Fly connectome to simulation: can a wiring map become a fly?
Trace a neuron through microscope slices, turn synapses into a graph, give the graph a firing rule, add a body and test the model against a real fly.
Lesson summary
A connectome is the map of a fly, not the fly: the rules, the body and the tests are what make it move.
- A neuron is traced by following its cut profiles slice to slice, and people check the joins.
- Contacts have a direction, so the wiring is a directed graph with counts on its edges.
- Each map covers one specimen and one part of the nervous system, so check which fly it is.
- The wiring says where a signal can go; a firing rule with leak and threshold says whether it does.
- A body needs an interface to motors and fresh senses coming back, or it acts on stale news.
- A model earns trust by predicting what happens when you silence a real neuron.
At slice 6 the tracer jumped onto the neighbour. What should it have followed instead?
Slices are so thin that a cell barely moves between them, so the piece that overlaps the last slice is the same cell. Nearest can belong to a neighbour.
The taste cell makes three contacts onto the relay. What goes in the relay’s row, under the taste cell?
A contact sends one way, from the side holding the vesicles. Taste cell to relay is 3, and relay to taste cell stays 0.
Why can the brain-only map not tell you how the fly walks?
FlyWire mapped one female brain. The circuits that run the legs are in the nerve cord, which the male CNS map adds.
Same wiring, same two touches. Why did the fly drink with the slow leak but not the fast one?
With the slow leak the pair added up to 1.08 and the relay fired. With the fast leak the first taste had nearly drained, so the pair peaked at 0.73.
With the plug on her smell lead pulled out, why did she walk right past the sugar?
Her brain held her first reading, 71 mm, the whole way, so the command never changed. With a fresh reading every 0.4 s she slowed and stopped at the drop.
Both model flies drank sugar. What finally told them apart?
With the relay silenced the models called it differently: A stayed dry and B drank through its extra wire. The real fly stayed dry, which ruled B out.
Stage 1 of 8
Is a scanned brain a working brain?
A fruit fly’s brain is the size of a poppy seed.
This fly walks to sugar and drinks, but she only exists on a screen.
Her brain was built from a real fly’s brain, cut into thousands of slices and scanned one at a time.
The fly it came from is gone; that’s her empty dish.
So is this her, running on a computer?
Can a wiring map become a fly?
A head like hers, drawn open, is packed with fine threads.
Each thread is a single cell, long and branching, called a neuron.
Three of them carry a taste of sugar: the taste cell, the relay, and the mouth cell.
Where the taste cell presses on the relay, the signal crosses over: that contact is a synapse.
Trace every neuron through the slices, mark every synapse, and that whole wiring map is her connectome.
Ours are three made up for teaching, but a real fly’s brain has about 140,000 neurons.
That’s too many to watch, so we’ll follow one taste of sugar, this pink drop, through three cells.
It passes from the taste cell to the relay to the mouth cell; each hand-over asks a question.
Some answers come straight from the scan, but the rest, someone has to build.
So we start with the scan: following the taste cell up through the stack of slices.
Stage 2 of 8
How do you follow one neuron through a stack of slices?
Here’s that stack of slices from her head, but the screen shows one at a time.
Each grey blob is a made-up piece of a cell, and the pink one is the taste cell.
A program called a tracer follows it up, picking the piece nearest the last one.
Each pick adds a bead on the right, so turn the wheel to the top.
Wait… that top half isn’t the taste cell.
At slice 6 it stepped onto a neighbour, so let’s roll back and look closely.
By the dashed outline of the taste cell one slice down sit two pieces. Which is the taste cell?
This one, though farther: the cell bends, so its piece stretches sideways, leaving the teal neighbour nearer.
But the taste cell’s piece covers 31% of the outline, and the neighbour’s only 7%.
Slices are so thin a cell barely moves, so the piece overlapping the last one most is the same cell.
Tracers still make wrong joins, so people proofread them; this one is ringed at slice 6.
Split it, so the tracer follows by overlap from there.
Now the taste cell runs whole, and the neighbour is a cell of its own.
Tracers fail the other way too: a fold hides slice 8, so this one stops and the top becomes a second cell.
One cell split in two is as wrong as two glued into one, so a proofreader joins them.
Higher up, a new batch came in with five flagged joins: some its own branches, but some glued-on neighbours.
So split only the glued-on neighbours before the next batch arrives.
Stage 3 of 8
If two neurons touch, does the signal go both ways?
Now the cells are whole, let’s see where they really touch, inside her head.
Our three: the taste cell, the relay, then the mouth cell.
Look closer where they touch: three contacts here, six here.
If touching were enough, a relay pulse would run back as easily as forward.
So test it: this electrode starts a pulse in the relay. Send it.
The taste cell stays dark: the pulse dies at all three contacts.
But it crosses all six, so the mouth cell lights.
Why? Only the taste cell’s side has bubbles: vesicles, round a T-shaped bar.
The relay’s side is a thick patch with nothing to release.
So synapses like these pass the signal one way, shown by these arrows.
A map needs only direction and count: the three fold into taste cell to relay, 3,
while the six fold into relay to mouth cell, 6.
So they fit a table: each row sends, each column receives.
The 3 goes in the taste cell’s row, under the relay.
But they touch at three spots: what goes in the relay’s row, under the taste cell?
Zero, because the relay’s pulse died at those three contacts.
The 6 drops under the mouth cell, and every empty square gets 0.
Fold it, and the 3 lands on a 0: the table isn’t a mirror, because contacts aren’t.
Count every pair this way, and you have the connectome: who sends to whom, how many times.
A real one is just bigger: about 140,000 rows by as many columns.
Now you count: each contact comes turned, so the sender isn’t always on top.
Find the vesicles, then file each contact in its sender’s row.
Stage 4 of 8
Which fly is in the map?
Whose wiring is it, though? A connectome is read off one particular fly.
The famous one, FlyWire, is one adult female: 139,255 neurons, each glowing where it sits.
So the cells that move her legs should be in there somewhere.
Go on, tap her leg and search her map for them.
Not one. Her leg’s trace died in a dark thorax.
Those leg cells sit back there, in her nerve cord, a fly’s spinal cord.
But FlyWire stops at her neck, so her map holds only her brain.
For legs you need another fly: this adult male, mapped in 2026.
His map covers his brain and his nerve cord, over 166,000 neurons.
So the same search in him lights up his leg neurons.
And our taste-to-mouth route? Its three cells are back: taste cell, relay, mouth cell.
In a real fly they meet low in the brain, just above the mouth.
So which map holds this route: hers, his, or both?
Both. The route drops into the same spot in each fly’s brain.
That spot is the feeding zone, or subesophageal zone.
So for feeding, her brain-only map is enough: in 2024, a whole-brain model built on it switched on its sugar neurons.
But walking runs through the nerve cord, and only his map has one.
So the fly comes first. Now you choose: questions drop down a chute between them.
File each question with a map that holds its wiring. Still open: is holding the route enough to make a fly drink?
Stage 5 of 8
If the wiring connects taste to mouth, why doesn’t the fly drink?
If holding the route is enough, then a fly that tastes sugar should just drink.
So here’s a real one on a rig, with sugar by her mouth.
Inside her head, the route is three cells: a taste cell, a relay with a glass bead, and a mouth cell.
A wire runs to the scope, so we can watch them work.
Go on, touch the sugar to her mouth.
And… nothing. One taste fills the relay’s bead with charge, up to 0.6, but it drains away.
The relay only fires if its charge reaches this line, its threshold.
So one taste falls short. What gets it over the line? Make her drink.
Two tastes, close together: the second lands before the first drains, so they add up.
The relay flashes: a spike. It runs to the mouth cell, and she drinks.
But right after a spike, the bead turns grey, and a taste landing then adds nothing.
Now keep the wiring and turn this dial, so the bead drains six times faster: that’s the leak. Two close tastes: does she drink?
No. Watch the same two tastes: the first has nearly drained when the second lands, so the bead stops at 0.73.
The dim trace is the slow leak from before; the ghost, the drink it gave her.
Same map, same sugar, different fly. The wiring says where a signal can go; a rule the map doesn’t hold decides whether it goes.
The same wiring and the same two tastes make the fly drink under a slow leak and never fire the relay under a fast one.
Now use the rule: get her three drinks before the drop runs out or dries.
Stage 6 of 8
How does a spike become a step?
She drank, but only because the sugar came to her, pinned on the rig.
To walk to it, a model of her needs a body.
Here’s one, in a virtual arena, sugar ahead.
Her smell reading runs up this thread to her brain.
Her brain answers back through this card to her legs.
But the plug is out, so her brain hears only the first reading.
So press run, and see where she goes.
She never slowed: her brain held 71 mm, so it never heard she’d got closer.
Plug it back in, and a reading goes up every 0.4 s.
She slows, the card turns to stop, and she drinks.
Each pink dot is a reading that got through. That’s feedback.
Now she starts facing away, so her first readings light turn.
That card is the interface: brain signals in, walk, turn or stop out.
FlyWire stops at her neck, so the card drives a body model.
Ours is made up, but Eon’s real 2026 demo had the same parts.
The catch: between readings, she obeys the last one.
So turn the dial to every 0.8 s. Grey dots are the 0.4 s run.
So does she still stop at the drop?
No. Her dots land twice as far apart as the grey ones.
The last lands 10.7 mm short, still far enough for full speed.
Held 0.8 s, that carries her 15 mm, so the next reading finds her 4.3 mm past.
Fresh readings aren’t free: each runs the brain model once, and you have 16 runs.
Set the dial and press run. She’ll drink, and the body works, but will you know the model is right?
Stage 7 of 8
How would we know the model is right?
No: walking to sugar and drinking is what the model was built to do, so that alone can’t tell you.
So set it beside her: model A, on the screen, is wired like her: taste cell, relay, mouth cell.
Go on, give both flies sugar.
Both drank. Then model B switched on and drank too, though an extra wire skips its relay.
So sugar can’t tell A from B: we have to change the fly.
Flip the relay’s switch: a clamp closes on it, in the models and in her. That’s silencing.
A’s taste stops at the silent relay, but B’s takes the extra wire, so B drinks.
So they disagree: whatever she does, one model is wrong.
So with her relay silenced, what will she do?
She tastes it, but it stops at her silent relay, so she stays dry.
That matches A, so B is ruled out, and A earns a tick.
But A isn’t safe: two more models, C and D, match her record too.
Your bench: this lamp fires the relay without sugar, and the screen shows each model’s answer free.
Each real run costs two of your four lab days, so only run where models disagree.
Rule out the rivals, then back the one model left standing.
Silencing the mouth cell rules out C, and lighting the relay rules out D.
So A passed four tests that could have ruled it out: support, not proof.
None of them tested learning, hunger, walking or feeling.
But this is how a model earns trust: it says what a real fly will do before anyone checks.
Stage 8 of 8
What did the map give us, and what did we add?
That was the last experiment, so let’s walk back, starting with the slices, where you follow the overlap, not the nearest.
Then synapses: at these, the signal crosses one way, so the relay back to the taste cell stays 0.
And that table comes from one fly, so check which fly a map is, and which part it covers.
But the scan couldn’t say if the relay fires. With a fast leak, two tastes reached only 0.73, so she didn’t drink.
Then, to walk, the model needs a body and fresh readings, because on a stale one she walked past the drop.
And to trust it, you change the fly. Silence the relay, and model B still drinks, but the real fly stays dry.
So that’s the whole trick: a map from one fly, then a firing rule, a body, and a test that could have failed.
Is the fly on that screen her, then, running on a computer?
No. The poppy seed gave us the map.
Next time you read that a brain was uploaded, ask which parts came from the scan, and which parts someone had to build.
Now break it yourself. Silence a cell, cut a connection or change the tastes, and see if she still drinks.