Joe Strout shared on 4 Nov 2018

WOW! Apologies if this was posted before and I missed it, but I am so psyched about this breakthrough! The tech they built to scan at this scale is impressive, and the data — freely available, of course — should be 90% of what we need to actually upload a fruit fly! We'll need to actually convert all those images to a wiring diagram, using something like EyeWire or Google's recent automated tracing system. And there's plenty of work to be done in modeling, validating against other data, etc. But it's a huge step forward.

Introduction

How brain circuits allow animals to implement complex behavior remains a central mystery of neurobiology. Where available, neuronal “wiring diagrams” or “connectomes” (Lichtman and Sanes, 2008)—maps of the synaptic connectivity between the neurons in a circuit—have proved extremely useful for understanding circuit function (Ding et al., 2016, Jarrell et al., 2012, Kasthuri et al., 2015, Ohyama et al., 2015, Takemura et al., 2017b, Wanner et al., 2016). However, many neuronal circuits are brain-spanning, and access to whole-brain connectomes has been limited to a few small organisms, such as the nematode C. elegans, the larva of the fruit (or vinegar) fly, Drosophila melanogaster, and the tadpole larva of the tunicate Ciona intestinalis (Ohyama et al., 2015, Ryan et al., 2016, White et al., 1986).

The adult fruit fly has emerged as a key genetic model system for interrogating the neuronal substrates of sophisticated behaviors, such as place learning, flight control, courtship, grooming, and memory-driven action selection (Dickinson and Muijres, 2016, Hampel et al., 2015, Ofstad et al., 2011, Owald and Waddell, 2015, Pavlou and Goodwin, 2013). Given the morphological and physiological stereotypy and genetic accessibility of neuronal cell types in the fly brain, connectomes of circuits underlying these behaviors should translate well across individuals and significantly accelerate the dissection of the neuronal basis for behavior (Gruntman et al., 2018, Jovanic et al., 2016, Ohyama et al., 2015, Takemura et al., 2017a, Takemura et al., 2017b). However, at ∼8 × 107 μm3 and ∼100,000 neurons (Simpson, 2009), the brain of an adult fly is two orders of magnitude larger than that of the fruit fly larva, the next-largest brain imaged at synaptic resolution (Ohyama et al., 2015). This combination of scale and resolution has heretofore been unattainable by volume electron microscopy (EM), the only method capable of simultaneously resolving all neuronal branches and synapses in a given volume of brain tissue (Helmstaedter et al., 2008). Therefore, we built new hardware and software for high-speed acquisition and processing of serial section transmission EM (TEM) images and used this infrastructure to image a whole-fly brain at synaptic resolution (Figure 1).

We took multiple approaches to validate our whole-brain volume for tracing synaptic connectivity of brain-spanning neuronal circuits. Our efforts focused on the mushroom body (MB), which has been intensively studied for its role in associative memory formation and recall (Guven-Ozkan and Davis, 2014, Keene and Waddell, 2007). Olfactory projection neurons (PNs) provide the main sensory input to the MB; their connections with the MB intrinsic neurons, Kenyon cells (KCs), form a crucial stage of a fan-out fan-in network analogous to brain structures including the mammalian cerebellum (Farris, 2011, Stevens, 2015). Furthermore the logic of PN to KC connectivity ratios has been the subject of detailed experimental and theoretical analysis as a model for the construction of high dimensional sensory representations (e.g., Caron et al., 2013, Litwin-Kumar et al., 2017). KCs and PNs are brain spanning, morphologically stereotyped, and anatomically extremely well described at the light level (Aso et al., 2014, Jefferis et al., 2007, Tanaka et al., 2012), making them well suited for validating the accuracy of neural reconstructions in the volume. We have developed software tools to enable co-visualization, quantitative analysis, and rapid cell type identification by merging EM reconstructions with existing large-scale light microscopy (LM) databases of neuronal morphology (Chiang et al., 2011, Costa et al., 2016, Milyaev et al., 2012).

Independent tracing of Kenyon cell dendrites, which have some of the finest neurites in the fly brain (Yasuyama et al., 2002), provided a sensitive test of the consistency of neural reconstructions. Retrograde tracing from KC dendrites provided a complete enumeration of olfactory PN input to the MB and yielded an improved map of local circuitry in the calyx, the initial site for sampling and processing of sensory information in the MB. This revealed principles of coordinated organization that were invisible in previous work using light level data assembled from many different brains; for example we found a high degree of clustering of PN inputs, which may generate biases in PN-to-KC connectivity and therefore shape how olfactory PN input to the MB is sampled. Although the MB has been intensively studied, we also discovered a previously unknown, brain-spanning neuron that provides input to KCs and likely relays non-olfactory, multimodal information to the calyx. Finally, we show that KC dendrites make output synapses onto a small subset of available cell types, defining a specific local recurrent microcircuit in the calyx.

In conclusion, we describe the largest synaptic-resolution, whole-brain EM dataset obtained to date. We show that it enables efficient mapping and identification of both known and unknown neurons in adult Drosophila, a key model system for circuit neuroscience and, crucially, that it enables reliable and efficient determination of synaptic connectivity. We have made these data and supporting software freely available for download and immediate use by the scientific community.

Results

New Tools for Volume EM Data Acquisition

To meet the challenge of acquiring a whole fly brain, we used a variation of classical serial section TEM (ssTEM), in which images are acquired at high-speed with a TEM camera array (TEMCA) (Bock et al., 2011). Although sample handling for TEM is challenging (Figures S3A–S3D), in comparison to scanning EM-based methods, the intrinsically parallel nature of the electron optical image formed in TEM makes it relatively straightforward to achieve high-quality EM images at high-speed (reviewed in Briggman and Bock, 2012).

We built two second-generation TEMCA (TEMCA2) systems (Figures S2A and S3E), using high-speed sCMOS cameras and made a custom piezo-driven Fast Stage (Figures S2B, S3F, and S3G) (Price and Bock, 2016a). Image mosaics were needed because at typical magnification, each camera in the array had an ∼8 μm field of view (FOV), whereas each whole-brain thin section was ∼750-μm wide × ∼350-μm tall (Figure 1A). At 4 nm/pixel, this resulted in a large (∼187,500 × 87,500 pixel, ∼16 GB) stitched image mosaic for each section. Each FOV was typically acquired using four 35 ms frames, versus ∼1 s in conventional TEM imaging systems. The Fast Stage moved one FOV in 30–50 ms (including settle time; Figures S2C and S2D; Video S1), versus the ∼4 s in conventional systems. The combination of high-speed imaging acquisition and sample translation allowed a single whole-brain thin section to be imaged in less than 7 min, for a per-section throughput of ∼50 MPix/s (∼5 times faster than the first-generation TEMCA (Bock et al., 2011), and ∼40x faster than conventional TEM systems (Takemura et al., 2013). Since each sample grid (Figure S3H) typically supported three whole-brain sections (Figure 1C), and it takes ∼10 min to exchange grids and define target regions of interest (ROIs), per-grid throughput, which includes sample exchange and ROI definition by microscopists, was ∼27 MPix/s.