The Medicine Nobel Is A Photonics Prize

Blue light at 473 nanometres travels about 47 micrometres through mouse cortex before it is scattered. Not absorbed — scattered, which for a technique that needs to put a known number of photons onto a known population of cells is in some ways worse: the photons are not gone, they are somewhere you did not aim them. That single figure, measured and published a decade ago, is the reason an entire industry of implanted fibres, tapered probes, red-shifted proteins and wearable light engines exists. On 5 October the Nobel Assembly at Karolinska Institutet gave the 2026 Nobel Prize in Physiology or Medicine to the people who made that number matter.

The prize is shared by Karl Deisseroth, D. H. Chen Professor and professor of psychiatry and behavioral sciences and of bioengineering at Stanford University and a Howard Hughes Medical Institute investigator; Peter Hegemann of Humboldt-Universität zu Berlin; and Georg Nagel of Julius-Maximilians-Universität Würzburg, who was Professor of Molecular Plant Physiology at the Chair of Botany I there from 2004 to 2021 and remained a senior professor at the Institute of Physiology until the end of June this year. The announcement divides the work cleanly: Hegemann and Nagel found the protein in a single-celled alga, and Deisseroth turned it into a switch for nerve cells.

The chain runs through three papers. In 2002 Nagel, Hegemann and colleagues identified channelrhodopsin-1 in the green alga Chlamydomonas reinhardtii and showed it was a light-gated proton channel. In 2003, in PNAS, the same group characterised channelrhodopsin-2: a directly light-switched, cation-selective ion channel, expressed functionally in frog oocytes and in mammalian cells. In 2005 Edward Boyden, Feng Zhang, Ernst Bamberg, Nagel and Deisseroth put it into cultured hippocampal neurons and drove action potentials with pulses of blue light, on demand, at millisecond precision.

That 2005 result is where a protein became an instrument. Everything since has been an engineering problem, and most of the engineering has been optical.

Figure 1. Georg Nagel, who shares the 2026 Nobel Prize in Physiology or Medicine with Peter Hegemann and Karl Deisseroth. Credit: Rudi Merkl / Universität Würzburg.

Read the 2003 paper as a component datasheet and channelrhodopsin-2 is a remarkable one. Its action spectrum peaks at 460 nanometres, about 40 nanometres blue of channelrhodopsin-1. It is non-selective among the monovalent cations — lithium, sodium, potassium, rubidium, caesium all pass — and permeable to calcium, strontium and barium, while anions, magnesium and anything as bulky as tetraethylammonium are excluded. The opening time constant is 0.2 milliseconds or shorter. Closing takes between 10 and 400 milliseconds depending on the pH inside the cell, and full recovery from the desensitised state takes roughly two seconds in darkness.

The causal chain is short enough to state without equations. The protein holds a retinal chromophore. A photon of the right energy isomerises it, the isomerisation forces a conformational change in the surrounding protein, and that change opens a pore through the cell membrane. Cations flow down their electrochemical gradient, the membrane depolarises, and if it depolarises past threshold the neuron fires. No second messenger, no enzyme cascade, no catalytic amplification of the kind that turns one photon into a chain of signalling events in a rod cell. The photon and the ion current are separated by a single photocycle, which is exactly why the response is fast enough to be useful and why the thing behaves like a detector rather than a biological process.

It is also why the photon budget is unforgiving. There is charge gain — an open channel passes ions for milliseconds after the photon that opened it has gone — but no cascade gain. Every channel that opens had to be hit by its own photon.

In 2005 the light came through a microscope. Boyden and colleagues measured 8 to 12 milliwatts per square millimetre at the sample through a 20× objective, using a 450-to-490-nanometre band, with 5-to-15-millisecond pulses; the first evoked spike peaked 8.0 ± 1.9 milliseconds after the pulse, with trial-to-trial jitter of 0.5 ± 0.3 milliseconds. In a dish, delivering that is trivial. In a living brain it is the entire problem.

Those figures are for acute mouse cortical slices at 473 nanometres, with a scattering coefficient of 211 per centimetre against an absorption coefficient of 0.6, and an anisotropy factor of 0.86 — scattering dominates by more than two orders of magnitude. The authors' point at the time was that the scattering length is considerably shorter than optogenetics experiments had been assuming. Half the light is gone within the thickness of two cell bodies. By a fifth of a millimetre there is almost nothing left.

Three consequences follow, and they shape every product in this corner of the market. The light source has to be inside the skull, which means an implanted fibre and a chronic optical feedthrough. The irradiance at the target is not the irradiance at the fibre tip, so dose is estimated rather than known, and a stated milliwatt figure without the depth and wavelength it was measured at means very little. And the illuminated volume is set by scattering rather than by the optics, so "which cells did I actually stimulate" becomes a modelling question instead of a measurement.

The obvious photonics answer is to go to a longer wavelength, where tissue scatters less. The protein engineers obliged.

Chrimson's amber peak buys penetration. Its faster variant ChrimsonR, carrying a single K176R mutation, cuts the off-kinetics from 21.4 ± 1.1 milliseconds to 15.8 ± 0.4 milliseconds, enough, the authors report, for reliable red-light-driven spiking at frequencies of at least 20 hertz. The same work demonstrated something a spectroscopist would recognise as the real prize: two opsins driven independently in the same tissue, Chronos at 470 nanometres and Chrimson at 625 nanometres, with the blue channel held between 0.2 and 0.5 milliwatts per square millimetre to keep it from leaking into the red-shifted population. Channel separation, in other words, with crosstalk set by an action-spectrum overlap rather than a filter edge.

That is where the opsin toolbox and the light-source catalogue stop matching each other.

Look at what a supplier to this market sells. Doric Lenses, one of the standard vendors, lists connectorized laser-diode modules at 405, 450, 473, 488, 520 and 638 nanometres, with typical fibre-coupled powers from 60 to 140 milliwatts, FC/APC terminated into multimode 50-micrometre, 0.22-NA fibre, all Class 3B. The 473-nanometre line puts out a typical 85 milliwatts.

Note what is missing. The diode range steps from 520 straight to 638 nanometres. There is no amber line, because there is no cheap, compact, fibre-coupled diode at 590 nanometres — InGaN runs out before it and AlGaInP starts after it, and the gap falls exactly where the best red-shifted opsins peak. It is filled either by diode-pumped solid-state lasers at 589 or 593 nanometres, which work but are bulky, expensive and awkward to modulate quickly, or by LEDs. The same vendor's connectorized LED range covers 24 wavelengths from 365 to 940 nanometres, including 470, 595, 625 and 635, coupled into 200- or 400-micrometre 0.57-NA fibre or 960-micrometre 0.63-NA. The powers tell the story: 470 nanometres gives up to about 200 milliwatts into the 960-micrometre fibre, while 595 nanometres gives 2.0, 8.0 and 44 milliwatts into the 200-, 400- and 960-micrometre fibres respectively. An LED buys the wavelength and pays in étendue — a large Lambertian emitter has far more of it than a small fibre will accept — and note that the LED range is not offered into the 50-micrometre fibre the diode modules use at all.

This is a genuine, unglamorous photonics opportunity sitting in plain sight: a compact, fibre-coupled, modulatable amber source with laser-like brightness. Twenty years of protein engineering have moved the demand into a band the light-source industry does not serve well.

Figure 2. Where the opsins peak and where the light sources are, and how far blue light gets. Schematic drawn from values reported by Nagel et al. (2003), Klapoetke et al. (2014), Sahel et al. (2021), Yona et al. (2016) and the Doric Lenses catalogue — not from measured data. Credit: Photonics Insider.

The other response to scattering is to stop fighting it and control the geometry instead. A flat-cleaved fibre illuminates a blob whose size is set by tissue optics; a shaped fibre can do better.

The devices are tapered multimode fibres — 0.22 NA with a 2.2-degree taper on a 50-micrometre core, or 0.39 NA with a 2.9-degree taper on a 200-micrometre core — in which the mode content launched at the input determines where along the taper light leaks out. Change the launch angle and the emitting zone slides along the implant. It is modal engineering doing the work of a mechanical positioner, in a device thin enough to leave in a mouse for months, and it is the kind of problem a fibre-optics engineer will recognise immediately and a neuroscientist generally will not.

The clinical case to date is not in the brain at all. It is in the eye, where the light does not have to go through anything.

In 2021 a team led by José-Alain Sahel and Botond Roska reported in Nature Medicine that a patient blind from retinitis pigmentosa had partially recovered visual function after a single intravitreal injection of 5.0 × 10¹⁰ vector genomes of AAV2.7m8 carrying ChrimsonR fused to tdTomato. The opsin peaks around 590 nanometres. The other half of the therapy is an instrument: goggles with a 595-nanometre peak and a 15-nanometre half width, which capture the visual scene and project it onto the treated retina as monochromatic amber.

The engineering constraint there is dose, and the numbers are instructive. Maximum retinal irradiance was 13 milliwatts per square centimetre — 4 × 10¹⁶ photons per square centimetre per second at a 6-millimetre pupil — with the working range spanning two orders of magnitude below that. The team put the retinal figure 51 times below the thermal safety threshold, the maximum corneal irradiance of 4.7 milliwatts per square centimetre 850 times below anterior-segment limits, and classified the device in the exempt risk group of IEC 62471, capping use at four hours a day as a precaution. With the goggles the patient identified a single large object in 36 of 39 trials; without them, nothing. The trial is PIONEER, NCT03326336.

Read that as a photonics engineer and it is a familiar object: a head-worn projector with a narrowband source, a scene camera, a real-time remapping pipeline and a hard optical-safety budget it must never exceed. The novelty is the receiver, and the receiver is a protein that algae use to find light.

Nothing about the physics changed on Monday. What changed is the standing of the field: optogenetics is now the subject of a Nobel Prize in Physiology or Medicine, and prizes move money, graduate students and instrument budgets.

The thing worth noticing from this side of the fence is how much of what happens next is optical rather than biological. The proteins are good. The opening is sub-millisecond, the spectra have been pushed 45 nanometres into the red, and two populations can be driven independently in the same tissue. The limits that remain are a scattering length of 47 micrometres, an amber band the diode industry does not serve, the étendue of an LED against the acceptance of a 50-micrometre fibre, and a dose at depth that nobody can actually measure. Those are not neuroscience problems. They are laser, fibre and radiometry problems, and the people who solve them will not be the ones collecting the medal in Stockholm in December.