A gene editor rewrote every copy of a cholesterol gene in human embryos. Delivered a second way, it stopped them developing.

The rest of the day

27 more stories on this beat.

Each with its own sources. None of these is a link to the story above.

1. 02

   First patient dosed in an AI-designed drug's final trial

   Insilico Medicine said on Tuesday that it had given the first dose of rentosertib in GENESIS-IPF-3, a final-stage trial at Peking Union Medical College Hospital in Beijing. [2] Idiopathic pulmonary fibrosis is a disease in which lung tissue scars and stiffens for no known reason, and breathing gets steadily harder. [2] The drug blocks a protein called TNIK, which the company says its software picked out and which had never been linked to lung scarring. [2] Patients are treated for 52 weeks, and the lead investigator said approval would take three to four years. [2]

   Why it matters — Insilico's earlier 12-week study was small and short, so this is the first time a molecule designed this way has been put in front of the test that decides whether regulators approve it. [2]

2. 03

   Novartis's failure knocked 22% off a rival

   Dyne Therapeutics, a Massachusetts company with no news of its own that day, lost 22% of its value. [11] The trigger was Novartis saying its drug del-desiran had missed the main goal of a final-stage trial in myotonic dystrophy type 1, a muscle-wasting disease. [11] Dyne is developing a different drug for the same disease and uses the same measure: how fast a hand relaxes after squeezing. [11] Eric Schmidt of Cantor Fitzgerald blamed wide variation in that test and cut his odds on Dyne's drug from 50% to 25%. [11]

   Why it matters — Novartis paid $12 billion for Avidity Biosciences in February largely for this drug, and Jefferies analysts said the result would revive questions about how carefully the company checked what it was buying. [11][13]

3. 04

   An oral lupus pill beat a dummy treatment

   Beeline Medicines reported on Wednesday that afimetoran, a daily pill, produced significantly better responses than placebo in a mid-stage lupus trial after 48 weeks. [3] Lupus is an autoimmune disease in which the body attacks its own tissues, and the field waited decades for a new drug before GSK's Benlysta arrived in 2011. [3] The pill blocks two immune sensors, TLR7 and TLR8, that genetics research has tied directly to the disease. [3] Beeline released no figures beyond a p-value under 0.001. [3]

   Why it matters — Beeline was set up in April with $300 million and five drugs handed over by Bristol Myers Squibb, so this is the first read on whether that portfolio was worth buying. [3]

4. 05

   $225m to soften an immunotherapy's timing

   Solstice Therapeutics of Boston raised a $225 million first funding round on Tuesday to take porustobart, an antibody it licensed from China's Harbour BioMed, into mid-stage testing in bowel cancer. [4] The drug releases a second brake on the immune system alongside Merck's Keytruda, in a tumour type where immunotherapy alone barely works. [4] It is built to clear the body in four to five days rather than the two to three weeks of existing drugs of its class, so immune side effects last a shorter time. [4]

   Why it matters — Chief executive Caroline Loew said an anti-PD-L1 drug alone has a 0% response rate in this setting, and the combination reached 30% in an earlier Chinese study. [4]

5. 06

   Tyra's bladder cancer pill misses the analysts' bar

   Tyra Biosciences released its first mid-stage results on dabogratinib, a daily pill for early bladder cancer, and its shares fell 38% to $19.30 on Wednesday. [6] Among patients left with a single small tumour after surgery, 63% had it disappear at the highest dose. [6] Analysts had set 70% as the bar, because Johnson and Johnson's Balversa cleared 72% at three months in a comparable group. [6] Severe side effects hit 9% and 14% of the two dose groups. [6]

   Why it matters — Tyra designed the drug to be gentler than Johnson and Johnson's Balversa, which doctors criticised for its side effects, and is going ahead with a registrational trial next year. [6][7]

6. 07

   An itch drug fails a second time and switches disease

   Evommune said on 9 September that EVO756 had missed every goal in a mid-stage trial of 121 adults with moderate to severe eczema, and that it will not pursue the drug for that condition. [8] The same drug had already failed in chronic hives in June, at every dose tested. [8] EVO756 blocks MRGPRX2, a switch on mast cells that makes them release the chemicals behind itching and swelling. [8] The company will still run a mid-stage trial of it in migraine prevention, and shares fell 18% to $10.70. [8]

   Why it matters — Evommune's remaining hope in skin disease is a different drug, EVO301, which won a mid-stage eczema trial in February and does not enter its next study until mid-2027. [8]

7. 08

   $275m for a one-shot therapy for childhood epilepsy

   Encoded Therapeutics raised $275 million to finish testing ETX101 and build manufacturing for it. [5] Dravet syndrome is a severe epilepsy that starts in a baby's first year, caused by a fault in one copy of the SCN1A gene. [5] ETX101 is a single infusion designed to raise the working copy's output in the brain cells that calm activity down. [5] An earlier trial showed a 78% cut in seizures at the end of last year, and the company had cut 29% of its staff in February 2025 to pay for it. [5]

   Why it matters — Biogen paid $165 million up front last year for US rights to a rival Dravet drug from Stoke Therapeutics, which reports final-stage results in the middle of next year. [5]

8. 09

   A gene therapy tripled lifespan in cats

   Cats bred with Sandhoff disease lived 12.4 months on average after a high dose of gene therapy injected into a vein, against 4.3 months untreated. [15] Sandhoff, and the closely related Tay-Sachs, are caused by a missing enzyme, so fatty waste piles up inside nerve cells and children usually die at around four. [15] Earlier attempts had to inject the therapy into the brain or the fluid around it; this one went into the bloodstream and still reached the nervous system. [15]

   Why it matters — There is no approved treatment for either disease, and an injection into a vein is a far simpler procedure to offer families than brain surgery. [15]

9. 10

   Semaglutide users had fewer asthma attacks

   People with asthma who started semaglutide, the drug sold as Ozempic and Wegovy, had nearly 40% fewer asthma attacks than similar patients on an older diabetes pill, in UK medical records. [14] The same analysis found about 20% fewer flare-ups among people with COPD, the lung disease usually caused by smoking. [14] Chloe Bloom of Imperial College London led four comparisons, each of 20,000 to 22,000 people. [14] These are records of what happened, not a trial. [14]

   Why it matters — Asthma and COPD have never been measured as outcomes in any of the trials that got these drugs approved, so this is the first large look at a question nobody was funded to ask. [14]

10. 11

    Most AI-designed proteins failed as cancer receptors

    Researchers screened 1,758 protein binders designed by generative AI, and tested them as the grabbing end of engineered immune cells. [21] A CAR-T therapy takes a patient's own T cells, adds a receptor that recognises a protein on the cancer, and returns them. [21] Most designs failed in one of three ways: the receptor fired with nothing to grab, it reached for a part of the target that is hidden in a real cell, or it attacked cells it should have left alone. [21] The team then found sequence changes that fixed each fault. [21][22]

    Why it matters — Design software is judged on whether two molecules stick together in a tube, and these designed proteins had barely been tested inside a candidate therapy before. [22]

11. 12

    A $50m spinout aims a leukaemia drug at diabetes

    Kura Oncology launched a company called Caspian Therapeutics with $50 million, led by BVF Partners and joined by Eli Lilly and a type 1 diabetes charity fund. [9] The money takes KO-7246, a menin inhibitor, through the last laboratory work and into a first human trial. [9] Menin inhibitors were built to block a protein interaction that drives aggressive leukaemia, and two of them were approved in late 2025. [9] Kura says blocking menin also pushes the pancreas's beta cells to make more insulin, in mouse and human tissue. [9]

    Why it matters — Kura keeps half of Caspian, so the outside investors carry the cost of finding out whether the diabetes idea holds. [9]

12. 13

    A $410m fund locked up for fifteen years

    Luma Group closed LumaBio Fund I at $410 million, structured to hold its investments for fifteen years rather than the usual ten. [10] The New York firm has already backed more than ten companies from it, including Altos Labs, which works on resetting the age of cells. [10] One company Luma co-founded, Vaccine Company, was bought by Eli Lilly this year for up to $1.55 billion. [10] Founder Joshua Fink said the fund exists to carry companies from discovery through to selling a medicine. [10]

    Why it matters — A drug takes longer to answer its own question than a normal fund is allowed to wait, which is why investors keep selling companies before the answer arrives. [10]

13. 14

    Regulators halt shipments from an Indian ingredient maker

    The US drug regulator has stopped all shipments into the country from Shoolin Pharma in Gujarat, India. [18] An inspector found residue and stains on production walls, floors and equipment, corrosion on a production line, and an employee in open-toed sandals. [18] The company makes active ingredients for more than a dozen medicines, including anti-convulsant pills, sold to US compounding pharmacies. [18] The inspection was in April; the warning letter is dated 18 August and was posted last week. [18]

    Why it matters — Active ingredients are the part of a pill that does the work, and a single supplier's plant can sit behind medicines sold under many different names. [18]

14. 15

    $4m in prizes to find new uses for cheap drugs

    The US National Institutes of Health opened a competition offering up to 40 prizes of $100,000 each for evidence that an existing generic drug treats a costly condition it is not approved for. [19] There are more than 32,000 approved generic drugs in the US, built on over 1,300 active ingredients. [19] Once a drug loses its patent, no company can recover the cost of running trials for a new use. [19] Entries must show a clinical signal from trial or health-record data. [19]

    Why it matters — The prizes are a public attempt to buy evidence that the market will not pay for, and the results could feed into official labelling and insurance cover. [19]

15. 16

    One gene lifted antibody output by up to half

    Researchers at BeOne Medicines raised protein output in Chinese hamster ovary cells by 30% to 50% by adding extra copies of a single gene, Pabpc1. [24] These cells are the standard factory for making antibody drugs, and a company's cost per dose depends heavily on how much each batch yields. [24] Pabpc1 makes the messenger RNA inside the cell last longer, so each instruction is read more times before it is broken down. [24] Growth rates and product quality were unchanged, and toxic waste products fell. [24]

    Why it matters — The gain was largest for the most complicated antibodies, at 50% for bispecifics, which are exactly the ones that are hardest and most expensive to manufacture. [24]

16. 17

    A US project to make RNA drugs adjustable

    A team led by Harvard Medical School and MIT won a one-year pilot award of up to $4.4 million from ARPA-H, a US health research agency, to build a control layer into RNA medicines. [23] An mRNA molecule has untranslated ends that fold into shapes, and those shapes set how much protein the message makes. [23] When a small molecule binds one of those folds, the shape rearranges and the output changes, so a pill could turn an RNA drug up or down. [23]

    Why it matters — Gene therapies today are set once and left, so a dial would let a doctor stop or lower a treatment that has gone too far. [23]

17. 18

    Tau feeds itself through the cell's power plants

    An NIH-funded team led by Bingwei Lu at Stanford University found that tau, a brain protein that forms toxic tangles in Alzheimer's and frontotemporal dementia, enters mitochondria and makes them run their energy chain backwards. [25] Mitochondria generate a cell's energy by passing electrons along a chain; under stress that flow can reverse, which damages the cell. [25] The reversed flow then drives more chemical tags onto tau, which is what makes it clump, so the damage feeds itself. [25] Stress alone did not start it in flies and mice that had no tau. [25]

    Why it matters — The work was done in flies, mice, human brain tissue and cultured nerve cells, and it points at breaking the loop rather than clearing the tangles. [25]

18. 19

    A blood test in your fifties tracked later decline

    Researchers tested blood from 1,350 people in their fifties and sixties, none diagnosed with Alzheimer's, and measured the disease-linked forms of tau and amyloid-beta. [26] Those with higher levels did worse on thinking and memory tests over the following five years, with slower thinking speed and more memory loss. [26] The study was funded by the US National Institute on Aging and published in The Lancet. [26] Kristine Yaffe of the University of California, San Francisco said the disease starts years before symptoms. [26]

    Why it matters — Nearly every Alzheimer's drug has been tried in people who already have symptoms, and a midlife signal is what any prevention trial needs in order to pick who to enrol. [26]

19. 20

    Women's own conditions still lack trial data

    A piece in Nature Medicine argues that questions specific to women, such as endometriosis and pregnancy disorders, remain underfunded and understudied, and that waiting for randomised trials alone will not close the gap. [20] Endometriosis is a condition where tissue like the womb lining grows outside it, causing severe pain. [20] The authors point to work using 300 million laboratory tests from women's medical records, which found a sharp change in body chemistry at the start of menopause. [20] They argue records, registries and patient-led projects should be used alongside trials. [20]

    Why it matters — Real-world records can be searched in a fraction of the time a trial takes, but they show what happened to people rather than what a treatment caused. [20]

20. 21

    A virtual cell built from 38 million measurements

    A Chinese team built a computer model called ProteinTalks by measuring how proteins in breast cancer cells changed over time after being hit with 63 approved drugs, producing more than 38 million measurements. [31] Most attempts at a virtual cell rely on gene readings; this one tracked the proteins themselves, which are what actually do the work in a cell. [31] The model predicted which drug combinations would work, which proteins drive resistance, and which patients would respond. [31] It also transferred to tumour samples grown from patients and to clinical biopsies. [31]

    Why it matters — The competing-interest note lists shareholders and employees of two companies among the authors, which is normal in this field and worth knowing when reading the benchmark results. [31]

21. 22

    Steering a molecule generator towards what you want

    Researchers published a method that lets a molecule-generating model be retrained on the fly towards whatever properties a chemist actually wants. [30] These models produce new molecule shapes, sometimes fitted to a pocket on a target protein, but pushing them towards a set of desired properties has been hard. [30] The usual approach fixes the conditions before training starts, which does not match how drug projects work, since the target properties change case by case. [30] The authors report that models tuned this way beat the untuned versions across their test cases. [30]

    Why it matters — It is the same problem the AI binder screen ran into: a generator can make plausible molecules long before anyone can steer it towards useful ones. [30][21]

22. 23

    A designed binder now needs ten tries, not thousands

    Martin Pacesa, who won the 2026 Eppendorf Award for Young European Investigators, described a tool called BindCraft that designs proteins to stick to a chosen target. [32] AlphaFold predicts the shape a protein sequence will fold into. BindCraft runs that backwards: it starts from the shape needed to grip the target and searches for a sequence that will fold into it. [32] Older design methods meant drafting thousands of candidates and testing them all; he said five to ten now need screening. [32]

    Why it matters — Binding is how almost every drug works, and many proteins researchers care about still have nothing that sticks to them. [32]

23. 24

    A factory cell line for a Duchenne gene therapy

    NewBiologix of Switzerland agreed to build a stable producer cell line for a Duchenne muscular dystrophy gene therapy from Synastra Biotechnology in Turkey. [27] Duchenne affects about one in 5,000 boys, caused by faults in the DMD gene that stop the body making dystrophin, so muscle including the heart wastes away. [27] Treating the whole body needs some of the largest doses of viral carrier in the field, which is why manufacturing decides whether a therapy is affordable. [27]

    Why it matters — Synastra came out of a university and a venture fund in Turkey, and the deal shows rare-disease gene therapy programmes starting outside the usual countries. [27]

24. 25

    The new flu strain dodges antibodies, but shots still help

    Researchers testing subclade K, a mutated H3N2 flu lineage that emerged in 2025 and has been active in many regions, found it evades antibodies significantly better than its predecessors. [28] They traced both the immune evasion and its improved entry into cells to a single mutation, A186D, in the protein the virus uses to latch on to cells. [28] Seasonal flu causes roughly a billion infections and more than 250,000 deaths a year. [28] Vaccination still boosted the ability of blood samples to neutralise subclade K. [28]

    Why it matters — The work was done with harmless stand-in particles in the laboratory rather than in people, so it says what antibodies can do and not how many infections are prevented. [28]

25. 26

    A bone-building switch found in mice

    Scientists at Leipzig University increased bone strength in mice, both healthy ones and mice with bone loss, using a compound called AP503. [16] It activates GPR133, a receptor on the surface of bone cells that responds to physical force and to signals from neighbouring cells. [16] Mice with a damaged version of the receptor lose bone density early, which is what pointed the team at it. [16] AP503 was found by a computer-assisted search rather than by screening chemicals in a laboratory. [16]

    Why it matters — About six million people in Germany alone have osteoporosis, most of them women, and current drugs are limited by side effects over long periods. [16]

26. 27

    Chinese licensing deals keep coming

    More than 100 licensing deals between Chinese biotechs and drugmakers in the US and Europe have been signed since the start of 2025, with the peak in the first three months of this year. [29] GSK and Roche added two more this week, with Hutchmed and Simcere Pharmaceutical. [29] AstraZeneca, AbbVie and Roche each paid at least $550 million up front earlier this year for cancer and obesity drugs. [29] Eli Lilly has outpaced every other large drugmaker in dealmaking this year. [29]

    Why it matters — Solstice's $225 million round on the same day exists to develop an antibody licensed from a Chinese company, which is what these hundred deals look like a year later. [29][4]

27. 28

    A heart catheter approved for irregular rhythms

    Abbott won US approval for TactiFlex Duo, a catheter used to burn or freeze the small patches of heart tissue that cause atrial fibrillation, an irregular and often fast heartbeat. [17] Approval was based on the company's FlexPulse Identity study in patients whose episodes come and go. [17] Abbott cites an estimate that 10.5 million people in the US have a heart that beats too fast, too slow or irregularly, which can lead to a stroke. [17] Rollout begins this week. [17]

    Why it matters — It is Abbott's fifth approval in this part of cardiology in just over a year, in a field where several manufacturers are replacing heat with electrical pulses. [17]