How a brain becomes a map, in five steps

What does it actually take to turn a brain into a map? Not a metaphor, the physical steps. I had to learn this from scratch, so here it is the way I would have wanted it explained.

The method we are scaling is LICONN, published in Nature in May 2025 by Johann Danzl's lab at IST Austria with Google Research. The idea in one sentence: expand the tissue until a light microscope can see a synapse, image it, and let the same networks that trace electron microscopy trace it. Everything below is their method and their numbers unless I say otherwise. The slab counts and the choice of microscope for a whole brain are our plan, and I label them as such.

Here's how it works:

1. Fix and slice. The brain is fixed by perfusion and cut on a vibratome into slabs. LICONN used 50 micrometer slices and showed 300 micrometers with adaptations. A mouse brain is about 40 to 250 slabs depending on the thickness we settle on. Every axon that crosses a slab boundary has to be matched on the other side, and nobody has shown that at whole-brain scale. Step 5 depends on it.

2. Expand sixteen-fold. Each slab is anchored to a swellable hydrogel, its proteins are stained, and the gel is expanded about 16 times in water. Think of a drawing on a balloon: blow it up and the lines that were too close to tell apart separate. A 20 nanometer gap between two neurons becomes 320 nanometers, which a light microscope resolves. The stain marks proteins broadly, so the image looks like electron microscopy: membranes, synapses, mitochondria.

3. Image with light. The expanded gel is imaged plane by plane. LICONN used a spinning-disk confocal with a water objective and reached an effective voxel of about 10 by 10 by 25 nanometers. For a whole brain we plan light-sheet microscopes, which image a plane at a time and are built for volumes this large. At LICONN's spinning-disk rate a brain is about 400 microscope-years by our arithmetic. Light sheets are far faster at lower resolution, and whether one holds LICONN's resolution sets the capital cost. Either way it needs a factory, not a lab bench.

4. Trace, then fix the tracing. The best published pipeline traces each neuron through the stack. Today that is PATHFINDER, Google's 2025 system, which starts from its flood-filling networks and adds automated repair on top. Separate models find the synapses from their protein labels. The trace still has mistakes, a neuron cut in two or two joined into one, and today people find and fix them by hand. That is most of the cost of a brain map. It can be automated because the fixes made to the public maps were logged, three million edits on the fly brain and a million on the mouse cubic millimeter, and about 700,000 of them have been curated into a public benchmark. We trained a model on that benchmark and ran a repair pipeline on the public mouse map. Closing the gap that remains is our core work.

5. Put it back together, labeled. Slabs are registered to their neighbors so neurons continue across them. Antibodies or reporter lines in the same gel label the synaptic proteins, the cell types and the receptor we are asked about, so the finished map says not only who connects to whom but which receptor each neuron carries. LICONN imaged one drug-target receptor subunit this way, GluN1 of the NMDA receptor. No G protein-coupled receptor, the class most brain drugs bind, has been imaged in expanded tissue at this resolution yet. That is among the first things we have to measure.

What is a discovery and what is a factory

I find it useful to sort each step into one of two bins.

Steps 2 and 3 are published and work at small scale. Making them work on hundreds of slabs, uniformly, with the same expansion factor and stain in every one, is a manufacturing problem. Hard, and the kind of hard that money and engineering solve. Sixteen-fold expansion of a whole mouse brain is about two liters of gel.

Step 4 is where the science is. A model that fixes the tracing as well as a person does not exist yet. Ours and Google's each close part of the gap. The data to train better ones exists and grows with every brain mapped.

Steps 1 and 5 hide the risk nobody talks about: registration across slab boundaries at whole-brain scale. If axons cannot be matched across cuts, the map is 250 maps. The Allen Institute's ExA-SPIM, which imaged a whole expanded mouse brain in about a day in 2025, suggests the sample handling is solvable, at three- to four-fold expansion rather than sixteen.

That said, nobody, including us, has run all five steps on a whole brain. The claim on this page is that each step exists, not that the chain does.

What it costs, step by step

The microscopes are capital, and capital for light is about a tenth of capital for electrons by E11 Bio's estimate. The chemistry is cheap per slab. The compute for step 4 is real but falls every year. The people in step 4, correcting by hand, are where $7.5 to 21.7 billion comes from today. That is the step we are replacing, and it is why what a brain map costs is a story about labor, not optics.

If you have expanded tissue, built a light sheet, or traced a neuron through a petabyte, you already know which of these steps I have understated. Please tell me which one!

Dan