Towards understanding the organizational principles of the fly brain neuronal network - Prof. Dr. Kei Ito
(Background behind our projects)
Why to study the fruit fly brain?
When I was a kid, I loved science-fiction movies. I was fascinated by the idea of warp engines that drive sci-fi spaceships. I therefore entered the Department of Physics and joined a lab of high-energy particle physics and gravity wave analysis for my bachelor thesis work. But then, after all, I realized that my talent is not good enough for building a truly innovative physics theory to fly faster than light. I therefore changed my target: A computer that can truly think like a human, again inspired by many computers and robots that appear in sci-fi movies. But this is still too demanding; I would not complete it before the end of my life. Then, how about a computer that can think like a decent animal?
I compared the animals that have been used for brain science. Mouse is still too complex. Nematode is a bit too simple. In between, an insect brain like that of a fly seemed to be a feasible target. For the master course, I therefore joined a lab that studies the fruit fly brain. I was too young; I thought that, after the two centuries of insect neuroscience history, there must have been enough knowledge to make models to simulate the fly brain neuronal network with a computer. It was in year 1986. I soon realized that the existing knowledge about neuronal circuits was still too preliminary. Before I build a computer model, I must investigate how actual neuronal circuits are organized in the insect brain.
The techniques for labeling neurons were still rather limited in 1980s. Screening of monoclonal antibodies was the most modern method at that time, which gave us useful antibodies such as 20C10 and nc82. But I wanted to get a broader overview about brain composition. I therefore chose the technique to label the proliferating neuronal stem cells (neuroblasts) and their progenies to understand when and how brain neurons are made (Ito and Hotta “Proliferation pattern of postembryonic neuroblasts in the brain of Drosophila melanogaster” Dev Biol, 149, 134-148, 1992),
Gene expression drivers for labeling and manipulating brain cells
Combining the advantages of various molecular-genetic techniques, a really useful approach was developed by Dr. Andrea Brand in early 1990s; the Gal4 enhancer-trap technique. By inserting the gene of the Gal4 expression driver protein into various sites of the genome, one can express the driver in many different specific cells, which can then drive the expression of other genes to visualize the cells, or to manipulate their functions. As a postdoctoral fellow I was involved in one of the first large-scale production of Gal4 expression driver lines at the lab of Dr. Gerhard Technau. I screened the lines that label different subsets of glial cells and made a systematic classification system, which is still used today (Ito, Urban, Technau. “Distribution, classification, and development of Drosophila glial cells in the late embryonic and early larval ventral nerve cord” Roux's Arch. Dev Biol, 204, 284-307, 1995).
The Gal4 driver lines made in each lab are usually distinguished by numbers, but the lines made by different labs may get the same number. To avoid confusion, we gave the name of our Gal4 lines “MZ”, such as the line “MZ317”, from the car license plate identifier of the city where we worked (Mainz in Germany).
The number of MZ lines we produced, about 330, was large enough to find useful lines for labeling all the types of glial cells in embryos, but far from enough for labeling much more different types of neurons in the adult brain. For this purpose, we would need at least about 5,000 driver lines. I explored the way to enable this really large-scape driver line generation project. But it was in year 1994. Only a few years after its initial development, the potential power of the Gal4 system was not yet widely recognized, and I was still a person who just finished a single postdoctoral period. There was no way to find a gigantic support for making thousands of Gal4 driver lines.
At that time there were many young scientists in Japan, about the same or a bit higher job-level than mine, who were interested in this new technique. So, I organized a consortium. Each group generates 500 or more Gal4 driver lines. Once done, one can freely use all the driver lines that are made by other groups. To avoid conflicts, each group determined the purpose of analysis, for example I would use the lines for identifying neurons, another group would use them for aging research, another group for the developmental study of a certain body part, and so on. Eight groups joined forces, and we eventually made 4,500 Gal4 driver lines, which we named “NP” lines after Nippon, the name of Japan in the local language.
We managed to complete the generation of driver lines in 1998. We later donated all the lines to the public fly stock center so that anyone can use the lines. Since then, this collection helped the analysis of diverse scientists in many areas of Drosophila research.
Making the catalogue of neurons using gene expression drivers
The Gal4 driver lines we generated turned out to be very useful for our work. We screened these lines for identifying neurons in various specific parts of the fly brain and made systematic catalogues of identified neurons. These include:
Olfactory system (e.g., Tanaka, et al. “Integration of chemosensory pathways in the Drosophila second-order olfactory centres” Curr Biol, 14, 449-57, 2004; Okada, et al. “GABA-mediated neural connections in the Drosophila antennal lobe” J Comp Neurol, 514, 74-91, 2009; Tanaka, et al. “The organization of antennal lobe-associated neurons in the adult Drosophila melanogaster brain” J Comp Neurol, 520, 4067-4130, 2012),
Mushroom body (Ito, et al. “The Drosophila mushroom body is a quadruple structure of clonal units each of which contains a virtually identical set of neurones and glial cells” Development, 124, 761-71, 1997; Ito, et al. “The organization of extrinsic neurones and their implications in the functional roles of the mushroom bodies in Drosophila melanogaster Meigen” Learn Mem, 5, 52-77, 1998; Tanaka, et al. “Neuronal assemblies of the Drosophila mushroom body” J Comp Neurol, 508, 711-755, 2008)
Visual system (e.g., Otsuna and Ito “Systematic analysis of the visual projection neurones of Drosophila melanogaster. I. Lobula-specific pathways” J Comp Neurol, 497, 928-58, 2006; Otsuna and Ito “Parallel neural pathways in higher visual centers of the Drosophila brain that mediate wavelength-specific behavior” Front. Neural Circuits, 8, 8, 2014)
Auditory system (Kamikouchi, et al. “Comprehensive classification of the auditory sensory projections in the brain of the fruit fly Drosophila melanogaster” J Comp Neurol, 499, 317-56, 2006; Kamikouchi, et al. “The neural basis of Drosophila gravity sensing and hearing” Nature, 458, 165-171, 2009)
Gustatory system (Miyazaki and Ito “Neural architecture of the primary gustatory center of Drosophila melanogaster visualized with GAL4 and LexA enhancer-trap systems” J Comp Neurol, 518, 4147–4181, 2010)
Somatosensory system (Tsubouchi, et al. “Topological and modality-specific representation of somatosensory information in the fly brain. Science 358, 615-623, 2017)
Being a pioneer: Map and go to the next
A great advantage of Gal4 expression driver system is that not only the anatomy of neurons but also their functions can be examined. Our neuron catalogue provided not only anatomical knowledge but also a toolset to analyze the functions of the identified neurons.
In some cases, we ourselves performed functional analyses. For example, we made one of the first large-scale behavioral screening by the selective activation of specific neurons (Flood, et al. “A single pair of interneurons commands the Drosophila feeding motor program” Nature 499, 83-87, 2013). The screening was done even before Channel-rhodopsin became available for flies, but it took many years to complete subsequent analyses by our colleagues for final publication.
Much more often, we provided the identified driver lines to other scientists for further analyses. Using the anatomical knowledge and the molecular-genetic resources we provide, these labs became able to perform advanced functional experiments of the neurons that have not been known before. For example, research on the fly learning center mushroom body had been limited to the analysis of its intrinsic neurons, the Kenyon cells. We published a systematic catalogue of its input/output neurons that connect the mushroom body lobes with other neuropils, and proposed a hypothesis that neurons that innervate specific compartments of the mushroom body lobes would have distinct roles in different aspects of mushroom body functions. Soon after our publication many labs started the analyses of these neurons, which led to the continuing research trend to investigate the specific roles of diverse mushroom body compartments.
This successful loop stimulated the generation of even larger collections of expression driver lines, such as the FlyLight project at the Howard Hughes Medical Institute Janelia Research Campus in late 2000s (e.g., Jenett, et al. “A GAL4-driver line resource for Drosophila neurobiology” Cell Rep, 2(4), 991-1001, 2012. We are not involved in this work, but our endeavor is mentioned as a pioneer of this approach.)
Our unique expertise is to identify and map so far unknown neurons in the brain regions that have hardly been analyzed before. There are many more labs that can perform sophisticated functional analyses of the identified neurons. Thus, instead of staying in a particular brain part to analyze the functions of the neurons we identified by ourselves, we leave this new research field (or a battlefield for publication) to other labs, and we ourselves continuously moved on to new brain regions. This “Map and go to the next” style has been the backbone of our approach.
Getting overview, fundamental principles, and compartment definition
Making a catalogue of neurons is important, but this is not the aim of our research. What is much more important is to get fundamental principles by examining the identified neurons.
Many parts of the brain appear like tangled spaghetti of neuronal fibers. By visualizing certain subsets of neurons, one can distinguish specific organizations that are hidden inside the spaghetti bowl. But one cannot talk about organization, if one examines only a handful neurons. Such neurons might be very good representatives of the overall organization, but they might otherwise be too specific ones, or just outliers. To obtain a systematic concept, one must examine as many neurons as possible to get the overview. By comparing these neurons and find common key features, we can obtain fundamental organizational principles of that brain part.
The organizational principle is most clearly expressed in the form of compartmentalized architecture inside each neuropil. Except for the glomeruli in the antennal lobe, most neuropils appear contiguous with no clear boundaries. But by visualizing specific sets of neurons, one can discern “hidden” compartment organization. Based on our catalogue of neurons, we have identified various compartment architecture and given them names. These include:
- alpha/beta prime lobes of the mushroom body that is distinct from the alpha/beta lobes,
- layered organization within each lobe formed by different subtypes of Kenyon cells,
- segmented compartments within each lobe formed by input/output neurons such as gamma1, gamma2, alpha 1, alpha2,,,
- the terms “visual projection neurons” and “optic glomeruli” as well as their specific compositions in higher visual centers,
- distinct parts of the lateral horn that receive specific kinds of olfactory signals,
- segregated zones in the Johnston’s Organ Neuron terminals that are specialized for vibration and deflection of antennae,
- segregated layers in the ventral nerve cord each of which receive signals from a specific type of somatosensory neurons,
among others. The compartments we identified later became the targets of functional analyses by many other labs.
Towards the understanding of terra incognita: define brain regions
Having revealed the basic architecture of most of the sensory-related neuropils, we have been working on the analysis of the remaining parts of the brain, which we call “terra incognita”. One big problem towards this goal was that many parts of the fly brain, or insect brain in general, have not been clearly defined. Boundaries between many brain parts were ambiguous, and some parts of the brain even had no name. This situation was problematic towards the forthcoming era of connectome.
To address this issue I organized a consortium of insect brain experts working on various species in 2007. After seven years of extensive discussion, the Insect Brain Name Working Group managed to establish the definition of the names and boundaries of all the neuropils, or brain regions, which completely cover the entire brain, based on the fly brain anatomy but taking the brain organization of broad insect species into account. We devised many new names to annotate the brain regions that have essentially been ignored, and also resolved conflicting names that have been used confusingly (Ito et al “A systematic nomenclature for the insect brain” Neuron, 81, 755-765, 2014). We anticipated the necessity of such a standard brain region definition for the forthcoming connectome era. Indeed, just a few years after its publication, this brain map became the indispensable backbone for annotating the locations and synapse numbers for the connectome database.
Towards the understanding of terra incognita: analyze ignored neuropils
The next step is to analyze neurons in each brain region that has hardly been investigated before. The Gal4 expression driver lines that we have generated long ago, however, are not very effective for addressing these brain parts. Because the expression patterns of these lines are determined by the activity of one certain gene, various neurons tend to be labeled by one driver line. Intermingled labeling makes it difficult to distinguish individual neurons. We therefore had a collaborative project with the HHMI Janelia research campus to generate much more specific driver lines using the Split-Gal4 system, which combines the expression patterns of two genes to achieve specific labeling only in the intersection of these patterns. We have screened more than 20,000 combinations of Split-Gal4 and established more than 1,000 lines that are quite specific for labeling neurons in terra incognita.
We also utilize the recent electron microscopy (EM) connectome data. EM data provide very high-resolution images of essentially all the neurons in the fly brain, but determining neuron types -- the sets of neurons that share common morphological and connectivity features -- was not that easy. We helped the Janelia FlyEM project to determine neuron types in terra incognita. We also devised six different ways of naming these neuron types, compared the pros and cons of different naming schemes, and determined the one that is most intuitive and easy to pronounce (a very important factor for communication among researchers). Eventually we determined and named 3,777 neuron types in terra incognita, which account for more than 70% of the whole neuron types in the hemibrain EM dataset (Scheffer. et al. “A connectome and analysis of the adult Drosophila central brain” Elife, 9, e57443, 2020).
The comprehensive EM neuron dataset provides us with typically between 2,000 and 6,000 neurons associated with each brain region of terra incognita. This is much much more than the neuron numbers that we have analyzed using Gal4 driver lines. We explored how to categorize these neurons in the most straightforward way and devised a hierarchical classification system.
Classification of neurons by itself is actually not our aim. Our aim is to identify hidden organization in each brain region. Neurons that share a common morphological feature -- forming a kind of compartment -- in one brain region may project differently in other brain regions. To reveal specific architecture in one neuropil, we must classify neurons specifically for that neuropil based on the arborization patterns within that neuropil. Thus, if we analyze five different neuropils, we will get five different hierarchical neuron classification, which are not consistent to each other. But the compartments are determined by the neuron classification for each neuropil, which are not affected by the different neuron classification in other neuropils. By repeating these steps, we will be able to determine compartment organization in all the brain regions, hopefully before I retire.
Definition of compartments has turned out to be quite robust. Back in 2008 we determined all the compartments in the mushroom body based on a limited number of neurons identified with Gal4 driver lines. Even after extensive studies by many researchers and the identification of hundreds of EM neuron data, the compartments we defined remain unchanged. Concerning the visual system, EM data have identified much more optic glomeruli than those that we first reported in 2006, but all the glomeruli we identified remain unchanged, and newly identified neurons follow the same organizational principles.
Connectome and neuronal organization
EM datasets with the information of input/output synapse locations provide us with a near-complete catalogue of neurons and their connectivity. Many advanced computational analyses became feasible. We however put more emphasis on the visual examination to reveal the organization of neuronal fibers. Neurons that are connected at one part of the brain and those connected at a different part of the brain are likely to have different roles in neuronal computation. But the lack of detailed location map makes it difficult to investigate such differences. We want to provide knowledge so that people can easily understand different locations of synaptic connections. Previously we provided community with the definition of brain regions. This greatly helped the mapping of synapses across brain. But people cannot yet easily distinguish synapses formed at different parts within each brain region. Our map of even smaller compartments inside neuropils will help improving the precision of connectome analysis.
Analyze in 3D
Human brain has very advanced visual recognition capability. Morphology of neurons are inherently three dimensional (3D), but 3D visualization of neurons has not been straightforward. We are very keen on analyzing neurons in 3D. Even my first paper published in 1992 presented 3D stereograms, at that old time generated manually for the view for each eye. Since then, we always use the most cutting-edge visualization techniques available at that period. We even develop visualization software tools by ourselves.
Recently we are relying heavily on 3D virtual reality (VR). By virtually expanding the fly brain to the size of a small building and examining individual neuron branches from inside the brain, we can distinguish many features that are not recognizable with simple projection views. By combining computational neuron clustering and 3D-VR visual examination, we can identify fine differences inside the neuropils. By overlaying EM data and light microscopy (LM) neuron data of driver lines and antibodies and examining them in 3D-VR, we can identify specific neuron types even if the LM data appear heavily intermingled.
To convey the 3D anatomical knowledge we have obtained, we must provide very clear images of neurons so that the audience can distinguish the key morphological features that we want to explain. This requires a sense of art. Scientists are professional photographers, because we make our living by producing impressive images for publication. Unfortunately, we cannot provide 3D experiences in published papers, which are printed on 2D sheets of paper (or displayed on a regular 2D computer screen). To convey 3D structural information of neurons with convincing and easy-to-understand 2D images, all the approaches that have been accumulated among architectural, landscape, product, street photo and portrait photographers are useful.
Side story -- Cell lineage-dependent brain organization
Complex systems such as cars, airplanes and buildings can be understood in a more straightforward manner by learning how they are built from individual parts. Likewise, complex organization of the brain neuronal circuits can be interpreted more easily, if one learns how they are built from the ancestor cells. My first research at the PhD period therefore focused on the proliferation pattern of the neuronal stem cells (neuroblasts). I found that the whole central brain is made by about 100 neuroblasts, and among them four neuroblasts seem to be specialized for making the mushroom body (Ito and Hotta, 1992).
To prove this, I must label one of the neuroblasts and visualize its progeny. The technique available at that time -- injecting dyes into neuroblasts -- works well for embryogenesis, but it cannot visualize the progeny in the adult. During the larval period the volume of the nervous system expands 1,000 times; injected dyes become diluted below the detection level.
Thanks to the ever-advancing molecular genetic methods, I utilized the yeast-derived flippase gene to induce genetic recombination randomly in the dividing cells. Using this system I visualized the progeny of each of the four mushroom body neuroblasts and confirmed that they are indeed specialized for this brain structure. I also found about 20 other cases, where a neuroblast generates distinct types of brain cells. These led to the principle of lineage-dependent brain organization. (Ito, et al. “The Drosophila mushroom body is a quadruple structure of clonal units each of which contains a virtually identical set of neurones and glial cells” Development, 124, 761-71, 1997).
The next goal is to trace the progeny of all the ca. 100 neuroblasts in the brain. It took a very long time to complete this. After 16 years we managed to achieve this (Ito, et al. “Systematic analysis of neural projections reveals clonal composition of the Drosophila brain” Curr. Biol, 23, 644–655, 2013). As I proposed back in 1997, all the cell lineage produces distinct sets of progeny, each of which contributes to the formation of specific neuronal circuits in the fly brain.
Over the time, this principle tends to be overinterpreted -- a particular cell lineage is specialized for a particular brain function, or a particular brain function is correlated with distinct cell lineages. This, however, is a too simplified view. We showed that each neuroblast produces a distinct set of neuronal projections. But this single set may contribute to not only one but several -- sometimes many -- different parts of the brain neuronal circuits. In a few extreme cases, a subset of a single neuroblast progeny even migrates far away from its original location during the pupal stage, so that the same cell lineage forms two cell clusters in the anterior and posterior brain (Ito, et al. 2013).
This overinterpretation may perhaps have occurred, because many people focus on a few “organized” brain structures such as the mushroom body, central complex and antennal lobes. These structures are each formed by only several cell lineages. They appear organized, because they are simple. On the contrary, each part in terra incognita is contributed by as many as 34 cell lineages (Ito, et al. 2013). These brain parts appear diffused and disorganized, because they are constructed in much more complex manners.