Link to the code: brain-emulation GitHub repository

Connectome-seq: Mapping Synapses With Barcodes Instead of Electron Beams

Electron microscopy has been the gold standard for connectomics since the field began. To map a synapse at nanoscale resolution — to definitively confirm which axon terminal contacts which dendritic spine — you need the resolving power of an electron beam. This is why the C. elegans connectome took over a decade to complete in the 1980s, why the Drosophila whole-brain connectome required years of effort from the FlyWire consortium, and why a complete human connectome remains one of science’s most distant technical horizons.

A study published in Nature Methods in April 2026 introduces a different strategy. Connectome-seq uses molecular barcodes — short DNA sequences uniquely assigned to each neuron — to map connectivity through sequencing rather than imaging. Instead of tracing axons through thousands of serial electron micrograph sections, the method captures synaptic fragments biochemically and reads out which barcodes are paired at each synapse. The result: single-synapse resolution connectivity maps produced at approximately one-tenth the cost of EM-based approaches.

The Barcoding Strategy

The core concept in connectome-seq is neuronal barcoding. Each neuron in the tissue of interest is given a unique DNA sequence — a short oligonucleotide barcode — through viral delivery of barcoded expression cassettes. The barcode integrates into the cell’s genome and is expressed continuously.

At synapses, pre- and postsynaptic membranes are in close contact. The connectome-seq protocol exploits this proximity: a biochemical proximity ligation step captures paired barcodes from neurons whose membranes are within synaptic distance of each other. This creates paired barcode molecules — short double-stranded DNA constructs in which one strand comes from the presynaptic neuron and one from the postsynaptic neuron.

These paired barcodes are then extracted and sequenced using high-throughput next-generation sequencing. Each sequencing read represents a synapse: the presynaptic neuron’s barcode paired with the postsynaptic neuron’s barcode. Aggregating millions of reads across a tissue sample produces a connectivity matrix — which neurons connect to which, with estimated synapse count as a proxy for connection strength.

The method was demonstrated on mouse hippocampal circuits, mapping over 1,000 individually barcoded neurons at single-synapse resolution. The resulting connectivity graph revealed previously unknown direct synaptic connections between cell types thought to be connected only through intermediary neurons.

Cost and Scale Advantages Over Electron Microscopy

The cost comparison is stark. Current EM-based connectomics of a 1mm³ mouse cortex tissue block — a small volume containing roughly 100,000 neurons — requires large-format electron microscopy acquisition, petabyte-scale image storage, and months of AI-assisted proofreading. The estimated cost runs into the millions of dollars per cubic millimeter at current production rates.

Connectome-seq’s cost scales with sequencing depth, not tissue volume or neuron count. As sequencing costs have fallen by orders of magnitude over the past decade (following Moore’s Law-equivalent trajectories for sequencing throughput), the cost per synapse in connectome-seq decreases accordingly. The April 2026 implementation achieves approximately 10-fold lower cost per neuron compared to EM-based methods at equivalent scale.

Sensitivity is also improved. The MAPseq2 enhancement — a concurrent development published in the same timeframe — increased sensitivity by roughly 10-fold over the previous sequencing-based approach, meaning the method captures a higher fraction of true synapses rather than missing weakly expressed connections.

What It Found in Mouse Hippocampus

The study’s biological findings are as significant as its methodological contribution. In the mouse hippocampal CA1 circuit — one of the most studied neural circuits in neuroscience — connectome-seq identified direct synaptic connections between excitatory pyramidal neurons and a specific class of inhibitory interneuron that prior anatomical studies had not detected. These connections had been invisible to standard fluorescence microscopy because the synapses were too sparse and small to resolve at light-microscopy resolution, and the circuit had not been prioritized for EM-based analysis.

The newly identified connections alter the predicted dynamics of CA1 inhibitory circuits. Computational models that did not include these connections had underestimated the feedforward inhibition onto a subset of pyramidal neurons during theta-rhythm oscillations. With the new connectivity data, the models better reproduce the experimentally measured firing patterns in awake-behaving mice.

This is a concrete example of why synapse-level resolution matters: circuit-level models built on incomplete connectivity data make incorrect functional predictions, even when the models are otherwise well-constructed.

Implications for Human Connectome Mapping

The human brain contains approximately 86 billion neurons and an estimated 100 trillion synapses. Mapping it at single-synapse resolution by EM would require petabytes of image data and thousands of human-years of proofreading labor, even with AI assistance. The AI segmentation advances covered earlier have dramatically reduced the proofreading burden, but the imaging itself remains a fundamental bottleneck.

Connectome-seq does not eliminate the bottleneck. Current implementations require viral barcoding of target tissue, which for human application would require either postmortem delivery (limiting the window between death and tissue processing) or in vivo viral delivery prior to death (which introduces ethical and safety considerations). The proximity ligation step also requires careful biochemical optimization for different tissue types and ages.

But the method opens a path that did not previously exist: a sequencing-based route to connectome mapping that scales with molecular biology rather than with imaging infrastructure. As viral delivery methods improve and sequencing throughput continues to increase, the feasibility of connectome-seq for larger mammalian brains — and eventually for preserved human brain tissue — will improve correspondingly.

The preservation work from Song et al. (bioRxiv March 2026) demonstrated that ultrastructural preservation of large mammal brains within a 14-minute post-mortem window is achievable. If postmortem connectome-seq can be reliably performed on preserved tissue, the two methods together could define a practical pathway for whole-brain human connectome mapping that is not dependent on EM imaging scale-up.

Relationship to Sequencing-Based Connectomics History

Connectome-seq builds on MAPseq (Multiplexed Analysis of Projections by Sequencing), developed by Bhavan Bhatt and colleagues at Cold Spring Harbor Laboratory, which used barcoded viruses to map long-range axonal projections rather than local synaptic connections. MAPseq established that sequencing-based neuronal tracing was feasible but was limited to projection mapping (which neuron projects where) rather than synapse-level local connectivity.

The April 2026 study extends this to local synaptic circuits by adding the proximity ligation step, which captures paired barcodes at actual synaptic contacts rather than at axonal termination zones. This is the technical advance that enables single-synapse resolution.

The lineage from MAPseq through connectome-seq illustrates how molecular biology-based approaches to connectomics are maturing alongside EM-based methods. The two approaches have complementary strengths: EM captures spatial structure, protein composition, and organelle-level detail at each synapse; connectome-seq captures connectivity topology at scale and cost. A complete picture of brain wiring will likely require both.

Official Sources