Organization of descending neurons in Drosophila melanogaster

Neural processing in the brain controls behavior through descending neurons (DNs) - neurons which carry signals from the brain to the spinal cord (or thoracic ganglia in insects). Because DNs arise from multiple circuits in the brain, the numerical simplicity and availability of genetic tools make Drosophila a tractable model for understanding descending motor control. As a first step towards a comprehensive study of descending motor control, here we estimate the number and distribution of DNs in the Drosophila brain. We labeled DNs by backfilling them with dextran dye applied to the neck connective and estimated that there are ~1100 DNs distributed in 6 clusters in Drosophila. To assess the distribution of DNs by neurotransmitters, we labeled DNs in flies in which neurons expressing the major neurotransmitters were also labeled. We found DNs belonging to every neurotransmitter class we tested: acetylcholine, GABA, glutamate, serotonin, dopamine and octopamine. Both the major excitatory neurotransmitter (acetylcholine) and the major inhibitory neurotransmitter (GABA) are employed equally; this stands in contrast to vertebrate DNs which are predominantly excitatory. By comparing the distribution of DNs in Drosophila to those reported previously in other insects, we conclude that the organization of DNs in insects is highly conserved.

A conserved feature of motor control across the animal kingdom is the anatomical separation between circuits that control rhythm generation, which are found in spinal cord/ventral nerve cord/segmental ganglia (collectively called body ganglia), and the circuits that integrate sensory information and initiate movement, which are found in the brain. These two circuits are connected by descending neurons (DNs), which have their cell bodies in the brain and carry sensory processing and motor-related information to the body ganglia, and ascending neurons (ANs), which have their cell bodies in the body ganglia and carry motor-related and sensory feedback information to the brain.
Because insects demonstrate highly differentiated motor repertoires while utilizing relatively few neurons, they are an excellent model system for unraveling the general principles of motor control. Individual DNs have been characterized in many species including cockroach (Periplaneta americana) [1][2][3] , cricket (Gryllus bimaculatus) 4,5 , moth (Bombyx mori ) 6,7 , blowfly (Calliphora erythrocephala) [8][9][10] , locust (Schistocerca gregaria) [11][12][13] , and fruit fly (Drosophila melanogaster) [14][15][16] . Characterization of individual DNs provides much insight into descending motor control in insects. But to take advantage of the numerical simplicity of insects, a more comprehensive approach is necessary. Two studies have taken such an approach: One study in cricket 17 and another in cockroach 18 both used retrograde labeling from the cervical connective to quantify the number and distribution of DNs in cricket 17 . These studies showed that the DN population is similar in these two insect species and provided an anatomical framework for understanding information flow in the insect brain 18 .
Although the anatomical studies in cricket 17 and cockroach 18 have identified how DNs are organized, the lack of genetic tools in these insects have limited our progress in understanding descending motor control 19 . In contrast, these tools are readily available in Drosophila melanogaster. The availability of genetic tools as well as recent technical developments have made it possible to assess or manipulate the activities of identified neurons. This ability to probe identified neurons in vivo using either functional imaging [20][21][22] or electrophysiological recording 23,24 has led to a comprehensive understanding of the circuit basis of many computations in the brain. These include computations underlying sensory processing such as motion detection 25,26 and olfactory processing 27,28 and cognitive functions such as associative learning 29 and spatial memory formation 30 .
In contrast to these advances in our understanding of sensory processing and decision-making in Drosophila, the neuronal circuit for descending control of movement remains understudied. Recent work has begun to Scientific RepoRts | 6:20259 | DOI: 10.1038/srep20259 characterize the role of individual DNs in specific tasks. For instance, recent studies have identified DNs that are crucial for backwards walking 14 , courtship song production 15 , and evasive take-off 16 . However, a comprehensive description of descending motor control is missing.
As a first step towards a comprehensive understanding of DNs in Drosophila, here we present an anatomical survey of DNs. We performed retrograde labeling of axons via the fly's cervical connective to estimate the number and distribution of DNs. We also describe the projections of DNs, and the distribution of DNs according to expression of specific neurotransmitters. We find that the number and distribution of DNs is similar to that observed in other insect species, suggesting evolutionary conservation in the number and organization of DNs amongst insect species. Our study is the first (to our knowledge) to present the distribution of DNs in both the supraesophageal and subesopheageal zones, as well as the first study to present the organization of DNs according to the neurotransmitter they employ. We present our results in the context of studies of DNs in other insect species and in vertebrates.

Results
Backfilling of the axons through the cervical connective labels a small number of somata in the brain in a distinctive pattern. These are the DN cell bodies (by definition). Backfilling also labels the neuropil in a characteristic pattern. Neuropil labeling reflects DN dendrites and axon collaterals, as well as the axons of ascending neurons (ANs). We first present our estimate of DN cell bodies and their distribution in the brain. Next, we describe the neural processes.
Drosophila has ~1,100 DNs. The total number of cell bodies labeled in the brain varied between 837 and 907, with a mean of 878.5 DNs (± 29.8 STD, n = 5) ( Table 1). This variation likely reflects differences in the efficacy of the labeling procedure rather than an individual-to-individual variation in the number of DNs. Thus, higher numbers are likely to be closer to the actual number of DNs. Following the method of Okada and colleagues 18 , we also estimated the total number of DNs as a sum of the largest number of DNs observed for each cluster (see below) yielding an upper estimate of 1,113 DNs. Our method of backfilling will also label neck motor neurons that have their cell body in the brain as well as neurons that project through the cardiac recurrent nerve, resulting in a small overestimate for the number of DNs. In blowflies, it has been estimated that there are ~20 neck motor neurons 31,32 and 16 neurons that project through the cardiac recurrent nerve 25 . Thus, these other populations make only a small contribution towards the overall cell count.
The narrow range of the numbers of labeled DNs across individuals and their stereotyped distribution suggests a high labeling efficiency. To estimate labeling efficiency, we performed retrograde labeling in flies in which a small number of DNs are genetically labeled. We chose e49-Gal4/tsh-Gal80; UAS-mCD8-GFP/+ flies because e49-Gal4 labels a small number of DNs in several DN clusters (see Supplementary Fig. S1 online). Because the genetically labeled DNs are isolated from other genetically labeled neurons in this line, we were able to unequivocally conclude that there are exactly 18 genetically labeled DNs in this line. We found that in the fly in which we labeled the largest fraction of DNs (out of 5 flies of this genotype that we bulk-labeled), we labeled 15 of the 18 genetically labeled DNs implying a labeling efficiency of 83%. In this same fly, a total of 883 DNs were labeled suggesting 1,060 DNs. This number is comparable to the number of DNs that we estimated by summing the maximum number per cluster.
The above experiment suggested a second method for measuring labeling efficiency: If the labeling efficiency were 100%, the number of genetically labeled axons in the cervical connective should be equal to the number of double-labeled cell bodies in the brain and thoracic ganglia. To facilitate axon counting, we employed tsh-Gal80 which represses Gal4 in most ANs, thus reducing the number of labeled axons in the cervical connective. By comparing the number of double-labeled neurons in the brain and thoracic ganglia to axons in the cervical connective, we were able to conclude that the labeling efficiency is high but we were not able to quantitatively assess the labeling efficiency (see Supplementary Fig. 1 for details).
The number of DNs reported in this study is remarkably similar to the numbers of DNs reported in previous studies performed in cockroach 18 and cricket 17 . After accounting for the fact that the gnathal (formerly subesophageal) ganglia are not fused to the cerebral (formerly supraesophageal) ganglia in the cockroach and cricket as they are in Drosophila, the 235 pairs of DNs in cockroach and approximately 200 pairs of DNs in cricket match the 412 neurons (206 pairs) we found in the cerebral ganglia. DNs are organized in 6 clusters. In this section, we describe the distribution of DNs. Because the distribution of Drosophila DNs resembles the distribution of DNs in cricket 17 and cockroach 18 , we will relate the distribution of Drosophila DNs to their descriptions in cricket 17 and cockroach 18 . DNs have cell bodies distributed across 6 clusters ( Fig. 1b-d). In the following section, we describe these clusters in order from their b-anterior to b-posterior positions in the brain, where the prefix "b-" denotes the body axis rather than the embryonic neuroaxis (see methods and Ito et al. 33 for more details). The number of neurons in each cluster, and the corresponding number of neurons in cricket and cockroach is reported in Table 1.
b-Anterior clusters. There are three anterior clusters: the anterior optic tubercle (AOTU) cluster, the anterior ventrolateral protocerebrum (AVLP) cluster, and the periesophageal (PENP) cluster (Fig. 1b,c). All three clusters contain relatively few DNs. The AOTU clusters and the AVLP clusters are paired (one in each hemisphere). The AOTU cluster (Fig. 1b, olive green) is located lateral to the vertical lobe of the mushroom body and medial to the anterior optic tubercle. Neurons in this cluster send projections through the medial antennal lobe tract (mALT) toward the ventromedial neuropils. Based on both the soma location and the projection of the primary neurites, this cluster corresponds to the cluster i5 described in cricket 17 and cockroach 18 . In the cricket, an additional cluster was identified, i5n, which was medial to cluster i5 and whose neurites projected along a distinct but parallel tract. Although we could not distinguish two distinct clusters of soma in our study, two sets of neurites projecting from this cluster are labeled (data not shown), suggesting that the AOTU cluster in Drosophila may also be subdivided into two separate clusters.
The AVLP cluster (Fig. 1b, dark blue) is located between the anterior ventrolateral protocerebrum and the antennal lobe. The AVLP cluster does not have an obvious equivalent in cockroach 18 , but may be analogous to cluster i6 in the cricket 17 .
The PENP cluster (Fig. 1b, golden yellow) is located between the prow of the periesophageal neuropil and the antennal lobe. The location of this cluster corresponds to cluster i7 previously reported in the cockroach and cricket 17,18 .
Pars Intercerebralis (PI). The PI cluster (Fig. 1c, red) is located between the hemispheres of the superior medial protocerebrum. Neurons in this cluster send projections into the median bundle 34 . Of these neurons, 14 have been characterized in Drosophila as insulin producing cells (IPCs) that project through the cardiac recurrent nerve to the corpora cardiac and the associated aorta, proventriculus, and crop, rather than through the ventral nerve cord to the thoracic ganglia 34,35 . However, a previous study in the locust has identified at least one DN from the PI cluster which innervates the thoracic ganglia 11 . In addition, studies in the cockroach have also identified DNs with soma located in the PI 2 . Gnathal Ganglia (GNG). The GNG cluster is the largest DN cluster, containing an average of 483.2 neurons (± 61 STD, n = 5) ( Fig. 1b-d, lavender). This cluster is further subdivided into a medial cluster and two lateral clusters. The medial cluster is found near the b-ventral (n-posterior) surface of the gnathal ganglia, while the lateral clusters are found lateral to the gnathal ganglia and b-ventral (n-posterior) to the saddle. There were 187.8 neurons (± 62.9 STD, n = 5) in the medial GNG cluster and 110.5 neurons (± 33.5 STD, n = 10) in each of the lateral GNG clusters. Because of ambiguity in assigning neurons to medial versus lateral clusters, in Table 1 we report these clusters as a single cluster. The number of GNG-DNs we report here is similar to a previous study in the locust 36 , which reported a total of 153 neurons labeled through introduction of cobalt chloride into one of the two cervical connectives.
Superior Medial Protocerebrum (SMP). The SMP cluster consists of cell bodies distributed in the posterior superior medial protocerebrum, the superior intermediate protocerebrum, the posterior lateral protocerebrum, the inferior bridge, and the inferior and superior clamp (Fig. 1d, cyan). In the cricket and cockroach studies, the fact that the authors only labeled one connective allowed the authors to divide this cluster into four clusters -i1(or c1) through i4 (or c4), where "i" vs "c" refer to whether the neurons project to the ipsilateral or contralateral side of the brain 17,18 . The cricket and cockroach studies also described a large soma projecting into the ocellar tract that separated clusters i3 and c3 from i4 and c4 clusters. However, in Drosophila, unlike in the cockroach, the ocellar nerve is found in the midline of the brain 3,37,38 ; thus the ocellar nerve was not a feasible landmark. Because of the lack of clear landmarks and thus a lack of obvious boundaries between the four clusters, in this study we report the number of neurons in the four clusters as a single number.
Neuropil labeled by backfill from cervical connective. In addition to the cell bodies, backfill through the neck also labels the axons and dendrites of DNs and axons of ANs. The pattern of neuropil labeling was distinct and consistent from brain to brain. We could not distinguish between DN and AN processes and therefore report the overall neuropil labeling. The density of labeling for different neuropil regions (averaged over 5 brains) is shown in Fig. 2a.
The labeling was densest in the posterior slope, located in the b-posterior region of the brain. In contrast, known associative areas of the brain-the mushroom body and the central complex-had no detectable labeling, suggesting that these regions neither send direct outputs to nor receive direct inputs from the thoracic ganglia.
Some sensory neuropils were sparsely labeled: these include the lobula (Fig. 2c), the optic tubercle (Fig. 2d), and the lateral horn (Fig. 2e). This sparse labeling of sensory neuropil suggests that these regions are innervated by very few DNs (or ANs), and that most sensory information represented in these brain regions is further processed before being relayed to the DNs. In contrast, the ventrolateral protocerebrum, the majority of whose volume is composed of optic glomeruli that receive output from optic lobes, is densely labeled 39,40 (Fig. 2f). Previous studies in both Drosophila 41 and Calliphora 42 have also identified individual DNs whose dendrites innervate the optic glomeruli. The AMMC (antennae mechanosensory and motor center, Fig. 2g), which processes mechanosensory information, is also densely labeled. This is in agreement with previous studies which have identified descending neurons that carry information from the AMMC to the thoracic ganglia 41,43 .
The distribution of neuropil labeling is consistent with that observed in the cockroach 18 and suggests that, as in the cockroach, both direct and indirect pathways connect cephalic sensory processing to behavior 18 in Drosophila. Thus, information from AMMC and optic glomeruli can be directly communicated to DNs. At the same time, a lack of labeling in mushroom body and central complex implies that processing in these centers affect behavior indirectly; in this case the effect of sensory input on DNs is separated by at least two synapses.
Distribution of DNs by neurotransmitter. Because the neurotransmitter used by a given neuron is an important determinant of its function, we wanted to estimate the distribution of DNs by neurotransmitter. To label all DNs that utilize a given neurotransmitter, we perform labeling in flies in which the neurons utilizing a given neurotransmitter is also labeled (Fig. 3a). For instance, Fig. 3b-d shows projections of the anterior, medial, and posterior portions of the brain in a Cha-Gal4, UAS-GFP fly. Multiple studies have shown that Cha-Gal4 labels most cholinergic neurons 44,45 . When this is combined with red dextran dye ( Fig. 3e-g), the neurons that are labeled with both GFP and red dextran dye appear yellow ( Fig. 3h-j,m). These neurons are the DNs that express the neurotransmitter in question (in this case cholinergic DNs).
We performed experiments analogous to the ones described in Fig. 3 for acetylcholine (n = 4), GABA (n = 4), glutamate (n = 4), octopamine (n = 4), and serotonin and dopamine (n = 4). We found that DNs as a population use all the neurotransmitters we tested ( Table 2). As in the counts for total number of neurons, we assumed that variability in the number of DNs reflected variability in labeling efficiency; therefore, we report the maximum number of DNs per cluster rather than the mean. We first describe the distribution of DNs that belong to the major excitatory and inhibitory neurotransmitter types followed by DNs which do not use these major neurotransmitters (Fig. 4, Table 2 and Supplementary Table S1).
Major neurotransmitters: Cholinergic and GABAergic DNs. The major excitatory neurotransmitter in the Drosophila brain is acetylcholine 45,46 . To label cholinergic neurons, we performed bulk-labeling in Cha-Gal4, UAS-GFP flies (Figs 3b-m and 4a). To label DNs which express the major inhibitory neurotransmitter, GABA 47-49 , we employed two methods: First, we performed experiments in Gad1-Gal4;UAS-mCD8-GFP flies. Second, we performed labeling in a standard lab strain (w 1118 ) and used an anti-GABA antibody to label GABAergic neurons. We found that for all clusters except the GNG cluster, more neurons were labeled by the antibody method than by Gad1-Gal4. This is consistent with previous reports which show that Gad1-Gal4 does not label all GABAergic neurons 50 . Therefore, we report our results with the GABA antibody (Fig. 4a, Table 2, and Supplementary Fig. S2 and Table S1). We found that all clusters contain both cholinergic and GABAergic DNs (Fig. 4a, Table 2, and Supplementary  Table S1). However, the majority of the DNs in two of the anterior clusters (AOTU and PENP) were cholinergic. A greater fraction of the SMP DNs was GABAergic, while a greater fraction of the GNG DNs was cholinergic (especially in the lateral GNG). Overall roughly 40% of the DNs are cholinergic and 40% are GABAergic.
Minor neurotransmitters. There are 56 glutamatergic DNs (Fig. 4b and Supplementary Fig. S3). Glutamatergic neurons are labeled in Vglut-Gal4;UAS-CD8GFP flies, in which Gal4 is driven under the control of the vesicular glutamate transporter (Vglut). These are distributed in all clusters except for the AOTU and the PENP cluster, but most are found in the SMP and the GNG clusters (9 and 18 DNs, respectively, per hemisphere). Of these, at least 3 pairs of DNs in the SMP cluster and 5 pairs of DNs in the GNG cluster may be neck motor neurons, assuming homology between the blowfly Calliphora and Drosophila 31 . Consistent with this study, a previous study in the honeybee also found 5 DNs with glutamatergic-like reactivity located in the ocellar tract (the midline of the SMP cluster) and 16 additional DNs elsewhere in the SMP 51 .  The next most common neurotransmitter used is serotonergic (Fig. 4b and Supplementary Fig. S4). Serotonergic DNs were labeled using Ddc-Gal4 which labels both dopaminergic and serotonergic neurons and labels up to 30 DNs (summed over all clusters). Out of these, 2 DNs located in the GNG have been confirmed to be dopaminergic by performing labeling in TH-Gal4 flies in which only dopaminergic neurons are labeled 52 . Eighteen of the serotonergic DNs are in the SMP cluster and occur in a location similar to serotonergic DNs reported in locust 53 . The other 10 are present in the GNG cluster. Two of the lateral GNG serotonergic DNs have been previously reported in the moth (Manduca sexta) and described as also innervating the labial neuromere of the GNG 54 . There are also reports of a pair of medial GNG serotonergic DNs in the blowfly (Calliphora) 53 .
Assuming that most DNs only employ a single neurotransmitter, 85% of the DNs can be attributed to utilizing one of the five neurotransmitters we characterized. Some of the remaining 15% of DNs represent small diameter DNs which use minor neurotransmitter and are not labeled. We think that this is unlikely because we have shown in a previous study that small diameter dopaminergic neurons are reliably labeled 52 . Moreover, all the octopaminergic DNs that have been labeled genetically 55 were also identified in our study. Therefore, it is more likely that the remaining 15% of DNs use neurotransmitters not characterized in this study: histamine, tyramine or peptidergic neurotransmitters. For instance, previous studies in the moth Manduca sexta have found DNs that express FMRFamide in the midline of the GNG 56 . Our strategy for labeling DNs also label neurosecretory cells, such as those found in the PI and in the pars lateralis, whose location coincides with the SMP cluster. These neurosecretory cells secrete neuropeptides such as Drosophila insulin-like peptide and FMRF, and thus they may not express any of the neurotransmitters examined in this study 35,57 .

Discussion
We present here, what is to our knowledge, the first comprehensive description of the number and distribution of DNs in Drosophila. In the following, we discuss the organization of DNs in the context of studies that describe the organization of DNs in other invertebrates and vertebrates.
Our results suggest that Drosophila have ~1,100 DNs that are distributed across 6 clusters. Using a genetic strain in which a small number of DNs were labeled ( Supplementary Fig. S1), we were able to show that the labeling efficiency is high. Similarly, the number and distribution of octopaminergic DNs in our study matches the description of octopaminergic DNs 55 in another study that employs a completely different approach. Thus, our estimate of ~1,100 DNs is likely to be close to the actual number of DNs. Previous studies have also shown that there are ~3,600 axons which traverse the cervical connective in Drosophila 58 . Based on our work we estimate ~1,100 of these are descending axons, while ~2,500 are ascending axons that represent a combination of ascending motor and sensory input into the brain from the thoracic ganglia.
Our results also show that in Drosophila, DNs employ multiple neurotransmitters and no DN cluster exclusively expresses any single neurotransmitter type. A hallmark of descending motor control is that, irrespective of size and complexity of the movements being controlled, descending control systems employ DNs of multiple neurotransmitter type 59 ; the use of multiple neurotransmitter type is considered important for flexible control of behavior. Thus, it is not surprising that the use of multiple neurotransmitter types by the DN population is also observed in the fly. In this study, we provide the number and location of neurons which employ each of the different neurotransmitters. This will facilitate our understanding of how different neuromodulatory DNs affect a fly's motor behavior.
A comparison of our study to previous studies suggests a high degree of conservation in the number and organization of DNs across arthropods. The number of DNs in the cerebral ganglia of flies is similar to the number reported in cricket 17 and cockroach 18 , while the number of DN cell bodies we found in the gnathal ganglia is similar to the number previously described in locusts 36 . There is also a clear homology (see Fig. 5a) in the spatial distribution of DNs in the brain. The similarity between cricket 17 , cockroach 18 and fly (this study) is also  supported by a study in another holometabolous insect, the moth Bombyx mori, which found three different groups of DNs in the cerebral ganglia 60 , corresponding to the AOTU, SMP, and PENP clusters described in this study. Additionally, we reviewed most (if not all) reports of DNs in insects and were able to assign them to one of the clusters (Fig. 5b and Table 3). Given the overall conservation of the structure and function in the arthropod brain 33 and the reports of many homologous neurons such as the giant fiber neurons and the neck motor neurons 61 , the conservation in the distribution of DN clusters across insect orders is not surprising. But the remarkable conservation in the number of DNs is surprising given that the number of neurons in other structures such as antennal lobe 62 , optic lobe 62 and mushroom body 63 varies by several-orders of magnitude across different insect species. This conservation might reflect the fact that the number of muscles is similar across insects. Since most insect muscles are innervated by 1 to 3 motor neurons (no more than 13 motor neurons) 64 , insect motor systems likely have similar level of complexity. A similar result was also observed in comparing analogous brain regions in bumblebees to honeybees 65 . The authors found that sensory areas of the brain and mushroom body scaled with the size of the insect, but the central body (also referred to as central complex), which is associated with movement control, was smaller relative to brain size in the larger insects.
Cricket (order: orthoptera) and cockroach (order: Blattodea) are both hemimatabolous insects, while Drosophila is a holometabolous insect 33 . Since hemi-and holo-metabolous insects diverged at least 280 million years ago 33,66,67 , the similarities in the number and distribution of DNs among these insects implies a high level of conservation. In addition to the conservation in DN numbers across the insect class, the number of supraesophageal DNs in lobster have been estimated to be around 600-700 68 , suggesting that DN numbers are conserved across arthropods.
The organization of DN processes is also evolutionarily conserved among insects. Although we were not able to distinguish between axonal and dendritic processes, the overall organization of the labeled neural processes is strikingly similar to the cockroach study 18 . Both our study and the cockroach study suggest that DNs receive input from regions of the brain that are innervated by outputs from mushroom body and central complex. Equally importantly, neither DNs nor ANs innervate the central complex or mushroom body, implying that these neuropils do not directly affect motor output. Thus, one important pathway in the insect brain for information flow from sensory circuits to motor circuits is through the mushroom body and central complex to the DNs. There is also direct sensory input into descending neurons (DNs): The labeling presented in this study is consistent with visual inputs into DNs from optic glomeruli and mechanosensory input from AMMC. Similarly, the large number of DNs and dense labeling in the GNG is consistent with the prominent role of GNG in motor control.
In contrast, our study raises three important differences between descending motor control in vertebrates versus arthropods. The number of DNs in arthropods is 3-orders of magnitude smaller than in the vertebrates, which often possess upward of a million DNs 69,70 . This difference is unlikely to be due to the difference in the number of muscles because the number of muscles is surprisingly similar between insects and mammals 62 : the 296 skeletal muscles possessed by locust is comparable to the 316 muscles present in some primates and exceeds the number of muscles in rodents 71 . The difference between the number of DNs in the vertebrate and invertebrate model systems likely originates in the differences in the neuronal control of muscles 72,73 . A vertebrate skeletal muscle can be innervated by hundreds of motor neurons 64 . The dominant mechanism for regulating the amount of force generated by a vertebrate muscle is through the sequential recruitment of motor units. In contrast, arthropod muscles are typically innervated by only 1 to 3 motor neurons and there are rarely more than 13 motor neurons innervating a given muscle 64 . Arthropod muscles produce a gradation of force by activating individual muscles in a graded fashion. Because of the differences in the neural control strategies of muscle contraction, the total number of motor neurons in vertebrates is at least three-order of magnitude greater than in invertebrate systems: There are ~50,000 motor neurons 74 innervating limbs in a human compared to ~50 in Drosophila 75 . In comparison to this three-orders of magnitude difference in the number of motor neurons, the ratio of interneurons in the body ganglia to motor neurons only scales modestly: the ratio is 25:1 for macaque 76 , 14:1 for mouse and 8:1 for turtle 77 ; a similar ratio of 10:1 has been reported for locusts 78 . Thus, it is plausible that the large difference in DN numbers between vertebrates and invertebrates arise from the differences in the number of motor neurons.
Second, comparative study of DNs in vertebrates has revealed that several descending tracts such as reticulospinal tracts and vestibulospinal tract are common to all vertebrates 79 . Others have evolved during the course of vertebrate evolution. There is also a great variation in the relative number of neurons in each descending tract in different species 69,79 . These differences in the organization of vertebrate DNs contrast with the relatively conserved DN numbers in arthropods. Thus, the evolution of DNs within the vertebrate phyla and invertebrate phyla seems to follow a fundamentally different plan.
Finally, the two major clusters of DNs in flies, SMP and GNG, contain comparable numbers of excitatory cholinergic neurons and inhibitory GABAergic neurons. This contrasts with the largely excitatory neurons which comprise DNs in mammals. The corticospinal tract, rubrospinal tract and the vestibulospinal tract are all glutamatergic and thus presumably excitatory 80 . Only the reticulospinal tract has a significant fraction of GABAergic neurons; although, even in this case the 59% glutamatergic neurons dominate the 20% GABAergic axons 80 . This difference in neurotransmitter might reflect a fundamentally different logic underlying descending control in these two phyla.
These differences do not imply that vertebrate and invertebrate motor systems are fundamentally different. Both motor systems control rigid, articulated skeletal system using muscles which are in turn controlled by neurons. The architecture underlying neural control of movement is similar in both systems and involves descending control of central pattern generators. With respect to DN pathways, there are many functional parallels. Both the vertebrate brainstem and the gnathal ganglia DNs contain tonically firing neurons which are strongly activated during locomotion 52,81 . They also both play a crucial role in regulating indirect aspects of motor control such as . This is shown in comparison to the distribution of DN clusters in Drosophila (right), a holometabolous insect, as described in this study. The gnathal ganglion, which in holometabolous insects such as Drosophila is fused to the cerebral ganglia, is shown shaded in lavendar. (b) The approximate locations of insect DNs previously described in the literature is shown in the context of the clusters described in this study. The numbers correspond to rows listed in Table 3. respiration and posture and interact with other tracts in the brain in a parallel, hierarchical, but recurrent fashion 81 . Similarly, functions subserved by DNs in the cerebral ganglia parallel functions carried out by DNs which originate from the cortex or red nucleus in mammals. Many previously characterized sensory specific DNs in insects also have cell bodies located in the cerebral ganglia, such as the descending neurons of the ocellar and vertical system (DNOVs) 82 , the descending contralateral motion detector (DCMD) 13 , and the target selective descending neurons (TSDNs) 83 . In particular, the TSDNs operate similarly to the DNs in the vertebrate corticospinal tract: they process behaviorally relevant sensory information (visual stimuli resembling prey) and relay information directly to the wing motor centers using a population vector representation. Taken together, there are many features of descending motor control in invertebrate systems that suggest that they operate on principles similar to that in vertebrates. Thus, a detailed description of the DN population and its distribution in different  parts of the brain in the Drosophila will inform future models of how different descending tracts from the brain cooperate to control movement.

Methods
Fly stocks. Flies were reared on standard cornmeal-agar and raised at 25 °C. In all experiments, adult female flies at least 18 hours post-eclosion were used. Fly stocks were obtained from the Bloomington Stock Center, with the exception of E49-Gal4 84 (a gift from Kristin Scott). The neurotransmitter used by DNs was inferred from double labeling by both the dextran and a label for the neurotransmitter. To label GABAergic neurons, we used anti-GABA antibody. The genotype of the flies used to label the other neurotransmitters employed by different DNs are as follows: for cholinergic DNs, the genotype was Cha-Gal4,UAS-GFP; for serotonergic DNs, the genotype was Ddc-Gal4/+ ; UAS-mCD8-GFP/+ ; for octopaminergic neurons, the genotype was Tdc2-Gal4/UAS-mCD8-GFP; UAS-mCD8-GFP/+ ; for glutamatergic neurons, VGlut-Gal4/VGlut-Gal4; UAS-mCD8-GFP/UAS-mCD8-GFP.

Nomenclature and terminology.
We have adhered to the nomenclature recommended by the consortium "The Insect Brain Name Working Group" 33 . Because published work (Ref. 20) describes this nomenclature in details, we describe our naming conventions in brief. Consistent with the conventions suggested by the working group, the location of cell bodies is described with respect to the body axis (specified with a prefix b-). However, to maintain consistency with previous studies in other insects, which use the embryonic neuroaxis, we also specify the neuroaxis and add the prefix "n-" (as opposed to "b-") when relevant.
To describe major brain regions, we used neuromere based definitions (cerebral ganglia and gnathal ganglia) rather than esophagus-based definition (supraesophageal and subesophageal), to facilitate comparison of the anatomy between holometabolous insects such as Drosophila and the hemimetabolous insects, such as the cockroach and the cricket. The cerebral ganglia consist of the protocerebrum, deutocerebrum and tritocerebrum while the gnathal ganglia consists of the maxillary, mandibular and labial neuromeres. In holometabolous insects, such as Drosophila, the deutocerebrum and tritocerebrum are located in the subesophageal zone, while in the hemimetabolous insects, they are located in the supraesophageal zone.
DN clusters were named according to the neuropil adjacent to them. The consortium study describes the major neuropil as Level 1 neuropil which is then further subdivided into Level 2 neuropil. Three of the six clusters of DNs were named according to the Level 2 neuropil they were found adjacent to: anterior optic tubercle (AOTU), anterior ventrolateral protocerebrum (AVLP), superior medial protocerebrum (SMP). The gnathal ganglion (GNG) cluster was named for the Level 1 neuropil because no Level 2 divisions have been described yet. The periesophageal neuropil cluster (PENP) was named for the Level 1 neuropil rather than the three structures it was adjacent to because of the difficulty in distinguishing the Level 2 structures (specifically the flange, the prow, and the cantle). The remaining cluster of neurons, the Pars Intercerebralis (PI) cluster, has been previously described in the literature, and has not been reassigned a name by the Insect Brain Name Working Group.

Analysis of Density of Labeling.
To quantify the intensity of neuropil labeled by dextran, the confocal stack was sampled at 10 μ m intervals. The neuropil was manually divided into regions of interest on each slice of the 10 μ m substack. The mean intensity of all the pixels within each region indicated on each slice of the 10 μ m substack was computed, resulting in a single mean intensity for the volume of interest. The neuropils quantified were those listed as Level 2 neuropils in Ref 20, except for the inferior neuropils, mushroom body, central complex, lateral complex, and periesophageal neuropils, which were consolidated according to their Level 1 supercategories for convenience.