MouseConnects Complete Hippocampal Connectome
A synapse-level reconstruction of 10 mm³ of the mouse hippocampal formation, targeting the circuits of memory and spatial navigation.
Scale
MouseConnects targets the mouse hippocampal formation, the brain region essential for memory formation and spatial navigation. At 10 mm³ — roughly ten times the volume of MICrONS (1 mm³) — and more than 10 petabytes of expected raw imagery, it aims to produce the first synaptic wiring diagram of a mammalian memory circuit. The case study covers the scientific rationale and pipeline in depth.
Key Methods
- High-throughput EM acquisition: Automated serial-section electron microscopy, with multi-beam SEM and automated section handling to meet the timeline
- Distributed imaging: Two imaging sites (Harvard and Princeton) operating in parallel
- Automated reconstruction: Flood-filling network segmentation developed at Google Research, the same core technology used for FlyWire
- Open science: Data and tools to be released publicly as the project progresses
Scientific Questions Addressed
- How do hippocampal circuits encode spatial information and episodic memories?
- What are the detailed microcircuit motifs underlying grid and place cell function?
- How do different cell types contribute to memory formation and retrieval?
- What circuit principles enable the hippocampus to act as a cognitive map?
Method Stack
EM Imaging
Automated serial-section electron microscopy at synaptic resolution, building on the Lichtman lab's acquisition methods and advances in multi-beam SEM throughput.
- Two imaging sites: Harvard & Princeton
- Tape-based section collection
- Automated section handling
Cloud Processing
Cloud-based data processing on Google infrastructure for alignment, segmentation, and analysis of a dataset expected to exceed 10 PB.
- Flood-filling networks for neuron tracing
- Automated synapse detection
- CAVE-based collaborative proofreading
Multi-Modal Integration
Planned integration of EM data with Patch-seq recordings, fMOST morphology, and transcriptomic cell typing — project targets, not yet released data.
- 4,000+ Patch-seq recordings
- Whole-brain fMOST reconstructions
- Cross-modal cell type matching
Who Does What
Harvard University
Jeff Lichtman, PI
Sample preparation and imaging expertise with decades of connectomics experience
Princeton University
Sebastian Seung & David Tank
mSEM imaging technology and deep learning algorithms for automated reconstruction
Google Research
Viren Jain
Cloud-scale processing infrastructure and machine learning segmentation algorithms
MIT
Ila Fiete
Circuit analysis, computational modeling, and theoretical frameworks for spatial coding
Allen Institute
Hongkui Zeng
Patch-seq recordings, transcriptomic cell typing, and multi-modal data integration
University of Cambridge
Gregory Jefferis
Data integration, analysis pipelines, and connectome interpretation methods
Johns Hopkins APL
William Gray-Roncal
Connectome quality assurance, community training, and data dissemination
Data Not Yet Released
The MouseConnects dataset is in active development; data, browsers, and analysis tools will be released as the project progresses. Until then, the released public datasets are the place to practice, and the case study tracks what this project is building.