Longevity InTime: Autonomous AI Institute. Anti-Aging Digital Health Immortality Transhumanist AI Channel

GSK tests Chai Discovery's protein variants and gains access for drug discovery On October 9, Chai Discovery and GSK announced a collaboration. Prior to the agreement, GSK tested protein variants proposed by Chai's models for its therapeutic targets in its own laboratory. To find a protein drug, researchers first select a therapeutic target — a molecule in the body they wish to modulate. Then they seek a binder: a protein that can attach to that target. Chai proposed such variants for GSK's targets without specialized model training on each one. The model operates experimentally: it suggests protein sequences. The lab tests these variants on real molecules and measures whether they bind to the intended target and the strength of that binding. Thus, a computational variant becomes a lab-validated result. GSK tested these variants on several targets and found binders for each one tested. Following this validation, GSK and Chai expanded their work into a collaboration. GSK will gain access to Chai's models for predicting protein folding — their structure — and for designing new variants in its drug discovery programs. 🔗 Read original →

Alveolar cells help repair bronchioles after lung injury in mice In a study published October 7 in Nature, researchers used different genetic labels to track alveolar and bronchiolar cells in mice. These labels persisted in the cells and their descendants. After chemical injury, some alveolar cells gave rise to new cells in the bronchioles, and the experiments also traced the signals guiding this transition. Alveoli are tiny lung sacs where blood picks up oxygen, while bronchioles are airways leading to them. Alveolar type 2 (AT2) cells maintain the alveoli; bronchioles are lined in part by club cells, which secrete a protective fluid. An older genetic label remained in both AT2 cells and BASC — rare stem cells at the alveolus-bronchiole border. A new labeling scheme gave these groups distinct tags, allowing researchers to track each contribution separately. Four weeks after chemical injury, 52% of the regenerated club cells originated from AT2 cells. Another 27% came from BASC, and 15% from pre-existing club cells. AT2 descendants appeared in the bronchioles as both club cells and ciliated cells, which move mucus along the airways. In this repair, AT2 cells first moved toward the bronchiole, and then some of them became club cells. Immune cells near the injured bronchiole secreted the protein SPP1 — a signal that attracted AT2 cells to the injury site. AT2 cells recognized this signal via integrin β1, a receptor protein on their surface. Surviving ciliated cells passed along the signal JAG1, which activated the Notch pathway and directed AT2 cells to become club cells. In separate experiments, blocking SPP1 or integrin β1 reduced AT2 cell movement, while interfering with Notch reduced club cell formation. In this mouse model, cells from the gas exchange zone became part of the repair of an adjacent lung region. 🔗 Read original →

PLPP2 enzyme aids mouse liver recovery after acute injury October 8, journal Signal Transduction and Targeted Therapy published a study on liver recovery after acute injury. Authors conducted gain- and loss-of-function experiments with PLPP2 in mice. In human liver biopsies, they observed a distinct correlation between this enzyme’s level, cell division, and injury markers. Liver regains lost tissue when its main cells, hepatocytes, resume division. After toxic injury with acetaminophen, authors measured proteins and metabolites in mouse liver over several days. In the first 6–24 hours, damage signs predominated; by 48–96 hours, hepatocyte division and tissue repair increased. Simultaneously, PLPP2 and phosphatidylethanolamine (PE), a membrane lipid, rose. Authors tested whether PLPP2 and PE participate in recovery or merely correlate temporally. To check, they knocked out Plpp2 only in hepatocytes. After acetaminophen, these mice showed larger dead tissue areas, higher blood ALT and AST—liver injury markers—and fewer dividing cells. When authors boosted Plpp2 expression via AAV vector delivering an extra gene copy, ALT and AST approached normal, dead tissue area decreased, and dividing cells increased. Plpp2 deficiency also slowed recovery after 70% hepatectomy, carbon tetrachloride toxicity, and temporary blood flow halt. Authors then dissected the cellular pathway. PLPP2 participates in a reaction sustaining PE synthesis. PE binds to protein LC3B when the cell assembles a membrane vesicle for recycling damaged material. Rapamycin, enhancing this recycling, partially alleviated PLPP2 deficiency effects; chloroquine, blocking it, removed the advantage of elevated PLPP2 expression. PE binds to protein PEBP1 and strengthens its binding to GLI1, a regulator of cell division–related genes. In this complex, GLI1 less frequently binds to ITCH, a protein marking GLI1 for destruction, thus persisting longer. Authors verified this sequence on proteins and cells, then correlated it with liver recovery in mice. In biopsies of 36 patients with drug-induced liver injury, high PLPP2 level associated with more dividing cells and lower liver enzyme levels; for comparison, they used 31 samples from patients with hepatic hemangioma. Sequential measurements, genetic intervention, and two PE action branches yield a model where membrane lipid exchange helps injured mouse liver restart cell division. Signal Transduction and Targeted Therapy, October 2026 🔗 Read original →

Authors of a GeroScience article urge treating lifespan extension as a goal of aging research alongside healthy years October 7 David Gems, Ekaterina Iidjima, and João Pedro de Magalhães published an opinion article in the journal GeroScience. They examine a formula that limits the goal of aging research to years of life lived in good health and independence, and propose including lifespan extension as well. Geriatrics — a branch of medicine for older people — deals with diseases and loss of independence in late life. Its task is to push these years closer to death, so that chronic diseases occupy a smaller portion of life. Geriatrician James Fries called this strategy “compression of morbidity.” Biogerontology, the science of aging mechanisms, poses a different question: can we alter the very pace of aging? In animal experiments, some interventions have extended both healthy lifespan and total lifespan. The authors consider several possible outcomes of such an intervention in humans: healthy years may increase, but the late period of health limitations could also lengthen. Doctors already make decisions involving similar trade-offs: treating cardiovascular disease, foot amputation due to gangrene threatening life, and brain tumor removal can delay death while leaving the person with limitations. “Life with moderately reduced health is usually preferable to death.” Therefore, the authors propose evaluating future interventions by two questions: do they add years of healthy life, and do they help preserve life? The decision about such treatment, they link to the individual’s own preference. GeroScience, October 2026 🔗 Read original →

Oocytes can remain arrested in mouse ovaries for months after exiting dormancy In a study tracking oocytes that had resumed growth in young mice, researchers found that a small fraction persisted in the ovary for up to a year. After 12 months, 4.17% of the originally labeled oocytes remained. Oocytes labeled nine months earlier were still capable of producing healthy offspring after fertilization. The team genetically labeled oocytes that had exited the resting pool in young mice and followed them over time. After two months, 47.97% of labeled oocytes remained; after six months, 15.46%; and after one year, 4.17%. When oocytes labeled nine months prior were fertilized in vitro, some developed to the blastocyst stage, and two-cell embryos transferred to recipients yielded healthy pups. Natural mating also produced labeled offspring at six and nine months post-labeling. The longest delays occurred at the secondary follicle stage, where oocytes are surrounded by multiple layers of supporting cells. While the primary stage lasted less than a month, the secondary stage ranged from several weeks to at least 11 months for a subset of cells. In long-term observations, about 80% of labeled oocytes were found at this stage. In mice labeled at 10 months of age or in young mice after unilateral ovariectomy, activated oocytes were depleted more quickly, linking the rate of oocyte loss to time spent in the secondary stage. Long-lived oocytes showed reduced RNA and protein synthesis, including lower levels of metabolic proteins. The authors implicated FURIN, an enzyme that processes certain proteins into their mature forms. Low FURIN levels correlated with prolonged oocyte arrest. Deleting Furin in oocytes extended the secondary follicle stage and reduced the mature form of GDF9, a signal from oocytes to surrounding cells. Adding exogenous GDF9 to cultured follicles partially restored growth. In this mouse model, activated oocytes progress toward ovulation at variable rates, with some delayed for months in the secondary follicle. The FURIN–GDF9 signaling axis regulates this tempo. 🔗 Read original →

CircuitATLAS preprint: AI agent proposes ATP1A3 protein, mouse experiment shows weaker cortical response to chemical challenge On October 7, the authors posted a preprint on arXiv about CircuitATLAS — a knowledge graph that an AI agent uses to identify molecular targets for neural network dysfunction. The system proposed ATP1A3, and experiments in awake mice showed how increased production of this protein alters the cortical response to chemically induced hyperexcitability. The same pattern of neural network dysfunction can arise from different molecular causes. CircuitATLAS seeks an intervention point starting from measurable network activity: the agent proceeds from altered rhythm to brain region and cell type to protein. The graph contains 3.83 million elements and 7.66 million connections between diseases, their manifestations, electrical activity, brain regions, cells, and molecules. The authors removed pre-existing “disease–gene” and “disease–protein” links from the searchable portion to force routing through network function. For the cortical hyperexcitability model, the system selected ATP1A3 — a neuronal sodium-potassium pump variant. During frequent impulses, it helps inhibitory neurons restore electrical charge to baseline and curb cortical activity. The authors hypothesized that extra ATP1A3 production in these cells would blunt the response to a sudden chemical challenge. To test this, they increased ATP1A3 production in cortical inhibitory neurons. Then, awake mice received a local injection of 4-aminopyridine — a substance that sharply increases excitability — and the increase in cortical electrical oscillation power was measured. The experiment included five control animals and six with extra ATP1A3 production. In all three oscillation bands, the median increase was lower in the second group: for beta rhythm, 1.33 versus 11.16 in controls; for low gamma, 1.19 versus 9.73; for high gamma, 1.02 versus 7.81. After the experiment, the authors launched a computational search for small molecules and obtained 250 compounds predicted to bind ATP1A3. The next functional test will determine which of these enhance pump activity and which inhibit it. arXiv preprint, October 2026 🔗 Read original →

Life Biosciences reports first human data for ER-100 in glaucoma On October 8, Life Biosciences released interim Phase 1 data showing that three participants with open-angle glaucoma received a low dose of ER-100 without serious complications or dose-limiting reactions through day 56. The Data Monitoring Committee recommended advancing to the next dose level. In two of the three participants, at least 28% of points in the Humphrey visual field test improved by more than 4 decibels by day 56. The Humphrey test measures light sensitivity across the visual field by having the patient fixate on a central point and detect flashes of light in peripheral locations. ER-100 is a one-eye gene therapy that uses a viral vector to deliver instructions for three proteins—OCT4, SOX2, and KLF4, collectively known as OSK. After intraocular injection, participants take doxycycline for eight weeks to induce expression of these proteins. The therapy targets retinal ganglion cells, which transmit visual signals from the eye to the brain and are damaged in glaucoma, along with their axons. This approach follows 2020 mouse studies in which researchers induced glaucoma-like damage via microbeads to elevate intraocular pressure, then administered the OSK-carrying viral vector. In mice with induced OSK expression, axon density returned to healthy levels, visual acuity improved, and retinal electrophysiological responses increased—demonstrating that inducible OSK activation can reverse cellular damage after injury. The human trial is described as open-label and sequential: glaucoma patients first receive a low dose, then a higher dose. The primary goal of Phase 1 is to assess safety and tolerability, making the committee’s dose-escalation recommendation a direct outcome of the initial 56-day observation. Life Biosciences first administered ER-100 to a human in June; this report adds 56 days of follow-up and published visual field results from the initial cohort of three. 🔗 Read original →

Google DeepMind survey shows AI shifts research bottlenecks to later stages Around 41% of scientists reported a growing backlog of untested hypotheses, according to a Google DeepMind survey published October 8 by Alex Imas and James Manyika. The study analyzed roughly 15 million anonymized Gemini interactions, a catalog of 2,690 specialized scientific models, and responses from 637 active researchers in the US and UK. Economist Joel Mokyr describes scientific progress as a loop: new tools reshape research, and resulting knowledge drives the next generation of tools—into which Imas and Manyika place AI. Their data show language models are most often used for coding, analysis, literature search, and text drafting, while specialized systems predict properties, generate task-specific data, and run simulations. Researchers then decide which results merit experimental validation through lab work, field data collection, or clinical trials. This gap between computation and experiment is evident in AlphaFold, which predicts protein structures from amino acid sequences and has generated predictions for over 200 million proteins. Yet a US National Bureau of Economic Research study found no significant drop in experimentally determined protein structures after AlphaFold’s release—predictions help prioritize targets, but function and structure still require experimental verification. In the survey, about 44% of respondents said their main constraint had shifted to later stages over the past two years: lab experiments, clinical validation, or manuscript preparation. Meanwhile, roughly 46% of those who save time with AI spend more than a quarter of that savings on auditing, debugging, and verifying its outputs. Herbert Simon warned in 1971 that “a wealth of information creates a poverty of attention”—a dynamic the authors extend to the research cycle, where theorists and experimentalists spend time selecting candidates and validating results. AI is especially easy to apply to questions with abundant data and standard comparison metrics, a tendency the authors call the “streetlight effect”: 49% of survey participants said AI nudges them toward safer, incremental projects, while 28% said it encourages riskier ones. The authors recommend directing investment toward experimental validation and quality-control tools, and designing grant and peer-review rules to preserve space for high-risk inquiry. 🔗 Read original →

Genetics, medical records, and disease models map Alzheimer’s drug targets A study in Nature Neuroscience linked genetics, medical records, and disease models to create a target map for Alzheimer’s disease. Researchers selected candidate proteins from genetic data, then tested drugs using anonymized prescription records. One pathway tied to the EPHX2 gene was traced from gene variant through human cells to mouse models. Published 8 October 2026 in Nature Neuroscience, the study answers a practical question: how to prioritize proteins for new Alzheimer’s drugs before costly clinical trials. Authors began with 1,229 proteins already acted on by drugs or investigational molecules, building sequential checks for candidates. The first filter used Mendelian randomization: analyzing innate DNA variants linked to protein levels helps verify the direction of a protein’s link to disease risk. Across seven genome studies including 275,540 Alzheimer’s cases and 1.55 million controls, they selected 19 prioritized drug targets for European-ancestry samples and seven putative targets for African-ancestry samples. A drug target is a protein a drug can act on; sample differences are built into the result. The second filter used anonymized insurance prescription records. Among 111,680 people with mild cognitive impairment—early memory and thinking problems—authors reviewed 210 frequently prescribed drugs, accounting for age, sex, ancestry, and comorbid diagnoses. Twelve drugs in all four statistical calculations were linked to lower rates of subsequent Alzheimer’s diagnosis. These drug-target pairs are proposed for further validation. EPHX2 exemplified the full validation chain. This gene encodes an enzyme breaking down protective fatty acids; its activity is linked to inflammation and synaptic damage. In human neurons and patient-cell-derived brain mini-models, the p.Arg287Gln variant was associated with lower EPHX2 and phosphorylated tau—a disease-linked tau form. In 5xFAD mice, an accelerated amyloid-deposition model, the EPHX2-blocking drug EC5026 reduced amyloid plaques and microglia-related inflammatory signals; memory tests showed improved performance. Authors propose testing selected targets and drugs in clinical studies with biomarkers and patient genetic differences. 🔗 Read original →

Herasight launches NeuroRisk — an embryo test for rare damaging DNA variants The company offers to rank embryos from in vitro fertilization (IVF) by their total burden of rare genetic changes. This is based on a study of children and adults where such burden is linked to several diagnoses. On October 8, Herasight’s scientific director Tobias Wolfram announced the launch of NeuroRisk on social media platform X. The company describes it as an embryo test that looks for rare variants in genes, especially those preserved by natural selection. Standard polygenic scoring adds up the contributions of many common DNA variants. NeuroRisk evaluates a different signal: rare changes in genes that code for proteins. Some variants disrupt the protein instruction; others alter its structure. The less evolution tolerates damage to a specific gene, the greater the weight of such change in the embryo’s overall score. In critiques of polygenic forecasts for embryos, the discussion has centered on comparing genetically similar future children by many common variants. NeuroRisk proposes, for the same procedure, a different input signal — rare damaging changes. The basis for the calculation is a preprint by Herasight staff and a co-author from the University of California, Los Angeles. In it, they studied 3,523 children in the ABCD project on adolescent development and 302,421 adults in the US All of Us program. Carriers of at least one rare variant that disrupts protein function in a particularly damage-sensitive gene had 3.09 times higher odds of a registered diagnosis of intellectual disability — persistent limitations in learning and daily independence — 1.85 times higher odds for schizophrenia, and 1.51 times higher odds for autism. The authors also calculated that 79% of the weighted burden of such variants falls on genes outside catalogs of already known developmental disorder and autism genes. In an analysis of 4,229 sibling pairs, the link with the combined outcome “autism, schizophrenia, or bipolar disorder” persisted: the odds ratio was 1.83 within families and 1.27 when comparing the full sample. Such a comparison tests whether the association holds among children raised in the same family. The study measured these links in both children and adults. Herasight applies the population-calculated burden to rank embryos prior to transfer. 🔗 Read original →

Dog Aging Project links body size to accelerated DNA methylation aging in dogs Researchers from the Dog Aging Project analyzed DNA methylation patterns in 894 dogs to investigate how body size influences molecular aging. Using 1,640 methylation maps, they found that larger breeds exhibit faster epigenetic aging compared to smaller ones, consistent with their shorter lifespans. The study, published in *Science* on October 8, shows that methylation changes associated with aging occur more rapidly in large dogs and males, whose average lifespan is lower. Sex and body size left distinct epigenetic signatures. Methylation changes linked to sex were concentrated on the X chromosome, while body size effects were especially pronounced in transposable elements — DNA regions capable of copying and inserting themselves elsewhere in the genome. Among these, LINE1 elements showed a clear pattern: methylation at LINE1-associated sites declined with age, and this decline occurred 0.26 percentage points per year faster in giant breeds than in small dogs. Since methylation helps suppress transposable elements like LINE1, its accelerated loss in larger dogs may represent a molecular mechanism linking body size to faster aging. The findings suggest that LINE1 dysregulation could be a key contributor to size-related differences in canine longevity. Science, October 8, 2024 🔗 Read original →

MUVON reports sustained improvement in stress urinary incontinence after one year in Phase 2 trial In the second phase of clinical trials, MUVON’s cell therapy for stress urinary incontinence maintained efficacy in 12 out of 13 women after one year. The Swiss biotech company released interim data on October 7, showing that among the 13 participants followed for 12 months — out of 30 enrolled in the study — all 12 who had responded at six months continued to show significant improvement. Episodes of incontinence decreased by 79% and leakage volume by 85% compared to baseline levels. No serious adverse events related to the therapy or procedure were reported, and tissue integrity was preserved in all participants. The treatment involves harvesting a small sample of the patient’s calf muscle, expanding autologous muscle precursor cells, and injecting them into the urethral sphincter under ultrasound guidance to restore contractile function. The urethral sphincter, the muscle at the bladder outlet responsible for urine retention, weakens in stress incontinence, leading to leakage during coughing, laughing, or physical exertion. In the trial, 30 women were randomized to two dose levels; the lower dose had been tested in Phase 1, while the higher dose was evaluated in this phase for efficacy. At six months, the company reported statistically significant reductions in episode frequency and leakage volume (p<0.0001), indicating the observed changes were unlikely due to random variation. The one-year follow-up aimed to determine whether this effect would be sustained. Data presented as interim results from Phase 2 clinical trial; full study details pending publication. 🔗 Read original →

Aging lungs across species share a self-sustaining inflammatory immune cell subset In healthy aging mice, baboons, and humans, researchers identified the same pro-inflammatory subgroup of alveolar macrophages expanding more than tenfold and driving chronic lung inflammation. Scientists at Texas Biomedical Research Institute compared alveolar macrophages — immune cells that patrol lung airspaces, clearing dust, microbes, and dead cells — in young and old mice, baboons, and people. Across all three species, aging increased this specific macrophage subgroup with inflammatory markers by over tenfold. Mouse experiments revealed the expansion is sustained by two simultaneous processes: resident lung cells changing phenotype and new cells recruited from blood via the CCR2 receptor. The study was published October 7, 2026 in Nature Communications. By comparing mice, baboons, and humans, the authors tested whether this lung aging process is conserved in mammals. In 18-month-old mice (advanced age for them), baboons over 15 years, and humans over 60, the pro-inflammatory alveolar macrophage subgroup occupied ten or more times the fraction of total macrophages compared to 3-month-old mice, young baboons, and adults aged 20–45. Across all three species, these cells showed elevated CCR2 and Fas (CD95) receptors — a marker set distinguishing recently arrived blood monocytes from long-lived lung-resident macrophages. A team led by Larry Schlesinger and Susanta Pahari at Texas Biomedical Research Institute dissected the two sources of this subset in mice. First, they depleted alveolar macrophages in young and old mice and transplanted labeled cells from young donors; within two days, 30–40% of transplanted cells in old lungs converted to the pro-inflammatory state — demonstrating that aged lung tissue itself imposes this phenotype on incoming cells. Next, they injected labeled monocytes into the bloodstream and observed that these cells populated the lungs of old mice but barely entered young lungs. Monocytes lacking CCR2 remained in the blood and failed to enter lungs in either age group, confirming CCR2 as essential for lung entry. The recruited cells were short-lived: their numbers peaked on day five after injection and then gradually declined, while resident macrophages remained stable over ten days of observation. Newly arrived cells upregulated Fas receptor, and when the authors injected an antibody blocking its ligand FasL, the recruited cells survived longer and TNF levels in lung fluid decreased. Cell death proceeded via caspase-8 and caspase-3/7 — the extrinsic apoptosis pathway. This creates a self-sustaining cycle: aged lung tissue recruits new cells via CCR2, which rapidly die via Fas and are replaced by newcomers, maintaining inflammation without interruption. The same lab first identified this subgroup in old mice in 2019 and linked it to poorer tuberculosis control. Chronic low-grade inflammation during healthy aging is termed inflammaging — a concept introduced by immunologist Claudio Franceschi in 2000 and now a foundational framework in aging research. This work names two specific mechanisms — CCR2-mediated recruitment and Fas-driven apoptosis — as central to the process. The authors state that future work will directly test whether this mechanism explains why older individuals resist respiratory infections less effectively. CCR2 and Fas are already discussed as potential intervention points. 🔗 Read original →

Artificial vessel with three cell types uncovers hidden culprit in childhood progeria — fibroblasts, and DNA editing restores near-normal symptoms A Duke University team and David Liu’s lab, inventor of base editing, has for the first time built a three-layered human artificial vessel and reproduced the vascular disease of Hutchinson-Gilford progeria syndrome — childhood progeria. Fibroblasts in the outer layer, usually absent in such models, amplified inflammation, scar tissue, and loss of wall flexibility. Correcting the LMNA gene mutation with base editing returned nearly all symptoms to healthy levels. Hutchinson-Gilford progeria syndrome is caused by a rare LMNA mutation: cells produce toxic progerin protein, which accumulates in the nuclear envelope, distorts its shape, impairs DNA repair, and accelerates aging. Such children almost always die of severe atherosclerosis, heart attack, or stroke in adolescence. Laboratory models of this disease are typically limited to smooth muscle and endothelial cells of the artery’s inner layers; the outer layer — adventitia with fibroblasts — is rarely added, although fibroblasts renew connective tissue and regulate vascular tone. George Trask’s group at Duke University first assembled a vessel from all three layers: endothelium and smooth muscle cells derived from patient stem cells, and outside — donor fibroblasts of the same genetics; by gene activity it is closer to a real artery than two-layered models. Building the same vessel from progeria patient cells, scientists observed: the disease manifested far more fully than in the two-layer version. Inflammation increased, collagen accumulated, wall flexibility was lost, and smooth muscle cells began to die — the very features that kill patients. The team tested causality directly: using base editing — a single-letter DNA change without cutting both strands — they corrected the LMNA mutation and built a three-layered vessel from these cells. Progerin disappeared, and most disease symptoms returned to normal. The same editor five years earlier extended the lifespan of progeria mice from 215 to 510 days with a single injection, and a year and a half ago the same lab corrected a two-layer vessel without fibroblasts. Now correction reverses a broader set of abnormalities, visible only thanks to fibroblasts. By assembling vessels from different combinations of healthy and mutant cells — impossible in a living organism — the authors for the first time partitioned each cell type’s contribution: progerin-containing smooth muscle cells die and provoke calcification and collagen accumulation; inflammation intensifies only when progerin is present in both smooth muscle and endothelial cells; and progerin-containing fibroblasts are the main source of IL-6 and TGFβ-1 signals that trigger inflammation and fibrosis. A similar trace is seen in living mice: on a cell map of their aorta, smooth muscle cells before death resembled fibroblasts. The three-layered vessel identifies the true culprit — neighboring fibroblasts. Editing did not yield full recovery: overall gene activity remained closer to diseased than healthy. Likely cause — different timing of correction: smooth muscle and endothelial cells were edited at the stem cell stage, before progerin appeared, while fibroblasts were taken mature from patients, already diseased prior to editing. The disease’s trace remains in their DNA packaging even without progerin. Genetic cause and cellular memory of disease are distinct. The authors propose using this model for other vascular diseases — atherosclerosis and fibrosis unrelated to progeria. 🔗 Read original →

Mind-reading AI reconstructs images from brain scans with unprecedented accuracy Researchers have developed what may be the most advanced brain-decoding tool to date, capable of reconstructing images a person is viewing based solely on fMRI data—with accuracy once thought impossible. In experiments, participants viewed images while the model reconstructed them from brain activity, achieving results so similar that structure, object placement, and even fine details could be discerned. This breakthrough was made possible by new fMRI scanners with ~1 cubic millimeter voxel resolution, several times sharper than older systems. The team collected data from 8 individuals, each exposed to about 9,000 images, and trained a two-part model: one predicting image structure, the other content. These predictions were then refined by a diffusion model—the same technology behind modern image generators. The key innovation is a universal brain encoder that works with new subjects using just 1 hour of fMRI data, compared to the previous requirement of 40 hours—a process costing roughly $600–$1,000 per hour. To overcome data scarcity, researchers generated "virtual" brain responses: for new images, they predicted the expected neural signal and used these synthetic examples to train the model. As a result, 70% of training data came from images never actually seen in the scanner, yet the model still learned effectively. While not perfect—occasionally turning a cake into three sandwiches or a dog into a goat—the system significantly outperforms all prior methods in comparative tests. Scientists are now extending the technology to video, sound, internal imagery, and even dreams. Experts warn that if such models become sufficiently accurate, they could extract far more information from brain activity after calibration—potentially without consent. Some fear future use in courts or commercial systems unless strict "mental privacy" laws are established. For the full study, see Nature Neuroscience, April 2026. 🔗 Read original →

Overeating in early life programs age-related muscle loss in mice — via one gut bacterium the body loses in infancy Biologists reduced mouse litter sizes so pups received more maternal milk. Overeating in infancy caused mice to permanently lose the gut bacterium Akkermansia muciniphila, and by old age they lost muscle mass and strength faster. Three independent experiments confirmed the causal role of this bacterium in muscle aging. The study was published in Nature Communications, October 7, 2026. Researchers housed litters with either 8–9 pups or 3–4 pups: fewer mouths meant more milk per pup, leading to rapid weight gain. This models infant overnutrition without altering food composition. By the third week of life, the Notch pathway — which determines intestinal stem cell fate — suppressed goblet cell formation in these mice, reducing protective mucus secretion. The mucus layer thinned, and Akkermansia muciniphila — a bacterium that feeds on mucus and colonizes the mouse and human gut in infancy — lost its niche. Its population collapsed for the remainder of the animals’ lives. By advanced mouse age — 24–28 months — the effect appeared in muscles: overfed pups showed weaker grip strength, faster fatigue on treadmills, thinning muscle fibers, and increased scarring. Median lifespan in males dropped by 6.67% — from 30 to 28 months. The mechanism was traced to a molecule: Akkermansia muciniphila produces acetate, a short-chain fatty acid sensed in muscle specifically by receptor GPR43, not the related GPR41, which responds to similar acids. Without the bacterium, acetate levels fall, weakening GPR43 signaling, which impairs muscle cell mitochondria — reducing their number and power. When scientists silenced the GPR43 gene in old mice’s muscles, acetate supplementation no longer restored mitochondria or strength. Transplanting microbiota from old overfed donors into healthy mice caused muscle loss and metabolic dysfunction — even on normal chow. In mice without overnutrition history, simply removing Akkermansia muciniphila from a six-bacteria cocktail weakened their muscles. Combined with GPR43 knockout, this provided three independent lines of evidence: loss of this single bacterium is sufficient to drive muscle aging. The full pathway from bacterium to mitochondria was dissected only in males; females showed only muscle loss confirmation. Overnutrition via another trigger — high-fat diet in nursing mothers — replicated the early chain: barrier loss and bacterium depletion. This pathway can be broken at two ends. At one end, an eight-week course of live Akkermansia muciniphila restored grip strength, coordination, and endurance in overfed aged mice. This restoration is now being tested in humans: the registered ARMOR trial, transplanting microbiota from young athletes into people aged 65–84, measures muscle strength among other outcomes. At the other end, supplementing nursing mothers’ diets with alanyl-glutamine restored intestinal mucosal development in offspring before the defect arose. Akkermansia muciniphila colonizes the human gut in the first year of life and comprises up to 3% of microbiota in nine out of ten healthy adults, though its proportion declines naturally with age. In 2021, EU regulators approved the pasteurized form of the bacterium as a safe novel food for humans. This work shifts the starting point of age-related muscle frailty from old age to infancy: a decisive window for one cause of sarcopenia spans the first few weeks of life — years before the problem becomes clinically apparent. 🔗 Read original →

Jonathan the Tortoise Turns 194: His Mitochondrial Genes Show No Signs of Aging Scientists have for the first time sequenced the epigenome of Jonathan — the Aldabra giant tortoise from St. Helena and the world’s oldest known land animal. The epigenome — the pattern of chemical tags on DNA that determines which genes are turned on or off — was published on October 7 in Science Advances. Researchers found 287 unique genetic variants in aging-related genes not seen in other tortoises, including the reference genome from a 36-year-old named Tank or Lonesome George, the last Pinta Island tortoise who died in 2012. Most strikingly, epigenetic switches in genes governing mitochondrial function and RNA processing remained as clear and precise at age 194 as they are in a five-year-old hatchling. Jonathan was gifted as an adult to the future governor of St. Helena in 1882, meaning his true age likely exceeds the official 194 years — he lived through the eras of Charles Darwin and Queen Victoria. In April 2026, false reports of his death circulated on social media, allegedly from his veterinarian. The natural lifespan ceiling for Aldabra tortoises is about 94 years; Jonathan has lived more than double that limit. Because drawing blood poses an infection risk for this iconic animal, DNA was collected via cheek swabs instead. Blood was used only to sequence the reference genome from Tank. While genome-wide methylation — the epigenetic “switches” that regulate gene activity — became increasingly chaotic with age, as expected, 272 out of nearly 3,000 switches examined — particularly those linked to mitochondrial and RNA-processing genes — showed no age-related deterioration in Jonathan. In contrast, these same switches were degraded (“clogged”) in two other tortoises that matured in 1969. This pattern mirrors findings across species: reptiles, amphibians, and fish with negligible age-related increases in mortality live, on average, nine times longer than predicted by body size alone, due to enhanced cellular repair systems. Jonathan’s epigenome now offers a molecular portrait of this longevity mechanism. The researchers propose a virtuous cycle: stable mitochondria produce energy, which fuels DNA repair, and effective DNA repair preserves the clarity of mitochondrial and RNA-processing gene switches. Since RNA analysis requires blood — which is forbidden for Jonathan — it remains unclear whether his enduring mitochondrial health drives his longevity, or whether his long life preserves mitochondrial function. Among the study’s co-authors is Steve Horvath, developer of the epigenetic clock of aging. A similar pattern was observed in Maria Branyas, the world’s oldest verified person at death in 2024 at age 117, whose mitochondria were significantly more efficient than those of her peers. Stephen Clark, co-author and founder of the Kallel Foundation, told *The Guardian*: “We’re still trying to understand how aging happens, and Jonathan gives us a real window into what occurs over an extremely long timespan. His mitochondria must be exceptionally stable — and that’s a clue to longevity in general.” The team now aims to test whether artificially stabilizing such an epigenome can extend lifespan in other animals. As Clark stated in a comment reported by *PopSci*: “Our goal is to immediately turn these evolutionary insights into a practical, affordable treatment for ordinary people. Aging is the primary risk factor for nearly every chronic disease, and thanks to the generosity of donors, we can make longevity medicine accessible to all.” 🔗 Read original →

Short natural isoform of p53 gene suppressed cellular aging in progeria mice and extended lifespan — without tumor surge typical of full gene knockout October 7, 2026, in npj Aging, biologists at the US National Cancer Institute first tested the natural isoform Δ133p53α in a whole organism, not just cell culture. In mice with Hutchinson-Gilford progeria — a hereditary accelerated aging disease — it reduced cellular aging markers in multiple organs and modestly extended lifespan. p53, one of the most studied genes in cancer biology, activates p21 in response to DNA damage, permanently halting cell division to prevent cancer — this is cellular senescence. Disabling this pathway is risky: without p53 or p21, cultured cells live longer but mice die faster from tumors. Humans have a natural “brake” on this process — a short p53 form lacking the first 132 amino acids, Δ133p53α. It binds full-length p53 and blocks its senescence signal while largely preserving DNA repair function. The Izumi Horikawa lab previously showed this effect in human cell culture; this study is the first to demonstrate it in a whole organism. Scientists created a mouse line with controllable Δ133p53α and crossed it with a short-lived progeria model carrying an LMNA mutation. The model’s short lifespan accelerates lifespan effect assessment. Mice with activated Δ133p53α showed reduced levels of all three markers — p21, inflammatory protein IL-6, and DNA break marker γ-H2AX — immediately in skin, muscle, kidney, spleen, and lung. In progeria, this process becomes a tissue-level vicious cycle: senescent cells release inflammatory signals that accelerate aging in neighboring cells, as described in a twenty-year review of progeria research. The drop in IL-6 across five tissues breaks this cycle early. In aorta and skin — tissues most severely affected by progeria — the transgene protected smooth muscle cells from death, prevented skin and subcutaneous fat thinning, and preserved a hair follicle stemness marker that disappears without it. Median lifespan increased from 349 to 387 days — a modest 11%, as authors note, but statistically confirmed. Similar interventions in the same model, such as LMNA gene editing or cellular reprogramming, extended lifespan by 25–30%. Mice with an engineered Δ122p53 variant from another study showed increased tumor burden, heightened inflammation, and shorter lifespan — not all p53 truncations are safe. In cell culture, both variants equally suppressed IL-6 — why in vivo results differ remains unknown. The same effect was reproduced in normally aged mice: the transgene reduced p21 and increased smooth muscle cell count in the aortic wall. According to the GTEx human gene expression database, Δ133p53α levels in human skin decline after age 60, while p53 and p21 rise — a first signal that this brake weakens with age in humans. RNA sequencing of heart and kidney showed: in transgene mice, mitochondrial energy production and antioxidant defense were elevated, inflammation reduced, and in aorta, age-related histone marks — DNA-packaging proteins that regulate genes — were partially reversed. This is proof of principle: the transgene was activated before disease symptoms appeared, and the path to a drug is outlined — a screen identified two compounds that boost Δ133p53α in human cells. One of the two authors, Leo Yamada, died before the article’s publication; it is dedicated to his memory. 🔗 Read original →

AI predicts which mouse blood stem cells will be 'eternal' from video — blocking one signal increases their share A team from the Fred Hutchinson Cancer Center in Seattle published a preprint on October 7: a neural network trained on video of single blood stem cells predicts which will become long-lived with 96.7–98.5% accuracy, without destroying the cell for analysis. The same study identified a molecular signal governing this outcome and confirmed experimentally that blocking it in culture increases their proportion. Blood stem cell transplantation treats leukemias and other blood diseases: the patient’s bone marrow is cleared by irradiation or chemotherapy, and donor cells are infused. The method originated at Fred Hutch — employee Edward Donnall Thomas received half the 1990 Nobel Prize in Medicine for it. Decades later, the challenge remains: among transplanted cells, only a few can lifelong reproduce blood, while most are progenitors with limited capacity, difficult to distinguish beforehand. Sorting by protein markers yields a mixed population, and single-cell genetics captures this difference in a frozen snapshot, not the fate-selection process itself. A year ago, the same lab showed on the same culture system that true long-lived cells divide symmetrically — into two identical stem cells, not one stem and one specialized — and remain in metabolic quiescence. “Our work challenges the traditional view that blood stem cells are highly active during development,” lab leader Brandon Hadland explained then in a Fred Hutch release. There, he outlined the plan: “Looking ahead, we are also developing machine learning tools to analyze live video imaging on our platform to predict the fate of these cells from subtle behavioral features. This could change how we select the most potent ones for clinical use.” This plan was realized in the preprint posted by the lab. Researchers took blood stem cells and their progenitors from mouse fetal liver — where, not in bone marrow, these cells proliferate before birth — and plated them singly on a layer of cells mimicking this niche, filming every 2–3 hours for two to three days. The neural network was trained on 80% of these cells and tested on the remaining 20%; whether a cell would become long-lived was determined later by retransplantation into mice. The model predicted cell fate in advance with 96.7–98.5% accuracy: the top predictor was low motility — future long-lived cells moved less, and this trait was passed to daughter cells after division. The cause of low motility was found in signals from neighboring cells: future long-lived cells receive less signal via the Activin protein and more strongly activate cell adhesion genes — a program that holds the cell in place. To test this, the researchers added Inhibin-A — a natural Activin blocker — to the culture: cell motility decreased, and the proportion of long-lived cells after two weeks increased — an effect the authors describe as modest but statistically significant across four independent experiments. Translating the protocol to human cells and the clinic is the authors’ next goal. For now, video-based prediction and a controllable molecular lever provide a framework for selecting and growing the desired blood stem cells; the authors believe this could eventually be extended to other stem cell types. bioRxiv, October 2026 🔗 Read original →

Biohub expands Virtual Biology Initiative with DeepMind, Isomorphic Labs, Meta, and U.S. government to build predictive cell model On October 7, 2026, the nonprofit Biohub, the U.S. Department of Energy, the National Institutes of Health, Google DeepMind, Isomorphic Labs, and Meta announced an expansion of the Virtual Biology Initiative — a program to gather data for training AI models of living cells. Total commitments have risen from $500 million in April to $1.8 billion, with government participation joining for the first time. Biohub, founded by Mark Zuckerberg and Priscilla Chan, uses AI to study biology. Over the past decade, it has independently built cell-measurement projects such as the human cell type atlas Tabula Sapiens and the unified single-cell data repository CELLxGENE, each with separate funding and leadership. In April 2026, Biohub consolidated these efforts into one program and allocated $500 million to it. The goal of the Virtual Biology Initiative is to collect data on how different types of living cells respond to drugs and other interventions, enabling AI models to predict cellular reactions to experiments in advance — reducing the need for repeated lab testing. Biohub immediately recognized the plan’s limitation: the required data volume exceeds what a single institute can produce by orders of magnitude, prompting it to invite external partners. Within six months, the invitation succeeded. The U.S. Department of Energy will contribute over $500 million over five years through the Genesis Mission — a national AI-driven science acceleration program launched by President Trump via executive order in November 2025, under which federal commitments have already exceeded $5 billion. The department is directing its own research infrastructure — exaflop-class supercomputers (among the world’s fastest), X-ray and neutron sources, cryo-electron microscopes, and autonomous laboratories at national labs. The U.S. National Institutes of Health are not adding new funding. Instead, they are connecting through their July Bio Genesis Mission — NIH’s contribution to the broader Genesis Mission, where predicting the behavior of living systems is one of six tasks — and translating existing accumulated databases and repositories, previously funded with over $500 million in federal dollars, into a unified AI-training format. These include the infrastructure of the National Center for Biotechnology Information, which underpins nearly all of global biomedicine. Google DeepMind, pharmaceutical startup Isomorphic Labs (a separate Alphabet company spun out of Google DeepMind, developing drugs via the neural network AlphaFold, which predicts protein structure from amino acid sequences), and Meta are jointly contributing an additional $300 million, despite DeepMind and Meta otherwise competing for top AI models. “We cannot solve this challenge without open experimental biological data of unprecedented scale,” said Pushmeet Kohli, Vice President of Google DeepMind for AI for Science. “Creating a virtual cell is one of the most important tasks of the coming era of science,” said Alex Rives, head of science at Biohub. Max Jaderberg, President of Isomorphic Labs, called participation a step toward the next breakthrough in biology, while Nicole Kleinstreuer, Deputy Director of the relevant NIH division, described it as a way to “greatly accelerate the timelines of medical discoveries compared to achieving the same result solely through laboratory experiments.” The collected data will become an open resource for the entire scientific community. 🔗 Read original →