We Finally Have the Complete Wiring Diagram of a Brain. Here's What It Tells Us.
Imagine holding a map of every street, alley, and footpath in New York City — all 100,000-plus connections, down to the smallest synapse. Now imagine running traffic through that map to see where the jams form, where cars flow freely, and which intersections secretly control the whole system. That, in essence, is what neuroscientist Stavros Therianos has just done with a brain.
The brain in question belongs to a fruit fly larva — a tiny maggot just a few millimetres long, navigating its world in search of food. It doesn’t sound like much of a thinking machine. Yet its brain contains 3,013 neurons and 111,243 synaptic connections. In 2023, researchers completed the monumental task of mapping every single one of those connections using electron microscopy, producing the first complete brain wiring diagram — or connectome — of any insect. The question that immediately followed was deceptively simple: now that we can see the whole thing, what does the wiring actually tell us?
Background & Context
For decades, neuroscientists have debated how much of a brain’s activity is dictated by its wiring alone, as opposed to the subtler ingredients the wiring diagram doesn’t capture: the chemical identity of each synapse, the time constants of individual neurons, the wash of neuromodulators that changes from moment to moment. It’s an argument with real stakes. If the wiring largely determines function, then connectome projects — including the massive ongoing efforts to map mouse and human brain tissue — are not just descriptive exercises; they’re blueprints for understanding thought itself. If the wiring is more like a loose suggestion, then the connectome is less a map than a rough sketch.
The larval fruit fly offered the perfect test case. Small enough to model in its entirety, yet complex enough to contain identifiable circuits with known functions — including the mushroom body (the insect’s learning centre), the lateral horn (an olfactory hub), and the central complex (a navigation system) — this brain could finally answer a question that had been purely theoretical.
What the Researchers Did
Therianos took the connectome and turned it into a computational model, treating the entire brain as a frozen, rate-based neural network. Crucially, no parameters were tuned to fit any data. No single-neuron properties were fitted. The model simply took the reconstructed wiring — which neuron connects to which, and with how many synapses — and ran it. The goal was to isolate what the wiring diagram alone, stripped of all physiological detail, could produce.
What makes the study remarkable, however, is not just the model but the discipline of its controls. Therianos didn’t just compare the real connectome to a random network — that would only show it was non-random, which we already knew. Instead, he created a hierarchy of comparison networks. The critical one was an ensemble of a thousand rewired brains, each preserving every neuron’s number of connections (its in-degree and out-degree) and connection strengths, but scrambling where those connections actually went. If a property appeared in both the real connectome and these scrambled versions, it was driven by coarse statistics. If it appeared only in the real thing, it was driven by the precise placement of synapses. This ladder of null models became the study’s real power — it didn’t just report findings; it tested them to destruction.
What They Found
The results split cleanly into two categories, and the split itself is the headline.
First, the brain’s global dynamics — how strongly it amplifies signals, how many independent activity directions it preserves, how linear or nonlinear its processing is — were almost entirely statistical. The scrambled networks reproduced 96 to 98 percent of these properties. In other words, any brain with the same number of connections and the same connection strengths would behave roughly the same way at this macroscale. The coarse statistics set the regime.
Second, and far more interestingly, where neural activity actually travels is decided by the precise wiring. When the model was driven with sparse input — the equivalent of a faint scent entering the system — the real connectome confined the activity to just a fifth of the brain, channelling it into a compact olfactory pathway. The scrambled networks, by contrast, flooded nearly two-thirds of the brain with the same input. The exact placement of synapses acts as a gate, metering activity into functionally relevant circuits while keeping the rest of the brain quiet.
Then came the most striking finding. The mushroom body — the insect’s learning centre, famous for its role in associative memory — concentrates a disproportionate share of the brain’s leading dynamical modes. When you ask which neurons weigh most heavily in shaping the recurrent activity, the mushroom body dominates beyond anything its connection counts would predict. It was the only circuit that passed every statistical guard the study threw at it. The learning centre, it turns out, has a structural privilege written into the very geometry of its wiring.
Why It Matters
This separation — statistics for the regime, precise wiring for the routing — is more than an elegant result. It gives the neuroscience community a principled way to think about every connectome that follows. When the mouse connectome is completed, or when dense reconstructions of human cortex arrive, we’ll ask the same question: what does the wiring alone constrain, and what does it leave free? The methodology Therianos developed — the ladder of null models, the requirement that a circuit-level claim survive size-matched controls — provides a portable instrument for that inquiry.
The study also demonstrated what good science looks like when the instruments are powerful enough to be dangerous. Two initial findings — that the lateral horn carried a special dynamical role, and that the calyx’s sparse-coding expansion was wiring-specific — were retracted after the control hierarchy removed them. The lateral horn’s apparent importance vanished once compared against arbitrary neuron sets of the same size. The calyx expansion turned out to be generic to any sparse fan-out architecture, not a signature of the specific wiring. The retractions are not failures; they are what give the surviving results their credibility.
How It Could Change Our Lives
The most immediate impact is on how we build artificial neural networks. Anyone designing recurrent architectures — the kind used in time-series prediction, language processing, or robotic control — now has empirical evidence that coarse statistics (connection counts and weight distributions) determine the overall dynamical regime, while precise wiring governs routing and which subcircuits dominate. It means you can iterate on the broad structural parameters to set the scale of computation, but if you want to control where information flows, you need to care about the exact pattern of connections.
Longer term, the finding that the learning centre has a wiring-level structural privilege raises tantalising possibilities. If the mushroom body’s outsized role in the dynamics is a general principle — if learning centres across species concentrate influence through their wiring geometry — then understanding that geometry could inform everything from neuromorphic chip design to strategies for protecting cognitive function as brains age.
The Bigger Picture
This study sits at the threshold of a new era in neuroscience. For most of the field’s history, we studied brains like astronomers before the telescope — by inference, by lesion, by recording from a handful of neurons at a time. Complete connectomes change that. They let us ask structural questions at the scale of the whole system, not just the circuit. The larval fly brain, with its beautifully tractable size and its genuine behavioural complexity, is the first canvas on which these questions can be painted in full. It will not be the last.
Limitations & What’s Next
The model is deliberately stripped down — a frozen, rate-based operator with no action potentials, no time constants, no distinction between excitatory and inhibitory synapses, and no neuromodulation. It is not a simulation of a living fly larva, and Therianos is explicit about this. The findings describe what the wiring diagram alone can explain in a computational model, not what the animal’s brain is actually doing as it crawls toward a food source. The next step is to layer in the missing physiology — to ask whether the routing and mode-concentration patterns persist when the model becomes more biologically realistic. And the larger question — whether the same statistics-versus-geometry split holds in a mouse brain, a human cortical column, or any other connectome — is entirely open, waiting for the data and the discipline to answer it.
📄 Source: Therianos, S. (2026). Separating wiring-specific from statistical control of dynamics in a complete connectome. arXiv:2606.17745v1.