People asking why a single-celled alga swims toward light ended up handing brain science its most useful switch.
On 5 October 2026 the Nobel Prize in Physiology or Medicine went to Karl Deisseroth, Peter Hegemann and Georg Nagel "for their discoveries concerning light-gated ion channels and optogenetics".
The prize doesn't reward a conclusion about the brain. It rewards a tool. For the first time, researchers could pick out one type of neuron and, with a flash of light timed to the millisecond, make those cells fire or fall silent, then watch how an animal's behaviour changed. Whenever an earlier issue said "switch these cells on in a mouse and it…", this is usually what was behind it. What it changed is not what we know but how we can know it: from "this region lit up" to "this caused it".
Most of neuroscience's old methods were about watching. Brain imaging shows which region brightens when you're afraid; electrodes pick up which neuron fires when a rat turns a corner. But all of this is correlation: two things happening together doesn't tell you which one caused the other. Maybe the region is just along for the ride.
To prove causation you have to intervene: switch it off and see whether the thing still happens; switch it on and see whether the thing appears out of nowhere. The trouble is that different types of neuron are packed together like wires of different colours twisted into one cable. Pass current through an electrode and every nearby cell is stimulated; use a drug and it takes minutes to act and spreads across the brain.
In 1979 Francis Crick, co-discoverer of the DNA double helix, wrote a wish in Scientific American: neuroscience needed a way to switch off one type of cell while leaving all the others untouched. Later he suggested light might do it. When he made that wish, the answer was hiding somewhere that had nothing to do with brains.
Chlamydomonas is a green alga made of a single cell, with a small orange-red spot called an eyespot. It uses the eyespot to sense where light is coming from and beats its two flagella to swim toward it, so it can photosynthesise. From the 1980s and 1990s, Hegemann kept asking: how exactly does it "see" light?
He and Nagel (with their collaborator Ernst Bamberg) eventually found the answer: the alga's membrane carries a protein that, when struck by light, opens a tiny pore by itself and lets charged ions through. It was named channelrhodopsin. They reported the first one (ChR1) in 2002 and a second, more useful one (ChR2) in 2003: shine blue light on it and positive ions rush into the cell.
What makes it so neat is that one protein does two jobs: it's both the light sensor and the gate. Others had controlled neurons with light before, such as Gero Miesenböck's 2002 system, but those needed several genes at once. ChR2 needs just one.
A neuron fires when positively charged ions suddenly pour in, flipping its voltage upward into a spike called an action potential. Neurons & action potentials In 2005, in Deisseroth's lab at Stanford, with experiments led by Ed Boyden (then a PhD student) and Feng Zhang and in collaboration with Nagel and Bamberg, the team put the ChR2 gene into rat neurons: every flash of blue light made the neuron fire once, with millisecond precision. Two years later they borrowed another microbe's light-sensitive protein, halorhodopsin: yellow light drives negative ions in and holds the neuron down so it can't fire. Now they had both switches.
The last piece of the puzzle was choosing which cells. In front of the light-sensitive gene you can attach a "keycard" sequence, a promoter, that only certain neuron types will read, so only they build the protein; then an optical fibre a little thicker than a hair goes into the target region. Within one piece of brain tissue, only the chosen cell type takes orders from the light. Crick's wish had come true.
With a switch in hand, many "we think it works like this" ideas could be tested directly for the first time.
Is a memory really stored in a small group of cells? In 2012 Xu Liu, Steve Ramirez and colleagues in Susumu Tonegawa's lab gave a mouse a mild shock in one box while tagging the hippocampal cells active at that moment with ChR2. Hippocampus & entorhinal cortex Later they put it in a new, perfectly safe box and turned on blue light: the mouse froze at once, as if back in the frightening place. Lighting up that small group of cells was enough to bring back a fear memory. Such cells came to be called a memory engram, echoing Topic 4.
Does dopamine really report "how much better than expected"? Topic 6 showed that dopamine neurons fire like a reward prediction error: more firing when things turn out better than expected, and learning is driven by that difference. Reward circuit But that was only correlation. In 2013 Elizabeth Steinberg and colleagues used a clever trick: first rats learned "light means sugar"; then light and a tone came together. Because the sugar was already predicted by the light, there was no surprise, and rats normally fail to learn what the tone means. But if a burst of dopamine-neuron firing was added artificially with light at the moment the sugar arrived, the rats learned it. The difference signal doesn't merely accompany learning; it causes it.
One limit needs stating: artificial stimulation drives hundreds or thousands of cells in lockstep, far more uniformly than anything natural. "Switching it on can cause a behaviour" shows it is sufficient to produce the effect, not that the brain normally uses it that way. Good experiments test both directions, on and off, and compare against natural recordings.
People studying what goes on inside large language models (a field called interpretability) have walked the same road. An early method is the probe: check whether certain numbers in a layer correlate with "this sentence is about Paris", which is only correlation. Later methods intervene: swap the model's internal values at one location for the values from a different run, called activation patching, or simply set them to zero, called ablation, and see whether the output changes. A 2022 study used this kind of "causal tracing" to locate facts like "the Eiffel Tower is in Paris" in specific parts of the model's middle layers, and could rewrite them there directly. Artificial networks are far more convenient than brains: every number is visible and editable, no algae required. But the limit is the same: if changing a spot changes the output, it's sufficient to matter, not proof that the model normally relies only on it.
Each Nobel prize can go to at most three people, and optogenetics clearly took more than three pairs of hands. Controversy arrived with the announcement, centred on Ed Boyden: first author of the key 2005 paper, then a PhD student, later at MIT. Others often named include Zhuo-Hua Pan, who used ChR2 to make retinal neurons respond to light in early 2004, about six months before the Stanford experiment; Gero Miesenböck, who achieved light control of neurons earlier; and Ernst Bamberg, co-author on the ChR2 papers. The committee chair declined to comment on why these three were chosen. A common reading is that the committee honoured "those who discovered the molecule" plus "the person who turned it into a neuroscience tool".
How far is it from people? Optogenetics first needs a virus to deliver the gene into cells, then needs light delivered deep into tissue, and both are hard in a human brain. The closest application is the eye, because the retina is already transparent to light. In 2021 José-Alain Sahel and colleagues reported a 58-year-old man blinded by retinitis pigmentosa who had a light-sensitive protein gene injected into one eye and then wore special goggles turning images into amber light pulses. He became able to make out the white stripes of a pedestrian crossing and count objects on a table. This is one case from an early trial: proof of concept, still far from a routine therapy.
It also has alternatives and successors: chemogenetics switches cells with a designed receptor plus a matching drug, no fibre needed, but much slower; methods using ultrasound instead of light are being explored. The brain–computer interface issue (Topic 29) was about "reading"; optogenetics opened up "writing". In animals, writing is now routine; in people, it's only just beginning.