Biotech, in brief

Approving on smaller proof

The US regulator cleared Bayer's lung pill on 69 patients with nobody to compare them against. [1] It cleared AstraZeneca's breast cancer pill on how long patients went before a scan showed growth, counted from a blood test result. [3]

Testing before symptoms appear

AstraZeneca's pill is prescribed the moment a blood test finds an ESR1 mutation, before any scan changes. [4] A separate team reported that a protein called fibronectin breaks the brain's protective lining early in Alzheimer's, before memory goes. [6]

Computers designing the molecules

One team screened 1,758 computer-designed proteins as receptors for cancer-killing cells and found three reasons most of them fail. [13] Another released an open platform that predicts where a brain tumour will grow back, covering 79.4% of recurrences. [14]

Buying the laboratory suppliers

Revvity agreed to buy Human Cell Design, a French firm that grows human pancreas cells for laboratories. [9] The ten largest firms that manufacture medicines for other companies took $35.116bn in 2025. [12]

One pressure seen across the day runs through Biotech too.

• Proof that measures something elseBuilding ↑ · seen in 2 beats→

1. 1The US regulator cleared a lung cancer pill for untreated patients, on 69 people and no comparison group
2. 2A boxed warning on last week's breast pill
3. 3Novartis shareholders turn on its shopping
4. 4Nine in ten costly biologics have no copy coming
5. 5Chemotherapy's mark in children's healthy organs

The US drug regulator cleared the lung cancer pill sevabertinib on 9 September for people who have had no treatment for the disease at all. [1]

Watch AstraZeneca's Serena-4 trial, due to report before the end of the year, which tests the same breast cancer pill as a first treatment rather than as a switch. [4]

The drug is the same one already cleared for patients who had tried something else, and what changed this week is how early in the illness it may be given.

The same step. The pill is given the moment a blood test finds the mutation, months before a scan would show the cancer growing again

The receptor goes into the blood stem cell before it becomes a T cell, so the receptors that would attack the patient never form and never have to be deleted

The same chain read backwards. The mutations are written during treatment in childhood, and by the time a late illness appears there is nothing left to act on

The leak is one of the first steps in Alzheimer's, so blocking fibronectin acts ahead of the plaques and tangles today's drugs are aimed at

And when the clock starts at the moment a test goes off, some of the extra months on paper come from starting the count sooner.

A regulator read 69 scans, a laboratory watched cells in a dish, a mother in North Yorkshire watched a four-year-old finish six months in hospital.

The moment when acting helps most is usually the moment nobody can tell anything has begun.

Also seen in Mind & body.

The prescribing label for Etcamah, AstraZeneca's new breast cancer pill, was published on 11 September with a warning in a black box at the top. [3]

Taken with ribociclib, one of the three drugs it is meant to be combined with, it can set off a dangerous heart rhythm and cause sudden death.

Doctors must run a heart tracing before starting and keep checking it. The pill was cleared on 4 September as the 38th brand-new medicine of 2026. [2]

The regulator and its own outside advisers had both reviewed the trial negatively before AstraZeneca supplied more data. [4]

The label now attaches a heart risk to a pill meant for patients whose cancer is still under control.

Eight shareholders in Novartis, a Swiss drugmaker and one of the world's largest, said they will press the company over the firms it has been buying. [5]

The trigger was del-desiran, a drug for the muscle-wasting disease myotonic dystrophy, which missed the main goal of its final trial. [5]

The shareholders said it raised questions about the $12bn Novartis paid last year for Avidity Biosciences, the company that had been developing it.

The shares fell 11% in a day, wiping nearly $30bn and every gain made since January. [5]

It is the company's second big trial setback in days, and buying a late-stage drug is how large drugmakers normally replace the ones going off patent. [5]

Ninety per cent of the expensive biological medicines that lose patent protection by 2034 have no lower-cost version in development, according to a report cited by the Wall Street Journal. [5]

Biological medicines are grown in living cells rather than mixed from chemicals, which makes copying them slow and costly.

Those copies, called biosimilars, are what is meant to bring the price down once a patent runs out. [5]

Health systems budget for prices falling when patents expire. On these drugs there will be nothing to fall to. [5]

Researchers read single DNA molecules from healthy tissue in children who had been treated for cancer. [18]

They found that platinum chemotherapy, one of the oldest and most used families of cancer drug, pushed mutation levels in normal tissues up to those usually seen in adults.

In the liver it left a pattern of damage found in no other organ. Some of the new mutations were of a type linked to leukaemia. [18]

Childhood cancer survivors often develop other illnesses decades later, and this is a first look at the record the treatment writes into organs it was never aimed at. [18]

A trial across South Africa, Uganda, Zambia, Zimbabwe and Mozambique put 1,101 children aged two months to six years on an injected antibiotic for severe pneumonia, then switched some of them to amoxicillin tablets. [20]

World Health Organization guidance says five days of injections, which usually means five days in hospital.

Readmission or death within 28 days was 5.6% on plain amoxicillin, 6.9% on amoxicillin-clavulanate and 6.3% on injections alone. [20]

Pneumonia is one of the leading infectious killers of young children, and a shorter stay means less exposure to the resistant bacteria that live in hospitals. [20]

The blood-brain barrier is the tight lining that keeps most of what is in the blood out of the brain, and it starts leaking early in Alzheimer's disease. [6]

The leak is worst in people carrying the APOE4 version of a common gene.

A team reported on 11 September that a protein called fibronectin, made by support cells in the brain, piles up around the vessels and causes it.

Removing the fibronectin restored the barrier in laboratory models and in animals. [6]

Today's Alzheimer's drugs attack the plaques and tangles that appear later. This points at a step that happens before them. [6]

T cell treatments are usually built from a patient's own immune cells, which takes weeks and costs six figures. [8]

Cells from a donor can be made in advance and frozen, but they often attack the patient's own tissue.

A UCLA team instead put a cancer-recognising receptor into blood stem cells from donated umbilical cord blood and grew them into T cells in the laboratory, so the dangerous receptors never form.

One dose controlled ovarian cancer and melanoma in mice. [8]

It is an attempt to turn a made-to-order treatment into something frozen on a shelf. It has not been tried in a person. [8]

PCSK9 is a protein that strips away the receptors liver cells use to pull cholesterol out of the blood, so blocking it lowers cholesterol. [7]

Researchers at the University of Barcelona and the University of Oregon designed short folded pieces of DNA, called polypurine hairpins, that turn the PCSK9 gene down at source.

Their best one cut the protein by 87% in human liver cells and cholesterol by 47% in mice after a single injection. [7]

The PCSK9 drugs already on sale are antibodies or RNA and are expensive to manufacture.

This is a cheaper kind of molecule doing the same job, so far only in cells and animals. [7]

AI can now design a protein to stick to almost any target, and the open question is whether those designs survive inside a medicine. [13]

A team built 1,758 newly designed binders against three proteins found on blood cancers into the receptor a CAR T cell uses to spot its target.

Most failed for three reasons: the cell switched itself on with no target there, the binder could not reach the part it was aimed at, or it stuck to the wrong things. [13]

It puts a number on the distance between designing a protein and having a treatment, which is the step usually skipped when AI-designed drugs are described. [13]

Glioblastoma, the most aggressive common brain tumour in adults, almost always grows back, because its cells spread past the edge a scan can see. [14]

Radiotherapy guidelines deal with that by treating a fixed margin around the visible tumour, the same width for everybody.

A team released PREDICT-GBM, an open set of scans from 243 patients followed over time with a shared way of testing prediction models.

Their own model covered 79.4% of the places the tumour actually came back. [14]

The gain over the standard margin was small but real, and publishing the patients means rival models can now be judged on the same cases. [14]

A personalised messenger-RNA cancer vaccine is made from the mutations in one person's own tumour, so every dose is its own production run. [15]

Making them at scale means thousands of patient-specific batches running side by side, which is the opposite of how medicines are normally mass-produced.

The field moved forward last month when Merck and Moderna's personalised melanoma shot met its goal in a final-stage trial. [15]

The science question is being answered and the manufacturing question is not, and the second one decides how many people can ever be given it. [15]

The epigenome is the set of chemical tags sitting on top of DNA that tell each cell which genes to use, and it shifts with stress, diet, infection, pollution and age. [16]

Epigenome editing aims to change those tags rather than rewrite the genetic code underneath.

Fyodor Urnov of the Innovative Genomics Institute at Berkeley co-founded Tune Therapeutics, which is running trials of an epigenome editing treatment for long-term hepatitis B in the liver. [16]

A change to the tags can in principle be undone, where a cut to the DNA cannot, and that difference decides what is safe to try. [16]

Buruli ulcer is a bacterial disease that eats away skin and soft tissue, and it is common in parts of West and Central Africa.

Nobody knows how the bacterium reaches people. [20] Researchers combined 110 environmental, ecological and social measurements with records of where the bacterium and the disease turn up.

Seasonal swings in ultraviolet light and in temperature were the strongest predictors of cases. They published maps of where conditions would suit it worldwide. [20]

The maps point at places where the disease may already exist and go unreported, which is where surveillance would find it. [20]

The Trump administration made four senior appointments at the US Food and Drug Administration permanent on 8 September. [22] Michael Davis, a psychiatrist, will direct the centre that reviews drugs.

Karim Mikhail, formerly of Merck, will direct the centre that reviews biological medicines. Bret Koplow, an agency lawyer since 2011, takes the tobacco centre.

Three of them had been doing the jobs in an acting capacity. Heidi Overton was nominated to run the whole agency late last month. [22]

These are the offices that decide which medicines reach patients, and they had been run by stand-ins. [22]

Revvity, an American maker of laboratory instruments and reagents, agreed to buy Human Cell Design, a French company that grows human pancreas beta cells for research. [9]

Beta cells are the ones that make insulin.

Laboratory-grown human versions let a company test a diabetes or obesity drug on human tissue before it reaches an animal or a person.

The deal is expected to close in the last quarter of 2026. [9]

Testing on human cells rather than mouse cells is meant to catch failures earlier, and the tools for doing it are being bought up by the big instrument firms. [9]

Schrodinger sells software that predicts how a molecule will behave before anyone makes it, and it also takes stakes in the drugs its software helps design. [10]

It has now spun out a new company, Tectora Therapeutics, handing it two small-molecule programmes, SDGR-4594 and SDGR-8139, whose targets have not been disclosed.

Tectora will work in secret for now. Schrodinger's head of therapeutics research said it means to keep expanding arrangements like this. [10]

It is a toolmaker taking a share of the medicine rather than a fee for the tool, which changes what the software is worth to its owner. [10]

Clean Cells, a French laboratory that checks medicines grown in living cells for stray viruses, launched a sequencing-based version of that test and bought Xegen, the French firm that had been analysing the results. [11]

Sequencing produces enormous files, and turning them into an answer a regulator will accept is the hard part. Xegen's chief executive and his two staff are joining.

European rules have been pushing this kind of testing since 2024. [11]

Reading the data was the step being contracted out, and it is the step the regulator's question actually turns on. [11]

Most drugmakers no longer own all the factories that make their products. They hire firms called contract development and manufacturing organisations instead. [12]

The ten largest took a combined $35.116bn in 2025, up 6% from $33.127bn the year before, and eight of the ten grew.

Most large drug companies expect to lean on them harder in coming years, even though what they charge is rising. [12]

Every gene therapy and antibody in today's news is made somewhere, and increasingly that somewhere belongs to a company whose name is not on the box. [12]

Clinical trials are slow and expensive, and much of the cost comes from finding patients, waiting years for hard results and working through complicated eligibility rules. [17]

A review in Nature Reviews Bioengineering proposes running them with AI at four stages, from gathering and cleaning the patient data at the start to checking each AI tool against the exact job it is doing.

It also sets out where regulators and ethics sit in that. [17]

It is a proposal rather than a result, and it names the awkward part plainly: an AI used in a trial has to be validated for the specific question it is asked. [17]

Gene editing has moved from cutting DNA, which damages cells, to methods that change single letters without a cut. [19]

Those handle small errors well and cannot insert a large piece of DNA.

A review out this month covers the enzymes that can: recombinases and transposases, many of them found by searching genetic databases by computer and then improved in the laboratory. [19]

Diseases such as cystic fibrosis have hundreds of different causing mutations, and writing in one whole working gene would treat them all at once instead of one at a time. [19]

Autumn, from Helperby in North Yorkshire, was diagnosed at 18 months with Hyper-IgE syndrome, a rare fault in the immune system that brings severe eczema and constant infections. [21]

She had chemotherapy to clear out her own bone marrow, then a stem cell transplant from a donor in Germany in December, then more than 30 days in a hospital room that only her parents could enter.

Six months of treatment in all. She started school on Tuesday. [21]

The treatment for the condition is to destroy a child's own immune system and rebuild it from a stranger's cells. [21]