Abstract

Electron microscopy (EM) has sufficient resolution to map neural circuits at the level of individual neurons and the chemical synapses between them. Ongoing advances in EM imaging hardware and software have permitted increasingly large volumes of neural tissue to be imaged, using a variety of EM-based methods. We recently developed a second-generation transmission EM camera array (TEMCA2), capable of acquiring high quality images with sustained throughput of _{50 MPix/s at 4×4 nm/pixel, as well as an Autoloader for hands-free sample exchange, enabling days to weeks of automated imaging without manual intervention. We used this infrastructure to image the complete brain of a female adult fruit fly Drosophila melanogaster. The resulting dataset comprises 21 million images occupying 106 TB on disk. We also developed a downstream cluster-backed image processing pipeline to stitch, register, and intensity correct these images, enabling manual tracing of neuronal connectivity spanning the entire fly brain.}

We focused our pilot tracing efforts on the interface between the olfactory system and the mushroom body, the site of associative learning in the fly. The mushroom body contains 2,000 Kenyon cells (KCs) on each side of the brain, which receive a large olfactory input from the antennal lobe via second-order olfactory projection neurons (PNs). KC dendrites receive input from PNs in the calyx of the mushroom body in what is thought to be a random fashion. KC axons then project anteriorly in a bundle called the pedunculus, to the lobes of the mushroom body where synaptic modulation underlying memory formation occurs. We reconstructed ~10% of the KCs in the calyx and their presynaptic PN inputs to generate a PN-to-KC connectivity graph.

We find that KCs that fasciculate with one another in the pedunculus are much more likely to receive input from a common PN, and are more likely to make axo-axonic synapses with one another in the pedunculus. Since KCs with shared PN input are expected to have more correlated spiking activity, the network structure we describe here could result in the pooling of olfactory information between more highly correlated KCs, prior to the arrival of that information in downstream circuits for associative learning.