FluoEM: Virtual labeling of axons in 3-dimensional electron microscopy data for long-range connectomics

Abstract

Volume electron microscopy (3D EM) has enabled the dense reconstruction of neuronal circuits in datasets that are so far about a few hundred micrometers in extent. In mammalian brains, most neuronal circuits are however highly non-local, such that a large fraction of the synapses in such a volume of neuropil originates from distant projection sources. The labeling and identification of such long-range axonal inputs from multiple sources within a densely reconstructed EM dataset has been notoriously difficult because of the limited color label space of EM. Here, we present FluoEM, a set of experimental and computational methods that allows the identification of multi-color fluorescently labeled axons in dense EM data without the need for artificially introduced fiducial marks or direct label conversion for EM. The approach is based on correlated imaging of the tissue and computational matching of neurite reconstructions, amounting to a virtual color labeling of axons in dense EM circuit data. We show that the identification of fluorescent light-microscopically (LM) imaged axons in 3D EM data from mouse cortex is faithfully possible as soon as the EM dataset is about 40-50 m in extent, relying on the unique trajectories of axons in dense mammalian neuropil. The method is exemplified for the identification of longdistance axonal input into layer 1 of the mouse cerebral cortex.