Dead Stars, Old Spikes and a Chair

A team led by astronomers at the University of Warwick's Centre for Exoplanets and Habitability reports what may be the first evidence of a second-generation planet¹. That means a world formed from material a dying star expelled, not from the disc of gas and dust around a newborn star. The result was published on 5 October in Nature Astronomy, with co-authors from the Instituto de Astrofísica de Canarias, MIT, the Harvard & Smithsonian Center for Astrophysics, the Space Telescope Science Institute and other institutions. No one has seen the planet directly. It is inferred from the chemistry of its star and stays a candidate until confirmed.

Planets are normally thought to form once. A young star is surrounded by an accretion disc², a flattened ring of leftover gas and dust that spirals inward, and dust grains in that disc clump into planets. Earth formed this way about 4.5 billion years ago. When a Sun-like star nears the end of its life, it swells into a red giant, sheds its outer layers, and leaves a white dwarf³: a small, dense, very hot core. Astronomers have long wondered whether some of the shed material could settle into a fresh disc around the white dwarf and build new planets. That is hard to test, because the resulting planets would be faint, close to their stars, and easy to confuse with old planets that survived the star's death.

Re-reading a 1999 spectrum⁴

Lead author Williams and colleagues went back to the archive with an updated database of atomic lines. Many of the mystery lines fit niobium⁵ (Nb), a metal used on Earth in steel alloys and superconducting magnets. "What Hubble is showing us in this white dwarf system is something we haven't seen before: a high abundance of the element niobium," Williams said, adding that she didn't recognise its signature when she first found it. Niobium had never been reported in any white dwarf. The team checked the Hubble result against archival data from the Far Ultraviolet Spectroscopic Explorer (FUSE)⁶, a NASA ultraviolet observatory that is now retired.

The team also used four months of observations from the Transiting Exoplanet Survey Satellite (TESS)⁷, which watches for the small dips in brightness caused by planets passing in front of their stars. The article doesn't say what TESS saw, or failed to see, or how it constrained the planet. From the combined data the team estimates a gas giant roughly the size of Jupiter orbiting about 6 million km from the white dwarf. For comparison, Mercury orbits the Sun at about 58 million km.

Williams is optimistic about the planet's future: "If the second-generation planet is there, I think it is likely to survive. Eventually the white dwarf will cool and then maintain a consistent temperature, with the planet in its stable habitable zone⁸ for millions of years." For a gas giant, "habitable zone" only means the range of distances where temperatures would allow liquid water. It doesn't mean the planet could support life.

APOE4¹ is the strongest common genetic risk factor for Alzheimer's disease. Joel Blanchard, a neuroscientist at the Icahn School of Medicine at Mount Sinai in New York City, puts it this way: in a room of 10 people with Alzheimer's, six would carry it. But having the variant doesn't mean getting the disease. Many carriers never develop dementia, and researchers still can't say exactly how the gene makes the brain more vulnerable. A study by Blanchard and colleagues, published September 24 in Cell, suggests one route. APOE4 appears to push the support cells around the brain's small blood vessels to become scar-producing cells, and that scarring may help amyloid-beta² build up.

APOE is the gene for a protein that carries fats in the bloodstream. It comes in three common forms, or alleles³, and APOE4 is the one tied most strongly to Alzheimer's. The puzzle is how a gene for a fat-carrying protein connects to a disease defined by plaques of amyloid-beta, a protein whose buildup is one of the first signs of Alzheimer's in the brain.

The lead for this study came from people who should have been at high risk but weren't. In 2024, a study of more than 3,500 people included some who carried two copies of APOE4 and had not developed Alzheimer's. Their DNA turned out to have a mutation that left their blood vessels with much less fibronectin⁴. Fibronectin is a sticky protein that acts like cellular superglue during wound healing. Those researchers proposed that too much of it could make it harder for the brain to clear amyloid-beta.

The team built an atlas of gene expression⁵, meaning which genes each cell is actively reading and turning into proteins, for the brain's blood vessels, one cell at a time. They didn't collect new tissue. Instead they combined data from three existing brain-wide maps, which gave them 66,000 vascular cells from 220 people. They then compared people with two copies of APOE4 against people with lower-risk APOE variants.

Two differences stood out. APOE4 carriers had fewer pericytes⁶, star-shaped cells that wrap around small blood vessels and keep them stable. They also had more myofibroblasts⁷, cells that normally appear after tissue injury, lay down scar tissue and produce large amounts of fibronectin.

The team also looked for what flips the switch. In the APOE4 models, signaling through TGF-beta⁸, an immune signaling molecule known for its role in wound healing and scarring, helped push pericytes toward the scar-producing state. Blocking TGF-beta signaling in the cell models and in aged mice restored the share of blood vessels wrapped by pericytes.

This year marks 100 years since two 1926 papers by Edgar Douglas Adrian and Yngve Zotterman showed that neurons communicate through trains of identical spikes¹. Carandini argues that this was the discovery of the neural code. Writing in The Transmitter, the neuroscientist Matteo Carandini argues that the field should celebrate the anniversary the way biologists celebrate the genetic code. To show what the two men actually saw, he has also re-processed their original recordings so that modern eyes can read them.

Adrian and his mentor Keith Lucas had already shown that nerves respond in an "all-or-none²" way. A current applied to a motor nerve made the attached muscle contract only if the current passed a threshold. Above that threshold, a stronger current produced no stronger effect. The property seemed to belong to the nerve itself and to apply to all nerves.

The obstacle was size. Measured inside a cell, a spike is tens of millivolts. Measured from outside the fibre, it is about a thousand times smaller, roughly tens of microvolts. The fix came from radio: the vacuum valve³ (vacuum tube). Groups at Harvard and Washington University had begun using valve amplifiers on nerve signals.

When Adrian first tried one, he was discouraged. The fluctuations in his motor-nerve recording looked random, like noise or equipment artifacts. Then he noticed they appeared only when the muscle was stretched. In his words, "The explanation suddenly dawned on me": he was recording sensory impulses travelling up the nerve from the muscle spindles⁴, the muscle's stretch sensors. These were the first recorded spike trains. But they mixed many fibres together, and Adrian suspected that with more work he could isolate one.

Carandini points out that the published traces are almost unreadable today. One reason is that time runs right to left. The bigger reason is the recording device, a capillary electrometer⁵. This is a fine glass pipette filled with mercury. Voltage moves the mercury slightly, and the movement was magnified under a microscope, projected and filmed. Adrian chose it because it was robust and survived the amplifier's occasional wild artifacts. But the mercury is slow to settle after each movement, so the signal is smeared out in time.

Recovering the true signal requires deconvolution⁶, a mathematical step that undoes a known blurring. Carandini compares it to how modern researchers infer spikes from slow calcium-indicator signals. In 1926, deconvolution could be done with a mechanical device, but more often it was left to the imagination of the experimenters and their readers.

The same 1926 work reported the first property of what neurons encode. Under steady stretch, firing slowed over time, which Adrian called "adaptation⁷." To show it, he plotted the nerve's firing rate against time, before, during and after the stretch.

Carandini calls this a watershed. It is the first graph of a neuron's firing rate against anything. Today such plots are routine. This kind is called a PSTH⁸, or peri-stimulus time histogram: firing rate plotted against time relative to a stimulus. Plots like it document a century of landmark findings:

A single three-dimensional shape that fills all of space without gaps or overlaps, but only in patterns that never repeat, has been announced. That makes it a 3D "einstein¹." The shape is called Chair44. It was reported in a preprint², a paper not yet peer reviewed, by Ioannis Tsiokos. He directs Automorph, which he calls an AI research lab, and he describes himself as "not a mathematician." He is the only human author. The result has also brought out a sharp argument about how AI-produced mathematics should be presented and checked.

The answer for flat surfaces came in 2023. David Smith, an amateur, found a hat-shaped tile. With Joseph Samuel Myers, Craig Kaplan and Chaim Goodman-Strauss, he proved in a paper later published in Combinatorial Theory that the hat is an aperiodic monotile³: a single tile that can cover the plane but never periodically. Such shapes are nicknamed "einsteins," from the German ein Stein, "one stone." Hats group into larger hat-like clusters, and those group into larger ones again. However far the tiling extends, the arrangement keeps changing. The find was popular enough that researchers held a two-day "Hatfest" at the University of Oxford.

In three dimensions there was already a candidate, the Schmitt-Conway-Danzer biprism⁴, a convex polyhedron known for decades. Kaplan says it "obeys the letter of the law of aperiodicity but not the spirit." In standard terms it is weakly aperiodic. No tiling with it repeats by a pure shift, but the tilings still have a screw symmetry⁵, a rotation combined with a slide along an axis, that carries the pattern onto itself forever. A strongly aperiodic tile allows no infinite symmetry of that kind at all. Finding a strong 3D example had been open for decades. Chair44 is claimed to close it. Kaplan says he is "pleased that we have the solution to what was, until recently, a nagging open problem."

The article's description of this step is inconsistent. One passage says four chairs make a bigger one. The figure caption and Goodman-Strauss both say eight. Volume settles it. A chair twice as large in every direction has 2³ = 8 times the volume, 56 small cubes instead of 7, so it takes eight copies. Each level is called a supertile⁶.

The mathematical claim now has backing from specialists who did not rely on the original paper's reasoning. That is a stronger position than a single AI-written preprint. The harder questions are about how the work gets done. Flicker calls it "a new important result" but warns that "our jobs get reduced to sifting through AI slop in search of nuggets of gold." He places it in a summer in which, he says, AI models have repeatedly cracked long-standing open problems, including one of the seven Millennium Prize Problems⁷. The article gives no detail on those claims.

Goodman-Strauss takes a more measured view. He expects Chair44 to be "the first example of something that we're going to be encountering a lot of." He suggests mathematicians revisit hard questions they set aside, since LLMs⁸ (large language models, the technology behind chatbots) may now help as tools, though not as a substitute for human reasoning. The unsettled part, he says, is "what's expected in terms of communication": who must write a proof that people can read and check before a result counts as known.

Nobody counted. Within a few decades of contact, the "Great Dying¹" had wrecked the societies that could have been counted. That was anthropologist Eric Wolf's name for the combined toll of epidemic disease, warfare and enslavement. Any estimate therefore has to work backward from indirect evidence, and its answer depends on the assumptions behind it.

The highest well-known figure came from anthropologist Henry Dobyns in 1983: about 112 million people, including 18 million in North America. He started from carrying capacity², meaning how many people each type of landscape could support. Then he used records of 16th-century epidemics to work backward. "In essence, Dobyns filled up the landscape to its maximum carrying capacity, so that all environmentally similar places were occupied by the same number of people," Milner said. Dobyns also assumed epidemics reached everyone, which only makes sense if the land really was full. Fieldwork in the 1970s and 1980s showed the Americas were settled unevenly, so both assumptions overshot.

Pottery and house counts differ from place to place. Robert Kelly, an archaeologist at the University of Wyoming, argues for a measure that works everywhere: "Everyone had to build fires for heat and for cooking." Charcoal from hearths, along with animal bone and plant remains, can be radiocarbon-dated. North American dates are collected in the Canadian Archaeological Radiocarbon Database (CARD)³.

Matthew Peros, a geographer at Bishop's University in Quebec, did this with co-authors in 2009, when CARD held about 25,000 dates. The curve rose sharply around 50 B.C. and peaked around A.D. 1150. "I had assumed it would be more like 1500 just following the arrival of the Europeans," Peros said. One possible reason for the decline after 1150 is that people gathered into denser settlements, where diseases such as tuberculosis killed more of them. Kelly and colleagues repeated the analysis in a 2025 study in the journal PNAS⁴, by which time CARD held more than 100,000 dates. They found the same peak, plus a slight recovery just before 1500. "The Indigenous population did not simply increase to a peak at 1500 CE and then decline," they wrote.

A rough calculation (illustrative, not the study's actual method) shows how much depends on these inputs. Growing from 1,000 people to 2.5 million is about 11 doublings. Spread over roughly 13,850 years, that works out to an average growth rate of about 0.056% a year. Change the start date or the size of the founding group, and the endpoint changes too. The authors note that an earlier arrival would push the totals higher, and so would a larger founding group. Footprints at White Sands⁵, New Mexico, dated to about 23,000 years ago, suggest an earlier arrival is possible.

Kelly argues that the timing could have changed history. Around A.D. 1000, Cahokia⁶, near modern St. Louis, covered 6 square miles (16 square kilometers) and held nearly 20,000 people. "They could have fielded thousands of warriors," he said. "It could have been a very different history."

The radiocarbon approach shows changes over time better than it gives totals. The totals depend on the founder and growth assumptions above. The article doesn't discuss the method's known biases, so these come from general knowledge. Older material survives less well, which can make recent periods look busier. Archaeologists also date some sites and periods more than others. Uneven stretches in the calibration curve⁷ used to convert carbon-14 into calendar years can also create bumps in the summed curve. The larger 2025 dataset makes the A.D. 1150 peak more robust, but does not remove these problems.