China clears the first CAR-T for a solid tumor — the wall cancer therapy has hit for 15 years

Biotech & Longevity 3 min 80 sources

A Shanghai biotech won the world's first approval for a CAR-T therapy against a solid tumor, cracking a problem that has stopped the field since 2010 — plus an AI lab that now wants to make its own drugs, a 40-year-old cancer target finally cornered, and a clue to why some brains shrug off Alzheimer's.

Key takeaways

China approved the first CAR-T therapy for a solid tumour, cracking a problem that had stopped the field since 2010 — though only in late-stage stomach cancer, and only in one country so far.
AI labs like Anthropic now want to develop their own drugs, not just sell software — but designing a molecule is the easy part; surviving human trials is where most drugs still fail.
A 40-year "undruggable" cancer target fell to a Roche drug in a late-stage lung-cancer trial, and a Dutch study found brain cells that may explain why some people resist Alzheimer's.

The wall CAR-T just climbed

For fifteen years, one of cancer medicine’s most powerful tools has only worked on one kind of cancer. That changed on June 22.

CAR-T therapy takes a patient’s own immune cells, re-engineers them in a lab to recognise a cancer, and puts them back to hunt it down [10]. Since the first use in 2010 it has produced eight approved medicines — every one of them for blood cancers like leukaemia and lymphoma [39]. Blood cancer cells float free in the bloodstream, easy for the engineered cells to reach. Solid tumours — the lumps that make up most cancer, in the stomach, lung, brain, pancreas — have resisted every attempt [20][49].

Now China’s drug regulator has approved satri-cel, made by Shanghai-based CARsgen, to treat advanced stomach cancer [39]. It is the world’s first CAR-T cleared for a solid tumour. “It is the first large market approval for any CAR-T,” said Carl June, the University of Pennsylvania scientist who led the original CAR-T work [39]. The therapy targets a protein called Claudin18.2 that sits on stomach-cancer cells but not on the healthy stomach cells around them [39] — the whole game in a solid tumour is finding a marker the drug can aim at that won’t also wreck healthy tissue. Zonghai Li, who spent his whole career in China, spotted that target while screening cancer cells at the Shanghai Cancer Institute and ran his first trial in 2017 [39].

The caveats are real: it was approved in heavily pre-treated patients who had run out of other options, and one country’s approval is not the world’s [39]. But the wall has a crack in it, and it did not come from Boston or San Francisco [39].

The AI companies want to make the drugs, not just the tools

Anthropic, the AI lab behind the Claude models, told STAT it will start developing its own drugs [54]. It joins a rush: Insilico Medicine, an AI-drug firm, signed discovery deals with the drugmakers Merck and Takeda [63], and a wave of “AI scientist” startups now pitch software that designs candidate molecules [63].

For years the AI firms sold the shovels — tools that drug companies used. Moving to make the drugs themselves is a different bet: that running your own programmes teaches you things selling software never will [54]. Whether an algorithm that proposes a molecule can shorten the decade-long, mostly-failing slog from idea to approved medicine is still unproven. Designing a plausible molecule is the easy end; the human trials are where roughly nine in ten drugs die.

A target 40 years in the crosshairs

KRAS is a gene that, when mutated, drives about a quarter of all cancers. It was called “undruggable” for four decades — the protein it makes is smooth and offers almost nothing for a drug to grab onto. This week Roche reported that its KRAS-blocking drug beat rivals from Amgen and Bristol Myers Squibb in a head-to-head phase 3 lung-cancer trial [40]. A phase 3 is the large, final human test before approval, so this counts as a real result, not a lab hope. A separate kidney-cancer drug, casdatifan, showed lasting responses in about a third of patients in an early trial [34] — a reminder that “a third” is often what genuine progress looks like, not a headline cure.

The story worth watching: brains that shrug off Alzheimer’s

Some people carry the physical damage of Alzheimer’s in their brains — the plaques and tangles — and stay mentally sharp anyway. A study from the Netherlands Institute for Neuroscience, published July 3, offers a clue why: a rare group of immature brain cells in these people seems to survive the damage instead of dying from it [62]. It is early, and resilience is hard to turn into a treatment. But it flips the usual question. Instead of only asking what causes the disease, it asks what protects against it — and that is a door the field has barely opened [37].

02 · Lesson · why it matters

Why the same weapon wins one war and loses the next

A cure that works in one place and fails in another is usually not a story about the cure — it's a story about the ground it was used on.

One tool, two verdicts

Here is the strange fact behind today’s news. CAR-T therapy — a patient’s own immune cells, re-engineered to hunt cancer — has been one of medicine’s great successes for blood cancers. Eight approved drugs since 2010. People given months to live have walked out cancer-free.

Point that same weapon at a stomach tumour, and for fifteen years it did almost nothing. Not a weaker version of the tool. The same tool. The engineered cells were just as clever, just as deadly. And they failed.

When one weapon wins decisively in one place and loses completely in another, the tool is not the interesting variable. The ground is.

Why blood was the easy war

A blood cancer floats. Its cells drift free in the bloodstream, one at a time, mixed into the very fluid the immune cells travel through. An engineered hunter cell can bump into almost any target it’s looking for. There’s nowhere to hide.

And blood-cancer cells wear a clear uniform. They carry a marker — a protein on the surface — that healthy cells nearby mostly don’t. So the drug can be told: kill anything wearing this. The target is reachable, and it is distinct. Those two things are the whole reason it worked.

The tumour builds a fortress

A solid tumour is a different kind of enemy, because it does something a floating cell can’t: it builds terrain.

It packs itself into a dense mass the hunter cells struggle to push into. It surrounds itself with a chemical moat — signals that switch off immune cells as they approach, so the ones that do get in arrive exhausted. And it rarely wears a clean uniform. The flags on a tumour cell are usually the same flags worn by the healthy tissue it grew from, so a drug told to “kill anything wearing this” would kill the patient’s stomach too.

The tumour isn’t just sitting there being hard to reach. It is actively engineering the ground it’s fought on — the reach, the moat, the disguise. That engineering, not the tumour’s raw danger, is what beat the weapon for fifteen years.

So look at what actually changed this year. The Shanghai team didn’t build a stronger cell. They found Claudin18.2 — a flag that stomach-cancer cells fly and healthy stomach cells don’t. The breakthrough wasn’t a better weapon. It was reading the terrain well enough to find the one honest target on it.

The variable we keep missing

This is a pattern that reaches far past cancer, and it’s an easy one to get backwards.

A drug works in a trial in one country and flops in another — and everyone argues about the drug, when the patients, the diets, the other illnesses were different. A teaching method transforms one school and does nothing at the next. A business model prints money in one city and collapses in the next town over. A policy that fixed one country’s problem makes another’s worse.

Each time, the instinct is to judge the tool. It worked, so it’s good; it failed, so it’s flawed. But often the tool was constant and the ground moved underneath it — the surrounding conditions that let it grip in one place and gave it nothing to hold in another. The success and the failure were both mostly written by terrain nobody was looking at.

You are on the ground too

It’s worth sitting with how close this is. Most of us will, at some point, be the patient — the one for whom a treatment either reaches its target or can’t. Whether it works may depend less on the drug’s brilliance than on the specific terrain of one body: which flags a tumour happens to fly, what else is going on inside.

And the habit of over-crediting the tool runs through ordinary life. When something worked for us — a diet, a routine, an investment, a way of raising a kid — we tend to credit the thing we did, and recommend it. We rarely credit the ground we did it on: the body, the income, the moment, the people around us that made it grip. The next person copies the tool onto different terrain and it slides right off, and they wonder what they did wrong.

What the crack in the wall really shows

The satri-cel approval is a real achievement, and its limits are real too — late-stage patients, one country, one tumour type. But the deeper lesson is quieter than “cancer breakthrough.”

It’s that a weapon’s power was never the whole story. The same cells that couldn’t touch a tumour for fifteen years can touch this one now, because someone finally read the ground carefully enough to find the one place it was exposed. The tool didn’t get smarter. Our reading of the terrain did.

Carry that, and you hold your own successes and failures a little more loosely. When something works, ask what ground it worked on before you trust it to work again. You are almost never seeing all of it.

03 · Lab · your turn

Deploy the Therapy

Rehearse how a treatment's success is decided by the terrain it's used on — reach, moat, and a unique target — not by cranking the dose.

04 · Hope · carry this

A wall that held for fifteen years just cracked — not because someone built a bigger weapon, but because someone kept reading the ground until they found the one place it was open. The hardest problems aren't permanent; most are just waiting for a patient enough eye.