Connectomics

Connectomics is the mapping of the complete set of connections in a nervous system. The term covers two very different enterprises that share a name: macroscale connectomics, which uses diffusion MRI to trace fibre bundles between brain regions in living people, and synapse-resolution connectomics, which uses electron microscopy on fixed tissue to identify every neuron and every synapse in a volume. The second is the one relevant to Whole brain emulation and to circuit neuroscience, and it is destructive, slow, and — for the first time in the mid-2020s — producing complete brains.

The word was proposed in 2005, by analogy with the genome, for a comprehensive structural description of a nervous system.¹ The practice is much older: the complete wiring diagram of Caenorhabditis elegans, reconstructed by hand from electron micrographs over more than a decade, was published in 1986 and remains the only nervous system whose full anatomy has been known for long enough to test what such knowledge is worth.²

§Scales and methods

Macroscale connectomes derive from diffusion-weighted MRI, which infers the orientation of white matter tracts from the directional diffusion of water. Resolution is on the order of a millimetre, tractography algorithms are prone to false positives at fibre crossings, and the output is a region-by-region matrix rather than a circuit. The NIH Human Connectome Project, launched in 2010, produced the reference datasets for this scale and made population-level comparison of connectivity routine. Macroscale maps are clinically useful — they guide electrode placement in Deep brain stimulation and inform hypotheses about psychiatric disorders as disorders of connectivity — but they are three orders of magnitude too coarse to identify a synapse.

Synapse-resolution connectomics works from tissue that has been chemically fixed, stained with osmium and other heavy metals, embedded in resin, and sectioned or milled. Three imaging strategies dominate: serial-section transmission electron microscopy, in which ultrathin sections are collected on tape or grids and imaged in parallel; focused ion beam scanning electron microscopy, which mills away the block face between images and gives isotropic resolution at lower throughput; and serial block-face scanning electron microscopy using a diamond knife inside the microscope chamber. All require in-plane resolution of a few nanometres to resolve synaptic vesicles and the ~20 nm gap between membranes.

Reconstruction, not imaging, was long the bottleneck. Tracing neurites by hand through thousands of sections is prohibitively slow; the field became tractable when convolutional segmentation models, notably the flood-filling networks developed at Google, reduced the error rate enough that human effort shifted from tracing to proofreading.³ Even so, the adult fly connectome required a large community of proofreaders working over years.

§Development history

1. 1986
   C. elegans
   White, Southgate, Thomson and Brenner publish the nematode's complete nervous system: 302 neurons and on the order of 7,000 connections, reconstructed by hand from electron micrographs over more than a decade.
2. 2005
   The word
   Sporns, Tononi and Kötter propose 'connectome' for a comprehensive structural description of a nervous system, by analogy with the genome.
3. 2010
   Human Connectome Project
   The NIH funds large-scale diffusion and functional MRI mapping in healthy adults, establishing macroscale connectomics as a standard method.
4. 2018
   Automated segmentation matures
   Flood-filling networks push automated neurite tracing to error rates low enough that reconstruction of large volumes becomes practical.
5. 2020
   Fly hemibrain
   A Janelia Research Campus team releases a dense reconstruction of about 25,000 neurons in the central brain of Drosophila, with synaptic connectivity.
6. 2023
   Larval fly brain
   Winding and colleagues publish the complete synaptic connectome of a Drosophila larva, roughly 3,000 neurons — the first whole brain of an insect.
7. 2024
   Human cortical millimetre
   Shapson-Coe, Lichtman, Jain and colleagues release H01, a cubic millimetre of human temporal cortex containing tens of thousands of cells and around 150 million synapses in about 1.4 petabytes of imagery.
8. 2024
   Adult fly connectome
   The FlyWire consortium publishes the wiring diagram of an entire adult fruit fly brain, roughly 140,000 neurons, together with cell-type annotation.
9. 2025
   MICrONS mouse cortex
   A cubic millimetre of mouse visual cortex is published with structural reconstruction co-registered to prior functional imaging of the same neurons.

§What has been mapped

Complete connectomes exist for C. elegans — including both sexes and several developmental stages — for the Drosophila larva, whose roughly 3,000 neurons were mapped in 2023,⁴ and for the adult fruit fly. The fly is the important milestone: an animal with courtship, learning, navigation, and flight, whose full synaptic wiring is now a downloadable dataset.⁵ Partial but dense reconstructions exist for larval zebrafish and for the mouse and human cortical volumes noted above; the human sample, a cubic millimetre of temporal cortex, occupies about 1.4 petabytes and contains on the order of 150 million synapses.⁶

For mammals the unit of achievement is still the cubic millimetre. The MICrONS mouse dataset is the most informative because the same neurons were imaged functionally, using calcium indicators, before the tissue was fixed and sectioned; it is therefore possible to ask how a cell's tuning relates to its inputs. That combination, not the wiring alone, is what circuit neuroscience wanted.

§Throughput, cost, and the scaling problem

The human-brain arithmetic is unforgiving. A cubic millimetre at electron-microscope resolution generates on the order of a petabyte; a human brain is roughly a million cubic millimetres. Linear extrapolation gives a dataset near a zettabyte, comparable to a substantial fraction of global data storage, before any of it is segmented. Imaging time scales similarly. Progress on this is real — multi-beam electron microscopes, faster milling, better compression, and cheaper segmentation have each moved throughput by large factors — but the gap is six orders of magnitude, and no announced instrument closes it.

Alternative strategies aim to sidestep the imaging cost entirely. Sequencing-based approaches developed in Anthony Zador's group tag individual neurons with random RNA barcodes and read out projections by sequencing rather than imaging, trading synapse-level detail for throughput. George Church and collaborators have proposed related molecularly annotated schemes. Synchrotron X-ray tomography can survey large volumes of stained tissue at coarser resolution without sectioning them, which is useful for targeting where to slice. None of these yields a synapse-level map of a mammalian brain.

§Why a wiring diagram is not a mind

The connectome is a static structural description. It does not carry synaptic sign, strength, or short-term dynamics; it does not record which neuromodulators a cell responds to; it does not capture extrasynaptic signalling, gap-junction coupling strength, or gene expression. Cornelia Bargmann and Eve Marder made the argument early and directly: the same anatomy supports many functional configurations, and knowing the anatomy constrains but does not determine the computation.⁷ This is the same underdetermination that separates a connectome from a theory of the Neural correlates of consciousness, and it is why possession of a wiring diagram does not by itself decide questions about Substrate independence or Machine consciousness.

The nematode is the standing test of what a wiring diagram is worth, and the test has not been passed. Its 302 neurons have been mapped since 1986, and four decades of effort have not produced a simulation that reproduces the animal's behavioural repertoire — the missing quantities being exactly synaptic sign and strength and the extrasynaptic neuropeptide signalling that reconfigures the circuit. No connectome of any animal has yet been used to reproduce that animal's behaviour.

A widely cited demonstration of the limit comes from outside biology. Eric Jonas and Konrad Kording applied standard neuroscience analyses, including a complete connectome, to a 6502 microprocessor running video games and found that the methods failed to recover how the chip works.⁸ The point is not that connectomes are useless — they are the substrate on which every circuit hypothesis is built — but that structure underdetermines function.

One brain, one time point

Every published whole-animal connectome is a snapshot. The fly datasets each come from a single individual fixed at a single moment; the nematode reconstruction was assembled from sections of more than one animal, so it is a composite rather than any one worm's wiring. Variation between individuals of the same species is substantial in mammals and non-trivial even in flies, and synapses turn over on timescales of days. A connectome is a snapshot, not a constant.

§Applications

Circuit neuroscience is the immediate beneficiary. Fly connectome data has already been used to predict responses of unrecorded neurons and to identify circuits for specific behaviours, with predictions tested by Optogenetics and electrophysiology. In mammals, dense reconstruction has revealed connectivity rules — which cell types contact which, and how selectively — that sparse recording could not establish. Structural priors of this kind feed back into applied work: population models used in Neural decoding and in Brain–computer interface research, and circuit models of the hippocampus underlying attempts at a Memory prosthesis, all depend on assumptions about connectivity that dense reconstruction can now check. Connectomic methods have begun to be applied to cortical Organoids as well, where the question is how far self-organized tissue reproduces the wiring statistics of a real cortex.

Connectomics also underwrites the argument for Brain preservation: if the connectome plus molecular annotation is the information that matters, then a preservation method that demonstrably retains it converts death into a storage problem. Whether that "if" is true is exactly what the Bargmann–Marder objection challenges, and it bears directly on Cryonics, on Mind uploading, and on what would have to survive for the preserved person to survive with it, a question treated under Personal identity and continuity. Nothing in current connectomics settles which level of detail an emulation would need.

§Outlook

The stated near-term targets are a whole mouse brain at synapse resolution — an effort several groups and funders, including the US BRAIN Initiative, have framed as the next milestone — and larger human cortical volumes with molecular labelling. A mouse brain is roughly a thousand times the volume already reconstructed and would be the first mammalian connectome in full. Whether it arrives in the 2030s depends less on microscopes than on segmentation cost and on whether funders sustain a project whose output is a dataset rather than a discovery.

§See also

• Whole brain emulation
• Mind uploading
• Brain preservation
• Neural decoding
• Neural correlates of consciousness
• Optogenetics
• Organoids
• George Church

§References