Take a cornea out of an eye, drop it in saline, and within a few hours it is cloudy and half again as thick. Nothing has attacked it. That is simply the tissue at rest.
So the window you are reading through is not clear the way glass is clear. Glass is clear because of what it is. Your cornea is clear because of what something is doing to it, continuously, and has been doing since before you were born.
The leak
Most of the cornea is stroma: collagen fibrils in an array so regular, and so much finer than a wavelength of light, that what each one scatters is cancelled by its neighbours. Threaded between them are proteoglycans carrying fixed negative charges, and fixed charge holds water the way a sponge does.
The stroma is therefore always straining to swell, and the strain has been measured: at normal thickness it pulls fluid inward at about 50 mmHg. Let the fibrils drift apart and the cancellation fails. Thickness and clarity are one variable, not two.
The pump
On the cornea’s back surface, facing into the eye, is a single layer of cells one cell deep — the endothelium — whose work is to move that water back out. It runs on sodium-potassium ATPase, bicarbonate, carbonic anhydrase. No. 57 described machinery like this holding a gradient up for signalling; here the same running cost buys nothing but clarity.
The arrangement has a name, the pump–leak mechanism, and the point of it is that transparency is not a state the cornea is in. It is a rate held against a load, the way a bailed boat floats.
The honest part: the pump has been understood in outline since the early 1970s, and how it physically moves water is still contested — ions first with water following, or water dragged through the junctions between cells by the current itself. A prominent paper for the second was retracted. Fifty years on, we can measure the bailing exactly and still argue about the bucket.
A workforce that only ever shrinks
You are born with about six thousand of these cells per square millimetre. By your twenties it is near 3,000. By the seventh and eighth decades, about 2,600, falling around 0.6% a year. Somewhere near 400–500 per square millimetre the pump can no longer beat the leak, and the cornea swells for good.
Every number in that sequence is smaller than the one before. No. 123 gave a hurricane a computed ceiling it almost never reaches; this is the opposite shape, a floor you approach all your life and cannot climb back from. It is also the inverse of last issue’s gut, where the defence held because its residents divided faster than they washed away. Here nobody divides at all.
They are not unable. They are forbidden.
The received line is that these cells cannot divide, and it is the wrong word. Put them in a dish, block transforming growth factor β, and they divide well enough that people are growing them for transplant.
In the eye they are held in G1 by two things at once. They sit in aqueous humour, which is rich in growth-inhibiting signals, TGF-β2 among them. And they are pressed against each other on every side, and that contact itself enforces the arrest. A complete monolayer is a monolayer whose cells have been told to stop.
Which is the loop, and it is nastier than “these cells can’t regrow.” The condition that makes the layer work is the condition that forbids it to rebuild. Intactness is the off switch. No. 97 found ground that excludes its own occupant; this excludes the occupant’s replacement.
The healing that restores the ban
Kill some of them and the survivors do not divide into the gap. They stretch — each cell flattening and spreading until the bare patch is covered. Fewer cells, larger, less regular, still a continuous sheet.
That is a real repair; the barrier closes. But contact has been restored, so the arrest has been restored, and the count is permanently lower. The wound heals by re-imposing the rule that stopped anyone replacing the dead.
Taking tissue away
Which suggests something backwards, and surgeons now do it. In Descemet stripping only, the diseased central endothelium — about four millimetres of it — is peeled off and nothing is put back. Removing it releases the contact inhibition at the wound’s edge; healthier cells migrate in from the periphery; the cornea clears. In one long-term series, twelve of thirteen eyes did. Rho-kinase inhibitor drops speed it up.
It needs a good periphery to draw from, so it is not for everyone. But set it against the arithmetic of corneal donation — roughly one cornea available for every seventy needed worldwide — and the shape is worth sitting with. The treatment is a subtraction. You get the pump back by taking tissue away and letting the survivors off the leash.
≈
One loop I’m watching
Next: a crystal you make colder by shining a laser at it. It absorbs a photon, borrows a quantum of its own heat to top that photon up, and emits one carrying more energy away than arrived. Repeat, and the solid refrigerates itself — 91 K, no moving parts, no coolant, nothing but light in and brighter light out.