Biotech, in brief

Bayer, the German maker of medicines and farm chemicals, said on Friday 2 October it will spend $2.2 billion on a medicine plant in New Albany, Ohio. [1]

The plant is due to start making medicine ingredients in 2031, and a second site for finished medicines should follow in 2034. [1]

Bayer expects about 600 permanent jobs there, plus about 1,500 temporary jobs while it is being built. [1]

The US now brings in 35% of the money Bayer makes from medicines, up from 20% in 2018. [1]

Bayer wants to double its US medicine sales by 2030, relying mainly on Kerendia for kidney disease, Nubeqa for prostate cancer and asundexian, an experimental drug against strokes. [1]

US medicine prices are far higher than in other rich countries, and President Trump has been pressing drugmakers to cut them. [1]

Three days before Bayer's plan, on Tuesday 29 September, a new US tax of 100% on some imported medicines came into force. [2]

The tax left out generic drugs, the cheap copies of older medicines that fill 90% of US prescriptions. [2]

About 40% of those generics are made in India. One Dr. Reddy's factory in Hyderabad makes nearly a billion doses a month, and 75% go to US patients. [2]

Trump said in July he plans a 100% tax on imported generics from 2028, and 200% a year later, after his term has ended. [2]

Roche, the Swiss drugmaker, pledged $50 billion for the US last year, around the time other big drugmakers made similar promises. [3]

Ashley Magargee, who runs Roche's US arm Genentech, says part of that $50 billion was spending that was already planned. [3]

Why it matters. The next sign is whether the tax on imported generics that Trump described for 2028 is actually ordered; for now the cheap pills enter the US without it. [2]

Roche, the Swiss drugmaker, told investors on Monday 28 September that it has started building research labs run largely by AI. [4] It wants its AI tool, Target Nexus, to help shape 80% of decisions on which research to pursue by the end of 2026. [4] From late 2025 to mid-2026, 40% of its drug-pipeline decisions already had a recorded AI or computing contribution. [4] Its US arm Genentech has tested AI agents from Anthropic running a protein test in the lab. [3]

Why it matters. Roche also said more than 80% of its final-stage trials have succeeded so far in 2026, against 65% in 2025. [4] Genentech's chief says the tools add to scientists rather than replace them, while several senior Genentech scientists have left in a research shake-up. [3]

The FDA, the US medicines regulator, has published a transparency roadmap listing what it plans to make public over the next year. [6] It plans a rule to announce every major application to sell a drug when it is filed. [6] It also plans to publish companies' trial reports for approved drugs, with trade secrets removed. [6] In 2025 it began publishing its rejection letters, starting with more than 200 from 2020 to 2024. [6]

Why it matters. An FDA review in 2015 found companies left out 85% of the agency's safety and efficacy concerns when they announced a rejection. [6] In August the FDA also agreed to share non-public information with the SEC, the US stock-market regulator, so companies cannot describe FDA decisions selectively. [6]

A trial called EVITA gave 109 people in Denmark and the US either 1,000mg of vitamin C a day or a dummy pill for a year. [7] All had a blood disorder that can turn into cancer, or an early, low-risk blood cancer. [7] Vitamin C did not slow the growth of the abnormal blood cells, which was the trial's main question. [7] But after nearly three years, 11 people on vitamin C had died, against 24 on the dummy pill. [7]

Why it matters. The death count was an extra finding the trial was not built to test, so its leaders say a larger final-stage trial must confirm it. [7] They say it is too soon to recommend anything, and the vitamin C group had more stomach and gut problems. [7]

PARP inhibitors are cancer pills used in ovarian, breast, prostate and pancreatic cancer, and they come with a raised risk of a later blood cancer. [12] Researchers studied blood samples from 10,156 patients tested at Memorial Sloan Kettering, a New York cancer hospital. [12] They found the risky blood-cell changes after PARP pills were largely explained by earlier carboplatin, a chemotherapy drug. [12] The changes grew less in patients born with faults in a DNA-repair system. [12]

Why it matters. These later blood cancers are deadly: fewer than one patient in ten survives five years. [12] PARP pills are now given earlier in treatment, so knowing who is most at risk matters to more patients; the study was published in Nature Genetics. [12]

Researchers at Oxford Population Health checked clinic readings from Our Future Health, the UK's largest health research programme. [9] They found 37% of adults who joined between 2022 and 2025 had high blood pressure. [9] Of those, 59% did not know it and were taking no medicine for it. [9] Half of the people already on blood-pressure pills still had readings of 140/90 or higher. [9]

Why it matters. High blood pressure is linked to around half of UK heart attacks and strokes, the British Heart Foundation says. [9] The authors argue that England's treatment rules, which ask for several readings before a diagnosis, leave many people untreated. [9]

Fierce Biotech, a drug-industry news site, counted 21 layoff announcements from drug and biotech companies between July and September. [10] That compares with 17 in the spring, 33 in the first three months of 2026 and 62 in the same months of 2025. [10] The biggest was Novo Nordisk, the Danish maker of Wegovy, which said 4,000 more staff had gone. [10] A recruiter at the staffing firm Kelly says she sees signs of recovery. [10]

Why it matters. Massachusetts, a biotech centre, lost about 3,600 drug-research jobs in 2025, its first yearly fall in more than two decades, a trade group says. [10] A recruiter at Kelly says hiring centres are growing in Texas, North Carolina and the Midwest. [10]

BioPharma Dive, a drug-industry news site, has reported how a chemistry AI agent called ChemAgent went wrong in a 2025 test. [5] Asked whether a molecule harms the liver, it twice fed a tool a badly written chemical formula and got errors. [5] Instead of stopping, it switched to a general AI tool and gave a confident answer. [5] The answer, that the molecule carried a liver risk, was wrong. [5]

Why it matters. Drug companies are giving AI agents more steps in a row, so one wrong answer can steer which experiment comes next. [5] Srijit Seal, a scientist at a chemical-safety AI company, says wrongly calling a compound safe would be far worse, because tests get skipped. [5]

Ed Patrick, an anaesthetist in England, used Google's Gemini AI to search medical papers when his mother Ann had non-stop seizures in May. [8] He suspected a lack of vitamin B6, possibly caused by carbidopa-levodopa, the common Parkinson's drug she had taken for years. [8] Doctors in Nottingham gave her B6 through a drip, and the seizures stopped within minutes. [8] Her neurologist says only 14 similar cases have ever been reported. [8]

Why it matters. Her hospital team has changed its own practice, and Parkinson's UK, a charity, supports trials to test the link. [8] Doctors quoted by the BBC warn patients not to change medicines or supplements without medical advice. [8]

Researchers at the University of Wisconsin-Madison studied imidazole propionate, a molecule some gut bacteria make from histidine, a building block of protein in food. [11] In mice, the molecule reaching the brain increased the build-up of amyloid and tau, the two proteins that clump in Alzheimer's disease. [11] In blood from almost 1,200 volunteers, people with the highest levels lost memory and thinking skills much faster. [11] A gene variant carried by about 43% of them goes with higher levels. [11]

Why it matters. The human data show a link, not proof that the molecule causes the decline. [11] The team hopes for a drug that lowers it in the blood, the way statins lower cholesterol, but none exists yet. [11]

Alpha-thalassaemia is an inherited disease in which the body makes too little of one part of haemoglobin, the protein that carries oxygen in blood. [14] Its most severe form usually kills a baby before birth. [14] Researchers writing in Nature Genetics found the DNA switches that silence zeta-globin, a stand-in part used only in the early embryo. [14] Gene edits that turned it back on kept mice with the severe form alive until late in pregnancy, and worked in patients' blood cells in the lab. [14]

Why it matters. Today the only lasting cure is a stem-cell transplant, which needs a matched donor most patients do not have. [14] The faulty genes are common, carried by up to 70% of people in some regions because they protect against malaria; the edits have not been tried in people. [14]

Researchers at the University of Colorado Anschutz studied high-risk myelodysplastic syndromes, or MDS, blood cancers in which the bone marrow stops making healthy blood. [13] They found the stem cells driving the disease lean heavily on NAD, a molecule cells need to make energy. [13] Blocking NAMPT, an enzyme that recycles NAD, weakened those stem cells in patient cells and animals, while healthy stem cells coped. [13] Between 10,000 and 20,000 people in the US are diagnosed with MDS each year. [13]

Why it matters. High-risk MDS can turn into acute myeloid leukaemia, an aggressive blood cancer that is hard to treat. [13] The team plans studies in people of drugs that block NAMPT; so far the work is in the lab. [13]

About 100 million people in the US have fatty liver disease, and 20% to 25% of them go on to a worse, scarring form called MASH, the American Liver Foundation estimates. [15] Researchers co-led by Cedars-Sinai, a Los Angeles hospital, found an enzyme called UBE2N falls in liver cells as the disease worsens. [15] It helps clear out damaged mitochondria, the parts of a cell that make energy. [15] Restoring it in mice cut liver fat, inflammation and scarring. [15]

Why it matters. MASH has no cure, and care rests mainly on changes to diet and exercise plus a few medicines. [15] The work is in mice, published in Nature Metabolism, and no treatment built on it exists yet. [15]

Self-amplifying RNA vaccines carry instructions that copy themselves inside cells, so a smaller dose can make more of the target protein. [17] But the cell's own antiviral defences often shut them down. [17] Researchers reported in Nature Communications that adding B2, a protein from the Nodamura virus, hides the RNA from those defences. [17] It raised how much protein the RNA made without blocking the immune signals a vaccine needs. [17]

Why it matters. That clash has held back these vaccines, which can protect at lower doses than ordinary mRNA shots. [17] The tests were in cells; no vaccine built this way has been tried in people. [17]

Leucine is an amino acid, a building block of protein, found in milk, meat, beans and lentils. [16] Researchers at the University of Cologne in Germany found it stops cells breaking down certain proteins on the outside of mitochondria, the cell's energy makers. [16] Keeping those proteins lets mitochondria make more energy. [16] Leucine does it by slowing SEL1L, a protein that marks faulty proteins for disposal, the team reported in Nature Cell Biology. [16]

Why it matters. The team warns that SEL1L also clears damaged proteins, so changing it could backfire. [16] Some lung cancer cells with changes in how they handle leucine survived better, which ties the finding to cancer research too. [16]

Researchers at ETH Zurich, a Swiss university, working with Brigham and Women's Hospital and MIT in the US, bonded a tough water-based gel to titanium. [18] The gel behaves like cartilage, the smooth tissue lining joints, and the titanium anchors into bone. [18] A hip joint made this way kept moving smoothly through 500,000 loading cycles at 3,000 newtons. [18] An artificial spinal disc made the same way absorbed more than 80% of the energy of an impact. [18]

Why it matters. Gels like this have rarely reached patients, because they would not stay fixed to the hard parts of an implant. [18] So far it has been tested in the lab, not in people, and three of the authors have filed a patent. [18]

Some people who need many blood transfusions make antibodies against donor blood, so a safe match gets hard to find. [19] Researchers in Saudi Arabia and Bahrain modelled a bank of red blood cells grown from stem cells for such patients. [19] Using records from 3,719 patients, they found three donor cell lines could match 134 of the 153 patients they could analyse. [19] Most of the rest carried antibodies to one blood marker, called c. [19]

Why it matters. Red blood cells grown from stem cells are still experimental, so this is a plan for how few lines a bank might need. [19] The authors say gene editing could close the gap left by the c marker. [19]