolution, first examined1 them, he could not immediately account for what he was looking at. Each molar carried a smooth, rounded groove running around it, as if something had been clenched in the mouth and worked against the tooth, year after year, until the enamel gave way.
The site is Leang Bulu Bettue, a cave complex in the Maros-Pangkep karst. The jaw fragment is catalogued as Maros-LBB-1a, and it is, at present, the only human skeletal element ever recovered from the Pleistocene of Sulawesi, an island of some 174,000 square kilometers. There was almost nothing to compare it to.
Until now, the working assumption in archaeology has been that people started using mind-altering plants in earnest only after they settled down to farm, somewhere around 12,500 years ago, when barley beer, coca, and psilocybin mushrooms begin turning up in the record. Betel nut, the seed chewed by roughly one in ten people alive today for its buzz, was thought to be younger still: the earliest solid chemical evidence for its use came from a Bronze Age site in Thailand, about 3,500 years old. The wear on the Sulawesi jaw undoes both timelines. Whatever this forager was doing with their mouth, they were doing it before agriculture reached the region at all.
Almost nothing is actually known about how people first found psychoactive plants, or why. Betel nut’s active compound, a muscarinic agonist called arecoline, is generally assumed to need slaked lime, a caustic paste of pulverized shells or limestone, to become bioavailable enough to feel anything from it. Modern users still prepare their quid this way: a sliced areca seed, a smear of lime, wrapped in a leaf of the betel vine, Piper betle. Without the lime, conventional wisdom held, there wasn’t much of a drug there at all.
]]>That should have ended the story. Since 2010, when a sliver of DNA from a finger bone in a Siberian cave revealed an entirely new kind of ancient human, genome sequencing has been the default way to confirm a Denisovan identity, the kind of evidence that settles an argument. What settled this one instead was protein: the two skull fragments and the forearm piece were confirmed as Denisovan through molecules built from the same genetic code as DNA but far more stubborn about surviving, and in this cave, they had outlasted DNA completely.
The stakes reach beyond one cave. Denisovans are known mostly through their genes: people living today across Southeast Asia and Oceania, in Papua New Guinea, the Philippines, and Australia among other places, carry some of the highest percentages of Denisovan ancestry found anywhere. But the physical evidence of that population, actual bone and tooth, has been almost entirely missing from the region where the genetic signal runs strongest. Southwestern China has sat inside that gap since Denisovans were first identified. Bianfu Cave closes part of it, with more confirmed Denisovan material than any site outside Siberia.
Of the sixty thousand-odd fragments screened, twenty-two looked, by eye, like they might be hominin. Three were.
How Do You Identify a Fossil Without DNA?
Screening tens of thousands of unidentifiable bone splinters for the rare one or two that might be human is normally done with a technique called ZooMS, zooarchaeology by mass spectrometry, which reads short fragments of collagen protein well enough to sort a bone by animal family without a full genetic workup. It is fast and cheap compared with genome sequencing, but even ZooMS has limits: at Denisova Cave and at Baishiya Karst Cave on the Tibetan Plateau, running it blind across the whole assemblage turned up hominin bone at a rate of about one fragment in a thousand. Running that same approach across all 60,000 Bianfu Cave fragments would have meant processing thousands of bones for a chance at one or two hominin fragments, a slow, expensive proposition no lab budget was built for.
So the team looked first, by hand, at bone size, cortical thickness, and surface texture, the kind of read a trained osteologist does before any lab work begins, and narrowed 60,000 fragments down to 22 candidates worth testing. Of those 22, three came back with collagen markers specific to the genus Homo: two curved pieces of parietal bone, the plates that form the side and roof of the skull, one from a layer dated to roughly 148,000 to 134,000 years ago and cataloged as BFD767, the other from an older layer near 167,000 to 150,000 years ago and cataloged as BFD769; and a partial radius, one of the two forearm bones, cataloged as BFD771 and pulled from the same younger layer as BFD767. Two teeth from that younger layer, a lower premolar and a lower molar, went through the same analysis and came back Denisovan as well.
DNA is a fragile molecule, two strands loosely paired and easy to break apart with heat, water, and time. Collagen, the structural protein that makes up most of bone, is built differently: three long chains wound into a tight triple helix, chemically tough enough to survive well past the point where DNA has fallen apart into unreadable fragments. That toughness is why protein has become the main way of confirming Denisovan identity outside the cold, dry conditions of Siberia. The mandible from Xiahe on the Tibetan Plateau, the mandible dredged from the Taiwan Strait near Penghu, and now the bones from Bianfu Cave were all identified this way, with no DNA recovered from any of them. Even the most complete Denisovan skull ever found, the cranium from Harbin in northeastern China, gave up its species identity through protein first; a trace of Denisovan DNA eventually turned up, but only in the calculus built up on its teeth, not in the bone itself.
Proteins carry genetic information too, just written a step removed from DNA. A change in a single DNA letter can change a single amino acid in the protein that DNA codes for, and researchers can read that change directly in a protein sequence even after the DNA that produced it is long gone. In the Bianfu Cave specimens, that readout came from a collagen chain called COL1A2, at one specific position out of more than a thousand. More than 99.999 percent of a large modern human reference database carries the amino acid arginine there. Every confirmed Denisovan tested so far, from Siberia to Tibet to Taiwan, carries lysine instead.
Every one of the five Bianfu Cave specimens carries it too.
A few other variants added detail rather than proof. Peptides from the enamel of both teeth carried an abundance of a protein called amelogenin Y, present only in male individuals, so at least two of the Bianfu Cave Denisovans were men. Another enamel variant, in a protein called ameloblastin, is shared with East Asian Homo erectus, with the Harbin and Penghu fossils, and with a scatter of living people who carry the derived form today, most of them from Southeast Asia and Oceania, the same populations whose genomes point back toward exactly this part of the map.
Dating the bones directly turned out to be its own small puzzle. Uranium-series measurements on the three hominin bones and the carbonate crusted onto them gave minimum ages of around 102,000 years for the two specimens from the younger layer and about 143,000 years for the parietal from the older one, younger in both cases than the layer they came from. Bone keeps absorbing uranium from groundwater long after an animal dies, so a bone’s own uranium clock only ever proves a minimum age, never the true one. The more reliable numbers come from the sediment around the bones, dated by luminescence, and from the layered mineral crusts nearby, dated independently by the same uranium method applied to a substance that behaves as a closed system. Those put the younger layer at roughly 148,000 to 134,000 years old and the older one at 167,000 to 150,000, the figures the excavation team treats as accurate. The radius raises a different kind of question, less about when than about what kind of arm it was attached to.
]]>That primate-only model is one of three versions of a genomic language model called GPN-Star, short for genomic pretrained network with species tree and alignment representations, published this week in Nature1 by a team at UC Berkeley led by Yun Song. The assumption behind PhyloP and PhastCons, and behind most of the field before them, is that more evolutionary distance means more signal: compare a piece of human DNA against hundreds of species reaching back hundreds of millions of years, and whatever never changed is whatever evolution refused to let change. GPN-Star was built to test that assumption directly, training separate copies of the same model on three different slices of the tree of life, one spanning roughly 600 million years of vertebrate evolution, one spanning about 100 million years of mammals, and one confined to other primates. For a surprising amount of human biology, the shortest of those three windows carried the most information, not the longest.
Primates share a common ancestor with humans a little under 65 million years back, roughly a tenth as deep as the vertebrate alignment GPN-Star’s broadest version was trained on.
The human genome was fully sequenced more than twenty years ago, three billion base pairs read end to end, and most of what those letters mean is still unclear. An estimated one to two percent of the sequence codes for protein. The rest is a mix of evolutionary debris, sequence that no longer does anything, and regulatory elements, the switches that decide when a gene turns on, in which tissue, and how strongly. Sorting the debris from the switches, one base pair at a time, is the problem that determines whether a clinician can tell a patient what a mutation in their own DNA actually means, and it is the problem GPN-Star was built to attack.
Song, a professor of computer science and statistics at Berkeley and an investigator at the Innovative Genomics Institute, framed the goal in practical terms. Laboratories have built increasingly creative ways to test what a given mutation does, he said, but they still “cannot experimentally test every single variant in the genome,” so the model’s job is to tell biologists which of the millions of candidates are worth testing first. GPN-Star approaches that goal by training on data that has already been through a phylogenetic sorting process, a whole-genome alignment mapped onto a species tree, rather than on raw, unaligned sequence. Competing models trained the harder way include Evo 2, built on more than 100,000 species spanning every domain of life using 2,000 NVIDIA H100 processors running for months, and Nucleotide Transformer, a 2.5-billion-parameter model trained on 128 processors for a month. GPN-Star, at 200 million parameters, trains in days on eight NVIDIA A100 GPUs.
]]>Ancient DNA researchers use a specific kind of damage to judge whether a sample is genuinely old. Cytosine bases near the broken ends of a DNA fragment convert into a form the sequencer reads as thymine, an error called deamination that accumulates the longer a molecule sits in the ground. More deamination is supposed to mean more time has passed. It is one of the field’s basic yardsticks, the reason a headline can credibly claim a genome is fifty thousand years old rather than five. Crawford Lake breaks that yardstick.
The finding comes from a paper in Molecular Ecology1 led by Tyler Murchie, of the Hakai Institute, and Matthew Emery, an assistant professor of anthropology at Binghamton University, with Hendrik Poinar, a professor of anthropology at McMaster University, as co-senior author. The team was not chasing a methods curiosity. They wanted a full ecological history of Crawford Lake, recovered not from pollen grains, beetle wings, or charcoal flecks, the proxies lake scientists have relied on for decades, but from environmental DNA shed by every organism that lived, died, or defecated near the water: plants, animals, fungi, and bacteria, all mixed into the sediment and sequenced together. Sedimentary ancient DNA, or sedaDNA, can do something those older proxies cannot: put a name, sometimes a species-level name, on organisms that left no visible trace at all.
The averages make the pattern clear. Reads assigned to Canada goose, Branta canadensis, showed a mean cytosine-to-thymine deamination rate of 32.4 percent. Permafrost-preserved mammoth bone from Alaska’s Tanana Valley, roughly fourteen thousand years old, averaged 18.8 percent. Permafrost sediment from the Klondike in Yukon, twenty to thirty thousand years old, averaged 11.9 percent. Only the oldest ancient DNA on record, bone and coprolites pushing past seven hundred thousand years, reached comparable numbers, around 40 to 45 percent, and a few individual Crawford Lake goose samples, pulled from sediment as young as five hundred years, hit 52 percent on their own. The lake’s DNA fragments were shorter too: goose DNA averaged fifty-one base pairs against fifty-four for the permafrost bone, and Crawford Lake’s plant DNA, corn, sunflower, and pine, averaged in the low forties.
Murchie, Emery, and their colleagues point to chemistry rather than time. Permafrost preserves DNA the way a freezer preserves food, by halting the reactions that break it down, while lake mud runs on a different chemistry entirely. Crawford Lake’s water column has measured between pH 5.8 and 8.5, acidic enough at its lower end to speed up the hydrolysis that strips bases off a DNA strand, though the team is careful to note that pH alone cannot explain damage this severe. Water content, temperature, mineral absorption, and microbial activity probably all played a part. So, they suspect, did a goose’s digestive system: much of this DNA did not fall into the lake so much as pass through a bird first.
]]>Isotope chemists can read this because different tissues in the body stop recording diet at different points in a life. A tooth crown mineralizes in childhood and then goes chemically silent, a small frozen archive of what a kid ate. Bone keeps remodeling for decades, so its isotope signature is a rolling average of a person’s more recent years. Panicum miliaceum, broomcorn millet, uses a different, more efficient photosynthetic pathway than wheat or barley (C4 rather than C3), one that leaves a distinctly heavier carbon isotope signature in the tissue of whoever eats it regularly. Comparing tooth against bone lets researchers catch a dietary shift that happened somewhere in the gap between them, even without knowing exactly when.
This woman, cataloged in the new study as MIL1_234, is one of 192 individuals from more than a dozen Bronze Age cemeteries in southern Poland whose remains were analyzed for the project, published in the journal Science Advances,1 which combined freshly generated isotope data with everything previously published on these sites. Her case matters because of the date attached to it. Her bone collagen carbon value, -16.03 per mil, sits well above the -18.0 per mil threshold Polish researchers use to flag millet in the diet, and her radiocarbon date centers on roughly 1588 BCE, plus or minus 48 years. The previous benchmark for the earliest substantial millet consumption in this part of Europe, established in 2021 from carbon isotope work at these same sites, stood at 1460 BCE, give or take 40 years. MIL1_234 pushes that back by more than a century, past even the oldest millet grains yet dug out of Polish soil, found at Lipnik and dated to around 1303 BCE. Her date also lands close to the earliest millet remains found anywhere in Europe, a set of charred grains from Vinogradnyi Sad in southern Ukraine dated to roughly 1546 BCE, hinting that the crop reached Central Europe from the east earlier than the physical seeds alone had suggested.
She isn’t alone in this pattern. A man buried in a separate grave at the same site, feature 631, cataloged as MIL1_233 and radiocarbon dated to about 1468 BCE, shows the same signature: a childhood on wheat and barley, with the shift to millet occurring sometime in his adult life, at the latest within five years of his death. His case comes with its own small complication. A protein test on his tooth enamel, run for this study to determine biological sex, failed to detect the male-specific peptide researchers normally look for, which would technically classify him as female. Earlier DNA analysis and the shape of his bones both say male. The team kept the DNA-based call, and noted, almost in passing, that this is not the first time protein-based sex testing has disagreed with genetic testing on the same skeleton.
The same pattern, a childhood on wheat and barley followed by millet later in life, turns up in at least one person at nearly every cemetery in the study with isotopic evidence of the crop. The lone exception is Kraków Nowa Huta Mogiła, where only two people were tested, too few to say either way.
]]>None of that clay traveled far to get there. A 1992 petrographic study of pottery from Munhata, another Yarmukian-culture site in the Jordan Valley, traced its raw material to terra rossa, the reddish soil eroding off the slopes above the valley floor. The archaeologist Yuval Goren found no evidence that anyone had sought out a special deposit; potters were using what had simply washed down to them. The usual explanation for why people started firing that soil at all leans on gradually accumulated motives: the appeal of prestige vessels and feasting displays, the pressure of larger populations needing to store and cook more food, generations of incidental experience lining baskets and hearth-pits with clay before anyone thought to bake it on purpose. A new study by Amos Frumkin, a geomorphologist at the Institute of Earth Sciences at the Hebrew University of Jerusalem, published in the journal Geomorphology,1 argues that something more specific and more sudden may also have been at work: that the same disaster which buried the valley in clay might also have shown its future residents, by accident, exactly what fire does to it.
Frumkin calls this the natural kiln hypothesis, and it is bluntly literal. Deep in the ashes of an intense wildfire, exposed clay can reach the five hundred to seven hundred degrees Celsius needed to harden it permanently, the same transformation a kiln performs on purpose. If wildfires swept the hills above the valley with enough frequency during the right window of time, and the clay that resulted ended up conspicuously baked in places where people were already living, then whoever eventually built the first deliberate kiln may not have invented the underlying principle so much as recognized it.
The case for those wildfires rests on unglamorous material: cores of lake mud, and cave formations. In a long core drilled through the bed of the Dead Sea, and in another through Lake Hula, researchers count microscopic charcoal fragments layer by layer, a direct record of how much burned nearby and when. In the Soreq and Har Nof caves near Jerusalem, the carbon and strontium locked into slow-growing speleothems record, less directly but at finer chronological resolution, when the hillsides above them lost their plant cover and their topsoil. Frumkin lines these records up across two intervals separated by more than a hundred thousand years, and in both, the charcoal spikes, the vegetation isotopes swing sharply, and the soil erodes away.
What the fires and erosion left behind in the valleys was clay, in enormous quantity. Thick layers of it, called the Fazael Formation in the Jordan Valley and the Ze’elim Formation around the Dead Sea, pile up in places to tens of meters, recycled wholesale from the stripped hillsides above. Upslope, the effect was closer to eviction: a large Pre-Pottery Neolithic settlement at Motza, in the Judean highlands, appears to have been abandoned right around the peak of the fires, its local soils exhausted. Downslope, the opposite happened. Sha’ar Hagolan, Jericho, Netiv Hagdud, and Gilgal, among the best-known Neolithic villages in the region, all sit on freshly redeposited soil, in exactly the spots where a catastrophe on the hills became a resource in the valley.
The relevant interval here is what geologists call the early Holocene, roughly 11,700 to 7,000 years ago. Its most intense pulse of fire shows up in the Hula core at approximately 9,500 years ago, when charcoal counts roughly tripled over what came before. A charcoal spike dated independently to about 9,600 years ago turns up in a lake core from the Ghab Valley in northwest Syria, hundreds of kilometers to the north, which is the detail Frumkin leans on to argue that the fires were regional and climatic, not a side effect of Neolithic farmers burning off vegetation nearby. The timing at least holds together: a landscape saturated with fire-baked clay well before pottery-makers show up in the archaeological record at Sha’ar Hagolan, rather than after.
]]>Melon’s standing in Chinese culture runs deeper than agriculture. Its abundance of seeds made it, for centuries, a folk emblem of fertility and continuity, and Song-era potters shaped whole lines of celadon ware after its rounded body and pale flesh. It would be easy to assume that a fruit this embedded in the culture, this tied up with the idea of abundance, tasted the way melon tastes now. The two Gugang seeds, cataloged in the new study as GG1 and GG4, complicate that assumption from two directions at once: neither genome carries the markers of a sweet dessert fruit, and neither confirms that China ever bred its own melon independently in the first place.
The seeds come from Shuomen Gugang, a medieval port that connected inland China to trade networks across the western Pacific and Indian Ocean during the Song dynasty (960 to 1279 CE). A companion seed from the same waterlogged layer was radiocarbon dated directly, placing the deposit at roughly 1020 to 1150 CE. The analysis, carried out by an international team anchored at Durham University in the UK and Washington University in St. Louis, was published this month in the Proceedings of the National Academy of Sciences.1
Cucumis melo is not a small crop. Global cultivation passed 22 million tons in 2024, and yet its route into East Asia has been argued over for more than a decade, with the argument reversing at least once already. Susanne Renner, a biologist at Washington University in St. Louis and one of the study’s senior authors, has spent more than twenty-five years working on the genus. Renner has said that early researchers assumed an African origin for melon simply because so many wild relatives of the genus Cucumis grow there, and diversity was long read as a signal of origin. Her own biogeographic work, drawing on roughly one hundred Cucumis samples collected across Africa, Australia, and Asia, pointed somewhere else: toward a set of overlooked relatives scattered around the Indian Ocean and Australia, and toward Asia, not Africa, as melon’s actual homeland.
Genomic surveys since then have converged on three known domestication events for melon: once in Northeast Africa, and twice, independently, in India. China has always been the loose thread. Archaeobotanists working the lower Yangtze have recovered Cucumis seeds spanning the Neolithic through the Bronze Age, a run of nearly seven thousand years, and that long record has kept alive the possibility of a fourth domestication, centered in China and separate from the Indian and African lineages. The trouble is that Cucumis seeds are notoriously hard to sort by eye. Domesticated, feral, and wild seeds overlap in shape, and centuries underground blur what differences exist. Nobody was going to settle this by measuring seed outlines under a microscope.
The genome answers what the seed shape could not. Both GG1 and GG4 cluster with the agrestis lineage, the broad East Asian branch of cultivated melon still grown across China today, rather than with any separate, unsampled group that would mark an independent Chinese domestication. That points toward introduction from the existing Asian domestication pool, most likely South Asia, rather than a homegrown Chinese origin. The authors stop short of closing the door entirely. No wild Cucumis melo population has ever been confirmed inside China, and an extinct or unsampled progenitor can’t be ruled out from two low-coverage genomes alone. The relationship between these Song-era seeds and the far older Neolithic material from the same region remains, in the paper’s own phrasing, unresolved.
Xinyi Liu, an archaeologist in the anthropology department at Washington University in St. Louis and another of the study’s authors, has pointed to two plausible corridors for that introduction: a northern route along the Silk Road, skirting the Tibetan Plateau, or a southern one, the so-called Tea Horse Road, threading through the mountains of southwest China and the Himalayas toward South and Southeast Asia. The archaeological record favors the south. Morphologically identified melon seeds turn up in a tomb at Wuwangdui dated to 238 BCE, just before the Han dynasty began, roughly nine centuries before comparable remains appear at a northern site closer to 700 CE. Liu has argued that gap makes a southern introduction the stronger candidate. If that holds, melon may have reached China twice: once overland, centuries before the Han, and again by sea, arriving at ports like Gugang alongside the same tropical cargo, lychee, kiwifruit, grape, and olive, that turned up in the shipwrecks themselves.
]]>A heel-strike is exactly what it sounds like: the heel touches down first, before any other part of the foot, and only then does the weight roll forward across the sole toward the toes. Every person reading this does it, at every step, and has done it since somewhere around the age of two, regardless of how long or short their legs happen to be. It is such a stable, unremarkable piece of anatomy that biomechanists studying the origins of human bipedalism have mostly looked past it, chasing flashier explanations involving the pelvis, the spine, the size of the gluteal muscles. Landing on the heel seemed like a detail.
New data, published today in the Proceedings of the National Academy of Sciences1 by a team led by Nicholas Holowka, an assistant professor of anthropology with a joint appointment at the Huck Institute of Life Sciences at Penn State, argue that it is not a detail. Heel-striking, in their account, is a genuine adaptation, one that appeared somewhere along the hominin lineage and had to be paid for twice over: once in energy spent, and once in force absorbed by bone.
The team’s clearest evidence for how strange the human heel-strike really is comes from how little it varies. Nine barefoot adults, walking at their own pace across a rigid runway at the University at Buffalo while an eight-camera motion-capture system tracked reflective markers on their legs and pelvis, produced a foot-strike angle, the angle the foot makes with the ground at the instant of contact, that ranged across the whole group by only 10.3 degrees, from 12.3 to 22.6. A single chimpanzee, walking on a comparable runway, could swing, individually, across 2.4 to 8.6 times that range from one step to the next. Chimp A’s own range, all by himself, was 39.7 degrees. Chimp C’s was 23.8. Chimp B, oddly, was the tightest of the three at 11.1 degrees, closer to human consistency than either of the other two, but centered on an angle that never once crossed into heel-strike territory. He had found his own fixed setting, just not the one that includes a heel.
What that fixed human setting buys is efficiency, and the researchers finally have a number for it. Eleven of the human volunteers walked on a treadmill while breathing into a K5 portable respirometry system, which tracks oxygen consumption and carbon dioxide output breath by breath, first using their normal heel-first gait and then using a deliberately altered one: contacting the ground near the outside edge of the forefoot, close to the base of the pinky toe, before letting the heel drop, a posture the researchers call a midfoot-strike. Volunteers needed about ten minutes of practice to manage it without stumbling. Walking heel-first cost an average of 2.59 joules per kilogram per meter traveled. Walking midfoot-first cost 3.46, a jump of 26 to 41 percent, and one the authors note is roughly six times larger than the metabolic penalty measured in an earlier study of runners switching from heel-strikes to midfoot-strikes. The same volunteers who saved that energy heel-striking also absorbed considerably more force doing it.
Loading rates, how fast the impact force arrives after the foot touches down, were 121 to 162% higher when volunteers heel-struck than when they used a midfoot-strike.
Impact peak forces followed the same pattern, 168 to 206 percent higher for heel-strikes, even though the overall maximum force each foot experienced across the full stance phase barely moved, drifting by less than two percent in either direction. The difference lives in the first few milliseconds of contact: a sharp spike that a midfoot-strike smooths away almost entirely.
Nathan Thompson, an associate professor at the New York Institute of Technology’s College of Osteopathic Medicine and a co-author on the study, put the mechanism in domestic terms.
“Imagine you are trying to sneak across a creaky wooden floor,” he said. “You tend to walk on the balls of your feet, because this reduces the rate of loading on the floor and creates less creaking. It’s a softer way to walk.”
That, more or less, is what the chimpanzees were doing, especially on two legs. Chimps A and C both used fewer heel-strikes when walking bipedally than when they were on all fours, and bipedal loading rates in the chimpanzees were 57.9 to 138.1 percent higher on the strides where they heel-struck than on the ones where they used a midfoot-strike instead. “It seems that when walking on two legs, chimpanzees prefer the ‘softer’ way to land on their feet,” Thompson said. A 2003 study of plantar pressure in seven bonobos found the same avoidance, more midfoot contact bipedally than quadrupedally, and researchers observing the terrestrial behavior of wild orangutans, who spend most of their lives in trees but do sometimes come down and walk upright, have reported similar reluctance to heel-strike. Something about standing on two legs seems to make heel-striking a worse bargain if you are an ape and not a hominin.
Nicholas Holowka, assistant professor of anthropology at Penn State, explains the differences between human and chimpanzee walking styles and why humans evolved to walk heel first. Credit: Ben Manning / Penn State
]]>Sudan Memory1 is the first mass digitization project ever attempted in Sudan. Its roots go back to January 2013, when Dr. Badreldin Elhag Musa, executive director of the Sudanese Association for Archiving Knowledge, contacted archival institutions in the UK about Sudanese cultural heritage he considered at risk, not from war, since there wasn’t one yet, but from the ordinary decay that threatens any uncatalogued archive: paper rotting, negatives degrading, tape demagnetizing. That inquiry eventually reached Professor Marilyn Deegan at King’s College London, who would go on to lead the project as its managing director. By December 2017, with an initial £800,000 from the British Council’s Cultural Emergency Fund, later topped up to £1,170,000, and from 2019 a further $655,000 from the Aliph Foundation, formal digitization work had begun across institutions in Khartoum and beyond.
By 2022, the project’s team, spread across partner institutions in Sudan and collaborators in the UK, had digitized roughly 148,000 items drawn from forty-three public and private collections: photographs, manuscripts, museum objects, film and audio recordings, oral histories, and a 3D interactive reconstruction of what the historic port of Suakin might have looked like around 1900. Sixty thousand of those items, described in both English and Arabic, went up on a public website. The rest existed only in the project’s own storage, subject to agreements that varied collection by collection.
Nobody involved in 2013 was digitizing against a war that hadn’t happened yet. That war arrived in April 2023, when fighting broke out across Sudan and cultural institutions including the National Museum and the Khalifa House Museum were looted and damaged. But the years in between, the ordinary years of scanning manuscripts and cataloguing photographs under a government that fell in 2019, then a pandemic, then a fragile transition, turned out to matter more than anyone planning the project in 2013 could have known. A new history of the project, compiled by researchers at King’s College London from fourteen interviews conducted between January and May 2025 with the people who built it, keeps circling back to a single, under-examined idea: that digitizing something is not the same as de-materializing it. The digital copy still has to live somewhere. It still has to move through customs, survive a bad internet connection, and get carried, sometimes literally, from one place to another.
]]>Twenty centimeters is a shallow plough, and it accounts for a good deal about the site. It lifts sherds, brick fragments and the occasional inscribed piece into the plough zone and shifts them a short distance sideways, and it comes nowhere near the tops of buried walls. Nimrud, ancient Kalhu, was never intensively reoccupied after the Assyrian empire fell in 612 BC. The survey recorded material running from the third millennium BC through to the Islamic period, roughly the mid-seventh century AD onward, but nothing later ever settled on the site with the weight of the Assyrian town, so the houses and streets of the ninth to seventh centuries BC sit under farmland rather than under a medieval quarter, which is not the case at Nineveh, where modern Mosul has spread across a substantial part of the ancient city.
The lower town covers 340 hectares inside nearly eight kilometers of fortification wall.
In the 180 years since Austen Henry Layard opened his first trenches there in 1845, three buildings in all of that area have been dug at any scale: Fort Shalmaneser at the south-east corner, the Town Wall Palace, and the palace in the sector known as PD5. Everything Nimrud is famous for came off the acropolis, and it came off in quantity. Max Mallowan’s excavations in the 1950s recovered a stele on which Ashurnasirpal II (r. 883–859 BC), who made Kalhu his capital in 879 BC, records ten days of feasting for 69,574 people at the city’s inauguration, a figure given to the last digit. There is no comparable number for how many people actually lived in the lower town during any Assyrian decade; the estimates in circulation come from assumed densities multiplied by the walled area rather than from excavated houses. The Nimrud Lower Town Archaeological Project, run by Daniele Morandi Bonacossi and Francesca Simi1 of the Department of Humanities and Cultural Heritage at Udine, is an attempt to recover the street plan, the neighborhoods, the workshops and the water supply of an imperial capital using methods that mostly do not involve digging. The question underneath the fieldwork is whether a city of this size can be read from its surface, from sherd counts, a terrain model and the magnetic contrast between buried fill and the soil around it, well enough to say where a trench should go and, past that, well enough to say how the place was organized.
The starting point was photographs taken to look at something else. Declassified imagery from U-2 flights and the CORONA and HEXAGON satellite programs, dated between 1967 and 1974, covers northern Iraq at usable resolution, and its value now lies partly in its age, since it predates most of the land leveling, canal cutting and village expansion of the last fifty years. Jason Ur, an archaeologist at Harvard who works on Mesopotamian settlement, drew on this class of imagery in a 2013 paper on the morphology of Neo-Assyrian cities, and at Nimrud it showed the fortification circuit, the gates, a street system and blocks of denser and sparser occupation inside the walls. But a line on a photograph is only a proposal about a line in the ground. Before the 2025 survey could test any of it, the team needed a modern map, which they built with differential GPS and a drone: new benchmarks, an orthophoto, a digital terrain model, a site plan.
]]>That tomb is a small version of a larger problem, and both of them sit inside a new study of 149 ancient genomes from Gansu published in Cell1 on September 3. The work was led by Qiaomei Fu at the Institute of Vertebrate Paleontology and Paleoanthropology in Beijing, with archaeologists from the Institute of Cultural Relics and Archaeology in Gansu Province and collaborators elsewhere. The team generated 186 sequences from 185 samples, obtained 62 direct radiocarbon dates, and after quality filtering ended up with 149 usable individuals from eleven sites, spanning roughly 4,700 to 3,000 years before present. Ninety-five of those individuals come from Mogou alone.
Gansu is the obvious place to ask the question they were asking. The province sits at a junction: the Inner Asian Mountain Corridor to the west, the Loess Plateau to the east, the Mongolian Plateau north, the Tibetan Plateau south. By about 4,000 years ago, crops and animals first domesticated in southwest Asia were showing up there. Wheat. Barley. Cattle. Sheep and goats. Some of the earliest copper and bronze objects recovered anywhere in China turn up in the same region, and a few cultural features at these sites recall the early Bronze Age Andronovo horizon far to the west. Nine of the eleven newly sampled sites show western-derived crops or livestock in use.
So the genomes should show western Eurasian ancestry. They don’t.
Principal component analysis places every one of the newly reported Gansu individuals inside the range of East Asian variation, with no drift toward western Eurasian or Central Asian populations. ADMIXTURE finds no meaningful western component. D-statistics comparing the Gansu groups against four reference ancestries used as western Eurasian anchors, Anatolian Farmer, Iranian Farmer, Eastern Hunter-Gatherer and Caucasus Hunter-Gatherer, return affinities no different from what any other East Asian population produces. Three independent approaches, one answer.
This is not how it went in Europe. There, the arrival of farming came bundled with people: the Neolithic expansion of Anatolian farmers, then the Bronze Age expansion of Yamnaya agropastoralists, both leaving genetic signatures visible thousands of years later. Central Asia shows something similar. Gansu shows the opposite. The domesticates crossed the continent and the people who domesticated them apparently did not, or at least not in numbers that left a trace.
The authors are careful about how far that goes. Absence of evidence at the population level is not proof that no individual with western Eurasian ancestry ever set foot in Gansu. What the data rule out is a demographic event large enough to register. Whoever carried the wheat east, they were not founding populations. The plausible mechanisms are cultural transmission along the Inner Asian Mountain Corridor, or exchange through intermediary groups on the Mongolian Steppe, and the study notes that direct archaeological evidence for a steppe route remains thin.
]]>The furnace was in the Rooiberg valley in what is now South Africa’s Limpopo province, roughly 360 kilometres to the south-southeast. Somebody looked at this piece of industrial waste, decided it was interesting, and carried it that entire distance.
We can say that with some confidence because tin carries a geological signature. The sheet has a tin isotope ratio (δ124Sn of 1.4 per mil) that is distinctively characteristic of Rooiberg cassiterite, and a lead model age of about 2050 million years, matching the Bushveld Large Igneous Province that hosts the Rooiberg deposits. The object is a curiosity in the literal sense. Someone found it curious.
That sheet is one of the odder finds in a new study by Jay Stephens, Carla Klehm, Phillip Segadika and David Killick, published in Azania: Archaeological Research in Africa,1 which works out where the copper and tin at three Iron Age sites in east-central Botswana actually came from. The three sites are Bosutswe, a three-hectare hilltop settlement that began as a cattle post in the seventh century AD and was abandoned by the seventeenth; Khubu la Dintša, a much smaller settlement about 14 kilometres away, occupied roughly cal AD 1220 to 1445; and Lose, a hilltop site about 100 kilometres south, occupied from around AD 900 to 1500. They are the only three sites anywhere that have produced Lose pottery, a locally made ware with incised triangular motifs borrowed from Mapungubwe styles. Lose is therefore a site, a pottery type and a period name (c. AD 1200 to 1650) all at once, which is worth holding in mind for what follows.
Bosutswe was well connected long before any of this. Glass beads from the Indian Ocean trade, cowrie shells from the coast, a single gold spiral, chickens (Gallus gallus) and black rats (Rattus rattus) all turn up in its deposits. What has been much harder to see is what these communities traded with each other, inside the African interior, rather than with the Swahili coast.
Metals are the way in, but only two of them. Iron makes up about 7.1 percent of the earth’s crust, which means fingerprinting all its possible sources is not a research programme, it is a fantasy. Gold is the opposite problem: rare, but scattered across an enormous number of tiny deposits, and endlessly recyclable. Summers counted 1119 pre-colonial gold mines in what is now Zimbabwe and Botswana against 143 copper mines, and that figure excludes alluvial gold. Copper, at 75 parts per million of the crust, and tin, at 2.5 ppm, sit in the sweet spot. They are common in archaeological assemblages and scarce in the ground, and since 2014 the team behind this study has been building a reference database of lead isotope ratios for southern African ore deposits, now over 800 entries.
The Bosutswe metals sort into four groups: 14 unalloyed copper objects, two pieces of “dirty copper” with trace tin, 25 bronzes, and the lead-tin sheet. Nearly all of it is jewellery or the debris of making jewellery. Beads, clips, wound-wire helices, bangles.
The copper in those objects comes from almost everywhere. Cobalt-bearing deposits on the Central African Copperbelt and the Kipushi deposit on the Congo-Kinshasa and Zambia border, more than 950 kilometres away. Copper Queen in northwestern Zimbabwe, over 500 kilometres. Musina and Phalaborwa in South Africa, 380 and 515 kilometres. Rooiberg copper, from small deposits in the same valley as the tin. Selebi Phikwe and Dukwe in Botswana, 150 and 140 kilometres. And an unidentified vein-type deposit somewhere in the Transvaal Supergroup rocks of northern South Africa.
There is no dominant source. That is the finding. Contemporary Mapungubwe, 350 kilometres east, drew its copper almost exclusively from Phalaborwa; Bosutswe has exactly one unambiguous piece of Phalaborwa copper in the whole sample.
]]>All the skulls faced east. Each came with its own articulated fore- and hindlimbs, the feet still in anatomical order, which is what archaeologists mean by a “head-and-hoof” deposit and which usually indicates the hide was still attached when the animal went into the ground.
Directly east, about five metres away, stood a small timber building.
This is the site at Land East of Kings Warren, Red Lodge, in the chalk country of west Suffolk, excavated by Archaeology South-East in 2016 ahead of a housing development and published this year in Britannia1 by Angus Forshaw. The excavation turned up material from the Early Neolithic through to the Middle Ages. The Roman-period parts are the ones worth sitting with, because they are built on something much older.
That is a small result from a small prototype. The larger claim behind it is that the instrument doing the looking cost roughly a twentieth of the hyperspectral rigs previously brought onto excavations, runs on a Raspberry Pi 4, and can be powered off a cordless drill battery.
The problem it addresses is old and specific. Excavation is the identification of stratigraphic units followed by their removal in sequence, which means the whole enterprise depends on an archaeologist being able to see where one unit stops and the next begins. Usually that works. In dark earths it does not. Where organic content is high, everything in the profile is black, and the contrasts that define stratigraphic events are either faint or absent in visible light. The units get sorted out later, in post-excavation analysis, which is a poor consolation given that the deposits themselves are gone by then.
Sorte Muld, on Bornholm, is the case where this bites hardest. The name means black earth, and the site earned it: more than a meter of cultural deposits accumulated during occupation running from roughly AD 0 to AD 1000, laid down by a settlement that was a power center in the Baltic and is best known for more than 3,400 gold foil figures dating to the sixth and seventh centuries. Buildings there were rebuilt repeatedly. The busier a given part of the complex was, the messier its sedimentology. Some of the resulting layers are a few millimeters thick. Whether a given one was even visible on a given day could depend on how wet the soil happened to be.
The profile the team worked1 on was reopened during a 2019 excavation, catalogued as Margrethe Watts, BMR1191, to supply material for dating and environmental analyses. The imaging happened there.
The video shows the LEDMSI flash illuminating the soil at different wavelengths. Credit: David Stott, Moesgaard
]]>The tomb belongs to King Kaolie of Chu, who died in 238 BC. It is the first properly excavated royal Chu tomb of the Warring States period, and AMS radiocarbon dating of plant material from the bundles themselves returned an age of roughly 2200 cal BP, consistent with the king’s death date rather than with any later intrusion. Fan Yang, Yong Ge and colleagues took1 twelve bundles at random, cleaned them in the lab, and unwrapped them.
The wrapping leaves came out identifiable. Oblong to obovate, short acute tip, wavy-toothed margins, pinnate venation with the tertiary veins running perpendicular to the secondaries: that is Quercus dentata, the daimyo oak, and the team backed the leaf-shape match with a comparison of epidermal cell patterns against modern specimens. The cords were harder. There were no diagnostic hemp fibres, which rules out Cannabis sativa, the obvious candidate. What the cords did yield was phytoliths, silica bodies that form inside plant cells and survive after the tissue rots. Bilobate and saddle forms dominated, the signature of the grass subfamilies Panicoideae and Arundinoideae. In two cord samples a good share of those phytoliths were still articulated, meaning they sat in their original cellular arrangement rather than washing in from surrounding sediment. The cordage was made from grass, probably from more than one kind.
Inside: 456 seeds from ten of the bundles. Rice, Oryza sativa, made up 43.6 percent. Foxtail millet, Setaria italica, 28.5 percent. Broomcorn millet, Panicum miliaceum, 26.1 percent. Also present were a single seed of Zanthoxylum bungeanum and one of Malva sp., both aromatics used as spice, and six seeds of Scirpus juncoides, a weed of flooded rice fields that probably came along with the harvest.
So far this reads like lunch. A leaf packet of rice and millet, tied with grass rope, put in a tomb. Anyone who has eaten zongzi at the Dragon Boat Festival knows the object.
]]>Windmill Way is a 29-metre rockshelter on a sandstone ridge in southeast Cape York Peninsula, and the alcove at its northern end is a miserable place to live. The ceiling is under a metre high. You would have to crawl. What that alcove contained, though, was thousands of fragmented animal bones, many of them spirally fractured, many of them gnawed, some still carrying dried flesh. The obvious reading was carnivore accumulation. Dingoes drag carcasses into low sheltered spaces and chew them there. The excavators reasonably concluded that most of the faunal assemblage was not the product of human hands at all.
Then someone noticed a blackish-brown ball stuck to the root end of a single macropod incisor. Small, neat, deliberately shaped. Not the kind of thing that happens to a tooth by accident.
That observation, plus the identification of an unambiguous bone point, forced a reassessment. Every tooth and jaw fragment from the excavation went back under inspection. Ten teeth showed clear macroscopic resin. Two more were flagged as possible. Twelve in total, and Lynley Wallis and her colleagues have now published the analysis in Australian Archaeology.1
All twelve are mandibular incisors. Six come from agile wallabies (Notamacropus agilis), three from spectacled hare wallabies (Lagorchestes conspicillatus), two from northern nail-tail wallabies (Onychogalea unguifera), and one is too broken to identify with confidence, though its width points to N. agilis as well. All three species live in the country around the shelter today. All three are represented in the gnawed bone assemblage. The most economical explanation is that whoever made these objects picked the teeth out of the debris already lying on the shelter floor, which is a strange and rather intimate image: someone sorting through the leavings of dingo meals for the right sort of tooth.
Radiocarbon dating went directly to the resin on two specimens, returning ages of roughly 1,170 to 1,290 calibrated years before present. That detail is worth pausing on. Nearly every comparable ornament from Australia comes from a burial and is dated by proxy, from the age of the human remains rather than the object. These two resin balls were dated as themselves.
]]>Fly a laser scanner over that forest and you find a city underneath.
That part is no longer surprising. Airborne lidar has been reshaping Maya archaeology since the Caracol survey in 2011, and by now the pattern is familiar. Pulses of laser light go down, most bounce off leaves, a fraction slip through gaps and hit the ground, and enough of those ground returns come back to build a digital terrain model at half-meter resolution. Strip away the vegetation digitally and the buried landscape appears: plazas, causeways, canals, house mounds, low walls running for kilometers through what looks from the ground like undifferentiated jungle.
What Sara Eshleman and colleagues did in a new paper in Scientific Reports1 is look at the part everyone throws away.
The vegetation returns. All those pulses that hit the canopy and never made it to the ground are, for archaeological purposes, noise to be filtered out. But they encode the three-dimensional structure of the forest in considerable detail: how tall it is, how the leaf mass is distributed vertically, how bumpy the upper surface is, how much light gets through. Foresters and ecologists have been using exactly this data for years to estimate carbon and biomass. Nobody had systematically correlated it with the archaeology sitting directly beneath.
The team ran that correlation across a 47 square kilometer study area covering two large Maya centers, Wari Camp and Gran Cacao, plus the Central Rio Bravo wetland field complex, an 8 square kilometer expanse of rectilinear canals cut into the floodplain next to Wari Camp. Occupation ran from about 2900 to 1000 BP, with the population peak during the Classic Period, 1400 to 1000 BP. The wetland complex itself was in use from roughly 2030 to 840 BP. The canals, in other words, went out of service more than eight centuries ago.
Three categories of ancient feature were mapped: settlement structures, wetland field systems, and upland wall features. Rather than trusting a single archaeologist’s eye, two researchers independently digitized every human modification they could identify and assigned confidence levels to each call. Those judgments were combined into continuous indices, so what enters the statistical models is not “there is a house mound here” but “here is how much confident evidence of settlement exists in this grid cell.” Topographic variables went in alongside: slope, curvature, aspect, elevation, local and regional topographic position. Topography has long been understood as the dominant control on vegetation in this region, so it needed to be held accountable before anything could be attributed to the Maya. Spatial error models handled the autocorrelation that inevitably shows up when you sample a landscape comprehensively.
The result is that each of the three feature types leaves a different mark on the forest above it.
]]>Another, ARJB V0522, from the Saint Eusebius church cemetery in Arnhem, was around ten years old. One facet joint on T5 already shows the pitting and contour change that pathologists associate with cartilage failure.
These two are among 143 individuals, aged from four to twenty-five at death, analysed by Alex Tutwiler and Rachel Schats of Leiden University in a paper published in the International Journal of Paleopathology.1 The question they set out to answer is one that Dutch historical records, for all their bureaucratic thoroughness, cannot touch. The archives tell us how many children worked, at what age they started, how many hours they put in and what they were paid. They say almost nothing about what the work did to the bodies doing it.
Three cemetery assemblages, three different lives. Keyserkerk in Middenbeemster is rural and low-to-middling in status, its economy built on dairy cattle and cheese; the burials there date from 1829 to 1866, and 51 individuals from the site met the study’s criteria. Eusebiuskerk in Arnhem is urban and poor, drawn from the outer northern portion of the churchyard where the lower social strata were interred; it dates from 1626 to 1829 and contributed 49 individuals. Nieuwe Kerk in Delft is the control of sorts: 43 individuals from graves in the church choir, dating from 1624 to 1814, belonging to families wealthy enough to be buried inside the church and almost certainly exempt from work. Arnhem had tobacco processing, textile manufacture, brick-making and paper mills. Delft had trade, and children who studied.
What Tutwiler and Schats looked for were three kinds of change, all of them things that clinicians normally see in middle-aged and older patients. Facet joint osteoarthritis is degeneration of the small paired synovial joints at the back of each vertebra, the ones that let the spine bend and twist while bearing up to a third of the load passing through a spinal segment. Schmorl’s nodes are the disc-through-endplate herniations described above, linked clinically to compressive and shearing force. Vertebral body margin modifications are new bone forming as a lipped rim around the edge of the vertebral body, most often in response to disc degeneration. None of these is caused by labor alone. All three have genetic, developmental and traumatic contributions. What makes them interpretable here is age. In a fifty-year-old the signal is drowned out by ordinary senescence. In a nine-year-old there is much less noise.
Middenbeemster came out worst on every measure the study applied. Overall lesion prevalence there reached 62.6 percent in the upper thoracic vertebrae and 68.4% in the lower thoracic, higher than any spinal region at either of the other sites. The gap held across every age category, from the youngest children through the early young adults, and it held for Schmorl’s nodes and for margin modifications assessed separately.
The upper thoracic figure is the strange one. That stretch of spine, T1 through T6, is the least mobile part of the column, splinted by the ribcage and comparatively protected from the flexion and rotation that wreck the lumbar region. Children in a dairy village were wearing it out anyway. Tutwiler and Schats attribute this to the muscles that anchor there rather than to the vertebrae themselves. The rhomboids and trapezius originate on the spinous processes of the upper thoracic spine and stabilize the shoulder girdle, and they are exactly the muscles a small body recruits when it lifts something heavier than it should. Beemster children were milking cows and taking part in cheese-making by the age of nine, according to an agricultural survey from 1800. Cheese-making meant hauling milk basins, churning, and lifting and pressing moulds, done in a hunched posture, repeatedly, by people who did not yet have the body mass to do it without strain.
]]>