NeuromedinU is a potent regulator of food intake and activity in mammals. In Drosophila, neurons producing the homologous neuropeptide hugin regulate feeding and locomotion in a similar manner. Here, we use EM-based reconstruction to generate the entire connectome of hugin-producing neurons in the Drosophila larval CNS. We demonstrate that hugin neurons use synaptic transmission in addition to peptidergic neuromodulation and identify acetylcholine as a key transmitter. Hugin neuropeptide and acetylcholine are both necessary for the regulatory effect on feeding. We further show that subtypes of hugin neurons connect chemosensory to endocrine system by combinations of synaptic and peptide-receptor connections. Targets include endocrine neurons producing DH44, a CRH-like peptide, and insulin-like peptides. Homologs of these peptides are likewise downstream of neuromedinU, revealing striking parallels in flies and mammals. We propose that hugin neurons are part of an ancient physiological control system that has been conserved at functional and molecular level.https://doi.org/10.7554/eLife.16799.001
Multiple studies have demonstrated functional conservation of fundamental hormonal systems for metabolic regulation in mammals and Drosophila. This includes insulin (Ikeya et al., 2002; Rulifson et al., 2002), glucagon (Kim and Rulifson, 2004), and leptin (Rajan and Perrimon, 2012). In addition to these predominantly peripherally released peptides, there is a range of neuropeptides that are employed within the central nervous systems (CNS) of vertebrates and have homologs in invertebrates, e.g. neuropeptide Y (NPY), corticotropin-releasing hormone (CRH) or oxytocin/vasopressin (Nässel and Winther, 2010; Nässel and Wegener, 2011; Grimmelikhuijzen and Hauser, 2012; Mirabeau and Joly, 2013; Jékely, 2013).
Among these, neuromedinU (NMU) is known for its profound effects on feeding behavior and activity; NMU inhibits feeding behavior (Howard et al., 2000), promotes physical activity (Novak et al., 2007; Chiu et al., 2016), and is involved in energy homeostasis (Nakazato et al., 2000; Ivanov et al., 2002) and stress response (Hanada et al., 2001; Zeng et al., 2006). Hugin is a member of the pyrokinin/PBAN (pheromone biosynthesis activating neuropeptide) peptide family and a Drosophila homolog of NMU that has recently gained traction due to similar effects on behavior in the fly: increased hugin signaling inhibits food intake and promotes locomotion (Melcher et al., 2006; Schoofs et al., 2014; Bader et al., 2007b). In mammals, distribution of the NMU peptide, NMU-expressing cells and NMU-positive fibers is wide and complex. High levels of NMU have been reported in the arcuate nucleus of the hypothalamus, the pituitary, the medulla oblongata of the brain stem, and the spinal cord (Domin et al., 1987; Ballesta et al., 1988; Howard et al., 2000; Ivanov et al., 2004). The number of neurons involved and their morphology is unknown. In Drosophila, the distribution of hugin is less complex, yet similar: the peptide is produced by neurons in the subesophageal zone that have hugin-positive projections into the ring gland, the pars intercerebralis and ventral nerve cord (Melcher and Pankratz 2005) (Figure 1). While comparisons across large evolutionary distances are generally difficult, these regions of the fly brain were suggested to correspond to aforementioned regions of NMU occurrence based on morphological, genetic and functional similarities (Ghysen, 2003; Hartenstein, 2006). Consequently, NMU/hugin has previously been referred to as a clear example of evolutionary constancy of peptide function (Taghert and Nitabach, 2012).
Although functional and morphological aspects of neurons employing either neuropeptide have been extensively studied in the past, knowledge about their connectivity is fragmentary. While large-scale connectomic analyses in vertebrates remain challenging, generation of high-resolution connectomes has recently become feasible in Drosophila (Ohyama et al., 2015; Berck et al., 2016; Fushiki et al., 2016; Schneider-Mizell et al., 2016). We took advantage of this and performed an integrated analysis of synaptic and G-protein-coupled receptor (GPCR)-mediated connectivity of hugin neurons in the CNS of Drosophila. Our data demonstrates that hugin neurons employ small molecule transmitters in addition to the neuropeptide. We identify acetylcholine as a transmitter that is employed by hugin neurons and find that it is required for their effect on feeding behavior. Next, we show that hugin neurons form distinct units, and demonstrate that clusters of neurons employing the same neuropeptide are remarkably different in their synaptic connectivity. One unit of hugin neurons is presynaptic to subsets of median neurosecretory cells (mNSCs) in the protocerebrum. In parallel to the synaptic connectivity, mNSCs also express the G-protein-coupled receptor PK2-R1, a hugin receptor, rendering them likely targets of both fast synaptic transmission and neuromodulatory effects from hugin neurons. These mNSCs produce diuretic hormone 44 (DH44, a CRH-like peptide) and Drosophila insulin-like peptides, both of which have mammalian homologs that are likewise downstream of NMU (Wren et al., 2002; Malendowicz et al., 2012). Endocrine function is essential to ensure homeostasis of the organism and coordinate fundamental behaviors, such as feeding, mating and reproduction, and acts as integrator of external and internal sensory cues (Swanson, 2000). Consequently, connections between sensory and endocrine systems are found across species (Yoon et al., 2005; Tessmar-Raible et al., 2007; Strausfeld 2012; Abitua et al., 2015). We show that hugin neurons receive chemosensory input in the subesophageal zone (SEZ), thereby linking chemosensory and neuroendocrine systems.
The hugin gene is expressed in only 20 neurons in the Drosophila CNS. This population comprises interneurons, which are confined within the CNS, as well as efferent neurons, which leave the CNS. The interneuron type can be subdivided into those projecting to the protocerebrum (hugin-PC, eight neurons) or the ventral nerve cord (hugin-VNC, four neurons). The efferent type can be subdivided into those projecting to the ring gland (hugin-RG, four neurons) or the pharynx (hugin-PH, four neurons) (Figure 2A) (Bader et al., 2007a). Based on these morphological features, we first reconstructed all hugin neurons in an ssTEM volume covering an entire larval CNS and the major neuroendocrine organ, the ring gland (Figure 2B; see Materials and methods for details). We then localized synaptic sites, which could be readily identified as optically dense structures (Prokop and Meinertzhagen, 2006). Comparing neurons of the same class, we found the number as well as the distribution of pre- and postsynaptic sites to be very similar among hugin neurons of the same class (Figure 2C–E, Video 1). Presynaptic sites are generally defined as having small clear core vesicles (SCVs) containing classic small molecule transmitter for fast synaptic transmission close to the active zone (Prokop and Meinertzhagen, 2006). Efferent hugin neurons (hugin-RG and hugin-PH) showed essentially no presynaptic sites (<1 average/neuron) within the CNS and we did not observe any SCVs. For hugin-RG neurons, membrane specializations resembling presynaptic sites were evident at their projection target, the ring gland. These sites did contain close-by DCVs but no SCVs and had in many cases no corresponding postsynaptic sites in adjacent neurons. Instead they bordered haemal space indicating neuroendocrine release (Figure 2—figure supplement 1A). The configuration of hugin-PH terminals is unknown as their peripheral target was outside of the ssTEM volume. For the interneuron classes (hugin-PC and hugin-VNC), we found SCVs at larger presynaptic sites, indicating that they employ classic neurotransmitter in addition to the hugin peptide (Figure 2—figure supplement 1B,C). Hugin-PC and hugin-VNC neurons’ projections represent mixed synaptic input-output compartments as they both showed pre- as well as postsynaptic sites along their neurites (Figure 2D,E).
All hugin neurons receive inputs within the SEZ [previously called subesophageal ganglion (SOG)], a chemosensory center that also houses the basic neuronal circuits generating feeding behavior (Hückesfeld et al., 2015). However, only the hugin-PC neurons showed considerable numbers of synaptic outputs in the SEZ, consistent with their previously reported effects on feeding (Schoofs et al., 2014; Hückesfeld et al., 2016) (Figure 2E).
The existence of presynaptic sites containing SCVs in addition to large DCVs led to the assumption that hugin-PC and hugin-VNC (possibly also hugin-PH neurons) employ small molecule neurotransmitters parallel to the hugin neuropeptide. To address this, we checked for one of the most abundantly expressed neurotransmitter in the Drosophila nervous system: acetylcholine (ACh) (Yasuyama and Salvaterra, 1999; Salvaterra and Kitamoto, 2001). In the past, immunohistochemical and promoter expression analyses of choline acetyltransferase (ChAT), the biosynthetic enzyme for ACh, were successfully used to demonstrate cholinergic transmission (Barnstedt et al., 2016; Miyamoto, 2012; Yapici et al., 2016). We used both, anti-ChAT antibody as well as a ChAT promoter analysis, and investigated co-localization with hugin neurons. In the EM data, we found hugin neurons to have comparatively few SCVs, suggesting only low amounts of small transmitters. In addition, ChAT is preferentially localized in the neuropil and less so in the somas (Sámano et al., 2006). Consistent with this, we found that ChAT immunoreactivity in hugin cell bodies was relatively low and varied strongly between samples. Therefore, we quantified the anti-ChAT signal to show that while ChAT levels were in some cases indiscernible from the background, overall highest levels of ChAT were found in hugin-PC and hugin-VNC/PH neurons (Figure 3A). Note that while hugin-PC and hugin-RG neurons were easily identifiable based on position and morphology, hugin-PH and hugin-VNC neurons usually clustered too tightly to be unambiguously discriminated and were thus treated as a single mixed group. Similar to the immunohistochemical analysis, the ChAT promoter (ChAT-GAL4) drove expression of a fluorescent reporter in all hugin-PC neurons plus a subset of hugin-VNC/PH neurons (Figure 3B). Hugin-RG showed weak ChAT signal with either method, consistent with these neurons lacking SCVs in the EM data.
These findings suggested that ACh may be a co-transmitter in hugin neurons. We previously demonstrated that RNAi-induced knockdown of the hugin neuropeptide rescues the phenotype of feeding suppression caused by induced activation of hugin neurons in behavioral and electrophysiological experiments (Schoofs et al., 2014). Here, we present the knockdown of ChAT using an established UAS-ChAT-RNAi line (Plaçais et al., 2013; see Materials and methods). Under unimpeded conditions (i.e., without ChAT knockdown), activation of hugin neurons leads to severe decrease of food intake in intact larvae. This decrease in food intake was rescued by knockdown of ChAT in hugin neurons to a similar degree as the knockdown of the hugin neuropeptide itself (Figure 3C).
In addition to a general decrease in food intake, activation of hugin neurons leads to a decrease in cycle frequency of pharyngeal pump motor activity (Schoofs et al., 2014). We used extracellular recordings of the antennal nerve (AN) in isolated CNS for precise monitoring of motor patterns of the pharyngeal pump (Schoofs et al., 2009). As for the food intake, knockdown of ChAT in hugin neurons also rescued the suppressive effect of hugin neuron activation on pharyngeal pumping (Figure 3D). Taken together, these data clearly demonstrate that ACh plays a functional role in hugin neurons. Moreover, this suggests that hugin neuropeptide and ACh have to be employed together in order to regulate feeding behavior.
Reconstruction of hugin neurons and localization of synaptic sites revealed that neurons of the two interneuron classes, hugin-PC and hugin-VNC, were reciprocally connected along their main neurites to ipsilateral neurons of the same class (Figure 4, Figure 2—figure supplement 1E,F). These connections made up a significant fraction of each neuron's total synaptic connections, implying that their activity might be coordinately regulated.
We therefore further explored the different classes within the population of hugin-producing neurons, asking whether hugin classes establish functional units or whether they are independently wired. To this end, we reconstructed 177 pre- and postsynaptic partners of hugin neurons (Figure 5A, see Materials and methods for details). First, we found that neurons of the same hugin class were connected to the same pre- and postsynaptic partners. Furthermore, most synaptic partners were connected exclusively to neurons of a single hugin class (Figure 5B). Second, pre- and postsynaptic partners of each hugin class resided in different parts of the CNS (Figure 5C; Video 2). For hugin-RG and hugin-PH, the vast majority of synapses were made with interneurons, 93 ± 4% and 97 ± 3%, respectively. This percentage was lower for hugin-PC (66 ± 6%) and hugin-VNC (81 ± 2%). To our knowledge, none of these interneuron partners have been previously described, making it difficult to speculate on their functions at this point. Non-interneuron partners will be described in the following sections. In summary, these findings show that neurons of each hugin class form complex microcircuits that are largely separate from one another.
Hugin neurons have a significant number of their incoming synapses (63 ± 22%) within the SEZ. This region of the CNS is analogous to the brainstem and is a first-order chemosensory center that receives input from various sensory organs (Ghysen, 2003). In addition, subsets of hugin neurons were recently shown to be responsive to gustatory stimuli (Hückesfeld et al., 2016). We therefore searched for sensory inputs to hugin neurons and found a total of 68 afferent neurons that made synaptic contacts onto hugin neurons (Figure 6A). Two major groups emerged: a larger, morphologically heterogeneous group consisting of afferent neurons projecting through one of the pharyngeal nerves (the antennal nerve) and, unexpectedly, a second, more homogeneous group entering the CNS through abdominal (but not thoracic) nerves. We observed that the reconstructed afferent presynaptic partners of hugin neurons covered different parts of the SEZ. Thus, we sought to cluster these afferent neurons by computing the similarity in spatial distribution of their synaptic sites, termed synapse similarity score.
Clustering based on synapse similarity score resulted in seven different groups, each of them covering distinct parts of the SEZ (Figure 6B; Video 3; see Materials and methods for details). To address the issue of the origin of identified sensory inputs, we compared our data with previous descriptions of larval sensory neurons. It is well established that abdominal nerves innervate internal and external sensory organs of the peripheral nervous system. This includes proprioceptive (chordotonal), tactile, nociceptive (multi dendritic neurons) and a range of sensory neurons whose function is still unknown (Hwang et al., 2007; Ghysen et al., 1986; Bodmer and Jan, 1987). To our knowledge, no abdominal sensory neurons with projections into the SEZ such as the one observed presynaptic to hugin have been described. However, the majority of afferent neurons synapsing onto hugin neurons stems from the antennal nerve. This pharyngeal nerve carries the axons of gustatory receptor neurons (GRNs) from internal pharyngeal sensilla as well as those of olfactory receptor neurons (ORNs) and other GRNs from the external sensory organs (Figure 6C,D) (Colomb et al., 2007; Vosshall and Stocker, 2007). ORNs can be unambiguously identified as they target specific glomeruli of the antennal lobe (Vosshall and Stocker, 2007), but no such sensory neurons were found to directly input onto hugin neurons (Figure 6—figure supplement 1).
The GRNs likewise target restricted regions of the SEZ neuropil, but this is not as well characterized as the antennal lobes (Colomb et al., 2007; Miyazaki and Ito, 2010). The antennal nerve neurons that contact the hugin cells show the morphology of this large, heterogeneous population of GRNs (Colomb, 2007; Kwon et al., 2011). We thus compared our clustered groups with previously defined light microscopy-based gustatory compartments of the SEZ (Colomb, 2007). Groups 2 and 6, which cover the anterior-medial SEZ, likely correspond to two areas described as the target of GRNs from internal pharyngeal sensilla only (Figure 6D). The remaining groups were either not previously described or difficult to unambiguously align with known areas. Our division into groups is also reflected at the level of their connectivity to hugin neurons: sensory neurons of group 1 have synaptic connections to both hugin-PC and hugin-VNC neurons. Groups 2–5, encompassing the previously described pharyngeal sensilla, are almost exclusively connected to hugin-PC neurons. Group 6 sensory neurons make few synapses onto hugin-RG neurons. Group 7, encompassing the abdominal afferent neurons, is primarily presynaptic to hugin-VNC (Figure 6E).
The efferent type hugin neurons, hugin-PH and hugin-RG, show little to no sensory input. In contrast, the interneuron type hugin neurons, hugin-PC and hugin-VNC, receive a significant fraction of their individual incoming synaptic connections (up to 39%) from sensory neurons. Summarizing, we found two out of four types of hugin neurons to receive synaptic input from a large heterogeneous but separable population of sensory neurons, many of which are GRNs from external and internal sensory organs. Hugin-PC neurons were recently shown to be activated by bitter gustatory stimuli but not salt, fructose or yeast (Hückesfeld et al., 2016). Our data strongly suggests that this activation is at least partially based on monosynaptic connections to GRNs. Moreover, the heterogeneity among the population of sensory neurons suggests that hugin-PC neurons do not merely function as simple relay station but rather fulfill an integrative function, for example between multiple yet-to-be-identified modalities or various external and internal sensory organs.
NMU has been well studied in the context of its effect on the hypothalamo-pituitary axis. We therefore looked for similar motifs among the downstream targets of hugin neurons. The cluster of hugin-PC neurons projects their neurites from the SEZ to the protocerebrum, terminating around the pars intercerebralis. Median neurosecretory cells (mNSCs) in this area constitute the major neuroendocrine center in the CNS, homologous to the mammalian hypothalamus, and target the neuroendocrine organ of Drosophila, the ring gland (Hartenstein, 2006).
Three different types of mNSCs produce distinct neuropeptides in a non-overlapping manner: 3 mNSCs produce diuretic hormone 44 (DH44), 2 mNSCs produce Dromyosuppressin (DMS) and 7 mNSCs produce Drosophila insulin-like peptides (Dilps, thus called insulin-producing cells [IPCs]) (Figure 7A) (Park et al., 2008). We found that hugin-PC neurons make extensive synaptic contacts onto most but not all of the mNSCs (Figure 7B; Figure 2—figure supplement 1G,H). mNSCs of the pars intercerebralis derive from the same neuroectodermal placodes and develop through symmetric cell division (de Velasco et al., 2007). Among the mNSCs, IPCs have been best studied: they have ipsilateral descending arborizations into the SEZ and project contralaterally into the ring gland (Rulifson et al., 2002). In contrast, morphology of DH44- or DMS-producing mNSCs has been described in less detail. Our reconstruction showed that all reconstructed mNSCs have the exact same features, rendering them morphologically indistinguishable (Figure 7C). To assign identities to the reconstructed mNSCs, we hypothesized that similar to hugin neurons, the three types of mNSCs would differ in their choice of synaptic partners. We therefore reconstructed all presynaptic partners and calculated the connectivity similarity score between the mNSCs. Clustering with this similarity in connectivity resulted in three groups comprising 3, 2, and 7 neurons, coinciding with the number of neurons of the known types of mNSCs. We thus suggest that the group of three represents DH44-producing cells, the group of two represents DMS-producing cells and the group of seven represents the IPCs (Figure 7D).
On this basis, hugin-PC neurons make extensive synaptic contacts to the IPCs but less so to DMS- and DH44-producing mNSCs. In accordance with hugin-PC neurons using ACh as neurotransmitter, IPCs were previously shown to express a muscarinic ACh receptor (Cao et al., 2014). Whether additional ACh receptors are expressed is unknown. Overall, synapses between hugin-PC neurons onto mNSCs constitute a large fraction of their respective synaptic connections (hugin-PC: up to 35%; mNSCs: up to 17%). In support of a tight interconnection between hugin neurons and these neuroendocrine neurons, we found that most of hugin-PC neurons’ presynaptic partners are also presynaptic to mNSCs (Figure 7E). These findings demonstrate that the neuroendocrine system is a major target of hugin neurons.
Unlike the small molecule messengers used for fast synaptic transmission, neuropeptides – such as hugin – are thought to be released independent of synaptic membrane specializations and are able to diffuse a considerable distance before binding their respective receptors. However, it has been proposed that neuropeptides released from most neurons act locally on cells that are either synaptically connected or immediately adjacent (van den Pol 2012). We therefore asked whether the synaptic connections between hugin-PC neurons and mNSCs would have a matching peptide-receptor connection. The hugin gene encodes a prepropeptide that is post-translationally processed to produce an eight-amino-acid neuropeptide, termed pyrokinin-2 (hug-PK2) or hugin neuropeptide (Meng et al., 2002). This hugin neuropeptide has been shown to activate the Drosophila G-protein-coupled receptor (GPCR) CG8784/PK2-R1 in mammalian cell systems, but the identities of the target neurons expressing the receptor remain unknown (Rosenkilde et al., 2003). To address this, we used two independent methods to generate transgenic fly lines, CG8784-GAL4::p65 and CG8784-6kb-GAL4, driving expression under control of putative CG8784 regulatory sequences (Figure 8A; Figure 8—figure supplement 1). Both CG8784-GAL4 lines drive expression of a GFP reporter in a prominent cluster of cells in the pars intercerebralis. Double stainings show that this expression co-localizes with the peptides produced by the three types of mNSCs: Dilp2, DH44, and DMS (Figure 8B–D; Figure 8—figure supplement 1). To support the receptor expression data, we performed calcium (Ca2+) imaging of the mNSCs upon treatment with hug-PK2 (Figure 8—figure supplement 2). Indeed, calcium activity of the mNSCs increased significantly after treatment with concentrations of 1 µM hug-PK2 or higher. These findings support the hypothesis that hugin-PC neurons employ both classical synaptic transmission and peptidergic signaling to target neurons of the neuroendocrine center.
Neuropeptides are produced in the soma and packaged into dense core vesicles (DCVs) before being transported to their release sites (van den Pol, 2012). Exploring the spatial relationship between DCVs and synapses, we observed that for both interneuron type hugin classes (hugin-PC and hugin-VNC) DCVs localized close to but not exclusively at presynaptic sites (Figure 9A,B). This was often the case at local swellings along the main neurites which featured multiple pre- as well as postsynaptic sites, as well as close-by DCVs. It is conceivable that such complex local synaptic circuitry might enable local peptide release. We found 98% of the synapses between hugin-PC and mNSCs to have DCVs in proximity to the presynaptic sites, opening up the possibility of co-transmission of hugin peptide and ACh (Nusbaum et al., 2001) (Figure 9C). However, most DCVs were probably too distant from presynaptic sites to be synaptically released, suggesting para- and non-synaptic release (Morris and Pow, 1991; Maley, 1990) (Figure 2—figure supplement 1D).
Taken together, these findings show that the neuroendocrine system is indeed a major downstream target of hugin neurons and that this is achieved by a combination of synaptic and GPCR-mediated neuromodulatory transmission (Figure 9D).
Almost all neurons in Drosophila are uniquely identifiable and stereotyped (Vogelstein et al., 2014; Manning et al., 2012). This enabled us to identify and reconstruct a set of 20 peptidergic neurons in an ssTEM volume spanning an entire larval CNS (Ohyama et al., 2015). These neurons produce the neuropeptide hugin and have previously been grouped into four classes based on their projection targets (Figure 2A) (Bader et al., 2007a). We found that neurons of the same morphological class (a) were very similar with respect to the distribution of synaptic sites, (b) shared a large fraction of their pre- and postsynaptic partners and (c) in case of the interneuron classes (hugin-PC and hugin-VNC), neurons were reciprocally connected along their axons with other neurons of the same class. This raises the question why the CNS sustains multiple copies of morphologically very similar neurons. Comparable features have been described for a population of neurons which produce crustacean cardioactive peptide (CCAP) in Drosophila (Karsai et al., 2013). The reciprocal connections as well as the overlap in synaptic partners suggest that the activity of neurons within each interneuron class is likely coordinately regulated and could help sustain persistent activity within the population. In the mammalian pyramidal network of the medial prefrontal cortex, reciprocal connectivity between neurons is thought to contribute to the network’s robustness by synchronizing activity within subpopulations and to support persistent activity (Wang et al., 2006). Similar interconnectivity and shared synaptic inputs have also been demonstrated for peptidergic neurons producing gonadotropin-releasing hormone (GnRH) and oxytocin in the hypothalamus (Campbell et al., 2009; Theodosis, 2002). Likewise, this is thought to synchronize neuronal activity and allow periodic bursting.
Previous studies showed that specific phenotypes and functions can be assigned to certain classes of hugin neurons: hugin-VNC neurons increase locomotion motor rhythms but do not affect food intake, whereas hugin-PC neurons decrease food intake and are necessary for processing of aversive gustatory cues (Schoofs et al., 2014; Hückesfeld et al., 2016). For hugin-RG or hugin-PH such specific functional effects have not yet been described (for summary see Table 1). One conceivable scenario would be that each hugin class mediates specific aspects of an overarching 'hugin phenotype'. This would require that under physiological conditions all hugin classes are coordinately active. However, we did not find any evidence of such coordination on the level of synaptic connectivity. Instead, each hugin class forms an independent microcircuit with its own unique set of pre- and postsynaptic partners. We thus predict that each class of hugin-producing neurons has a distinct context and function in which it is relevant for the organism.
Data presented in this study provide the neural substrate for previous observation as well as open new avenues for future studies. One of the key features in hugin connectivity is the sensory input to hugin-PC, hugin-VNC and, to a lesser extent, hugin-RG. While hugin-PC neurons are known to play a role in gustatory processing, there is no detailed study of this aspect for hugin-VNC or hugin-RG neurons (Hückesfeld et al., 2016). Sensory inputs to hugin neurons are very heterogeneous, which suggests that they have an integrative/processing rather than a simple relay function.
Hugin neurons also have profound effects on specific motor systems: hugin-PC neurons decelerate motor patterns for pharyngeal pumping whereas hugin-VNC neurons accelerate locomotion motor patterns (Schoofs et al., 2014; Hückesfeld et al., 2016). For hugin-PC, we have demonstrated that this effect is mediated by both synaptic and hugin peptide transmissions. For hugin-VNC, this effect is independent of the hugin neuropeptide, suggesting synaptic transmission to play a key role (Schoofs et al., 2014). Suprisingly, we did not find any direct synaptic connections to the relevant motor neurons. However, the kinetics of the effects of hugin neurons on motor systems have not yet been studied at a high enough temporal resolution (i.e. by intracellular recordings) to assume monosynaptic connections. It is thus well conceivable that connections to the respective motor systems are polysynaptic and occur further downstream. Alternatively, this may involve an additional non-synaptic (peptidergic) step. A strong candidate for this is the neuroendocrine system which we identify as the major downstream target of hugin-PC neurons. Among the endocrine targets of hugin, the insulin-producing cells (IPCs) have long been known to centrally regulate feeding behavior (Erion and Sehgal, 2013). It is not known if insulin-signaling directly affects motor patterns in Drosophila. Nevertheless, increased insulin signaling has strong inhibitory effects on food-related sensory processing and feeding behavior (Wu et al., 2005a; Wu et al., 2005b). Whether the neuroendocrine system is a mediator of the suppressive effects of hugin-PC neurons on food intake remains to be determined.
The first functional description of hugin in Drosophila was done in larval and adult (Melcher and Pankratz, 2005), while more recent publications have focused entirely on the larva (Schoofs, 2014; Hückesfeld et al., 2016). One of the main reasons for this is the smaller behavioral repertoire of the larva: the lack of all but the most fundamental behaviors makes it well suited to address basic questions. Nevertheless, it stands to reason that elementary circuits should be conserved between larval and adult flies. To date, there is no systematic comparison of hugin across the life cycle of Drosophila. However, there is indication that hugin neurons retain their functionality from larva to the adult fly. First, morphology of hugin neurons remains virtually the same between larva and adults (Melcher and Pankratz, 2005). Second, hugin neurons seem to serve similar purposes in both stages: they acts as a brake on feeding behavior – likely as response to aversive sensory cues (Hückesfeld et al., 2016). In larvae, artificial activation of this brake shuts down feeding (Schoofs et al., 2014). In adults, removal of this break by silencing of hugin neurons leads to a facilitation (earlier onset) of feeding (Melcher and Pankratz, 2005). Such conservation of neuropeptidergic function between larval and adult Drosophila has been observed only in a few cases. Prominent examples are short (Lee et al., 2004, 2008) and long neuropeptide F (Beshel and Zhong, 2013; Wang et al., 2013), both of which show strong similarities with mammalian NPY. The lack of additional examples is not necessarily due to actual divergence of peptide function but rather due to the lack of data across both larva and adult. Given the wealth of existing data on hugin in larvae, it would be of great interest to investigate whether and to what extent the known features (connectivity, function, etc.) of this system are maintained throughout Drosophila’s life history.
A neural network is a highly dynamic structure and is subject to constant change, yet it is constrained by its connectivity and operates within the framework defined by the connections made between its neurons (Getting, 1989). On one hand, this connectivity is based on anatomical connections formed between members of the network, namely synapses and gap junctions. On the other hand, there are non-anatomical connections that do not require physical contact due to the signaling molecules, such as neuropeptides/-hormones, being able to travel considerable distances before binding their receptors (van den Pol, 2012). Our current integrated analysis of the operational framework for a set of neurons genetically defined by the expression of a common neuropeptide, positions hugin-producing neurons as a novel component in the regulation of neuroendocrine activity and the integration of sensory inputs. We show that most hugin neurons receive chemosensory input in the subesophageal zone, the brainstem analog of Drosophila (Ghysen, 2003; Schoofs et al., 2014). Of these, one class is embedded into a network whose downstream targets are median neurosecretory cells (mNSCs) of the pars intercerebralis, a region homologous to the mammalian hypothalamus (Hartenstein, 2006). We found that hugin neurons target mNSCs by two mechanisms. First, by classic synaptic transmission as our data strongly suggest that acetylcholine (ACh) acts as transmitter at these synapses. Accordingly, subsets of mNSCs have been shown to express a muscarinic ACh receptor (Cao et al., 2014). Whether additional ACh receptors are expressed is unknown. Second, by non-anatomical, neuromodulatory transmission using a peptide-receptor connection, as demonstrated by the expression of hugin G-protein-coupled receptor PK2-R1 (CG8784) in mNSCs. Strikingly, while PK2-R1 is expressed in all mNSCs, the hugin neurons have many synaptic contacts onto insulin-producing cells but few to DMS and DH44 neurons. This mismatch in synaptic vs. peptide targets among the mNSCs suggests an intricate influence of hugin-producing neurons on this neuroendocrine center. In favor of a complex regulation is that those mNSCs that are synaptically connected to hugin neurons additionally express a pyrokinin-1 receptor (PK1-R, CG9918) which, like PK2-R1, is related to mammalian neuromedinU receptors (Alfa et al., 2015; Cazzamali et al., 2005; Park et al., 2002). There is some evidence that PK1-R might also be activated by the hugin neuropeptide, which would add another regulatory layer (Cazzamali, 2005).
The concept of multiple messenger molecules within a single neuron is well established and appears to be widespread among many organisms and neuron types (Burnstock, 2004; Nusbaum et al., 2001; Merighi, 2002; Brezina, 2010; Li and Kim, 2008). For example, cholinergic transmission plays an important role in mediating the effect of NMU in mammals. This has been demonstrated in the context of anxiety but not yet for feeding behavior (Telegdy and Adamik, 2013; Tanaka and Telegdy, 2014). There are, however, only few examples of simultaneous employment of neuromodulation and fast synaptic transmission in which specific targets of both messengers have been investigated at single-cell level. In many cases, targets and effects of classic and peptide co-transmitters seem to diverge (e.g. (Sun et al., 2003; Li and van den Pol, 2006; Stein et al., 2007). In contrast, AgRP neurons in the mammalian hypothalamus employ neuropeptide Y, the eponymous agouty-related protein (AgRP) and the small molecule transmitter GABA to target pro-opiomelanocortin (POMC) neurons in order to control energy homeostasis (Cansell et al., 2012). Also, reminiscent of our observations is the situation in the frog sympathetic ganglia, where preganglionic neurons use both ACh and a neuropeptide to target so-called C cells but only the neuropeptide additionally targets B cells. In both targets, the neuropeptide elicits late, slow excitatory postsynaptic potentials (EPSPs) (Jan and Jan, 1983). It is conceivable that hugin-producing neurons act in a similar manner by exerting a slow, lasting neuromodulatory effect on all mNSCs and a fast, transient effect exclusively on synaptically connected mNSCs. Alternatively, the hugin neuropeptide could facilitate the postsynaptic effect of acetylcholine. Such is the case in Aplysia where a command-like neuron for feeding employs acetylcholine and two neuropeptides, feeding circuit activating peptide (FCAP) and cerebral peptide 2 (CP2). Both peptides work cooperatively on a postsynaptically connected motor neuron to enhance EPSPs in response to cholinergic transmission (Koh et al., 2003).
In addition to the different timescales that neuropeptides and small molecule transmitters operate on, they can also be employed under different circumstances. It is commonly thought that low-frequency neuronal activity is sufficient to trigger fast transmission using small molecule transmitters, whereas slow transmission employing neuropeptides requires higher frequency activity (Nusbaum et al., 2001). Hugin-producing neurons could employ peptidergic transmission only as a result of strong excitatory (e.g. sensory) input. There are, however, cases in which base activity of neurons is already sufficient for graded neuropeptide release: Aplysia ARC motor neurons employ ACh as well as neuropeptides and ACh is generally released at lower firing rates than the neuropeptide. This allows the motor neuron to function as purely cholinergic when firing slowly and as cholinergic/peptidergic when firing rapidly (Whim and Lloyd, 1989). However, peptide release already occurs at the lower end of the physiological activity of those neurons (Weiss et al., 1993; Vilim et al., 1996). It remains to be seen how synaptic and peptidergic transmission in hugin neurons relate to each other.
The present study is one of very few detailed descriptions of differential targets of co-transmission and – to our knowledge – the first of its kind in Drosophila. We hope these findings in a genetically tractable organism will provide a basis for elucidating some of the intriguing modes of action of peptidergic neurons.
The mammalian homolog of hugin, neuromedinU (NMU), is found in the CNS as well as in the gastrointestinal tract (Ballesta et al., 1988). Its two receptors, NMUR1 and NMUR2, show differential expression. NMUR2 is abundant in the brain and the spinal cord, whereas NMUR1 is expressed in peripheral tissues, in particular in the gastrointestinal tract (Mitchell et al., 2009). Both receptors mediate different effects of NMU. The peripheral NMUR1 is expressed in pancreatic islet β cells in humans and allows NMU to potently suppress glucose-induced insulin secretion (Alfa et al., 2015). The same study also showed that Limostatin (Lst) is a functional homolog of this peripheral NMU in Drosophila: Lst is expressed by glucose-sensing, gut-associated endocrine cells and suppresses the secretion of insulin-like peptides. The second, centrally expressed NMU receptor, NMUR2, is necessary for the effect of NMU on food intake and physical activity (Zeng et al., 2006; Peier et al., 2009). In this context, NMU is well established as a factor in regulation of the hypothalamo-pituitary axis (Wren et al., 2002; Malendowicz et al., 2012) and has a range of effects in the hypothalamus, the most important being the release of corticotropin-releasing hormone (CRH) (Hanada et al., 2001, 2003). We show that a subset of hugin-producing neurons targets the pars intercerebralis, the Drosophila homolog of the hypothalamus, in a similar fashion: neuroendocrine target cells in the pars intercerebralis produce a range of peptides, including diuretic hormone 44 which belongs to the insect CRH-like peptide family (Cabrero et al., 2002) (Figure 10). Given these similarities, we propose that hugin is homologous to central NMU just as Lst is a homologous to peripheral NMU. Demonstration that central NMU and hugin circuits share similar features beyond targeting neuroendocrine centers, e.g. the integration of chemosensory inputs, will require further studies on NMU regulation and connectivity.
Previous work on vertebrate and invertebrate neuroendocrine centers suggests that they evolved from a simple brain consisting of cells with dual sensory/neurosecretory properties, which later diversified into optimized single-function cells (Tessmar-Raible et al., 2007). There is evidence that despite the increase in neuronal specialization and complexity, connections between sensory and endocrine centers have been conserved throughout evolution (Yoon et al., 2005; Strausfeld 2012; Abitua et al., 2015). We propose that the connection between endocrine and chemosensory centers provided by hugin neurons represents such a conserved circuit that controls basic functions like feeding, locomotion, energy homeostasis and sex.
Indisputably, the NMU system in mammals is much more complex as NMU is found more widespread within the CNS and almost certainly involves a larger number of different neuron types. This complexity, however, only underlines the use of numerically smaller nervous systems such as Drosophila's to generate a foundation to build upon. Moreover, NMU/NMU-like systems may have similar functions not just in mammals and Drosophila but also other vertebrates such as fish (Chiu et al., 2016; Li et al., 2015) and other invertebrates such as C. elegans (Maier et al., 2010). In summary, our findings should encourage research in other organisms, such as the involvement of NMU and NMU homologs in relaying chemosensory information onto endocrine systems, and more ambitiously, to elucidate their connectomes in order to allow comparative analyses of the underlying network architecture.
Reconstructions were based on an ssTEM (serial section transmission electron microscope) data set comprising an entire central nervous system and the ring gland of a first-instar Drosophila larva. Generation of this data set was described previously (Ohyama et al., 2015). Neurons’ skeletons were manually reconstructed using a modified version of CATMAID (http://www.catmaid.org) (Saalfeld et al., 2009). Hugin-PH (pharynx) neurons were first identified by reconstructing all axons in the prothoracic accessory nerve, through which these neurons exit the CNS toward the pharynx. Similarly, hugin-RG (ring gland) neurons were identified by reconstructing all neurosecretory cells that target the ring gland. To find the remaining hugin neurons, neighbors of already identified hugin neurons were reconstructed. Among those, the remaining hugin neurons were unambiguously identified based on previously described morphological properties such as projection targets, dendritic arborizations, relative position to each other and prominent landmarks like antennal lobes or nerves (Bader et al., 2007a, Bader et al., 2007b). The mapped synaptic connections represent fast, chemical synapses matching previously described typical criteria: thick black active zones, pre- (e.g. T-bar, vesicles) and postsynaptic membrane specializations (Prokop and Meinertzhagen, 2006). Hugin inputs and outputs were traced by following the pre- and postsynaptically connected neurites to the respective neurons’ somata or nerve entry sites in sensory axons. Subsequently, all sensory and endocrine neurons synaptically connected to hugin neurons were fully reconstructed. Interneurons were fully reconstructed if (a) homologous neurons were found in both hemispheres/-segments (did not apply to medially unpaired neurons) and (b) at least one of the paired neurons was connected by a minimum of three synapses to/from hugin neurons. Neurons that did not fit either criterion were not fully reconstructed and thus excluded from statistical analysis. This resulted in the reconstruction 177 synaptic partners that together covered 90%/96% of hugin neurons' above threshold pre-/postsynaptic sites (Figure 5—figure supplement 1). The same parameters were applied to the reconstruction of synaptic partners of median neurosecretory cells (mNSCs). Morphological plots and example synapse's volume reconstruction were generated using custom python scripts or scripts for Blender 3D (www.blender.org). The script for a CATMAID-Blender interface is on Github (https://github.com/schlegelp/CATMAID-to-Blender). See supplemental neuron atlas (Supplementary files 1,2) of all reconstructed neurons and their connectivity with hugin neurons.
With the overall connectivity similarity score for vertices i and j in adjacency matrix A being the sum of over all connected partners k. and are variables that determine how similar two vertices have to be and how negatively a dissimilarity is punished. Values used were: and . To simplify graphical representation, we normalized the overall similarity score to the minimal (sum of over all k) and maximal (sum of over all k) achievable values, so that the similarity score remained between 0 and 1. Self-connections () and connections were ignored.
To calculate similarity of synapse placement between two neurons, we calculated the synapse similarity score (Figure 6D):
With the overall synapse similarity score for neurons i and j being the average of over all synapses s of i. Synapse k being the closest synapse of neuron j to synapses s [same sign (pre-/postsynapse) only]. being the linear distance between synapses s and k. Variable determines which distance between s and k is considered as close. and are defined as the number of synapses of neuron / that are within a radius of synapse k and s, respectively (same sign only). This ensures that in case of a strong disparity between and , will be close to zero even if distance is very small. Values used: == 2000 nm.
Percentage of synaptic connections was calculated by counting the number of synapses that constitute connections between neuron A and a given set of pre- or postsynaptic partners (e.g. sensory neurons) divided by the total number of either incoming or outgoing synaptic connections of neuron A. For presynaptic sites, each postsynaptic neurite counted as a single synaptic connection.
The CG8784-GAL4::p65 construct (Figure 8) was created using recombineering techniques (Warming et al., 2005) in P[acman] bacterial artificial chromosome (BAC) clone CH321-45L05 (Venken et al., 2009) (obtained from Children’s Hospital Oakland Research Institute, Oakland, CA), containing CG8784 within ~80 kb of flanking genomic context. A generic landing-site vector was created by flanking the kanamycin-resistance/ streptomycin-sensitivity marker in pSK+-rpsL-kana (Wang et al., 2009) (obtained from AddGene.org, plasmid #20871) with 5 ‘and 3’ homology arms (containing GAL4 coding sequences and HSP70 terminator sequences, respectively) amplified from pBPGUw (Pfeiffer et al., 2008). CG8784-specific homology arms were added to this cassette by PCR using the following primers (obtained as Ultramers from Integrated DNA Technologies, Inc., Coralville, Iowa; the lower case portions are CG8784-specific targeting sequences, and the capitalized portions match the pBPGUw homology arms):
This cassette was recombined into the BAC, replacing the coding portion of the first coding exon, and then full-length GAL4::p65-HSP70 amplified from pBPGAL4.2::p65Uw (Pfeiffer et al., 2010) was recombined into the landing site in a second recombination. Introns and exons following the insertion site were retained in case they contain expression-regulatory sequences, although they are presumably no longer transcribed. Correct recombination was verified by sequencing the recombined regions, and the final BAC was integrated into the third-chromosome attP site VK00033 (Venken et al., 2006) by Rainbow Transgenic Flies, Inc. (Camarillo, CA).
The CG8784-6kb-GAL4 (Figure 8—figure supplement 1) was created using standard restriction-digestion/ligation techniques in pCaSpeR-AUG-Gal4-X vector (Vosshall et al., 2000). An approximately 6 kb promoter fragment 5’ of the first coding exon was amplified using the following primers and inserted into a pCaSpeR vector (Addgene.org, plasmid #8378) containing a start codon (AUG) and the GAL4 gene (Figure 8—figure supplement 1).
This construct was integrated into the genome via P-element insertion.
For antibody stainings of CG8784-GAL4::p65, larvae expressing JFRC2-10xUAS-IVS-mCD8::GFP (Pfeiffer et al., 2010) driven by CG8784-GAL4::p65 were dissected in PBS. Brains were fixed in 4% formaldehyde in PBS for 1 hr, rinsed, blocked in 5% normal goat serum, and incubated overnight at 4°C with primaries: sheep anti-GFP (AbD Serotec #4745–1051), 1:500; rabbit anti-DH44 (Cabrero et al., 2002) (gift of Jan Veenstra), 1:1000; rabbit anti-DILP2 (Veenstra et al., 2008) (gift of Jan Veenstra), 1:1000; 1:1000; and rabbit anti-DMS (Schoofs et al., 1993; Park et al., 2008) (gift of Luc van den Bosch and Liliane Schoofs), 1:500. Tissues were rinsed and incubated overnight at 4°C in secondaries: Alexa Fluor 488 donkey anti-sheep (Jackson ImmunoResearch, #713–545-147) and rhodamine red-X donkey anti-rabbit (Jackson ImmunoResearch #711–296-152), both 1:500. Brains were rinsed and dehydrated through an ethanol-xylene series, mounted in DPX, and scanned on a Zeiss LSM 510 confocal microscope.
For antibody stainings of CG8784-6kb-GAL4, larvae expressing 10XUAS-mCD8::GFP (Bloomington, #32184) driven by CG8784-6kb-GAL4 were dissected in PBS. Brains were fixed in 4% paraformaldehyde for 30 min, rinsed, blocked in 5% normal goat serum, and incubated overnight at 4°C with primaries: goat anti-GFP-FITC (abcam, ab26662), 1:500; rabbit anti-DH44 (Cabrero et al., 2002) (gift of Jan Veenstra), 1:1000; guinea pig anti-Dilp2 (Bader et al., 2013) (Pankratz lab), 1:500 and rabbit anti-DMS (Schoofs et al., 1993; Park et al., 2008) (gift of Luc van den Bosch and Liliane Schoofs), 1:500. Tissues were rinsed and incubated overnight at 4°C in secondaries: anti-rabbit Alexa Fluor 633 (Invitrogen, A-21070) and anti-guinea pig Alexa Fluor 568 (Invitrogen, A-11075), both 1:500. Brains were rinsed, mounted in Mowiol (Roth, 0713), and scanned on a Zeiss LSM 710 confocal microscope.
For antibody stainings against choline acetyltransferase (ChAT), larvae expressing a YFP-tagged halorhodopsin (UAS- eNpHR-YFP; Bloomington, #41753) driven by HugS3-GAL4 (Melcher and Pankratz, 2005) as marker were prepared following the above protocol for CG8784-6kb-GAL4 stainings. Primary antibodies used: goat anti-GFP-FITC (abcam, ab26662), 1:500; mouse anti-ChAT (Developmental Studies Hybridoma Bank, ChAT4B1) (Takagawa and Salvaterra, 1996), 1:1000. Secondary antibodies used: anti-mouse Alexa Fluor 633 (Invitrogen, A-21046).
For investigation of ChAT promoter activity in hugin neurons, larvae expressing UAS-cd8a::mRFP (Bloomington, #27399) under the control of ChAT-GAL4 7.4 kb (Bloomington, #6798) and YFP directly under the control of the hugin promoter (hug-YFP; (Melcher and Pankratz, 2005) were prepared following the above protocol for CG8784-6kb-GAL4 stainings. Primary antibodies used: goat anti-GFP-FITC (abcam, ab26662), 1:500; mouse anti-RFP (abcam, ab65856), 1:500. Secondary antibodies used: anti-mouse Alexa Fluor 633 (Invitrogen, A-21046).
For quantification of ChAT antibody signals/ChAT promoter activity, samples were scanned on a Zeiss LSM 710 confocal microscope using a 63X objective (Zeiss). Settings were kept the same over all scans. Regions of interest were placed through the center of each hugin neuron’s soma, and the mean intensity was measured using ImageJ (https://imagej.nih.gov/ij/index.html) (Schneider et al., 2012). Hugin-PC and hugin-RG neurons were identified based on soma position and morphology. Hugin-VNC and hugin-PH could not be unambiguously discriminated as they were usually too tightly clustered. They were thus treated as a single group. For background normalization, an approximately 10×10 μm rectangle from the center of the image stack was chosen.
To investigate the role of acetylcholine as transmitter of hugin neurons, food intake and electrophysiological experiments were performed. Experimental procedures, materials and setups used in these assays been described extensively in Schoofs et al. (2014). The hugin and ChAT RNAi experiments presented in Figure 3 were performed together as part of a larger screen on neuronal populations and genes involved in larval feeding behavior. A portion of this screen, which did not include the ChAT RNAi data that we are now presenting here, was published previously in Schoofs et al. (2014). In the following, we briefly summarize procedures of Schoofs et al. (2014) for reader convenience. Please see that reference for more detailed description.
The following GAL4 driver and UAS effector lines were used: HugS3-GAL4 (Melcher and Pankratz 2005), UAS-dTrpA1 (Bloomington, #26263), UAS-LacZRNAi (gift from M. Jünger), UAS-HugRNAi1A (Schoofs et al., 2014) and UAS-ChAT-RNAi (TriP.JF01877) (Bloomington, #25856) (Barnstedt et al., 2016; Plaçais et al., 2013). OregonR and OregonR x UAS-dTrpA1 were used as control.
For the food intake assay, third instar larvae were first washed and starved for 30 min on RT. They were then transferred on yeast paste colored with crimson red and allowed to feed for 20 min. Experiments were performed at 32°C for dTrpA1-induced activation of hugin neurons and at 18°C as control condition. Afterwards larvae were photographed and the amount of food ingested was calculated as the area of the alimentary tract stained by the colored yeast divided by body surface area using ImageJ (https://imagej.nih.gov/ij/index.html) (Schneider et al., 2012). Data are represented as fold change between control condition (18°C) and dTrpA1-induced activation (32°C) normalized to the control.
For the electrophysiological assay, semi-intact preparations of third instar larvae were made in saline solution (Rohrbough and Broadie, 2002). En passant extracellular recordings of the antennal nerve (AN) were performed following previously described protocol (Schoofs et al., 2014). During the recordings, temperature of the CNS was alternated between 18°C (control condition) and 32°C (dTrpA1 activation). For analysis, fictive motor patterns of the pharyngeal pump (also: cibarial dilator musculature, CDM) were analyzed: fold change in cycle frequency between pairs of successive 18°C and 32°C sections of a recording was calculated.
Hugin-derived pyrokinin 2 (hug-PK2) was synthesized by Iris Biotech (Marktredwitz, Germany) using the amino acid sequence SVPFKPRL-NH2. The C terminus was amidated. The effect of hug-PK2 on calcium activity in median neurosecretory cells (mNSCs) was investigated using the calcium integrator CaMPARI (Fosque et al., 2015). To drive expression of CaMPARI in mNSCs, CG8784-6kb-GAL4 flies were crossed to UAS-CaMPARI (Bloomington #58761). Larval brains were dissected and placed in saline solution (Rohrbough and Broadie, 2002) containing either no, 100 nM, 1 µM or 10 µM hug-PK2. After 1 min of incubation, 405 nm photoconversion light was applied for 15 s. Afterwards, brains were placed on a poly-l-lysine-coated (Sigma-Aldrich, P8920) cover slide and scanned using a Zeiss LSM 780 confocal microscope. Settings were kept the same over all scans. Calcium activity was calculated as the ratio of the fluorescence of photoconverted (red) to unconverted (green) CaMPARI using ImageJ.
Genetic dissection of neural circuit anatomy underlying feeding behavior in Drosophila: distinct classes of hugin-expressing neuronsThe Journal of Comparative Neurology 502:848–856.https://doi.org/10.1002/cne.21342
The IGFBP7 homolog Imp-L2 promotes insulin signaling in distinct neurons of the Drosophila brainJournal of Cell Science 126:2571–2576.https://doi.org/10.1242/jcs.120261
Graded encoding of food odor value in the Drosophila brainJournal of Neuroscience 33:15693–15704.https://doi.org/10.1523/JNEUROSCI.2605-13.2013
Morphological differentiation of the embryonic peripheral neurons in DrosophilaRoux's Archives of Developmental Biology 196:69–77.https://doi.org/10.1007/BF00402027
Beyond the wiring diagram: signalling through complex neuromodulator networksPhilosophical Transactions of the Royal Society B: Biological Sciences 365:2363–2374.https://doi.org/10.1098/rstb.2010.0105
Arcuate AgRP neurons and the regulation of energy balanceFrontiers in Endocrinology 3:1–7.https://doi.org/10.3389/fendo.2012.00169
The Drosophila gene CG9918 codes for a pyrokinin-1 receptorBiochemical and Biophysical Research Communications 335:14–19.https://doi.org/10.1016/j.bbrc.2005.07.038
Architecture of the primary taste center of Drosophila melanogaster larvaeThe Journal of Comparative Neurology 502:834–847.https://doi.org/10.1002/cne.21312
Regulation of insect behavior via the insulin-signaling pathwayFrontiers in Physiology 4:1–6.https://doi.org/10.3389/fphys.2013.00353
Emerging principles governing the operation of neural networksAnnual Review of Neuroscience 12:185–204.https://doi.org/10.1146/annurev.ne.12.030189.001153
Sensory neurons and peripheral pathways in Drosophila embryosRoux's Archives of Developmental Biology 195:281–289.https://doi.org/10.1007/BF00376060
Mini-review: the evolution of neuropeptide signalingRegulatory Peptides 177 (Suppl):S6–S9.https://doi.org/10.1016/j.regpep.2012.05.001
Central actions of neuromedin U via corticotropin-releasing hormoneBiochemical and Biophysical Research Communications 311:954–958.https://doi.org/10.1016/j.bbrc.2003.10.098
The neuroendocrine system of invertebrates: a developmental and evolutionary perspectiveJournal of Endocrinology 190:555–570.https://doi.org/10.1677/joe.1.06964
Nociceptive neurons protect Drosophila larvae from parasitoid waspsCurrent Biology 17:2105–2116.https://doi.org/10.1016/j.cub.2007.11.029
Two neuropeptides colocalized in a command-like neuron use distinct mechanisms to enhance its fast synaptic connectionJournal of Neurophysiology 90:2074–2079.https://doi.org/10.1152/jn.00358.2003
Molecular and cellular organization of the taste system in the Drosophila larvaJournal of Neuroscience 31:15300–15309.https://doi.org/10.1523/JNEUROSCI.3363-11.2011
Drosophila short neuropeptide F regulates food intake and body sizeJournal of Biological Chemistry 279:50781–50789.https://doi.org/10.1074/jbc.M407842200
WormBook : the online review of C. elegans biologyNeuropeptides 212:1–36.
Distinct functions of neuromedin u and neuromedin s in orange-spotted grouperJournal of Molecular Endocrinology 55:95–106.https://doi.org/10.1530/JME-15-0018
Ultrastructural identification of neuropeptides in the central nervous systemJournal of Electron Microscopy Technique 15:67–80.https://doi.org/10.1002/jemt.1060150107
The Drosophila hugin gene codes for myostimulatory and ecdysis-modifying neuropeptidesMechanisms of Development 117:5–13.https://doi.org/10.1016/S0925-4773(02)00175-2
Costorage and coexistence of neuropeptides in the mammalian CNSProgress in Neurobiology 66:161–190.https://doi.org/10.1016/S0301-0082(01)00031-4
Emerging pharmacology and physiology of neuromedin U and the structurally related peptide neuromedin SBritish Journal of Pharmacology 158:87–103.https://doi.org/10.1111/j.1476-5381.2009.00252.x
Neural architecture of the primary gustatory center of Drosophila melanogaster visualized with GAL4 and LexA enhancer-trap systemsThe Journal of Comparative Neurology 518:4147–4181.https://doi.org/10.1002/cne.22433
Central effects of neuromedin U in the regulation of energy homeostasisBiochemical and Biophysical Research Communications 277:191–194.https://doi.org/10.1006/bbrc.2000.3669
The roles of co-transmission in neural network modulationTrends in Neurosciences 24:146–154.https://doi.org/10.1016/S0166-2236(00)01723-9
A comparative review of short and long neuropeptide F signaling in invertebrates: Any similarities to vertebrate neuropeptide Y signaling?Peptides, 32, 10.1016/j.peptides.2011.03.013, 21440021.
Drosophila neuropeptides in regulation of physiology and behaviorProgress in Neurobiology 92:42–104.https://doi.org/10.1016/j.pneurobio.2010.04.010
Development and structure of synaptic contacts in DrosophilaSeminars in Cell & Developmental Biology 17:20–30.https://doi.org/10.1016/j.semcdb.2005.11.010
Molecular cloning, functional expression, and gene silencing of two Drosophila receptors for the Drosophila neuropeptide pyrokinin-2Biochemical and Biophysical Research Communications 309:485–494.https://doi.org/10.1016/j.bbrc.2003.08.022
Drosophila cholinergic neurons and processes visualized with Gal4/UAS-GFPGene Expression Patterns 1:73–82.https://doi.org/10.1016/S1567-133X(01)00011-4
Divergent co-transmitter actions underlie motor pattern activation by a modulatory projection neuronEuropean Journal of Neuroscience 26:1148–1165.https://doi.org/10.1111/j.1460-9568.2007.05744.x
Arthropod Brains : Evolution, Functional Elegance, and Historical SignificanceCambridge, Mass: Harvard University Press.
Cerebral hemisphere regulation of motivated behaviorBrain Research 886:113–164.https://doi.org/10.1016/S0006-8993(00)02905-X
Neurotransmissions of antidepressant-like effects of neuromedin U-23 in miceBehavioural Brain Research 259:196–199.https://doi.org/10.1016/j.bbr.2013.11.005
Anxiolytic action of neuromedin-U and neurotransmitters involved in miceRegulatory Peptides 186:137–140.https://doi.org/10.1016/j.regpep.2013.07.008
Oxytocin-secreting neurons: A physiological model of morphological neuronal and glial plasticity in the adult hypothalamusFrontiers in Neuroendocrinology 23:101–135.https://doi.org/10.1006/frne.2001.0226
Molecular architecture of smell and taste in DrosophilaAnnual Review of Neuroscience 30:505–533.https://doi.org/10.1146/annurev.neuro.30.051606.094306
A new positive/negative selection scheme for precise BAC recombineeringMolecular Biotechnology 42:110–116.https://doi.org/10.1007/s12033-009-9142-3
Heterogeneity in the pyramidal network of the medial prefrontal cortexNature Neuroscience 9:534–542.https://doi.org/10.1038/nn1670
Simple and highly efficient BAC recombineering using galK selectionNucleic Acids Research 33:1–12.https://doi.org/10.1093/nar/gni035
Physiology and biochemistry of peptidergic cotransmission in AplysiaJournal of Physiology Paris 87:141–151.https://doi.org/10.1016/0928-4257(93)90025-O
Regulation of aversion to noxious food by Drosophila neuropeptide Y- and insulin-like systemsNature Neuroscience 8:1350–1355.https://doi.org/10.1038/nn1540
Localization of choline acetyltransferase-expressing neurons in Drosophila nervous systemMicroscopy Research and Technique 45:65–79.https://doi.org/10.1002/(SICI)1097-0029(19990415)45:2<65::AID-JEMT2>3.0.CO;2-0
Neuromedin U receptor 2-deficient mice display differential responses in sensory perception, stress, and feedingMolecular and Cellular Biology 26:9352–9363.https://doi.org/10.1128/MCB.01148-06
Ronald L CalabreseReviewing Editor; Emory University, United States
In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included.
Thank you for submitting your article "Synaptic transmission parallels neuromodulation in a central food-intake circuit" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor, Ronald L. Calabrese, and Eve Marder as the Senior Editor. The reviewers have opted to remain anonymous.
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
The paper by Schlegel et al. describes the circuitry of hugin-expressing neurons in the Drosophila larva. The neuron reconstructions and the analysis of connectivity are an impressive body of work. The authors combined it with the detailed analysis of the chemical connectivity of one class of hugin-producing neurons and their targets expressing the GPCR for hugin. The possibility of directly correlating individual neurons in light microscopy samples to the EM volume allowed the analysis of such molecular connectivity in the circuit at a cellular resolution. The 20 hugin neurons make up 4 distinct microcircuits: the activity of each of these is likely to underlie distinct features of feeding behavior.
There are some reservations about this paper that we ask the authors to address in revision.
1) The 4 hugin microcircuits likely have different functions, but are there any indications of what these might be? Can the authors provide any evidence that might begin to identify these functions? Is there evidence supporting a feeding-related function for each of the 4 hugin modules? Commonly, neuropeptides (like other transmitter molecules) subserve neural signaling in numerous unrelated contexts.
2) There is concern that the homology/analogy with the mammalian NMU neuron is overemphasized. Quoting from one of the expert reviews, "The authors compare the circuitry of hugin neurons to the connectivity of NMU neurons in vertebrates. Homologizing neuron types and brain regions across large evolutionary distances is difficult. The authors are apparently aware of these difficulties and sometimes use the term homology and sometimes analogy. These are of course not the same.
One criterion for establishing homologous neuron types across phyla would be that the neurons express a similar set of transcription factors and effector genes and are located in brain areas that as regions are possibly homologous (e.g., the spinal cord and the VNC).
The difficulty in comparing the fly hugin neurons to the NMU neurons in vertebrates is that in vertebrates the expression pattern of NMU is more complex. NMU neurons can be found in the hypothalamus, the pituitary, the brainstem, the spinal cord, and throughout the gastrointestinal tract. However, in the fly, hugin+ neurons are only found in the SEZ. These project to the VNC, PC and RG, but there are no hugin+ somata in these regions. The authors compare the projections of the hugin neurons to the distinct sets of NMU+ neurons in vertebrates. This is problematic. It is only the hugin+ cells in the SEZ in Drosophila that may be comparable to NMU+ cells in the brainstem in vertebrates. In this sense, Figure 1 is misleading. The authors must be more careful about this and explain more carefully what they are trying to compare to what and based on which criteria. Regarding a comparison of circuitry, it is even more problematic, since not much is known about the circuitry of brainstem NMU neurons in vertebrates.
The similarity in chemical connectivity of hugin/NMU-DH44/CRF neurons between the fly and vertebrates is intriguing. However, it is again difficult to directly compare these circuits, since there are no hugin+ neurons in the pars intercerebralis that would be equivalent to the hypothalamic NMU neurons."
3) Co-transmission as a theme is only partially supported and weakly contextualized. Quoting from one of the expert reviewers, "The conclusions regarding the functional consequences of the co-localized (not yet identified) small molecule transmitter and hugin peptide are only likely or possible ones, not firm ones. For example, the extent to which the influence of the hugin neurons, and particularly that of any one of the 4 modules, mimics the previously documented behavioral consequences of genetically manipulating the presence of the hugin peptide remains to be determined. Also, the proposed target neurons of neuronally released hugin peptide are the likely but not necessarily actual target neurons, despite the presynaptic presence of hugin peptide and postsynaptic presence of hugin receptors. This is because 1) the peptide release sites are not known, other than that they are not released into a synaptic cleft, 2) peptide access to receptors is unknowable from anatomy alone, 3) there might be additional, not yet identified hugin receptors, and 4) the presence of receptors alone does not prove a functional relationship because those receptors might be present to bind instead with different members of that peptide family released from other neurons. Concisely, anatomical results suggest but cannot prove functional consequences of peptidergic transmission. This latter point is in fact made explicitly/appropriately at several locations in the manuscript."
Another expert reviewer added, "The (putative) presence of both fast and slow synaptic transmission in these neurons isn't surprising, as it is something common to many neuropeptide neurons. Still, this is an important observation, and it should prompt a discussion about the role of neurotransmitters vs. peptide neuromodulators in feeding behavior. In the mammalian hypothalamus, GABA transmission is thought to be the main player in feeding regulation in the POMC and NPY neurons. Adding a few sentences on this would enrich the discussion. Importantly, the identity of the neurotransmitters involved in fast synaptic transmission isn't mentioned." (Is there any evidence that the small molecule co-transmitter is the same for each hugin neuron, both within and between modules?) "This seems a big omission given the scope of the article and that it would be straightforward experiment to do; I would encourage the authors to strengthen the manuscript by addressing these experiments at least in the context of a few of the Hugin microcircuits."
On this same point the first of the reviewers added, "The concept of co-transmission, a pivotal aspect of this paper as evident from its title, has been in the literature for several decades yet it does remain an under-developed area of study. Nevertheless, there are a number of published studies of the physiological influence of identified small molecule and neuropeptide co-transmitters on neural networks at the single cell level [none of which are referenced in this paper (?)]." Given the presence of this small but actual literature, laying out the context in which the hugin system can contribute would seem important, and its omission is unscholarly.
4) We would also like to see the authors address the relevance of the findings to adult feeding behavior. Assays for feeding behaviors in the larvae are still very primitive compared to those available in adults. Furthermore, larvae feeding is a fundamentally different process from adult feeling.
5) One expert reviewer stated "The description of the connectivity is only partial and the authors focus on the sensory inputs to the hugin-PC and hugin-VNC neurons. The description of the presynaptic circuitry of the other two hugin neuron types and of the postsynaptic circuitry of hugin-VNC neurons is incomplete. For example, hugin-RG and hugin-PH neurons have 39 and 23 presynaptic partners, respectively but the identities of there neurons are not described. Can the authors conclude something about what inputs these cells receive? Likewise, the postsynaptic partners of hugin-VNC neurons are not described in any detail. It seems that hugin-PC and hugin-VNC neurons also receive many input from non-sensory cells. What are these? The authors may of course choose what to show, but it should be stated clearly e.g., that the identity of these neurons is not known or that these circuits will be described elsewhere."
[Editors' note: further revisions were requested prior to acceptance, as described below.]
Thank you for resubmitting your work entitled "Synaptic transmission parallels neuromodulation in a central food-intake circuit" for further consideration at eLife. Your revised article has been favorably evaluated by Eve Marder (Senior editor), a Reviewing editor, and three reviewers.
The manuscript has been improved but there are some remaining issues that need to be addressed before acceptance, as outlined below. You will note that these mainly require editorial rewriting to reflect what appears still to be lack of consensus about what the essential message of the paper is, and still some tendencies to overstate what can be taken from the data. The Reviewing editor has summarized the results of the reviewer discussion, and also given you the entire reviews, for your information. We expect to make a final decision without returning the manuscript to the reviewers after receiving this next revision.
This is a very good revision of the previous submission that answers many of the concerns of the previous review, notably by adding new data implicating Hugin ACh co-transmission. Still there is some frustration that the authors have not precisely defined the focus of the paper, that they have pushed the homology with mammalian systems too far, that there are some irregularities in the RNAi data of Figure 3, and that the anatomical and genetic co-transmission data while strongly supportive are not backed up by cellular physiology studies performed in other systems. Quoting from the reviewer discussion:
1) '…the authors appear to be uncertain what is most important to emphasize (homology, cotransmission, connectomics, or feeding behavior) in their discussion/conclusions. I concur with the others that the homology argument is way oversold and the authors should be encouraged to back off considerably on that perspective; plus, doing so does not diminish the value of their work. The title of the paper is focused on cotransmission, which the authors do establish at the basic level (presence of cotransmitters, and their individual impact on feeding behavior when separately eliminated), but it seems to me that the heart of the paper is not about cotransmission or homology but about determining the input and output connectivity for the 4 classes of hugin-containing neurons. Should not the latter then be the focus of the paper's title?"
2) "As I argued in my first review, the authors cannot conclude that these circuits are homologous. The peptides (NMU/hugin) are orthologs (although one to many, since there is also a fly pyrokinin), the broad regions (SEZ, brain stem) are possibly homologous (but this is already highly contentious). But I don't think they can homologize a projection of a peptidergic cell to one area in the fly with cells expressing a homologous peptide in a (possibly) homologous brain region in the mouse. To claim circuit homology without even knowing the circuitry in a vertebrate is even more problematic. However, I think it is interesting to propose that there might be a common ancestry of these circuits, and there are clearly interesting parallels (including the neuropeptide targets of hugin/NMU). This could stimulate further work. However, it cannot be the main selling point of this paper."
3) "…the resolution of feeding experiments is unsatisfying, but I also recognize this is not an inherent fault of the paper because it is not its main goal. I do believe that the experiments added make it a better paper, although it seems that the RNAi data showed in that Figure 3 were quickly patched up for revision: some of it was previously published and the control is a mix of 3 different genetic backgrounds – this is highly unorthodox in the field and also I believe one should just replicate one's own experiment with new data instead of filling in old data, this is Drosophila, we are talking of a few crosses that take at best 2 weeks…."
The detailed reviews of the reviewers are appended and the authors should consider all the feedback provided in their revision.
The revised manuscript is greatly improved. The authors added new data to show that ACh and hugin are both involved in the effects of hugin neurons on feeding. The authors also improved the discussion about the evolutionary significance of their findings.
This is an impressive body of work. The main problem is that there is a disconnect between the known behavioral effects of hugin neurons and the circuitry. The incomplete circuits shown in the paper do not explain how hugin influences feeding and locomotion.
We learn that there are four classes of hugin neurons that form distinct micro-circuits. However, it is not clear from these microcircuits, how hugin exerts its effects on feeding and locomotion.
1) Hugin-VNC neurons can increase locomotion motor rhythms. These neurons have many partners in the VNC and receive sensory input from abdominal and antennal nerves. However, it is not clear how they exert their effect on locomotion.
2) Hugin-PC neurons modulate feeding behavior and are necessary for the processing of bitter gustatory cues. For these neurons the EM reconstructions identified direct sensory inputs from putative gustatory neurons. This is a case where the EM analysis is revealing and suggests a direct sensory input from bitter sensory neurons to hugin-PC. Hugin-PC neurons project to the pars intercerebralis and synapse on neurosecretory cells. These connections are analyzed in detail in the paper. The pars intercerebralis cells express a hugin receptor and are activated by pharmacologically applied hugin. However, it is not clear if these connections are involved in the regulation of feeding or not. The authors show that hugin-PCs also have outputs in the SEZ (unknown interneurons), a center that houses the basic neuronal circuits generating feeding behavior. This suggests that the physiological role of the connection between hugin-PCs and the pars intercerebralis neurons is unknown. The authors should at least speculate about the role of these neurosecretory neurons in the context of feeding/locomotion. Alternatively, could hugin neurons have additional roles, e.g., in the regulation of metabolism or diuresis?
3) Hugin-RG neurons project to the ring gland and have neuroendocrine release sites bordering haemal space. They receive input from unidentified interneurons. It is not discussed what may be the role of secreted hugin.
4) For Hugin-PH neurons there were no targets identified as these are outside the sampled volume. The input neurons are also mostly unidentified. A discussion of the potential role of these neurons is lacking.
It took me quite a while to put this puzzle together, and the authors don't help the reader to understand what the circuitry can and cannot explain. I suggest that the authors provide a summary table with the four hugin cell types, their known effects, identified partners, putative effects as suggested by the EM analysis, and maybe the unknowns/suggestions for further experiments.
This revision is a satisfying upgrade, particularly the identification of ACh as a small molecule cotransmitter for the hugin neurons. Many of the persisting issues that remain noteworthy relate to inappropriately equating anatomical results with physiological function. Some of these issues are readily repaired by modifying a conclusion from being firm (e.g. "indicates") to supportive (e.g. "suggests" or "supports the hypothesis that" or "is likely to", etc.). For example, the functional consequences of the co-localized ACh/Hugin are only likely or possible ones, not firm ones, because the appropriate experimental manipulations have yet to be performed. Similarly, other issues of this type would benefit from a slight change in the precision of the wording.
Regarding including appropriate literature citations for publications establishing neurotransmitter/neuropeptide co-transmission from identified neurons onto identified targets (see Author Reply to Reviewer Comments, Essential Revision #3, and the new Discussion), there is a strong selection from which to choose. I provide here an assortment (but not exhaustive list) of them, listed alphabetically, from several model systems published in well-regarded journals across the past 30 years. An even longer list of very nice publications can be generated regarding the functional consequences of this type of cotransmission from identified neurons at the level of network and/or behavioral output.
Blitz and Nusbaum, 1999, J Neurosci
Chalansani et al., 2010, Nat Neurosci
Ignell et al., 2009, PNAS [Flies]
Koh et al., 2003 J Neurophysiol
Li and van den Pol, 2006 J Neurosci
Qiu et al., 2016, eLife
Root et al., 2011, Cell [Root]
Sigvardt et al., 1986, J Neurosci
Stein et al., 2007, Eur J Neurosci
Sun et al., 2003, J Neurosci
Vilim et al., 1996, 2000, J Neurosci
Whim and Lloyd, 1989, PNAS
Results section paragraph two/Figure 2—figure supplement 1: How is it that there are "presynaptic densities" associated with clusters of dense cored vesicles (DCVs) and not small clear vesicles (SCVs) in the hugin-RG neuron terminals in the ring gland? Both in this manuscript and throughout the literature, it is established that presynaptic densities in neurons are the domain of SCVs and their release, and that DCVs are neither clustered close to the plasma membrane (or elsewhere) nor is there any ultrastructural identifier of their sites of release. Is my understanding of the vesicle cluster in that figure not accurate? If accurate, is there precedence for this relationship in the literature?
Subsection “Hugin classes form distinct units that share synaptic partners” and “Organizational principles of a peptidergic network” and elsewhere: Axo-axonic connections. Are there truly axo-axonic connections (as this type of connection is classical defined in the literature) onto hugin neurons? This type of synapse commonly refers to synaptic sites located close to specific transmitter release sites at axon terminals/boutons, where the axo-axonic synapse selectively influences only that transmitter release site and possibly nearby sites. These types of synapses are also electrotonically distant/isolated from the population of synaptic inputs which collectively determine whether or not the target neuron will fire an action potential, so they are not part of the synaptic integration mechanism that determines neuronal activity. There appears to be no explicit segregation in the text regarding axonal membrane vs. "synaptic integration membrane" or what would be called dendritic membrane in a vertebrate. However, as is common in invertebrate nervous systems, the hugin neurons do have neuropilar membrane on which there is an intermingling of inputs and outputs, as indicated in the second paragraph of the Results section.
Results section: "Hugin-PC and hugin-VNC neurons' projections represent mixed synaptic input-output compartments as they both showed pre- as well as postsynaptic sites along their neurites (Figure 2D,E)."
If there are truly axo-axonic synapses onto hugin neurons, then please provide more explicit evidence.
Overall, the core results of this paper well-establishes the (ACh/)hugin connectome and provides an interesting parallel to the distribution and behavioral influences of the mammalian hugin-equivalent (neuromedin U; NMU) peptide-containing neurons, which should be of interest to the broad eLife readership.
The authors' additions strengthened the manuscript and provided tools and avenues of investigation for other scientists in the field.
The ChAT experiments, especially the genetic data are quite interesting. While it is still far from clear how the two modes of transmission work and how they work together to mediate the Hugin neurons effect on behavior, I recognize this is outside the scope of this manuscript. It seems that the hugin neurons may be a good model to study this. The new parts of the text in the main body and discussion that address this were satisfactory.https://doi.org/10.7554/eLife.16799.023
- Michael J Texada
- Casey M Schneider-Mizell
- Haluk Lacin
- Feng Li
- Richard D Fetter
- James W Truman
- Albert Cardona
- Philipp Schlegel
- Anton Miroschnikow
- Andreas Schoofs
- Sebastian Hückesfeld
- Marc Peters
- Michael J Pankratz
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
We thank Jan Veenstra and Liliane Schoofs for their gifts of antisera, and Hubert Amrein, Ron Tanimoto, Leslie Vosshall, Christian Jünger, Barret Pfeiffer and Gerry Rubin for plasmids and fly lines. We thank SFB 645 and 704, DFG Cluster of Excellence ImmunoSensation and DFG grant PA 787 for financial support. We thank the Fly EM Project Team at HHMI Janelia for the gift of the EM volume, the HHMI visa office, and HHMI Janelia for funding. We also thank Lucia Torres, Gaia Tavosanis, Gáspár Jékely, Gregory Jefferis, Ingo Zinke, Scott Sternson, Christian Wegener, Volker Hartenstein and Nicholas Strausfeld for critical comments on earlier versions of this manuscript. The EM image data is available via the Open Connectome Project (http://www.openconnectomeproject.org).
- Ronald L Calabrese, Reviewing Editor, Emory University, United States
© 2016, Schlegel et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.