Three big drug programs died this week — and the one that hurts most did everything it was designed to do

Biotech & Longevity 4 min 80 sources

GSK, Roche and AstraZeneca all pulled the plug on programs for the hardest diseases. The Huntington's drug lowered the disease-causing protein exactly as planned — and patients still didn't get better.

Key takeaways

Three drugmakers — GSK, Roche and AstraZeneca — abandoned major programs this week for Alzheimer's, Huntington's and a fatal heart disease, all built on well-accepted theories.
The Huntington's drug tominersen did lower the disease-causing protein exactly as designed, yet patients didn't improve — a reminder that moving a lab number is not the same as helping a person.
The rare bright spot, a first-in-human stem-cell trial for Parkinson's, showed only that the approach is safe so far in eight patients — real progress, honestly small.

This was a week of quitting in biotech. Three drugmakers walked away from programs aimed at some of medicine’s hardest diseases — Alzheimer’s, Huntington’s, and a fatal heart condition. The news is usually breakthroughs. This week it was the graveyard.

Three programs, one week, all dead

On Thursday, GSK ended its neurodegenerative-disease partnership with the biotech Alector — a pact GSK entered in 2021 with a $700 million upfront payment and up to $2.2 billion promised [4]. Both drugs it bought into have now failed. One, latozinemab, missed its goal in a late-stage trial of frontotemporal dementia last year; the other, nivisnebart, was halted in April after a mid-stage Alzheimer’s trial was judged unlikely to work [4]. Both aimed at the same target — boosting a brain-protective protein called progranulin — and analysts now doubt the whole idea [4].

The same day, Roche told the Huntington’s disease community it was stopping two programs, including tominersen, developed with Ionis Pharmaceuticals [46]. Huntington’s is a cruel, inherited brain disease caused by a single faulty gene. And hours later, AstraZeneca said its heart drug Wainua — also partnered with Ionis — failed a pivotal trial in ATTR-CM, a condition where a misfolded protein clogs the heart muscle. The drug did no better than a placebo at preventing cardiovascular deaths, and AstraZeneca’s shares fell about 8% [47].

Two of the three failures were Ionis drugs. All three were built on a clean, logical theory of the disease. That is what makes this week worth stopping on.

The failure that should trouble you most

Tominersen is the one to sit with. It is an antisense drug — a short piece of genetic material designed to switch off a specific gene. The gene it targets makes the mutant huntingtin protein, which is the accepted cause of Huntington’s disease. The theory could not be cleaner: lower that protein, slow the disease.

And the drug worked — at the level it was built to work. In the 16-month trial, tominersen reduced the mutant protein, hitting its biomarker goal [46]. A biomarker is a stand-in you can measure — a lab number that’s meant to track the disease itself.

But the patients didn’t improve. The protein came down; the disease didn’t budge [46]. Roche was blunt about it: the reduction “didn’t come with a concurrent improvement in disease severity” [46]. The number moved. The person didn’t.

A pattern in what got abandoned

Look at all three and the same shape appears. Each drug was aimed at a target that everyone agreed was the right one — progranulin for dementia, mutant huntingtin for Huntington’s, a misfolded protein for ATTR heart disease. Each was, in a sense, correct about the target. And each still failed the patient.

That doesn’t mean the science was wrong. It may mean the drug came too late, after damage was done. It may mean the target is real but not the whole story. But it is a hard, recurring lesson in drug development: knowing the cause of a disease, and even moving it, is not the same as curing the disease. The failures cluster in the hardest, slowest diseases of the brain and heart for exactly this reason — there, the gap between the target and the patient is widest.

What a survivor looks like

The same week offered a quiet counterweight. Researchers at Lund University and Skåne University Hospital reported the first human trial of a stem-cell therapy for Parkinson’s disease, published in Nature Medicine [42]. Parkinson’s destroys the brain cells that make dopamine, a chemical that controls movement. The team grew fresh dopamine-making cells from stem cells and transplanted them into the brains of eight patients [78].

The result was cautious and real: no serious side effects from the cells in the first year, and signs the grafts survived [78]. “Replacing dopamine neurons that are lost in Parkinson’s disease has been a long-standing goal,” said Malin Parmar, who led the work [78]. But note the caveats — eight patients, an open-label study with no comparison group, and a design built to test safety, not to prove it works [42]. It took the field decades to reach this modest, hopeful first step.

And even real progress isn’t shared

One more report is worth closing on. The World Health Organization’s annual cancer review found that the striking scientific gains against cancer have barely reached millions of people [69]. In rich countries, 85% of those diagnosed with breast or childhood cancer survive at least five years; in poor countries, the figure falls below 30% [69]. Between 9% and 54% of the WHO’s top-priority cancer drugs are available in the lowest-income countries, against 68% to 94% in the richest; 23 countries have no radiation facilities at all [69].

The WHO’s cancer lead, Dr Andre Ilbawi, put the whole week in one line: “For years, the story told about cancer has been about scientific progress, new technologies, new treatment, new hope. That story is true… but it’s not the whole story.” [69]

02 · Lesson · why it matters

The number moved and the person didn't

We manage what we can measure, and then quietly start believing the measurement is the thing itself.

A drug that worked, on a patient who didn’t get better

The failure worth understanding this week isn’t the one that flopped. It’s the one that succeeded — and failed anyway.

Tominersen, the Huntington’s drug Roche just abandoned, was designed to lower a specific protein. That protein is the accepted cause of the disease. The drug lowered it. By every plan on paper, the patients should have started doing better. They didn’t.

Sit with how strange that is. The drug did exactly what it was built to do. The lab number fell. And the disease kept moving. This is not a story about a broken drug. It’s a story about the gap between a thing we can measure and the thing we actually care about.

The proxy is a stand-in, not the goal

Real diseases are slow, tangled, and hard to see. So science reaches for a proxy — a measurable stand-in that’s supposed to track the real thing. Lower the mutant protein, and you’ll slow the disease. Drop the cholesterol number, and you’ll prevent the heart attack. Raise the test score, and the child has learned.

The proxy is chosen because it’s honest and useful. Most of the time it holds. But a proxy is a bet — a bet that the thing you can measure and the thing you want are tightly bound. When the bet is right, moving the number moves the world. When it’s wrong, you get tominersen: the number obeys you, and reality shrugs.

The trap isn’t using proxies. You have no choice; you can’t measure “a disease” directly, only its shadows. The trap is forgetting that the proxy was ever a stand-in — starting to treat the shadow as the substance.

Why the target can be right and the drug still fail

Here’s the part that keeps drug developers humble. All three programs that died this week were aimed at targets nobody thinks were random. Progranulin really does protect neurons. Mutant huntingtin really does cause Huntington’s. The misfolded protein really does clog the ATTR heart.

So how do you aim at the correct cause and still miss? A few ways, and they all generalise. You can arrive too late, after the damage is done — the number was the cause, but not anymore. You can hit a real cause that turns out to be one of several. Or the proxy can be a genuine part of the disease that simply isn’t the lever — present at the scene, but not driving the car.

Being right about the target is not the same as being right about the patient. That’s the sentence a decade of neuroscience keeps writing in failed trials.

This is not a science problem. It’s a seeing problem.

Step back and the pattern is everywhere you look, because measuring-a-stand-in is how humans manage anything too big to hold directly.

A company watches quarterly revenue and forgets it was a proxy for a healthy business. A government watches GDP and forgets it was a proxy for people living well. A hospital watches wait times and forgets they were a proxy for care. A school watches test scores. In each case the number is real, useful, and — the moment it becomes the target — quietly dangerous, because now there are careers and budgets and billions riding on the stand-in itself.

And someone chose which number counts. Who decides that the mutant protein is the endpoint, that GDP is the measure, that this test is the one — that choice sets the terms everyone downstream lives under, and it comes to look like plain fact when it was a judgement all along.

What the whole looks like from here

The reader is inside this, not watching it. You are measured by proxies every day — a credit score standing in for your trustworthiness, a job title for your worth at work, a step count for your health. Some of those bets are good. Some have drifted so far that the number and the truth have quietly split, the way tominersen and the disease split.

The humility isn’t “measurement is bad.” It’s smaller and harder: every number you steer by is a stand-in for something you can’t see directly, and reality is under no obligation to agree that you picked the right one. The experts who chose these targets were careful, and they were surprised anyway. The next time a number tells you the problem is solved, the useful question is the one Roche was forced to ask this week — solved, or just the number?

03 · Lab · your turn

Read the Trial

Rehearse whether to trust a biomarker that drops or demand the real patient outcome, and feel how the same number can hide three different truths.

04 · Hope · carry this

Every dead program this week was a company choosing to tell the truth about what didn't work rather than sell a number that looked good — and that honesty, patient by ruled-out patient, is exactly how the field eventually finds the drug that does.