The dominant assumption in connectomics has been that electron microscopy is the only path to synaptic resolution. Resolving the nanoscale structures of individual synapses — the 20–40 nanometer cleft between pre- and postsynaptic membranes — requires the resolving power of an electron beam. Light microscopy, bounded by the diffraction limit of visible wavelengths, cannot reach that scale. This constraint has shaped the entire field: electron microscopes are massive, expensive, operationally demanding instruments. The result is that high-resolution connectomics has been confined to a handful of specialized facilities worldwide.

A paper published in Nature in 2026 by a collaborative research team introduces LICONN — Light-microscopy-based Connectomic reconstruction — a method that achieves electron-microscopy-equivalent synaptic resolution on standard light microscopes available in any modern neuroscience laboratory. The technique combines engineered hydrogel embedding, physical tissue expansion, and deep-learning segmentation to circumvent the diffraction limit without electron beams.

The implications for whole brain emulation are significant. The physics wall that makes non-destructive brain scanning so difficult has always had two components: resolution and access. LICONN does not solve the living-tissue problem — the tissue must be fixed and processed — but it removes the instrumentation bottleneck entirely. Any laboratory that can prepare brain tissue samples can now, in principle, produce connectome-quality data.

The Technical Problem LICONN Solves

Light travels in waves. When two structures are closer than roughly half the wavelength of the light being used, a conventional microscope cannot resolve them as separate objects. This is the Abbe diffraction limit. For visible light, it places a hard floor of approximately 200 nanometers on lateral resolution and 500 nanometers on axial resolution.

Synapses are smaller than this. The synaptic cleft itself is 20–40 nanometers. The dendritic spines that house postsynaptic densities are typically 100–500 nanometers in width at their narrowest. While the spine heads can sometimes be resolved with super-resolution techniques, the synaptic contacts themselves, the actual points of signal transmission, cannot be reliably mapped at light-microscopy resolution in intact tissue.

Electron microscopy bypasses this limit because electrons have wavelengths thousands of times shorter than visible light. A scanning or transmission electron microscope can resolve structures to 0.1–1 nanometer, well below the scale of individual synaptic proteins. This is why all nanoscale-resolution connectome reconstructions to date have required electron microscopes, from the C. elegans wiring diagram in the 1980s through the FlyWire fruit fly connectome and the Shapson-Coe petavoxel human cortex fragment published in 2024.

How LICONN Works

LICONN’s core innovation is physical expansion. Brain tissue is chemically fixed, then embedded in a hydrogel polymer network. When the hydrogel is exposed to water, it expands isotropically — uniformly in all three dimensions — magnifying the tissue by a factor of approximately 10 in each dimension. A structure that was 50 nanometers wide is now 500 nanometers wide, within the resolution range of a standard confocal microscope.

The expansion step is not new. Expansion microscopy was pioneered by Ed Boyden’s group at MIT and has been used for several years to improve resolution in fluorescence imaging. What LICONN adds is an engineered hydrogel formulation optimized specifically for connectomic reconstruction. The standard expansion protocols were developed for visualizing labeled proteins and structures; they cause subtle distortions in the extracellular matrix and lipid membranes that are acceptable for protein imaging but disqualifying for membrane tracing at synaptic precision.

The LICONN team redesigned the hydrogel chemistry to preserve membrane ultrastructure through the expansion process. The result is an expanded tissue sample in which cell membranes, organelles, and synaptic contacts are expanded but undistorted. The morphology of each structure is preserved at sufficient fidelity that synaptic connections can be identified and traced.

After expansion, the tissue is imaged on a standard confocal or light-sheet fluorescence microscope. The expanded volume produces raw image stacks that look, at the relevant scale, similar to electron micrograph stacks. These are then processed by a deep-learning segmentation pipeline. The neural network, trained on ground-truth EM data from the same tissue types, segments individual neurons and identifies synaptic contacts from their morphological signatures in the expanded light microscopy images.

Accuracy Against Electron Microscopy Ground Truth

The critical validation question is whether LICONN actually matches EM resolution, or merely approaches it. The Nature paper reports direct comparison experiments in which the same tissue volumes were mapped by both LICONN and serial-section EM. The comparison examined synaptic connection identification: which presynaptic terminals contact which postsynaptic structures, and where.

The match was not perfect — no biological imaging method is — but the discrepancy rates were within the range of EM-to-EM variability when the same volume is processed by different teams or protocols. Specifically, the false positive rate (LICONN identifying a synapse that EM does not confirm) and false negative rate (LICONN missing a synapse that EM identifies) were both below 5% in the test volumes, which is comparable to the disagreement rates between expert human annotators working from EM stacks.

This is the threshold that matters for connectomics. A 5% error rate per synapse is not zero, but it is low enough that statistical properties of circuits — which neurons are connected, how densely, with what approximate weight distributions — can be reliably extracted. For whole brain emulation, where the goal is a functionally accurate model rather than a structurally perfect replica, connection topology at this accuracy level is likely sufficient.

Cost and Throughput

The economics of LICONN are different in kind from EM-based connectomics, not just in degree. EM instruments cost $500,000 to several million dollars, require specialized maintenance, and are typically operated by facilities that serve multiple research groups on a shared basis. A single petavoxel EM acquisition run can take months. Storage costs for petabyte-scale image data are substantial.

LICONN runs on confocal and light-sheet microscopes that are standard equipment in thousands of neuroscience laboratories around the world. The hydrogel chemistry uses commercially available reagents. The deep-learning segmentation runs on GPU clusters comparable to those already used for imaging analysis in computational neuroscience labs.

The paper estimates throughput improvements of roughly an order of magnitude over EM-based connectomics at equivalent volume and resolution. The cost reduction is harder to quantify precisely because it depends heavily on labor and institutional context, but the authors describe LICONN as opening connectome-quality mapping to “any well-equipped neuroscience laboratory.”

This matters for the field because it decouples connectomic capability from access to specialized facilities. AI-guided segmentation has already reduced the proofreading bottleneck, shifting the rate-limiting step from annotation labor to acquisition throughput. LICONN shifts acquisition throughput from a specialized EM constraint to a general light microscopy problem. Both shifts together represent a substantial compression of the practical timeline for large-scale mammalian connectomics.

Relationship to Other Recent Advances

LICONN is one of several approaches that have emerged in 2025 and 2026 to address different components of the connectomics bottleneck. Connectome-seq — published in Nature Methods in April 2026 — uses molecular barcoding and sequencing rather than imaging to map synaptic connectivity. It achieves 10-fold cost reduction over EM for circuit-level topology, with a different trade-off profile: connectome-seq captures connectivity statistics efficiently but does not produce spatial images of the tissue. LICONN produces spatial images. The two methods address different downstream use cases.

SmartEM demonstrated AI-guided acquisition that accelerates EM itself by directing the electron beam to informative regions and skipping uninformative background. SmartEM remains an EM-based method, but reduces the raw scan time substantially. LICONN eliminates the EM instrument requirement entirely for resolution-limited applications.

Together, these three approaches represent a diversification of the connectomics toolkit. No single method is superior across all dimensions. For preserved tissue where spatial context matters, LICONN or SmartEM are appropriate. For large-scale circuit topology at low cost, connectome-seq offers an alternative route.

What LICONN Does Not Do

Several constraints on the method are worth stating directly.

First, LICONN requires fixed tissue. The tissue must be chemically preserved before expansion and imaging. This means LICONN cannot image living brains. It does not solve the fundamental problem that non-destructive in vivo synaptic-resolution imaging of a functioning brain remains physically impossible with current or near-future optical technology.

Second, the current implementation has been demonstrated at the scale of small tissue volumes — the validation data involves tissue blocks of the type used in typical EM connectomics studies, roughly cubic millimeters. Scaling LICONN to whole-brain volumes of a mouse (500 cubic millimeters) or rat (2,000 cubic millimeters) requires parallelization of both the imaging and the segmentation pipeline. The method does not in itself solve the scaling problem; it shifts the bottleneck to throughput rather than resolution.

Third, the expanded tissue cannot be re-used after LICONN processing. The embedding and expansion steps are destructive in the sense that the tissue cannot be recovered in its original state. For postmortem human brains, this means the LICONN sample and the tissue are consumed in the process.

Implications for Whole Brain Emulation

For the whole brain emulation roadmap, LICONN’s most significant contribution is accessibility. The institutional barriers to connectomic research have been a real constraint on how quickly the field can scale. Requiring specialized EM facilities means that only a few groups globally can produce high-resolution connectome data. Most neuroscience laboratories, even well-funded ones, cannot participate in data collection.

LICONN changes this. A group studying a specific brain region, or a specific disease model, can now generate connectome-quality data on their local equipment. This is likely to accelerate both data volume and scientific diversity in connectomics over the next several years — more groups, more regions, more species, more conditions.

The cross-species connectome transfer work demonstrated that connectivity data from one species can be used to control behavior in a virtual body of another species. The value of that line of work depends on having accurate connectivity data to transfer. LICONN makes accurate connectivity data easier to obtain across more species and more brain regions.

For human brain mapping specifically, the relevant near-term application is not whole-brain human connectomics — that remains a project of decades even with LICONN — but postmortem tissue research. If LICONN can reliably process preserved human brain samples, it opens a path to generating connectivity data from individual human brains without requiring EM access. Combined with advances in brain preservation protocols that maintain ultrastructural integrity through the postmortem window, this creates a more accessible pipeline for human connectome research.

TRL Assessment: TRL 3–4. The method is validated in laboratory conditions at the tissue level. Scaling to whole-brain volumes and human tissue applications requires further engineering development.

Official Sources

• Buchholz, T.-O. et al. (2026) — LICONN: Light-microscopy-based connectomic reconstruction at synaptic resolution. Nature. DOI: 10.1038/s41586-025-08985-1. https://www.nature.com/articles/s41586-025-08985-1
• Chen, F., Tillberg, P.W., & Boyden, E.S. (2015) — Expansion Microscopy. Science, 347(6221):543–548. DOI: 10.1126/science.1260088
• Shapson-Coe, A. et al. (2024) — A petavoxel fragment of human cerebral cortex reconstructed at nanoscale resolution. Science, 384(6696):eadk4858. DOI: 10.1126/science.adk4858
• Scheffer, L.K. et al. (2020) — A connectome and analysis of the adult Drosophila central brain. eLife, 9:e57443. DOI: 10.7554/eLife.57443
• Lichtman, J.W. & Denk, W. (2011) — The big and the small: Challenges of imaging the brain’s circuits. Science, 334(6056):618–623. DOI: 10.1126/science.1209168