by Nik Drummond and Alex Javier
How can animals detect a stimulus and know how to act towards it? How do they integrate what they have learned in their lifetime with what their evolutionary history has ‘taught’ them about said stimulus? From insects to mammals, the circuits that encode information about an animal’s chemical environment, or odors, appear to employ very similar computational logic (Masse et al., 2009; Wilson, 2013). For our model organism of choice, the tiny vinegar fly Drosophila melanogaster, a great deal is known about the olfactory circuit, especially its first two layers. Olfactory sensory neurons project information to the antennal lobe (AL), where their axons synapse onto the dendrites of olfactory projection neurons (PNs). The AL processes olfactory, thermosensory, and hygrosensory information. It comprises 58 glomeruli, 51 of which are dedicated to olfactory information. Each glomerulus is a target for a specific type of odour receptor, and information from each glomerulus is relayed by PNs to regions deeper in the brain, thought of as higher order brain centres. This is analogous to the mammalian olfactory system, where olfactory receptor neurons project to the olfactory bulb, synapsing with mitral and tufted cells which then feed olfactory information forward to the olfactory cortex and other higher brain areas.
In Drosophila, the two main targets of odour information are the lateral horn and the mushroom body (Fig. 1). The mushroom body is well studied and is responsible for encoding learned behaviour. Comparatively little is known about the lateral horn, although it is typically understood to encode innate behaviours. Amongst the latest work from our lab, using Electron Microscopy (EM) data, we have reconstructed all of the AL PNs, the second order neurons of the olfactory circuit, in Drosophila (Bates and Schlegel et al., 2020). Using this, we have created a full inventory of olfactory inputs to higher order brain areas, specifically the lateral horn, and have uncovered a number of interesting new findings relating to olfactory processing in Drosophila.
Previous efforts to characterise olfactory projection neurons have depended on sparse labelling, which is biased by, e.g., the availability of genetic tools to target a specific type of neuron or the likelihood of randomly labelling a specific type of neuron. Our EM dataset does not face this limitation and so provides us with the ability to completely reconstruct all PNs. Instead, we are limited by the time it takes to manually reconstruct neurons from greyscale EM image data. It takes a trained human ‘tracer’ ~60 hours of work to completely reconstruct an olfactory PN, though a novel flood-filling method from Peter Li at Google (Li et al., 2019) has provided a partial ‘auto-segmentation’ of our EM data, improving throughput by ~5-6x.
We found 346 olfactory PNs in total, which represent a staggering ~2-3x increase in the number of known olfactory PNs. These PNs can be split into two types. Uniglomerular PNs (uPNs) are well understood; they receive input in their dendrites from a single glomerulus and feed this information forward through the circuit. Previous work has identified 114 uPNs (Zheng et al., 2018), but we have identified 149, the majority of which are excitatory neurons that project first to the mushroom body, and then the lateral horn. The second type are multiglomerular (mPNs). The dendrites of these neurons innervate multiple glomeruli, presumably integrating olfactory information from across the odour space. An astonishing 197 mPNs were identified within our data set, compared to previous estimates of ~40 (Tanaka et al., 2012). mPNs also show a broad range of diversity, with some dendritic arbours only projecting to two or three glomeruli, and others dendritic arbours projecting across nearly the entire antennal lobe, receiving inputs from most glomeruli. Interestingly, mPNs do not seem to get sensory information exclusively from sensory neurons, but rather have lateral dendro-dendritic connectivity with uPNs within some glomeruli. Due to this, the odour response of mPNs is potentially much more complex than a summation across odour channels. A great deal of work is still needed to understand mPNs and elucidate the exciting mystery uncovered by our unique connectomics data.
Turning back to the uPNs, they can be divided into 78 types, based on their developmental lineage (stem cell of origin) and the glomerulus they innervate. It has long been thought that uPNs are highly stereotyped neurons, with the same circuit logic between brain hemispheres and even individuals. Here, we examined stereotypy between brain hemispheres by matching left and right hemisphere examples for 58 of the 78 uPN types. We found that the types of uPNs were identical between hemispheres. Furthermore, most glomeruli were represented by the same number of uPNs, with at most a ± 2 variance between left and right (Fig. 2).
During our reconstruction efforts, we annotated the synapses to and from PNs within the lateral horn and mushroom body. The axonal compartment of a neuron is typically where the neuron’s output synapses are. For PNs, this is the case within the mushroom body. In contrast, within the lateral horn, PN axons receive a large number of inputs, up to ~1,000 in some cases, which is about the number one expects to find on a medium-sized dendrite. Interestingly, a large proportion of these inputs are axo-axonal, from other PNs. We used a graph theory-based community detection algorithm to investigate what types of inputs converge onto individual PN axons. We see that the communities that form within this network represent specific subsets of the odour scene (e.g. pheromonal, food, or other odours). Axo-axonic connectivity between PNs may therefore be integrating ecologically similar odour channels. These roughly overlap with compartments of the lateral horn as defined by other studies (Frechter et al., 2019; Strutz et al., 2014; Jeanne et al., 2018).
We also identified a novel class of neuron representing a circuit motif unknown in both insects and mammals. This class is made up of 10 large, synapse-dense neurons that supply an average of 2.5% of input to the lateral horn. Since their axons reside in the LH and their dendrites are innervated by mushroom body output neurons (MBONs) as well as lateral horn output neurons (LHONs), we named them ‘centrifugal neurons’ (Fig. 3). These neurons receive a large amount of inputs on their dendrites by memory-readout neurons, MBONs, providing another example of interaction between learned and innate information. These could transmit memory-related input conveyed by the MBONs to the LH, which in turn could control and alter innate behaviour.
Understanding the lateral horn has only recently become more of a priority for the research community so most lateral horn targets of PNs have not yet been identified. In order to shed some light on what is downstream of these PNs, we selected 86 lateral horn neurons, including 56 lateral horn output neurons (LHONs) and 26 inhibitory local neurons (LHLNs). One thing that became apparent from our reconstructions is that, as with many insect neurons, the axon and dendrite compartments can be clearly defined, the dendritic compartments consistently lying closer to the neuron’s soma. Further, we see from the types of input these neurons receive that they are highly integrative in nature – roughly ⅓ of their dendritic inputs come from excitatory PNs, and another ⅓ come from local neurons, while the remaining ⅓ comes from a mixture of various other neuron types.
Another recently discovered class of neurons provides important input to neurons within the lateral horn. These “WED-PNs” are made up of 8 cell types, all inhibitory, and project to the LH from the wedge, a part of the ellipsoid body neuropile (Fig. 4). The wedge has recently been implicated in mechanosensation and potentially wind-sensing, so these neurons are presumably conveying mechanosensory input to the lateral horn. Following the reconstruction and upstream sampling from one neuron for each of the 8 cell types, we found that one of the largest sources of inputs to the WED-PNs are again memory-readout neurons, MBONs, suggesting yet another form of integration of learned and innate information in the lateral horn.
Our work has identified a number of interesting properties regarding the kinds of inputs received within the lateral horn. The way in which axo-axonic connectivity between the PNs is organised is indicative of integration of olfactory sensory information within ecologically similar groups across the odour space within the lateral horn. The WED-PNs, which receive olfactory inputs through MBONs, and output onto reinforcing dopaminergic neurons, could provide the basis for a circuit gating innate behaviours based on current internal states signalled by certain cues. The centrifugals, which receive information from multiple mushroom body compartments and have their axons in the lateral horn, particularly targeting PNs, could potentially be involved in gain modulation. The modulation of innate behaviour by memory could adapt the correct behavioural sequence to a particular context, favouring learned over innate behaviour. This is all indicative of highly integrative information processing within the lateral horn. Given the integration of information regarding the current odour space as well as from memory output neurons, this could eventually lead to the production of context-specific innate behaviours.
By leveraging our unique EM data set, our recent work has provided two interesting paths for future research. We have catalogued and identified the entire population of AL PNs, the olfactory inputs to the lateral horn and mushroom body. This has opened up questions regarding the functional nature of mPNs, for example, as well as the potential underlying structure and organisation of the lateral horn itself, in the form of the axo-axonic communities we have identified. The identification of two novel cell types which provide input to the lateral horn, WED-PNs facilitating mechanosensory information from the wedge and centrifugal neurons offering memory readout information from the mushroom body, further illustrates the intricate nature of information processing within the lateral horn. In the future, much exciting experimental work will shed further light on these circuit motifs. The integration of information from the mushroom body, the learning centre within the Drosophila brain, and how this interacts with innate behaviours, is of particular interest. Much remains to be done in order to understand the mechanism of information processing within the lateral horn and its role in mediating and producing innate behaviours. Our work here has taken a first step into uncovering some of the mysteries within this fascinating brain area.
References
Masse, N.Y; Turner, G.C; Jefferis, G.S.X.E., 2009. Olfactory Information Processing in Drosophila. Current Biology, 19 700-713
Wilson, R.I. 2013. Early Olfactory processing in Drosophila: Mechanisms and Principles. Annu. Rev. Neurosci., 36 217-241
Bates et al., 2020. Complete Connectomic Reconstruction of Olfactory Projection Neurons in the Fly Brain. Current Biology, 30 1-17
Li et al., 2019. Automated Reconstruction of a Serial-section EM Drosophila Brain with Flood-fillling Networks and Local Realignment. bioRxiv https://doi.org/10.1101/605634
Zheng et al., 2018. A Complete Electron Microscopy Volume of the Brain of Adult Drosophila melanogaster. Cell, 174 730-743.e22
Tanaka, N.K; Endo, K; Ito, K., 2012. Organization of Antennal Lobe-associated Neurons in Adult Drosophila melanogaster Brain. J. Comp. Neurol., 520 4067-4130
Frechter et al., 2019. Functional and Anatomical Specificity in a Higher Olfactory Centre. eLife, 8 e44590
Strutz et al., 2014. Decoding Odor Quality and Intensity in the Drosophila Brain. eLife, 3 e04147
Jeanne, J.M; Fisek, M.; Wilson, R.I. 2018. The Organisation of Projections from Olfactory Glomeruli onto Higher-Order Neurons. Neuron, 98 1198-1213.e6