Introduction

Olfaction is an anatomically shallow sensory system. In mammals and invertebrates just one synapse separates the sensory periphery from the central brain (Dolan et al., 2018; Liang et al., 2010; Owald et al., 2015; Shepherd, 2011; Su et al., 2009). In the olfactory system of Drosophila melanogaster, the first relay station of synaptic transmission is the antennal lobe (AL), which has a circuit architecture homologous to that of the vertebrate olfactory bulb (Boeckh et al., 1990; Sachse et al., 2021; Shepherd et al., 2021). The fly AL consists of approximately 58 spherical compartments, called glomeruli, which can be distinguished by size, shape and location (Bates et al., 2020; Gao et al., 2000; Grabe et al., 2015; Laissue et al., 1999; Vosshall et al., 2000). Each glomerulus receives stereotypic input from axon terminals of olfactory sensory neurons (OSNs), which have their cell bodies and dendrites located in the antennae or maxillary palps (Benton et al., 2006; de Bruyne et al., 1999; de Bruyne et al., 2001; Hallem et al., 2004; Shanbhag et al., 1999). All the OSNs innervating a given glomerulus express a typical repertoire of ligand-gated chemoreceptors (Benton et al., 2006; Couto et al., 2005; Fishilevich et al., 2005), which represent a wide range of specifications, binding either a single, few, or many distinct chemicals (Hallem et al., 2006; Hallem et al., 2004; Knaden et al., 2012; Münch et al., 2016; Seki et al., 2017; Wicher et al., 2021).

Most OSNs project bilaterally to the corresponding glomeruli in the left and right AL (Gaudry et al., 2013; Tobin et al., 2017). In the AL, OSNs convey odor signals to excitatory uniglomerular projection neurons (uPNs), which branch only within a single glomerulus, or to inhibitory multiglomerular PNs (mPNs) and inhibitory or excitatory local interneurons (LNs) (Ai et al., 2013; Bates et al., 2020; Cuntz et al., 2007; Kazama et al., 2008; Kazama et al., 2009; Kreher et al., 2008; Masse et al., 2009; Ng et al., 2002; Tanaka et al., 2012; Wilson, 2013). LNs innervate each several glomeruli and are the key modulatory neurons in the AL (Chou et al., 2010; Seki et al., 2010). The highly converging OSNs-to-PN signal transmission (Chen et al., 2005; Jeanne et al., 2015; Masse et al., 2009) is lateralized, activating ipsilateral uPNs more strongly than contralateral ones (Agarwal et al., 2011; Gaudry et al., 2013; Tobin et al., 2017). From the AL, uPNs and mPNs relay processed signal information to higher brain centers (Bates et al., 2020; Fiala, 2007; Galizia, 2014; Guven-Ozkan et al., 2014; Jefferis et al., 2007; Keene et al., 2007; Norgate et al., 2006; Strutz et al., 2014).

The stereotypic activity pattern elicited by distinct odorants encodes the odor space, represented in a so-called odotopic map of the AL according to the glomerular activation by distinct chemical classes. (Couto et al., 2005; Grabe et al., 2018; Grabe et al., 2015; Knaden et al., 2014; Laissue et al., 2008). Some odorants induce a fixed innate behavior (aversion or attraction), activating characteristically specific glomeruli (Gao et al., 2015; Grabe et al., 2018; Knaden et al., 2014; Knaden et al., 2012; Semmelhack et al., 2009). The encoding of hedonic valence already at the level of the AL is important for a fast odor coding. Most odorants are encoded in a combinatorial manner in the fly AL by activating multiple OSNs types expressing broadly tuned receptors and their glomerular circuits, including broadly tuned uPNs (de Bruyne et al., 2001; Galizia, 2014; Masse et al., 2009; Sachse et al., 2016; Seki et al., 2017; Silbering et al., 2007; Silbering et al., 2008; Szyszka et al., 2015). Certain chemoreceptors and their downstream glomerular circuits, however, have evolved a very high specificity and sensitivity to single or very few chemicals (Andersson et al., 2015; Haverkamp et al., 2018; Keesey et al., 2021). These narrowly tuned glomerular circuits often belong to dedicated olfactory pathways, called “labeled lines”, which process information regarding single odorants of particular importance for reproduction and survival (Datta et al., 2008; Dweck et al., 2015; Gao et al., 2015; Kurtovic et al., 2007; Stensmyr et al., 2012). An extreme example is the DA2 glomerulus, which responds exclusively to geosmin, an ecologically relevant chemical that alerts flies to the presence of harmful microbes, causing the fly to avoid laying eggs at these locations (Stensmyr et al., 2012). This dedicated olfactory pathway and its receptor sequence is conserved throughout evolution (Keesey et al., 2021; Keesey et al., 2019). Another example is glomerulus VA1v, which responds to methyl laurate, a pheromone that induces a strongly attractive response in female flies leading to aggregation behavior (Dweck et al., 2015). DL5, on the other hand, is an example of a broadly tuned glomerulus, innervated by OSNs activated by several odorants, like E2-hexenal and benzaldehyde (Knaden et al., 2012; Mohamed et al., 2019b; Münch et al., 2016; Seki et al., 2017). This functional diversity suggests differences in neuronal composition and synaptic connectivity between broadly and narrowly tuned glomeruli.

A survey of neuronal composition across glomeruli revealed great variation in the numbers of the different types of neurons innervating narrowly and broadly tuned glomeruli (Grabe et al., 2016). In general, narrowly tuned glomeruli are innervated by more uPNs and fewer LNs compared to more broadly tuned glomeruli (Chou et al., 2010; Grabe et al., 2016). In addition, narrowly tuned OSNs receive less global LN inhibition than broadly tuned ones (Grabe et al., 2020; Hong et al., 2015; Schlegel et al., 2021). For example, in female flies, the narrowly tuned glomerulus DA2 contains dendrites of 6-8 uPNs, whereas the broadly tuned glomerulus DL5 houses only 1 or 2 uPNs and has a higher number of innervating LNs. Interestingly, both glomeruli are innervated by the same number of OSNs (Grabe et al., 2016).

Little is known, however, about the microarchitecture of the synaptic circuitry in distinct glomeruli and, in particular, about ultrastructural differences between narrowly vs. broadly tuned glomerular circuits. Electron microscopy (EM) allows volume imaging with dense reconstruction of fine neurite branches and synapses in brain tissue at nanometer resolution, necessary to map synapses (Briggman et al., 2006; Cardona et al., 2009; Helmstaedter, 2013; Meinertzhagen, 2018; Rybak, 2013). The first ultrastructural insights into the synaptic connectivity of Drosophila olfactory glomeruli were obtained by studies based on serial section transmission EM (ssTEM) (Rybak, 2016; Tobin et al., 2017). Rybak et al. (2016) showed that all three basic classes of AL neurons make synapses with each other, while Tobin et al. (2017) revealed that the differences in number of innervating uPNs between the left and right DM6 glomeruli are compensated by differences in synaptic strength. With focused ion beam-scanning electron microscopy (FIB-SEM; (Knott et al., 2008)) a complete reconstruction of all neurons in the narrowly tuned, pheromone processing glomerulus VA1v was obtained (Horne et al., 2018). Recent technological innovations in ssTEM, FIB-SEM and automated neuron reconstruction have made connectome datasets of the adult Drosophila central nervous system available (Li, P. H. et al., 2020; Saalfeld et al., 2009; Scheffer et al., 2020; Zheng et al., 2018) and provided complete circuit descriptions of several brain centres (Auer et al., 2020; Bates et al., 2020; Coates et al., 2020; Dolan et al., 2019; Felsenberg et al., 2018; Hulse et al., 2021; Huoviala et al., 2020; Li, F. et al., 2020; Marin et al., 2020; Otto et al., 2020; Schlegel et al., 2021).

In an attempt to find answers to how highly specialized olfactory glomerular circuits of dedicated olfactory pathways differ in their signal integration from broadly tuned glomerular circuits, we compared the microarchitecture and synaptic circuitry of a narrowly and a broadly tuned glomerulus (DA2 and DL5). By using a correlative workflow combining transgenic markers with FIB-SEM, in order to identify these glomeruli, we reconstructed OSNs, uPNs and multiglomerular neurons (MGNs), mapped all associated synapses and compared the circuit organization of both glomeruli.

Results

Volume-based electron microscopy of two different olfactory glomeruli

To compare the synaptic circuitries of two olfactory glomeruli known to belong to either narrowly or broadly tuned glomerular types in Drosophila melanogaster, we mapped all synapses of glomeruli DA2 (right AL) and DL5 (left AL) in a single female fly (Figure 1A-B) with the aid of FIB-SEM. A partial reconstruction of a second DA2 in another fly was used to measure neuronal volume (see Methods). The reconstructions were based on high resolution (4×4×20 nm) datasets (Figure 1; Figure 1 – video 1), thus allowing reconstruction of the finest neuronal branches (∼20 nm diameter; Figure 1C-D) as well as mapping chemical synapses (example in Figure 1E) in the two volumes of interest (VOI). To restrict the imaging volume to the target VOIs, we employed a correlative approach for the first time for a Drosophila EM volume reconstruction. Glomeruli DA2 and DL5 were identified by their size, shape and location in brains of transgenic flies (Orco-GAL4; UAS-GCaMP6s) using the glomerular map of (Grabe et al., 2015). The flies expressed the green fluorescent protein GCaMP6 coupled with calmodulin and M13 (a peptide sequence from myosin light-chain kinase; Figure 1A-B). Subsequently, the identified glomeruli were marked by laser branding using a two-photon laser (Bishop et al., 2011). These fiducial marks were apparent under both light (Figure 1A-B) and electron microscopy (Figure 1C-D) and facilitated the delimitation of the VOIs during FIB-SEM scanning. We produced two complete FIB-SEM datasets: one for glomerulus DA2 and one for DL5 (pure imaging time for both glomeruli: ∼60 h) and a partial dataset for DA2 in a second fly.

Figure 1:

Skeleton based neuron reconstruction and synapse identification

We reconstructed all neurons within the two VOIs (example neuron: Figure 1F) and mapped all their synaptic connections using an iterative skeleton-based reconstruction approach, similar to previously reported procedures (Berck et al., 2016; Schneider-Mizell et al., 2016; Zheng et al., 2017) with the aid of the web-based neuron reconstruction software CATMAID (http://www.catmaid.org; RRID:SCR_006278; (Cardona et al., 2009; Schneider-Mizell et al., 2016); Figure 1 – video 1). Synapses were identified by their presynaptic transmitter release site, which in Drosophila is composed of a presynaptic density called a T-bar, surrounded by synaptic vesicles and apposed postsynaptic elements (Figure 1E), as previously described (Fröhlich, 1985; Huang et al., 2018; Li, P. H. et al., 2020; Rybak et al., 2016; Trujillo-Cenoz, 1969). All synapses observed in our FIB-SEM data sets were polyadic, i.e. each presynaptic site connected to multiple postsynaptic sites (See example in Figure 1E), a feature of insect brain synapses (Hartenstein, 2016; Malun et al., 1993; Meinertzhagen et al., 1991; Prokop et al., 2006; Rybak et al., 2016). Some synapses had up to 16 postsynaptic sites (Figure 2 – figure supplement 1B), i.e. one T-bar and 16 single synaptic profiles (i.e. sixteen 1:1 single output-input connections). Short neuronal fragments (<10 µm), which could not be connected to any neuronal fiber were designated as “orphans”. These fragments represented 4% of the total length of all traced neuronal fibers in DA2 and 6% in DL5 and contained about ∼12% of all synaptic contacts in both glomeruli.

Glomerular neurons: classification, description and inventory

Previous descriptions of the ultrastructural characteristics of the AL in Drosophila helped to classify AL neurons into 3 main classes (Figure 2A) Olfactory sensory neurons (OSNs), uniglomerular projection neurons (uPNs) and multiglomerular neurons (MGNs; cells that interconnect multiple glomeruli). MGNs are further subdivided into multiglomerular projection neurons (mPNs) and local interneurons (LNs) (Berck et al., 2016; Gruber et al., 2018; Horne et al., 2018; Li, P. H. et al., 2020; Rybak et al., 2016; Schlegel et al., 2021; Zheng et al., 2017). Most of the neuronal profiles within the MGN neuron class belong probably to inhibitory local neurons, as this cell type is the most numerous and broadly arborizing of the multiglomerular cell types in the antennal lobe (Chou et al., 2010; Lin et al., 2012). In addition, we observed a few neuronal fibers with an electron-dense and vesicle-rich cytosol, which we interpreted to be either peptidergic neurons (Eckstein et al., 2020; Nässel et al., 2006) or the contralaterally projecting, serotonin-immunoreactive deutocerebral (CSD) neuron, (Coates et al., 2020; Dacks et al., 2006; Eckstein et al., 2020; Goyal et al., 2013; Zheng et al., 2017). Except for these neuronal fibers containing abundant electron-dense vesicles, all other neuronal fibers were assigned to either OSNs, uPNs or MGNs based on their morphology (Figure 2A, B; see Methods).

Figure 2:

OSNs formed large, elongated synaptic boutons (Figure 2A), had the largest volume/length ratio of all three neuron classes (Figure 2 – figure supplement 1A) and displayed the lowest degree of branching intensity of all neurons in both glomeruli (Figure 2B). In agreement with what had been observed in other glomeruli (Rybak et al., 2016), the majority of output synapses made by OSN terminals were triads (1:3) and tetrads (1:4). The T-bars of OSN synapses exhibited a large variation in size and some of them were large enough to accommodate 16 postsynaptic contacts (Figure 2 – figure supplement 1B). The frequency of large T-bars was much higher in OSNs than in other neuron classes with an average polyadicity (average number of postsynaptic sites at each T-bar) of 6 (1:6; (Table 1, row 14). As OSNs had the greatest T-bar and output density along their axons (Table 1, row 10-11) they also displayed the largest synaptic ratios (both for the T-bars/input sites and output sites/input sites) of all neuron classes (Table 1, row 12-13), which was in line with previous observations (Rybak et al., 2016).

Table 1.

The uPNs exhibited the highest degree of branching intensity of the three neuron classes in both glomeruli (Figure 2A-B). They showed numerous very fine apical branches that frequently connected multiple times via spines to the same presynaptic site, leading to an entangled 3D shape typical for this neuron class (Figure 2A) (Rybak et al., 2016; Schlegel et al., 2021; Tobin et al., 2017). uPNs had the smallest volume/length ratio of all neuron classes (for the DA2: Figure 2 - supplement 1A). In addition to having many fine branches, uPN dendrites also had enlarged regions with almost no cytosol that were packed with large mitochondrial profiles extending over considerable distances. These enlarged profiles showed a larger degree of mitochondria fission (dividing and segregating mitochondrion organelles; personal observation) than the other neuron classes with rather round and compact mitochondria (Figure 2A; FIB-SEM image; see data availability). Seven uPNs were found in DA2, confirming light microscopy findings (Grabe et al., 2016). Two of them (PN#1, PN#2; see data availability) branched broadly and innervated the full glomerulus, and received more synaptic input than the other 5 uPNs (PN#3-#7; see Table S3), which branched exclusively in sub-regions of the glomerulus, with partial overlap. In addition to abundant clear small vesicles (∼20 nm in diameter) (Bates et al., 2020; Strutz et al., 2014; Yasuyama et al., 2003), uPN dendrites also displayed small electron-dense vesicles, as previously reported for PN axon terminals in the mushroom body calyx (Butcher et al., 2012; Yang et al., 2022). These electron-dense vesicles are packed with different types of neuropeptides that act as neuromodulators or co-transmitters (Croset et al., 2018; Eckstein et al., 2020; Gondré-Lewis et al., 2012; Li et al., 2017). In both glomeruli, uPNs had the highest neuronal synaptic input density and the lowest T-bar and output density of the three neuron classes (Table 1, row 9-11; DA2 and DL5 differences: see next section). The synaptic ratios (T-bars/input sites and output sites/input sites) were much lower for uPNs than for the other neuron classes (Table 1, row 12-13). The majority of uPN dendritic output synapses (feedback synapses) were tetrads in both glomeruli, with an average polyadicity of around 5 (lower than in OSNs; (Figure 2 – supplement 1; Table 1, row 14).

The majority of the neuronal fibers in both glomeruli belonged to MGNs (Figure 2A). MGNs exhibited variable morphology and ultrastructure, as expected, but shared also some ultrastructural features. Their synaptic boutons were formed by thin fibers, thus the volume/length ratio of MGNs was lower than that of OSNs but greater than that of uPNs (Figure 2 – figure supplement 1A). A similar relationship was found for the number of output sites and the T-bar density along MGN fibers, which were smaller than in OSNs but larger than in uPNs (Table 1, row 10-11). In contrast, branching intensity in MGNs was larger than in OSNs but smaller than in uPNs (Figure 2B). The synaptic ratio of output-to-input sites was around one (Table 1, row 12-13). MGNs had the lowest polyadicity (∼3) of the three neuron classes (Table 1, row 14) and their synapses were mainly triads (Figure 2 – supplement 1D). Interestingly, besides the abundant clear small vesicles (∼20 nm in diameter), some MGNs had small electron-dense vesicles, most likely housing the neuropeptide sNPF (Nässel et al., 2008).

DA2 is more densely innervated and has a higher synapse density than DL5

In our FIB-SEM datasets the volume of glomerulus DA2 was 45% smaller than that of glomerulus DL5 (1500 µm³ vs. 2640 µm³), which is in agreement with measurements based on light microscopy (DA2 = 1600 µm³, DL5 = 2900 µm³ (Grabe et al., 2016). We also confirmed that a similar number of OSNs (44-46 OSNs) innervated both glomeruli (Figure 2C), and that each glomerulus received OSN innervation from both the ipsilateral and contralateral antennae (Grabe et al., 2016; Vosshall et al., 2000). Also in agreement with (Grabe et al., 2016), the DA2 glomerulus was innervated by 7 uPNs whereas DL5 had a single uPN (Figure 2C). MGN cell numbers could not be counted in our study due to their multiglomerular morphology, which also prevented us from tracing MGN fibers to their soma due to our partial volume acquisition (see Methods).

To investigate differences between DA2 and DL5 we turned our attention to their glomerular innervation and synaptic composition. We measured the total length (sum in µm) of all neuronal fibers of each neuron class within the DA2 and DL5 (Figure 2C; Table 1, row 1). In addition, we counted all T-bars and their output sites (1:1 synaptic contacts) as well as all postsynaptic sites (input sites) for all neuron fibers together and for each neuron class individually (Table 1, row 2-4). We counted in total ∼ 14 000 synaptic contacts and 2648 T-bars in DA2 and ∼ 17 000 contacts and 3387 T-bars in DL5 (Figure 2C, Table 1, row 4). Most of these synapses were triads and tetrads (Figure 2 – figure supplement 1B-D). In order to compare DA2 and DL5 we normalized neuronal length and synaptic numbers to glomerular volume. We then analyzed (1) the innervation density, i.e., the length of neuronal fibers per glomerular volume (µm/µm³) and (2) the glomerular synaptic density (T-bar #, output site or input site #/µm³). Data are reported in total for all neuronal fibers of each neuron class (Table 1, row 5-8) and as an average for neuronal fibers of the respective neuron class (Figure 3). In addition, we compared (3) the average polyadicity for each neuron class (Figure 3) and (4) the average neuronal synaptic density (T-bar, output and input site density along each neuronal fiber) (#/µm) (Figure 3 – figure supplement 1B).

Figure 3:

We observed that the average neuron innervation density of OSNs was 20% higher in DA2 than in DL5 (Figure 3A), Table S1). Also the glomerular synaptic density of input sites, output sites and T-bars along OSNs was significantly higher in DA2 than in DL5 (Figure 3A). OSNs in DA2 formed therefore more input sites, and much more T-bars and output sites per glomerular volume than in DL5 (Table 1, row 7-8; relative differences: Table S1). In contrast, the density of input sites distributed along the length of OSN fibers was similar in DA2 and DL5, whereas T-bar and output site density along the OSN axons was significantly higher in DA2 (Figure 3 – figure supplement 1A).

We then asked if the DA2 glomerulus, due to its higher number of uPNs, also had a higher uPN innervation density and synaptic density of its postsynaptic sites and/or presynaptic sites compared to the DL5 glomerulus, which contains a single uPN. In the DA2, the fibers of the 7 uPNs had almost the same total length as the fibers of the single uPN in the more voluminous DL5 (4652 µm in DA2 vs. 5015 µm in DL5; Table 1, row 1). The DA2 uPNs had in addition a similar total number of input sites than the single uPN in DL5 (3887 vs. 3955; Table 1, row 2). As such, in DA2 the total innervation density of its 7 uPNs was higher as compared to the innervation density of the single uPN in DL5 (Table 1, row 5), even though the average innervation density of DA2-uPNs was lower (Figure 3B). The total glomerular input density of all uPNs was higher in DA2 as compared to DL5 (Table 1, row 6). On the other hand, the total glomerular synaptic density of the T-bars and output sites was similar in DA2 and DL5 (Table 1, rows 7-8). In line with these results, the neuronal density of T-bars and output sites was less in the DA2 uPNs compared to the DL5 single uPN, whereas the neuronal density of input sites was similar (Figure 3 – figure supplement 1B; Table 1, row 9-10). This caused almost twice as high synaptic ratios (T-bars-to-inputs and outputs-to-inputs) in the DL5 uPN relative to DA2 uPNs (Table 1; row 12-13).

We then hypothesized that DA2 will have a lower innervation density of MGNs (mainly LNs) than DL5 as it had been reported that DL5 is innervated by fewer LNs (Chou et al., 2010; Grabe et al., 2016). However, we observed the opposite: the innervation density of MGNs was significantly higher in DA2 than in DL5 (Figure 3C), with slightly higher total innervation density (Table 1, row 5). Interestingly, only the glomerular input density was significantly higher for DA2 MGNs compared to that found in DL5, not the glomerular synaptic density of output sites or of the T-bars (Figure 3C). However, the total glomerular synaptic density of input sites, output sites and T-bars were still higher in DA2 than in DL5 (Table 1, rows 6-8). Synaptic densities along the MGN fibers were similar in DA2 and DL5 (Figure 3 – supplement 1).

In summary, the DA2 glomerulus is more densely innervated than DL5 and has a more densely packed neuropil with more synaptic contacts relative to the DL5. The DA2 has a significantly higher innervation density and higher density of T-bars, output and input sites per volume (Figure 3D, Table 1, row 5-8). The degree of synapse polyadicity is also significantly higher in DA2 than in DL5 (Figure 3D, Table 1, row 14) due to a shift to higher polyadicity among OSN (Figure 3A) and MGN synapses (Figure 3C). OSNs show the strongest shift in polyadicity, with tetrads being the most abundant synapse type in DA2 whereas triads are the most abundant in DL5 OSNs (Figure 2 – figure supplement 1B).

Lateralization of OSN glomerular connectivity

In Drosophila melanogaster, the majority of olfactory glomeruli receive bilateral OSN input (Silbering et al., 2011; Stocker et al., 1990; Stocker et al., 1983; Vosshall et al., 2000) see scheme in Figure 4A). Recent studies have shown that ipsi- and contralateral OSNs are asymmetric in their synaptic connectivity to other neurons in the majority of the glomeruli (Schlegel et al., 2021; Tobin et al., 2017) and that ipsi- and contralateral OSNs activate uPNs in an asymmetric way (Gaudry et al., 2013; Tobin et al., 2017). However, not all glomeruli appear to have the same degree of lateralized OSN connectivity (Schlegel et al., 2021). At least for one narrowly tuned glomerulus (DA1), there is functional evidence that in female flies its uPNs are evenly activated by either ipsi- or contralateral antennal stimulation (Agarwal et al., 2011). We hypothesized that this lack of lateralization could be a feature of other narrowly tuned glomeruli.

Figure 4:

Ipsi- and contralateral OSNs in DA2 and DL5 were identified based on the location and trajectory of their axons (Figure 4B). In both glomeruli, ipsilateral OSN terminals were longer than their contralateral counterparts within the VOI, while polyadicity was stronger in contralateral axons. Synaptic density was not consistently higher or lower in ipsilateral OSNs compared to contralateral ones in DA2 and DL5 (Figure 4 – figure supplement 1).

We observed that the synaptic output of ipsi-vs. contralateral OSNs was asymmetric, with significant differences in the ipsi- and contralateral OSN output to either uPNs, OSNs or MGNs (Figure 4C, DA2 and DL5). In agreement with previous observations in other glomeruli (Schlegel et al., 2021), the output fraction to uPNs and OSNs was greater in ipsilateral OSNs than in contralateral ones (Figure 4C, DA2 and DL5). Vice versa, the OSN output to MGNs was greater in the contralateral glomerulus than in the ipsilateral side (Figure 4C, DA2 and DL5). However, the differences between the medians and means were smaller in DA2 than in DL5 (Figure 4C; differences between means: see data availability).

Our finding of less lateralized connections in the DA2 (Figure 4C, DA2 and DL5) was also observed in another narrowly tuned glomerulus (VA1v; Dweck et al., 2015) for which connectome data is available (Horne et al., 2018). In VA1v, the OSN output to uPNs and MGNs was significantly asymmetric in the same manner as in DA2 and DL5, i.e. with greater ipsilateral OSN output fractions to uPNs and OSNs and greater contralateral OSN output fraction to MGNs (Figure 4C). Asymmetry in the VA1v OSN output fractions was even less distinct than in DA2 (regarding both the difference between the median and the mean; Figure 4C and data availability). In VA1v, the OSN output fraction to OSNs was similar in ipsi- and contralateral OSNs (Figure 4C). In addition, the OSN input, from either sister OSNs or MGNs, was asymmetric in DL5 but not in the narrowly tuned glomeruli (Figure 4D). The inputs from uPNs to ipsi- or contralateral OSNs were not compared due to their low numbers.

In summary, our data add to the knowledge of lateralized connectivity within olfactory glomeruli and supports the hypothesis that narrowly tuned glomeruli have a lower degree of lateralization of OSN connectivity compared to broadly tuned glomeruli.

Glomeruli DA2 and DL5 differ in several features of their circuitry

Next, we asked whether the synaptic circuitries of DA2 and DL5 differ from each other. We counted each synaptic contact (Table S2 and S3) and categorized the distinct connection motifs according to the neuron class to which the output and input neuron belonged (Figure 5A; Table S2). Each connection motif (for example OSN>uPN, i.e., the OSN-to-uPN feedforward connection) was then assessed for its relative synaptic strength, i.e. how many synaptic contacts of this particular connection motif were found compared to the total number of synaptic contacts within the respective circuitry (Figure 5A-D; see Methods).

Figure 5:

We found that neurons from each class made synaptic contacts with each other in DA2 and DL5, as previously reported for other glomeruli (Berck et al., 2016; Horne et al., 2018; Rybak et al., 2016; Schlegel et al., 2021; Tobin et al., 2017). In both DA2 and DL5, OSNs provided the strongest relative synaptic output, i.e. 49% of all synaptic connections in DA2 and 43% in DL5 were formed by OSNs (Figure 5B-C). Thus, even though DA2 and DL5 had similar numbers of OSNs (44 and 46, respectively), those in DA2 provided a stronger circuit output (14% stronger; Table S2) than those in DL5 (Figure 5B-C). In both glomeruli the main OSN output partners were MGNs and uPNs, i.e. 27% of all circuitry connections in DA2 and 24% in DL5 were OSN>MGN connections and 20% in DA2 and 18% in DL5 were OSN>uPN connections (Figure 5B-C). In DA2, interestingly, each of the 7 uPNs received input from almost all OSNs and so could maintain a high degree of convergent signal transmission (Table S3). In contrast, OSNs received the lowest relative input of all neuron classes in DA2 and DL5 (7% and 8% respectively; Figure 5B-C). In line with previous observations in other glomeruli (Horne et al., 2018; Schlegel et al., 2021), OSNs also made abundant axo-axonic synapses with sister OSNs (2.6% in DA2 and 1.5% in DL5; Figure 5B-C). Thus, the relative synaptic strength of the OSN>OSN connection was 70% stronger in DA2 than in DL5 (Figure 5B-C; Table S2).

The uPNs in both glomeruli had the weakest relative output of all neuron classes within their circuitry, and this was even weaker (38%) in DA2 (Figure 5B-C; Table S2). In contrast, the relative synaptic input onto uPNs was greater in DA2 than in DL5 (33% vs. 28%, respectively; Figure 5B-C; 16% stronger in DA2; Table S2), which is in line with our finding that in DA2, the uPNs provide more input sites per unit of glomerular volume than in the DL5 (Figure 3B-C). In both glomeruli, the feedback connections from uPNs (depicted in Figure 5A), were almost exclusively directed towards MGNs, as previously reported for the broadly tuned DM6 and the narrowly tuned glomerulus VA1v (Horne et al., 2018; Tobin et al., 2017). However, the relative synaptic strength of the uPN>MGN connection was 40% weaker in DA2 than in DL5 (uPN>MGN: 10% in DA2 and 17% in DL5). Only a few cases of uPN>OSN synaptic connections were observed (a total of 16 in DA2 and 26 in DL5) representing a synaptic strength of 0.1% in DA2 and 0.2% in DL5 (Table S2). Finally, uPNs in DA2 also made 71 reciprocal synaptic connections (representing a synaptic strength of 0.6%; Table S2; Figure 5B), consistent with electrophysiological evidence for reciprocal synaptic interactions between sister uPNs (Kazama et al., 2009). The single uPN of the DL5 had 54 dendro-dendritic synapses (representing 0.4% of all DL5 synaptic contacts; Figure 5C), which were exclusively autapses, i.e. synapses formed by a neuron onto itself. Dendritic uPN autapses exist also in DA2-uPNs, but they were few: we observed only 14 autaptic uPN-uPN connections in DA2, which were mainly located at the two longest uPN dendrites (for further analysis of autapses see next section).

MGNs received the strongest input in both glomeruli (60% of the total input in DA2 and 64% in DL5; Figure 5B-C). This is in line with the observation that MGNs provided the majority of all traced neuronal fibers in each glomerulus and had the highest innervation density of all neuron classes; Table 1). The relative output strength of MGNs was similar in both glomerular circuits (∼40% of the total output in each glomerulus; Figure 5B-C). MGNs made many reciprocal synapses to each other, accounting for 23% of all synapses in both glomeruli (Figure 5B-C). The relative synaptic strength between MGN>uPN was stronger in DA2 (12%) than DL5 (10%) (Figure 5B-C; Table S2). The MGN>OSN feedback connection was relatively weak in both glomeruli (5% in DA2 vs. 6% in DL5; Figure 5B-C) but weaker (25%) in DA2 than in DL5 (Table S2).

We then looked at the fractional output and input of each neuron class (Figure 5E’, E’’). In both glomeruli OSNs had a similar proportion of their synaptic output onto uPNs (40%-41%), onto MGNs (55% in both) and onto sister OSNs (4%-5%) (Figure 5E’). From the uPNs perspective, over 93%-96% of their recurrent synaptic output was directed to MGNs in both DA2 and DL5, and few synapses were directed onto OSNs (∼1% of the uPN output; Figure 5E’). The uPN>uPN output fraction of the 7 uPNs in DA2 (reciprocal synapses) was twice the uPN output fraction (autaptic) of the single uPN dendrite in DL5 (6% vs. 3%; Figure 5E’). MGNs formed synaptic output mainly to other MGNs (58%-59% of the total MGN output in DA2 and DL5). Among MGNs we found also rare cases of autapses. The MGN>uPN output fraction was greater in DA2 (30%) than in DL5 (25%), whereas the MGN>OSN output fraction was smaller in DA2 (12%) than in DL5 (16%; Figure 5E’).

Turning to the input fractions of each neuron class, we found that in both glomeruli, OSNs received most of their input from MGNs (>50%). In DA2 the input fraction onto OSNs (MGN>OSN) was smaller than in DL5 (63% vs. 78%; Figure 5E’’). In contrast, the OSN input fraction from sister OSNs was greater in DA2 (35% vs. 20%; Figure 5E’’). In both glomeruli, the OSNs received only weak uPN input (2%) (Figure 5E’’). The input fractions onto the 7 uPNs, formed by uPNs, MGNs and OSNs, in the DA2 and the single uPN in DL5 were similar (Figure 5E’’). Most uPN input was delivered by OSNs (∼62% in both glomeruli) and less from MGNs (∼36%). The uPN input fraction from other uPNs in DA2 or the autaptic input from the single uPN in DL5 was small (2%; Figure 5E’’). In DA2 the MGNs received a smaller fraction of uPN feedback input than in DL5 (17% vs. 26%; Figure 5E’’) but a greater OSN input fraction (45% vs. 38%; Figure 5E’’). The fraction of MGN>MGN input was similar in both glomeruli.

To further explore whether the differences in circuitry between DA2 and DL5 reported here might represent features characteristic of narrowly tuned glomeruli, we analyzed connectome data from another narrowly tuned glomerulus (VA1v; (Horne et al., 2018). We calculated the relative synaptic strength between OSNs (n=107), uPNs (n=5) and MGNs (n=74) in the VA1v (Figure 5D; Table S2). We found that the two narrowly tuned glomeruli shared five circuit features that were different from the broadly tuned glomerulus DL5: (1), OSNs in VA1v, as reported above for DA2, displayed a stronger relative feedforward output to uPNs (22%) and to MGNs (32%) (Figure 5D). The uPNs and MGNs in VA1v, received a larger fraction of OSN input than in DL5 (Figure 5E’’). (2), the OSN>OSN synaptic output was four times stronger (6%) than in DL5 (1.5%; Figure 5B-D, Table S1). This was also reflected in the OSN output fraction to sister OSNs (10%), which in VA1v was more than twice that of DL5 (4%; Figure 5E’) and in the much greater OSN input fraction (38%) to OSNs in the VA1v than in DL5 (20%; Figure 5E’’). (3), in the VA1v the uPN>uPN relative synaptic output was more than twice that of DL5 (1% vs. 0.4% in DL5; Figure 5D), which is in accordance with a much greater uPN output fraction to uPNs (14%) in VA1v than in DL5 (3%) (Figure 5E’). (4), as observed before in DA2, VA1v uPNs had fewer feedback synapses onto MGNs than in DL5 (relative synaptic strength of uPN>MGN connection: 6% vs. 17%; Figure 5C-D), also reflected in a smaller output fraction from uPNs to MGNs in VA1v than in DL5 (81% vs. 96%; Figure 5E’). In agreement, the MGN input fraction from uPNs in VA1v was much smaller than in DL5 (10% vs. 26%; Figure 5E’’). (5), OSNs in VA1v received a smaller MGN input fraction than DL5 OSNs (60% vs. 78%; Figure 5E’’).

Besides relative differences (stronger or weaker) in DA2 and VA1v connection motifs compared to DL5, two connection motifs were stronger in DA2 and DL5 than in VA1v: (1) the MGN>uPN connection showed a synaptic strength of 12% and 10% in DA2 and DL5 vs. 8% in VA1v (Figure 5B-D, Table S2). In agreement with this, the MGN output fraction to uPNs (Figure 5E’, MGN output) and the MGN input fraction in uPNs was greater in DA2 and DL5 than in VA1v (Figure 5E’’, uPN input). (2), the relative synaptic strength in MGN>MGN motifs was similar between DA2 and DL5 (23%; Figure 5B-C), but weaker in VA1v (17%; Figure 5D, Table S2). This was also reflected in a smaller MGN output and input fraction from or to MGNs (Figure 5E’ and E’’).

In summary, the two narrowly tuned glomerular circuits studied here shared five circuit features when compared with the broadly tuned glomerular circuit (all glomerular circuit features in DA2, DL5 and VA1v are shown in Figure 6A). These features were (1) a stronger OSN>uPN and OSN>MGN connection, (2) a much stronger axo-axonic communication between sister OSNs, (3) a stronger dendro-dendritic connection between uPN dendrites, (4) less feedback from uPNs to MGNs and (5) less feedback from MGNs to OSNs (Figure 6B).

Figure 6:

Autapses in the large DL5 uPN connect distant regions of its dendritic tree

Autapses (synapses made by a neuron upon itself) have seldomly been reported in the Drosophila central nervous system (Horne et al., 2018; Takemura et al., 2015). In the DA2 glomerulus we found few autapses in uPNs and MGNs (Figure 5C; Figure 7A). In the dendritic tree of the single DL5 uPN, on the other hand, three observers registered 54 autaptic connections independently (see Methods). This represents 3% of the output connections of this neuron and 0.4% of all synaptic contacts in the whole glomerulus (Figure 7A; Figure 5C; E’). We found that these autapses were not distributed evenly along the dendritic tree of the DL5 uPN. Some dendritic branches received several autaptic inputs, whereas other had no autaptic input (Figure 7A) and we hypothesized that these autapses could connect distant parts of this very large dendritic tree. We thus analyzed the exact location and distribution of their presynaptic and postsynaptic sites (Figure 7A) We discovered a difference in the distribution of the pre- and postsynaptic elements of DL5 autapses. Whilst their presynaptic T-bars were evenly distributed at basal (strahler order: 5) and distal regions (strahler order: 1-4), 95% of their postsynaptic sites were located in the most distal region (strahler order 2-1; Figure 7B-C). We also calculated the geodesic distance (i.e., along-the-arbor distance) from pre- and post-synaptic sites to the basal root node, which is the node point where the uPN enters the glomerulus and is equivalent to the closest point to the soma in our reconstruction. The geodesic distance to the basal root node from the presynaptic site was significantly shorter than for postsynaptic sites (Figure 7 – figure supplement 1B). The pre- and postsynaptic sites of each autapse were either close to each other along the dendritic tree, or distant from each other (see examples in the dendrogram depicted in Figure 7D). Thus, the geodesic distance between pre- and postsynaptic sites, (see scheme in Figure 7E), as well as the number of branching points between pre- and postsynaptic partners, were bimodally distributed (Figure 7F-G). Autapses that connected distant dendritic branches were more frequent than those that connected close dendritic branches (Figure 7E-G). In summary, we found abundant autapses within the uPN dendrite of DL5 and they were unevenly distributed, with many output sites located in a few sub-branches connecting distal dendritic regions.

Figure 7:

Discussion

We hypothesized that specialized, narrowly tuned olfactory glomeruli differ in their ultrastructure and microcircuitry from broadly tuned glomeruli. By comparing data obtained with dense reconstructions of two narrowly tuned olfactory glomeruli with that of a broadly tuned glomerulus in Drosophila melanogaster, we found prominent features of narrowly tuned glomeruli involving synaptic composition, lateralization of sensory input and synaptic circuitry.

Glomerular circuit analysis: a correlative approach

The small size of olfactory glomeruli in Drosophila gave us the opportunity to reconstruct and analyze the dense connectome of entire glomeruli with volume-based electron microscopy in a reasonable time period. Here we developed a correlative workflow that combines transgenic neuron labeling with near-infra-red-laser-branding for precise volume targeting. We then used FIB- SEM (Bishop et al., 2011) to resolve glomerular networks at the synaptic level. A similar procedure was used recently to investigate single cellular organelles (Ronchi et al., 2021). An advantage of this approach is that it facilitates localization of the volume of interest with high precision and consequently limits to a minimum the volume to be scanned and reconstructed. At the same time, the limitation in volume is a drawback of our workflow, as it was impossible to reconstruct neurons back to their soma. This fact prevented the identification of individual neurons as in other connectome studies (Bates et al., 2020; Berck et al., 2016; Eichler et al., 2017; Horne et al., 2018; Scheffer et al., 2020; Schlegel et al., 2021; Xu et al., 2020; Zheng et al., 2018).

We provide data on innervation and synapse density of olfactory sensory neurons (OSNs), uniglomerular projection neurons (uPNs) and multiglomerular neurons (MGNs) in the Drosophila antennal lobe (AL). We observed a higher innervation density of all neuron types but mainly by uPNs and MGNs and in parallel higher density of synaptic contacts along OSN terminals in the narrowly tuned DA2 compared with DL5. These results suggest that narrowly tuned glomeruli have a more densely packed neuropil, with more numerous synaptic connections in the feedforward motifs OSN>uPN and OSN>MGN. Overall, our observations on synapse density were comparable with previous reports (Horne et al., 2018; Mosca et al., 2014; Rybak et al., 2016).

Specific features of narrowly tuned glomerular circuits

Our analysis revealed circuit features in the narrowly tuned glomerulus DA2 and VA1v that might be adaptations specific of such dedicated glomerular circuits. Nevertheless, future studies, analyzing precise numbers of synaptic connections in more individuals, combined with physiological studies and computational models are required to test this hypothesis.

The OSN>uPN feedforward connection is stronger in narrowly tuned glomeruli

Presynaptic OSN terminals provide the major input to uPNs in insect olfactory glomeruli (Chen et al., 2005; Hansson et al., 2000; Horne et al., 2018; Kazama et al., 2008; Lei et al., 2010; Rybak et al., 2018; Schlegel et al., 2021; Tobin et al., 2017). Here we showed that this connection is stronger in DA2 and VA1v than in DL5 (Figure 5 and 6). A strong OSNs>uPN synaptic connection will drive non-linear signal amplification, which improves signal detection at low odor concentrations (Bhandawat et al., 2007; Kazama et al., 2008; Masse et al., 2009; Ng et al., 2002). A larger number of synapses of this type could be an adaptation to improve this amplification effect, as shown by artificial increase of synaptic sites in the AL (Acebes et al., 2001) and in lateral horn dendrites (Liu et al., 2022).

Each of the 7 uPNs in DA2 received convergent synaptic input from almost all DA2-OSNs. This is in agreement with reports on the narrowly tuned glomeruli DA1 and VA1v (Agarwal et al., 2011; Horne et al., 2018; Jeanne et al., 2015) and for broadly tuned glomeruli (Chen et al., 2005; Kazama et al., 2009; Masse et al., 2009; Tobin et al., 2017; Vosshall et al., 2000). High OSN>uPN convergence is the main driver of highly correlated activity among uPNs in pheromone coding glomeruli in flies as well as moths (Kazama et al., 2009; Rospars et al., 2014). High convergence in the lateral horn improves signal transmission from uPNs to lateral horn neurons without sacrificing speed (Huoviala et al., 2020; Jeanne et al., 2015). In the mushroom body calyces, however, the high degree of convergence is only pursued for DA2 uPNs, which converge onto few Kenyon cells, whereas VA1v uPNs synapse randomly onto many dispersed Kenyon cells (Caron 2013; Zheng 2020; Li 2020), indicating diverse signal integration in the mushroom body.

From our study, we hypothesize that in narrowly tuned glomerular circuits, which have more uPNs, the maintained strong OSN>uPN convergence, improve signal transmission accuracy. Secondly, a stronger OSN>uPN connection might compensate for the lack of OSN>uPN signal transmission sites in the case of odorants activating OSNs in a single glomerulus.

Reciprocal connections between sister OSNs and sister uPNs are stronger in narrowly tuned glomeruli

The reciprocal OSN-OSN synapse is generally stronger in narrowly tuned glomeruli DA1, DL3 and DL4, compared to broadly tuned glomeruli DL5, DM6, DM3 and DM4 (Dweck et al., 2015; Ebrahim et al., 2015; Grabe et al., 2016; Knaden et al., 2012; Schlegel et al., 2021; Seki et al., 2017; Suh et al., 2004; Tobin et al., 2017). A high degree of axo-axonic synapses between sister OSNs was also found in VA1v (Horne et al., 2018; Schlegel et al., 2021)and DA2 but not in the DL5 (this study). Hence, we suggest that a strong OSN-OSN connection is a characteristic feature of the synaptic circuitry of narrowly tuned olfactory glomeruli. Axo-axonic connections have also been reported between gustatory and mechanosensory neurons in Drosophila larvae (Miroschnikow et al., 2018) and in the olfactory epithelium and the olfactory bulb of vertebrates (Hirata, 1964; Shepherd et al., 2021). In vertebrates, axo-axonic synapses between excitatory sensory neurons are involved in correlated transmitter release (Cover et al., 2021), reminiscent of correlated uPN activity due to reciprocal synaptic and electric coupling in the Drosophila AL and LH (Huoviala et al., 2020; Kazama et al., 2009). A strong OSN-OSN connection also has the potential to increase the correlation of OSN spiking events and therefore facilitate a robust OSN signal (de la Rocha et al., 2007).

Reciprocal dendro-dendritic synapses between sister uPNs are reported here for the DA2 have been reported previously also for glomeruli DM6, DM4, VA7 and VA1v (Horne et al., 2018; Kazama et al., 2009; Rybak et al., 2016; Tobin et al., 2017). These types of synapses enhance uPN signal correlation (Kazama et al., 2009), as reported for mitral and tufted cells of the vertebrate olfactory bulb, the circuit equivalent to PNs of insect ALs (Christie et al., 2005; McTavish et al., 2012; Shepherd et al., 2021). In Drosophila multiple uPNs could induce correlated PN depolarization events, which improve the signal-to-noise-ratio of PN signal transmission (Chen et al., 2005; Jeanne et al., 2015; Kazama et al., 2009).

In summary, our data give evidence that reciprocal OSN-OSN and uPN-uPN connections are a prominent feature of the synaptic circuit of narrowly tuned glomeruli. We suggest that those reciprocal OSN-OSN and uPN-uPN connections support correlation of neuronal activity and therefore boosts signal induced depolarization events. This will in turn enhance the signal-to-noise ratio (accuracy) and transmission probability of weak and/or irregular odorant input, increasing processing speed.

Less lateralization in the OSN bilateral connectivity in narrowly tuned glomeruli

In Drosophila, most OSN axons project bilaterally and form synapses in their corresponding glomerulus on both the left and right brain hemispheres (Couto et al., 2005; Kazama et al., 2009; Schlegel et al., 2021; Silbering et al., 2011; Stocker et al., 1990; Tobin et al., 2017; Vosshall et al., 2000). This is rarely observed in other insects and absent in vertebrates (Anton et al., 2003; Dalal et al., 2020; Galizia et al., 1998; Hansson et al., 2000; Masson et al., 1990; Parthasarathy et al., 2013; Stocker et al., 1983). In the mammalian olfactory system, bilateral comparison of olfactory input only occurs in higher brain centers (Dalal et al., 2020). In flies, bilateral sensory input enables them to discriminate odor sources of different spatial origin through bilateral comparison of olfactory stimulation (Borst et al., 1982; Duistermars et al., 2009; Gaudry et al., 2013; Mohamed et al., 2019a; Taisz et al., 2022). Asymmetric OSN connectivity, shown for many olfactory OSNs (Schlegel et al., 2021; Tobin et al., 2017) seems to be the origin of a bilateral contrast in the uPN response (Agarwal et al., 2011; Gaudry et al., 2013; Taisz et al., 2022; Tobin et al., 2017), and is most likely the key to precise odor source localization (Taisz et al., 2022). Bilateral comparison is also used in the lateral horn (a higher olfactory brain center in Drosophila) for odorant position coding (Mohamed et al., 2019a). However, not all glomeruli are similar in the magnitude of bilateral asymmetry with respect to their OSN connectivity (Schlegel et al., 2021) or their uPN responses (Agarwal et al., 2011).

In agreement with observations in other olfactory glomeruli (Schlegel et al., 2021; Tobin et al., 2017), we found that glomeruli DL5, DA2 and VA1v (data from: (Horne et al., 2018) have ipsilaterally asymmetric OSN synaptic output to excitatory uPNs and sister OSNs and contralaterally an enhanced OSN>MGN output (Figure 4). We believe that, in agreement with a recent study, these asymmetric connections determine a strong left-right-contrast in the uPN response, akin to a “winner-takes-all” principle (Taisz et al., 2022).

We also observed that the degree of bilateral OSN asymmetry in DA2 and VA1v was much weaker than in DL5 (Figure 4). Weakly lateralized OSN connectivity is perhaps insufficient to induce an adequate bilateral contrast necessary for odor source localization. Recent work supports this idea by showing the importance of the interplay of asymmetric OSN signaling and LN inhibition to enhance the bilateral contrast of uPN activity and to facilitate navigation (Taisz et al., 2022).

Why do these narrowly tuned glomeruli have weaker bilateral contrast than broadly tuned glomeruli? The answer could lie in the ecological significance of the individual odorants. Geosmin, encoded by glomerulus DA2 (Stensmyr et al., 2012), and the pheromone methyl laurate, encoded by glomerulus VA1v (Dweck et al., 2015), act at short distances, mainly when the fly is walking and not flying. Perhaps, the behavioral response to geosmin or methyl laurate does not need a precise odor source location. On the other hand, food odor detection at a distance, which happens mainly at flying conditions, needs continuous processing of odor position and body alignment to navigate towards the odor source (Demir et al., 2020; Thoma et al., 2015). The bilateral OSN projection onto uPNs in DA2 and VA1v potentially has a distinct function other than odor position coding and could, via the enhancement of the effect of convergence of OSN>uPN signal transmission, enhance odor signal amplification (Bhandawat et al., 2007; Jeanne et al., 2015; Kazama et al., 2009; Masse et al., 2009)

Distinct synaptic integration of local modulatory neurons in narrowly tuned glomeruli

MGNs are composed of multiglomerular projection neurons (mPNs) that project directly to the LH (Bates et al., 2020; Jefferis et al., 2007; Strutz et al., 2014) and inhibitory and excitatory local interneurons (LNs) that interconnect the AL glomeruli (Chou et al., 2010; Liu et al., 2013; Masse et al., 2009; Okada et al., 2009; Seki et al., 2010). Since LNs are the most numerous and broadly arborizing of the multiglomerular cell types in the AL (Chou et al., 2010; Lin et al., 2012), we focus our discussion on these. Multiglomerular LNs are crucial for the modulation of the OSN>uPN signal transmission (Chou et al., 2010; Galizia, 2014; Masse et al., 2009; Seki et al., 2010; Szyszka et al., 2015).

Previous observations have shown that glomeruli DA2 and VA1v have a lower number of innervating LNs (Chou et al., 2010; Grabe et al., 2016) and receive less global interglomerular LN inhibition than broadly tuned glomeruli (Hong et al., 2015). We therefore assumed that DA2 or VA1v would have a lower LN innervation density and less LN synaptic integration in their circuitry. However, we did not observe a general lower synaptic integration in DA2 (Figure 5) and found a greater MGN innervation density, and a higher density of input sites than in DL5. VA1v MGNs on the other hand received less synaptic input and provided less output in its glomerular circuit than MGNs in DL5.

Taking a closer look at particular synaptic connection motifs of MGNs we saw that narrowly tuned glomeruli had a relatively weak uPN>MGN feedback (Figure 6). uPN feedback onto LNs and their reciprocal connection (LN>uPN) were reported in Drosophila and other insects, such as honey bees, cockroaches and moths, but their function is still poorly understood (Boeckh et al., 1993; Sachse et al., 2002; Sun et al., 1997). In the honey bee reciprocal dendro-dendritic synapses between excitatory and inhibitory neurons enhance signal contrast and the reliability of true signal representations throughout the AL (Sachse et al., 2002; Yokoi et al., 1995). Here we could not differentiate the LN types involved in the uPN>MGN synaptic motif. However, the prevailing uPN>LN synapses involve mainly widespread pan-glomerular LNs in the adult (Horne et al., 2018) and larval AL (Berck et al., 2016), which are important for combinatorial coding (Galizia, 2014; Sachse et al., 2016). Thus, weaker uPN>MGN feedback in the narrowly tuned DA2 and VA1v circuits might be a compensatory mechanism to lower the computational demand of interglomerular communication for odor identity coding.

We also observed that OSNs received less MGN in the narrowly tuned DA2 and VA1v than in the DL5, suggesting that the OSNs in DA2 and VA1v receive relatively weak presynaptic inhibition. Pan-glomerular GABAergic LNs induce presynaptic inhibition at OSN presynaptic site (Berck et al., 2016; Schlegel et al., 2021). These inhibitory LNs are drivers of balanced glomerular gain control and are a key player for odor identity coding, balancing incoming and alternating odor intensities (Asahina et al., 2009; Galizia, 2014; Hong et al., 2015; Olsen et al., 2008; Root et al., 2008; Sachse et al., 2016; Silbering et al., 2008; Szyszka et al., 2015; Wang, 2012). Our data support these observations and provide an argument for why narrowly tuned OSNs receive much lower inhibition during AL stimulation with odorants activating other OSN populations (Hong et al., 2015). Even though DA2 and VA1v might receive less interglomerular inhibition, their OSN>MGN output is still strong, in agreement with studies showing that throughout the AL, global lateral inhibition mediated by LNs scales with general OSN activation (Hong et al., 2015; Olsen et al., 2008).

In summary, narrowly tuned circuits are probably influenced more strongly by intraglomerular than by interglomerular modulation. Narrowly tuned circuits perhaps have greater computational capacities in intraglomerular modulation of signal transmission, which could be important for example for PN fine-tuning and response adjustment (Assisi et al., 2012; Ng et al., 2002).

Above we discussed putative generic features of narrowly tuned glomerular circuits. Besides these circuit features, we found a strong MGN>MGN connection in the aversive glomerular circuits DA2 and DL5 in contrast to a much weaker MGN>MGN connection in the attractive glomerulus VA1v (Dweck et al., 2015; Knaden et al., 2014; Knaden et al., 2012; Mohamed et al., 2019b; Stensmyr et al., 2012). Why do aversive olfactory circuits have a stronger MGN>MGN connection than attractive circuits? In the larval Drosophila AL, reciprocal LN>LN synapses induce disinhibition induced by a strong connection between the pan-glomerular LNs and a bilateral projecting LN, the Keystone LN, which synapses strongly onto pan-glomerular LNs and selectively onto OSNs, which are activated by attractive food odors. This is thought to be a key feature to switch from homogenous to heterogeneous presynaptic inhibition and therefore to a selective gain control enhancing contrast between attractive and aversive odor activation (Berck et al., 2016). Such balanced inhibitory systems could also be present in the adult Drosophila AL, reflected in the strong LN>LN connection in DA2 and DL5. Disinhibition of interglomerular presynaptic inhibition in aversive glomeruli circuits might be important for the fly to stay vigilant to aversive odors, while perceiving attractive cues, for example during feeding conditions so that a fast switch in behavior can be initiated if necessary.

Autaptic connection within the dendritic tree of a single uPN

We observed autapses along the large dendritic tree of the single DL5-uPN. To our knowledge, this is the first report of bulk dendro-dendritic autapses in the Drosophila olfactory system, indicating a cell-type specific occurrence of autapses in the DL5-uPN as reported for other cell types in the optic lobe (Takemura et al., 2015). Autapses are also reported to be present at different frequencies in different types of neurons in the mammalian brain (Bacci et al., 2006; Bekkers, 1998; Bekkers, 2009; Ikeda et al., 2006; Saada et al., 2009; Tamás et al., 1997; Van der Loos et al., 1972). In Drosophila, most uPNs are cholinergic (Croset et al., 2018; Kazama et al., 2008; Tanaka et al., 2012; Yasuyama et al., 2003; Yasuyama et al., 1999) and the DL5-uPN autapses reported here might activate either nicotinic or muscarinic acetylcholine postsynaptic receptors. Muscarinic acetylcholine receptors have an inhibitory effect in the Kenyon cells of the mushroom body (Bielopolski et al., 2019), but mediate excitation in the AL (Rozenfeld et al., 2019).

What could be the function of these autaptic feedback loops within the DL5-uPN dendritic tree? Recent studies in vertebrates show that excitatory autapses enhance neuron bursting and excitability (Guo et al., 2016; Wiles et al., 2017; Yin et al., 2018). Autaptic inhibitory connections have been implicated in circuit synchronization, spike-timing precision, self-stabilization of neuronal circuits and feedback inhibition (Bacci et al., 2006; Bekkers, 1998; Ikeda et al.; Saada et al., 2009; Tamás et al., 1997; Van der Loos et al., 1972).

Autapses in the DL5 uPN form mainly long-distance feedback loops, connecting distinct dendritic subtrees and the basal dendrite region (closer to the soma) with distal branches. This spatial segregation is similar to the distribution of non-autaptic pre- and postsynaptic sites in Drosophila uPNs, where presynapses are located more frequently at basal dendrites than postsynapses (Rybak et al., 2016) and other insects, such as Periplaneta americana and moths (Lei et al., 2010; Malun, 1991; Sun et al., 1997). Dendro-dendritic autaptic feedback loops connecting basal to distal branches and distinct dendritic subtrees of a large dendritic tree might facilitate activity correlation between distant dendritic subunits, as described for non-autaptic, reciprocal uPN>uPN connections (Kazama et al., 2009). This could be important in a large compartmentalized dendrite that receives inhomogeneous excitation by several OSNs at distinct dendritic sites, in order to enhance synchronized depolarization events along the dendrite, supporting signal integration (Graubard et al., 1980; Tran-Van-Minh et al., 2015). Clustered autapses could mediate local signal input amplification for distinct dendritic subunits (Kumar et al., 2018; Liu et al., 2022). Autaptic contacts, finally, could be able to shift the uPN membrane depolarization towards the spiking threshold, and enhance the firing probability during activation.

In conclusion, we provide a comprehensive comparative analysis of the ultrastructure and synaptic circuitry of two functionally diverse olfactory glomeruli with distinct computational demands, processing either single odorant information in a dedicated olfactory pathway (DA2) or input regarding several odorants and taking part in combinatory coding across distributed glomeruli (DL5). Our work provides an opportunity to gain insight into variations in network architecture and provides fundamental knowledge for future understanding of glomerular processing. By comparing our data with those from another narrowly tuned glomerulus (VA1v), we distilled prominent circuit features that suggest that narrowly tuned glomerular circuits encode odor signals with a weaker left-right-contrast, improved accuracy, stronger signal amplification and stronger intraglomerular signal modulation relative to broadly tuned glomeruli. Our findings reveal the existence of autapses in olfactory glomeruli and indicate that dendro-dendritic autapses play an important role in dendritic signal integration.

Material and methods

Fly line and fly rearing

Flies of the genotype Orco-GAL4; UAS-GCaMP6s (Vosshall et al., 2000) were obtained from the Bloomington Drosophila Stock Center (https://bdsc.indiana.edu) and reared on standard Drosophila food at 25°C and 70% humidity on a 12 h:12 h day:night cycle. Seven-days old female flies were used. In these flies, Orco-positive olfactory sensory cells emit green fluorescence, making possible to identify individual glomeruli.

Brain dissection and fixation for Focus Ion Beam Microscopy-Scanning electron microscopy (FIB-SEM)

Two 7-day old female flies were anesthetized with nitric oxide (with Sleeper TAS; INJECT+MATIC, Switzerland) and decapitated with forceps. Heads were dipped for one minute in 0.05% Triton X-100 in 0.1M Sorensen’s phosphate buffer, pH 7.3 and transferred to a droplet of freshly prepared ice-cooled fixative (2.5% glutaraldehyde and 2.0% paraformaldehyde in 0.1M Sørensen’s phosphate buffer, pH 7.3; as in (Karnovsky, 1965). The proboscis was removed and the back of the head was opened to improve fixative penetration. After 5-10 minutes, the brain was dissected out of the head capsule and post-fixed for two hours on ice. Fixation was stopped by rinsing the brain several times in ice-cooled 0.1M Sørensen’s phosphate buffer, pH 7.3 (after (Rybak et al., 2016)).

Laser branding of glomeruli for identification during FIB-SEM microscopy

To identify the glomeruli of interest at the ultrastructural level and to limit to a minimum the volume of tissue to be scanned with FIB-SEM, near-infrared laser branding (NIRB,(Bishop et al., 2011). Glomeruli of interest were first located with light microscopy in brains of Orco-GAL4; UAS-GCaMP6s flies using a confocal microscope (ZEISS LSM 710 NLO, Carl Zeiss, Jena, Germany), a 40x water immersion objective (W Plan-Apochromat 40x/1.0 DIC VIS-IR, Carl Zeiss, Jena, Germany), a laser wavelength of 925 nm at 30% laser power and ZEN software (Carl Zeiss, Jena, Germany). Once glomeruli DA2 or DL5 were identified by means of location, shape and size the volume of interest (VOI) was tagged with fiducial marks (“laser-branded”) close to the borders of the glomerulus (Figure 1A-B), using an infrared Chameleon Ultra diode-pumped laser (Coherent, Santa Clara, USA) at wavelength 800 nm and at 75-90% of laser power). Two laser scan rounds were performed for each induced fiducial brand. DA2 (right AL) and DL5 (left AL) were laser-branded in the same fly. A second glomerulus DA2 was marked in the right AL of another fly.

Transmission Electron Microscopy

Brains were rinsed with 2.5% sodium-cacodylate buffer and post-fixed in 1% osmium tetroxide and 1% potassium ferrocyanide in cacodylate buffer for 2 hours. After rinsing with cacodylate buffer the brains were dehydrated with a graded acetone series (30%-100% acetone), including an additional en bloc staining step in-between, in which the brains were incubated in 1% uranyl acetate in 50% acetone for 30 minutes in the dark, and gradually infiltrated with Araldite (glycerol-based aromatic epoxy resins; Serva, Germany). In the final step, the tissue was embedded in pure resin and left in a 60°C incubator to polymerize for 48h. Resin blocks were trimmed with a Reichert UltraTrim microtome (Leica, Deer Park, USA) and the fiducial laser marks were then located in semi-thin sections. To check tissue quality before performing high-resolution volume-based electron microscopy, serial sections 50 nm in thickness were cut with a diamond knife (Ultra 45°, Diatome, Switzerland) on a Reichert Ultracut S ultramicrotome (Leica, Deer Park, Germany), collected on single slot grids (2 x 1 mm), and imaged with a JEM 1400 electron microscope (Jeol, Freising, Germany) operated at 80 kV. Digital micrographs were obtained with a Gatan Orius SC 1000 CCD camera (Gatan Orius SC 1000; Gatan, Pleasanton, USA) controlled with the Gatan Microscopy Suite software Vers. 2.31.734.0.

Focused Ion Beam-Scanning Electron Microscopy (FIB-SEM)

Before serial Focused Ion Beam milling and Scanning Electron Microscopy imaging (FIB-SEM; (Knott et al., 2008; Xu et al., 2017), the surface of the trimmed block was coated with a conductive carbon layer to prevent charging artifacts. A FEI Helios NanoLab G3 UC (FEI, Hilsboro, USA) was used for FIB-SEM process. The laser marks used to landmark the VOI were visible across the surface of the block. The VOI surface was protected via a local deposition of platinum using a gas injection system for subsequent ion and electron beam deposition. The material surrounding the VOI at the front and the side was removed to reduce re-deposition of material during FIB-SEM. Serial images across the entire VOI were generated by repeated cycles of milling slices orthogonal to the block surface via FIB and imaging via SEM the newly exposed surface. The tissue was milled with a focused beam of gallium ions using FEI’s Tomahawk ion column (accelerating voltage: 30 kV, beam current: 790 pA, milling steps: 20 nm). FEI’s Elstar electron column was used to create the backscattered electron contrast images using an In-Column Detector (accelerating voltage. 3kV; 1.6 nA; dwell time: 10 µs). The DA2 and DL5 volumes in the first fly were imaged with a resolution of 4.9 x 4.9 x 20 nm³/vox (DA2: 769 images with 4096 x 3536 pix; DL5: 976 images with 5218 x 3303 pix). The volume of a second DA2 in a second fly was imaged with a resolution of 4.4 x 4.4 x 20 nm³/vox (571 images with 4096 x 3536 pix). The milling/imaging cycles were controlled with the Auto Slice and View 4.0 software (FEI, Hilsboro, USA).

Image alignment, 3D reconstruction and segmentation

FIB-SEM image stacks were aligned by maximizing the Pearson correlation coefficient of the central part of two consecutive images using template matching from the openCV library (https://opencv.org). Dense reconstructions of the glomeruli were produced by manually tracing all neuronal fibers and by annotating all synapses within the two glomeruli, using a skeleton-based reconstruction procedure similar to previous approaches (Berck et al., 2016; Schneider-Mizell et al., 2016; Zheng et al., 2017). Up to five independent tracers and two reviewers participated in an iterative reconstruction process using the web-based reconstruction software CATMAID (http://www.catmaid.org; RRID:SCR_006278; (Saalfeld et al., 2009; Schneider-Mizell et al., 2016); Figure 1D, Figure 1 – video 1), performing a dense reconstruction of synaptic neuropil. In a second fly, neurons of a DA2 glomerulus were manually reconstructed with the volume-based reconstruction method TrakEM2 (Cardona et al., 2012), an ImageJ (Fiji) plugin (https://imagej.net/TrakEM2).

Neuron visualization

Reconstructed neurons were visualized using CATMAID 3D visualization (http://www.catmaid.org and using Blender 3D, an open-source 3D software (https://www.blender.org/; Figure 7 – figure supplement 1). Neuron data from CATMAID were imported and shaded by Strahler order using an existing CATMAID plugin for Blender (https://github.com/schlegelp/CATMAID-to-Blender; Schlegel et al., 2016). Volume-based reconstructions were visualized as surface shapes in CATMAID imported from TrakEM2 (https://imagej.net/TrakEM2).

Glomerular border definition

The definition of the boundary between olfactory glomeruli was based on the combination of several structural features: the spatial position of pre- and postsynaptic elements along OSN axons, the position of the majority of uPN postsynaptic sites, the faint glial leaflets scattered at the periphery of the glomerulus, and the fiducial laser marks (Figure 1B, D).

Neuron identification

Neuronal fibers were assigned to one of three pre-defined neuron classes: OSNs, uPNs, and MGNs. The classification was based on their 3D shape (Figure 2A), their branching intensity (Figure 2B), the average diameter of their fibers (neuronal profiles: Figure 2A - FIB-SEM image; exemplary volume-based reconstruction), the ratio of T-bars-to-input sites and the size of their T-bars, which were either “small” (few postsynaptic connections) or “large” (many postsynaptic connections Figure 2 – supplement 1B-D). In addition, several intracellular features helped to classify neuron classes: the shape and appearance of mitochondria, the size and electron density of vesicles and the amount of synaptic spinules (small filopodia-like invaginations of neighboring cells (Figure 2A - FIB-SEM image; (Gruber et al., 2018). OSNs and uPNs could be counted, due to their uniglomerular character, by means of the identification of the axons (OSNs) or main dendrites (uPNs) entering the glomerulus. The number of MGNs could not be counted because of their pan-glomerular projection patterns in the AL. Ipsi- and contralateral OSNs in DA2 and DL5 were identified based on the trajectory of axonal fibers and their entry location in each glomerulus, (example neurons: Figure 4B). Ipsilateral OSNs reach the glomerulus from the ipsilateral antennal nerve and leave the glomerulus towards the antennal lobe commissure (ALC: (Tanaka et al., 2012). Contralateral OSNs reach the glomerulus projecting from the ALC.

Data analysis

With the aid of the web-based software CATMAID (http://www.catmaid.org) the following properties were quantified: glomerular volume, neuronal fiber length (in µm), number of fiber branching points, number of synaptic input and output sites and T-bars (see data availability). In a second fly, the volume of neurons in DA2 was measured (Figure 2 – supplement 1A) with the aid of TrakEM2 (Cardona et al., 2012), an ImageJ (Fiji) plugin (https://imagej.net/TrakEM2). The following calculations were performed:

1. 

   1. calculated as a ratio: (1) the sum of all neuronal fibers of each neuron class or (2) all together (Table 1) or (3) for each neuron individually (Figure 3)

2. 

   1. calculated as a ratio: (1) the sum of all neuronal fibers of each neuron class or (2) all together (Table 1) or (3) for each neuron individually (Figure 3)

3. (Table 1; Figure 3 – figure supplement 1)

4. (represents the average for each neuron class; Table 1)

5. (represents the average number of postsynaptic sites at a T-bar of each neuron class; Table 1 and Figure 1E)

6. (Table S1; Table S2)

7. (Table S1; Table S2)

8. 

9. 

Graphs were made with the programming language R and RStudio (R Core Team, 2018) using the packages ‘ggplot2’ and ‘reshape’ (see data availability) or with Python (see data availability). EM and fluorescence images were visualized with ImageJ (Fiji) (http://fiji.sc/; (Schindelin et al., 2012) (and All figures were compiled with Adobe Illustrator CS5 software (Adobe Inc.).

Statistical analysis was performed with R Studio (R Studio Team, 2016) using the packages ‘ggsignif’. Differences between samples DA2 and DL5 or between ipsilateral and contralateral OSNs were tested for significance with a two-sided student’s t-test if sample size was normally distributed, or with Wilcoxon two sample test if the data was not normally distributed (noted in figure legend). Data is in all cases represented as mean + standard deviation.

Analysis of autapses

The location of autapses, the measurement of their geodesic distance (distance along the neuronal dendrite) and the number of branching points from point A (presynaptic site) to B (postsynaptic profile) was analyzed with Python using the package ‘neuroboom’ https://github.com/markuspleijzier/neuroboom (see also data availability).

Data availability

Datasets will be available through the public CATMAID instance: https://catmaid.ice.mpg.de/catmaid_2020.02.15/. Neurons are named according to their neuron classification. The neuroboom Python package was used for dendrogram analysis, available at https://github.com/markuspleijzier/neuroboom and https://pypi.org/project/neuroboom/. Further newly generated source codes can be made available upon request. Source data files are provided (see source data files).

Acknowledgements

The authors are most grateful to Katrin Buder for the support with electron microscopy sample preparation, and Veit Grabe for advice on two-photon imaging. Great thanks also to Albert Cardona for discussion on synaptic networks, him, and Tom Kazimiers (Kazmos GmbH) for instruction in the use of CATMAID. The neuronal reconstructions were conducted with the outstanding support of Damilola E. Akinyemi, Eckard E. Schumann, and Michael Adewoye. We thank Martin Nawrot and Magdalena Springer for constructive comments and discussions about autapses. The work was supported by Roland Kilper and Ute Müller (aura optics, Jena), the European Regional Development Fund, by funds from the DFG (grant no. 430592330), in the Priority Program ‘Evolutionary Optimisation of Neuronal Processing’ (DFG-SPP 2205) and by the Max Planck Society.

Supplementary figures

Figure 1 -- Video 1: FIB-SEM dataset of a DA2 glomerulus with highlighted uPN reconstruction

(see extra file).

The video shows a full FIB-SEM scan of a DA2 glomerulus at pixel resolution 4×4×20 nm, with the neuron trace of a single uniglomerular projection neuron (uPN#2) highlighted in yellow.

Figure 2 – figure supplement 1:

Supplementary Table S1:

Figure 3 – figure supplement 1:

Figure 4 – figure supplement 1:

Supplementary Table S2:

Supplementary Table S3:

Supplementary Table S4:

Figure 7 – figure supplement 1: