Detailed Structure of the Fruit Fly Brain

[BillTre]

A group of scientists have assembled transmission electron microscopic (EM) images from the serially sectioned brain of a female fruit fly (Drosophila melangaster, a model research organism) to make a database of the detailed structure of its brain.
Z. Zheng et al., “A complete electron microscopy volume of the brain of adult Drosophila melanogaster,” Cell, doi:10.1016/j.cell.2018.06.019, 2018.

Transmission electron microscopy allows synapses between individual neurons to be identified and for the details of complex processes of neurons to be reconstructed across different sections (3D reconstruction).
This little article shows a rotating 3D view of some of the neurons in the fly brain from this article.
This Science news article has a fascinating video that zooms in from a single section of the whole fly brain (low power) up to the high mag EM images that clearly show synapses. This might give you a feel for the amount of structural information involved.

This is a big step in technical capability.
Previously (1986), the nervous system (actually the whole worm) of the Caenorhabditis elegans (a much smaller simpler model research animal than a fruitfly) was serially sectioned and analyzed to identify all of its neurons, their structures and the synapses between them. From this information, the entire contectome (all the connections between all the different neurons in the animal) of C. elegans was assembled. This is information that would be important for any attempt to explain an organism's behavior with reference to the physical structure of its nervous system.
The C. elegans worm is very small and simple. The whole adoult animal has about 1,000 cells. The nervous system of the adult hermaphroditic, for example, has 302 neurons with about 7,000 synapses.
The fruit fly brain, in contrast, has about 100,000 (10⁵ neurons. All of the connections between the different neurons should be identifiable in this structural database.
The brain of a larval 5 day old zebrafish has been described as having about 100,000 neurons, but it continues to grow and add many more neurons as it enlarges greatly during its later development.
The human brain has been described as having 100,000,000,000 (100 billion, 10¹¹) neurons.

 

[jedishrfu]

How do they do this?

I saw one example where they showed the sliced photos and a researcher identified and marked the items of interest and then the computer would make a stacked 3D image from the researcher's marks. However, I thought maybe the technique has been automated now.

 

[BillTre]

I saw one example where they showed the sliced photos and a researcher identified and marked the items of interest and then the computer would make a stacked 3D image from the researcher's marks. However, I thought maybe the technique has been automated now.

I think I saw things done that way about 40 years ago.

Nowadays, I would think a lot of it is done automatically. The article did say something about 21 million images.
It would take me more than a couple of afternoons to get through those.

From the outside (not doing this personaly, but having heard some talks on it), I think of the procedure as something like:
Z-axis: one section is a step in the z-axis of amplitude = section thickness.
Determine tissue section shrinking due to histological processing.
A computer could do most of the initial imaging of individual sections, knowing section order, and keeping track of data.
The article indicates images are collected automatically from a single section (which would be a set of many images).
The individual images of a single section would be montagued together in someway, to make a single super large image (or image equivalent). Kind of like things like digital maps with their zoom in capabilities.
Alignment points between sections would have to be determined (or created ahead of time) and sections aligned.
Warping may be required for allignment since sections can definitely warp.

Detection of lines and other features would occur on individual sections. This are things Image-J can do.

Features shared across neighboring sections would be noted for use in reconstructing higher order entities.
The identification of some of these features would be strongly supported, some not so strongly supported.
There might be competing interpretations of the data.
Re-examine data, look for other relevant data (other sections?).
In a sense, it could be like automated science: a series of hypotheses (postulated structures extending over more than one section) and further tests seeking confirmation or refutation.

However I would guess that if indicators of something wrong were to arise (for example: if following the inside of a cell through sections, both sides of a single membrane were lead were labeled as the inside of a cell, than its a contradiction and a person would be alerted to resolve the problem.
Or maybe a person would focus on confirming the data for the particular cells they are interested in.

All of this requires really good histology (EM quality fixation throughout, every section cut perfectly, and all sections collected and documented) before any of the reconstructing can be started. These craftsman-like steps are not trivial.

 

[jedishrfu]

What the human expert did was identify the key structures possibly to train an algorithm to locate these in later images. Once identified then the 3D structure could be determined.

 

[Ygggdrasil]

From the methods section of the paper:

Neuron reconstructions are based on manual skeleton tracing of neuronal arbors and annotation of synapses from image stacks in CATMAID (http://www.catmaid.org) as described in Schneider-Mizell et al. (2016). All neurons included in analyses are reconstructed by at least two team members, an initial tracer and a subsequent proofreader who corroborates the tracer’s work.

Definitely not automated yet, which is why they traced only a limited portion of the fly's brain.