A dish is not a mouse is not a person

A molecule clears a tumour out of a lab dish. A week later it shrinks the same tumour in a mouse. The press release writes itself: new cancer breakthrough. Now bet your own money. Will it help a single human being? The honest answer, before any trial has run, is: almost certainly no. Most molecules that win in a dish and win in a mouse still fail in people — and they fail for reasons that have nothing to do with the lab work being sloppy. The lab work was fine. The problem is that a dish is not a mouse, and a mouse is not a person.

Three very different test beds

In a dish, you have cells in a flat pool of nutrient broth, bathed directly in your molecule at whatever dose you like. Nothing else is happening. No liver chewing the molecule up. No immune system. No blood supply to fight through. It is the cleanest possible question: can this molecule do the chemical thing at all? Useful — but it is the easiest test it will ever face.

A mouse is far harder. Now the molecule has to survive the gut or the bloodstream, get past the liver, reach the right tissue at a high enough dose, and not poison anything important on the way. A mouse has organs, a working immune system, a metabolism. This is why “works in a mouse” is real progress over a dish. But a mouse is small, short-lived, genetically nearly identical to its cage-mates, and often given its disease in a tidy lab-built form. It is a simplified, standardised animal.

A person is the hardest test bed of all, and in three specific ways.

Why people break the result

More complex. A human body has more tissues, a slower and different metabolism, decades of other conditions, and other drugs already on board. A dose that was clean in a mouse can hit a human liver or heart that a mouse simply doesn’t have in the same form. The molecule meets a far messier environment.

Slower to read out. A mouse lives about two years, so you can watch a disease run its whole course in months. Many human diseases take years to show whether a treatment changed anything. You can’t fast-forward a person. So the readout is slow, and slow readouts are where hopeful early signals quietly fade.

Far more varied. Lab mice are bred to be almost identical — same genes, same food, same cage, same lab-induced disease. Humans are the opposite: different genes, ages, diets, environments, disease severities, and histories. A molecule that helps the average tidy mouse can help some people, do nothing for others, and harm a few — and that spread only shows up once you test a varied crowd, never in a uniform one.

So a result in a dish or a mouse is not wrong. It is just answering an easier question than the one that matters. “It works in mice” is the beginning of a long, lossy road — not a finding about people.

The funnel, with numbers

Watch what attrition does to a hopeful batch. These rates are illustrative but realistic in shape — drug development really does lose this brutally.

Start with 100 compounds that all look good in a dish. Move them into animals.

• Works in a dish → works in a mouse. Maybe 40 still look good once a whole organism is involved. 100 → ~40.
• Mouse → safe in humans (Phase 1). About half clear the first human test for basic safety; the rest show toxicity that no mouse predicted. 40 → ~20.
• Phase 1 → shows a real signal (Phase 2). Here is the cliff. Most candidates that are safe turn out not to work in people. Roughly a third survive. 20 → ~7.
• Phase 2 → beats existing care (Phase 3). A promising signal must now prove it’s genuinely better than what patients already get. About half make it. 7 → ~3.
• Phase 3 → approved. Final hurdles — confirmation, manufacturing, review — take it down again. 3 → ~1.

Out of 100 dish winners, about 1 reaches a patient. You will rehearse this exact funnel in the lab in a moment, and you’ll find something uncomfortable: turning up the “this one looked so promising in mice” dial barely changes the survivor count. The filter doesn’t care how excited the press release was. It just keeps subtracting.

Why the bar has to be this high

It would be kinder, in the moment, to lower the bar — to call a mouse result a cure and let people try it. We don’t, and the funnel is why. If 99 of every 100 hopeful compounds would have failed or harmed people, then a system that waves them through hands out 99 useless or dangerous treatments for every 1 that helps. The high evidence bar is not bureaucracy. It is the only thing standing between a real effect and a coincidence that looked good in a simplified animal.

This is the caveat that sits under the entire subject — under every gene-editing headline, every longevity claim, every “scientists discover.” It is also why this course keeps returning to one question: was it shown in a dish or a mouse, or proven in people? You met the seed of this idea earlier — a drug binding its target in a dish is only the start of the road. This is the whole road.

On the whole

The gap between a dish and a person is really a gap between a clean question and a messy world. Simplify a system enough and almost anything looks like it works; the truth only shows when you put it back into the full, varied, slow, complicated reality it has to survive. That is not a fact about mice. It is a fact about how knowing works — we are always tempted to trust the tidy version because it flatters our hope. The discipline of waiting for the hard test, on the real thing, is how we keep from fooling ourselves. And since the real thing, in the end, is people like us, the patience is not abstract. It is the difference between a headline and a help.