a-html.html

Journal Pre-proof

The Drosophila connectome reveals Axo-Axonic Synapses on Descending Neurons

Cesar Ceballos, Juan Lopez, Ty Roachford, Daniel Sanchez, Sabrina Jara, Kelli

Robbins, Casey L. Spencer, Rodney Murphey, Rodrigo F.O. Pena

PII:

S2589-0042(26)00999-5

DOI:

https://doi.org/10.1016/j.isci.2026.115624

Reference:

ISCI 115624

To appear in:

iScience

Received Date: 15 September 2025
Revised Date: 5 February 2026
Accepted Date: 10 February 2026

Please cite this article as: Ceballos, C., Lopez, J., Roachford, T., Sanchez, D., Jara, S., Robbins, K.,

Spencer, C.L, Murphey, R., Pena, R.F., The Drosophila connectome reveals Axo-Axonic Synapses on

Descending Neurons,

iScience

(2026), doi:

https://doi.org/10.1016/j.isci.2026.115624

This is a PDF of an article that has undergone enhancements after acceptance, such as the addition

of a cover page and metadata, and formatting for readability. This version will undergo additional

copyediting, typesetting and review before it is published in its final form. As such, this version is no

longer the Accepted Manuscript, but it is not yet the definitive Version of Record; we are providing

this early version to give early visibility of the article. Please note that Elsevier’s sharing policy for the

Published Journal Article applies to this version, see:

https://www.elsevier.com/about/policies-and-

standards/sharing#4-published-journal-article

. Please also note that, during the production process,

errors may be discovered which could affect the content, and all legal disclaimers that apply to the

journal pertain.

The Drosophila connectome reveals Axo-Axonic
Synapses on Descending Neurons

Cesar Ceballos

, Juan Lopez

, Ty Roachford

, Daniel Sanchez

, Sabrina Jara

, Kelli Robbins

Casey L Spencer

, Rodney Murphey

, Rodrigo FO Pena

1,3

Department of Biological Sciences, Florida Atlantic University, FL 33458, Jupiter, Florida, USA

Harriet Wilkes Honors College, Florida Atlantic University, Jupiter, FL, 33458, USA

Lead contact

*Correspondence

penar@fau.edu

Summary

Axo-axonic synapses can veto, amplify, or synchronize spikes, yet their circuit-scale logic is

unknown. Using the complete electron-microscopy connectome of the adult male Drosophila, we

charted every axo-axonic input onto the 1,314 descending neurons that carry brain commands to

the ventral nerve cord. By definition, any synapse connected to a descending neuron within the

cord is axo-axonic. Thus, we uncovered the ascending-descending and interneurons-descending

axo-axonic relationship. Neurons with many partners (high-degree) integrate into the network

without clustering into an interconnected ‘rich-club’ of hubs. We identified an octet of

ascending neurons (AN08B098) whose axo-axonic input to the Giant Fibers (DNp01) predicted

modulation of the escape circuit. Immunostaining confirms their cholinergic identity, while

optogenetic activation confirmed that this excitatory cohort increases DNp01 excitability,

validating connectome-derived rules. Our work delivers a map of axo-axonic wiring in a

complete ventral nerve cord connectome and provides constraints for models of fast motor

control.

Keywords:

Connectome, Drosophila, Axo-axonic, Giant Fiber escape circuit, Network topology,

Electron-microscopy reconstruction, Motor control

Journal Pre-proof

Introduction

Neurons communicate through a variety of synaptic motifs. In flies, the canonical arrangement is

axo-dendritic, but a second configuration, axo-axonic, can exert unusually powerful control over

spike initiation and transmitter release

1,2

. Axo-axonic contacts (AACs) in invertebrates have been

known for some time

2–6

. Similar studies have also been conducted in vertebrates

7–9

. In the

mammalian cortex, exemplified by chandelier cells, recent studies show how inhibition modulates

the pyramidal axons at their initial segments

10,11

. In fish, excitatory spiral fibers and M-cells

control a similar circuit

. Although these examples suggest broader significance, the quantitative

wiring logic of axo-axonic synapses remained poorly characterized, primarily due to the absence

of a fully mapped system at the necessary synaptic resolution.

This limitation has been addressed by recent electron microscopy reconstructions of the complete

adult male nerve cord and female brain (MANC

13–15

and FAFB

) in

Drosophila melanogaster

now publicly available through the NeuPrint interface

. Early graph analyses of the FAFB data

revealed rich-club hubs and other mesoscale motifs

, but systematic exploration of the ventral

nerve cord (VNC), the insect analogue of the vertebrate spinal cord, is only beginning. At the heart

of the VNC, a pair of descending neurons, known as Giant Fibers (GFs), function as command-

like neurons that receive visual input in the brain and drive jump and flight motor neurons

18–20

These GFs receive a high number of AACs; thus, it is only natural to assume that an AAC-to-

behavior modulation is in place. This system is comparable to the descending escape circuits in

larger insects, which have been shown to be the target of axo-axonic connections, exemplified by

the input to the descending contralateral movement detector (DCMD) neurons in the locust central

nervous system

Within the VNC, a single class of neurons is ideally positioned for studying axo-axonic wiring and

their connection to the GFs: Descending neurons (DNs). Because DNs have their somata and

dendrites in the brain yet project their axons into the VNC

21–24

, any synapse made onto a DN within

the cord is, by definition, axo-axonic. In fact, we have manually curated 1,000 ascending-

descending neuron connections, where we observed clear presynaptic mitochondrial expression

(93%), a hallmark of axonal terminals

(example shown in Fig. S1, and curation details in Table

S1). The MANC connectome contains DN, ascending neurons (ANs), intrinsic interneurons (INs),

and motor neurons (MNs), offering a comprehensive substrate for network-wide analysis of this

motif. Although classic anatomical work suggested that each GF acts largely in isolation, our work

using the MANC connectome reveals an unexpectedly rich set of axo-axonic inputs onto the GF

axons, including direct contacts from other DNs, ANs, and INs. In the present study, we show that

the GF receives 97 distinct axo-axonic partners, with some offering a crucial role for axo-axonic

modulation, i.e., an octet of ascending neurons (AN08B098). Below, we will refer to these neurons

of interest as Axc.

Journal Pre-proof

Past light-microscopic surveys catalogued individual DN types and noted scattered DN-to-DN

contacts

, yet the circuit-level role of axo-axonic synapses has remained unexplored. The

outstanding issues are their prevalence, partner preference, and functional impact on fast behavior.

Specifically, it is unknown how frequently axo-axonic connections occur among the 861,591

possible DN pairs, and whether such a sparse motif can nevertheless mediate rapid, system-wide

control required for escape. We tackle these questions by mining the complete electron-

microscopy connectome of the adult male ventral nerve cord (MANC). Our analysis (i) enumerates

every axo-axonic input onto all 1,314 DNs, (ii) dissects the morphological relationship to DN

connectivity, (iii) shows that the DN network is small-world but not rich-club, partitioned into

thirteen functional communities, (iv) uses large-scale leaky integrate-and-fire simulations to

predict which highly connected AACs modulate the Giant Fiber escape pathway, and (v) confirms

those predictions with targeted optogenetic activation

in vivo

Our results provide the first system-level blueprint of axo-axonic wiring in the

Drosophila

adult

nervous system, suggesting general design principles for descending motor control that may extend

beyond insects. They also illustrate the functional power of the sparse axo-axonic network

described below and place the GFs at the center of a distributed control system rather than a simple

linear reflex arc.

Journal Pre-proof

Results

Global Distribution of Axo-Axonic Inputs to Descending Neurons

To uncover the prevalence of axo-axonic connections made onto DNs in the VNC, we used the

MANC dataset available through the NeuPrint+ interface and API

26–28

. Figure 1

shows

morphological reconstructions of some exemplary DNs of interest from Neuroglancer: DNp01

(GFs, black dot), DNd03 (blue dot), and DNg02 (red dot, Fig. 1

). For each neuron,

Neuroglancer overlays the 3D-EM reconstruction on the VNC volume. The GFs, whose axons are

nearly unbranched (green or blue), receive one of the highest numbers of axo-axonic inputs (n =

97; see NeuPrint+) in the dataset. DNd03 is a highly arborized neuron with a similarly large input

count (n=160), while DNg02 has less arborization but similar axo-axonic inputs (n=110).

Interestingly, all three share similar synaptic strength (i.e. the equivalent of synapse count from

the connectome MANC) despite differences in arborization (DNp01 2352, DNd03 2364, DNg02

1751; Fig. 1

). Neuroglancer presents transmission-EM micrographs at 500 nm resolution taken

from MANC; presynaptic active zones are marked by magenta dots, postsynaptic densities by cyan

dots, and partner axons are pseudo-colored to match the 3D rendering (Fig. 1

100

101

We quantified the distribution of axo-axonic connections (i.e. presynaptic DN, AN and IN) made

102

onto DNs reported by MANC. Figure 1

reveals a highly skewed distribution of partner count;

103

the median DN is innervated by a single presynaptic neuron (inter-quartile range 1–3), and 11%

104

of DNs receive input from only one cell with a synaptic strength of six or greater. A small set of

105

outliers, however, attracts hundreds of partners, including DNp01 (GFs) with 97 distinct axo-

106

axonic inputs. Strikingly, the GFs receive these connections despite possessing one of the simplest

107

axonal arbors in the dataset (Fig. 1

). All other DNs exhibit more extensively branched and

108

structurally complex axonal arbors, which are likely to account for their higher levels of

109

innervation. Total input strength follows a similar heavy-tailed pattern (Fig. 1

). The modal DN

110

receives fewer than 20 synapses, whereas the GF is again among the top of the list in terms of

111

individual contacts. Across the population, partner count and cumulative strength are tightly

112

coupled (Pearson r = 0.96, p < 0.001; Fig. 1

), indicating that neurons that recruit many partners

113

also receive proportionally stronger innervation. Together, these findings identify the distribution

114

of axo-axonic connections made with DNs in the VNC. We confirm that the synapses quantified

115

in the graph analysis correspond to active zones and demonstrate how neurons with markedly

116

different morphologies can converge on similar axo-axonic input profiles.

117

118

Using these findings, we next examine whether anatomical complexity serves as a strong, though

119

not perfect, predictor of synaptic strength for axo-axonic inputs onto DNs (Fig. 2). Representative

120

reconstructions illustrate the general pattern: neurons with short, unbranched axons receive few

121

contacts, those with intermediate arborization receive proportionately more, and highly branched

122

cells accumulate the greatest axo-axonic strength (Fig. 2

). A kernel density estimation (KDE)

123

plot of all 1,314 DNs reinforces this observation, showing a diagonal ridge in which increasing

124

cable length accompanies increasing total input strength (Fig. 2

). Linear, quadratic, and symbolic

125

Journal Pre-proof

regressions all yield significant positive slopes, confirming a monotonic relationship (Fig. 2

126

However, the shallow gradient and the presence of two striking outliers, a long axon with sparse

127

innervation and a short axon with dense innervation, indicate that branch length alone does not

128

determine synaptic investment (Fig. 2

). Thus, we find that a presynaptic neuron’s morphology

129

broadly predicts the degree of modulation received by a partner DN, although additional cell-type-

130

specific factors likely refine the final allocation of axo-axonic synapses.

131

132

Further exploration revealed a unique feature of connectivity. In the VNC, neurons will often form

133

synapses with multiple downstream partners (Fig. 3

). We found certain neurons, such as

134

members of AN08B098, had a disproportionately larger number of connections made onto a single

135

target type of DNs such as the GF than other neuron type. To determine whether these neurons

136

constitute a separable class, we quantified neuron output specialization for every presynaptic

137

neuron using three metrics: the fraction of total synapses made onto the dominant DN, the entropy

138

of the neuron type distribution, and the number of connections of the second largest neuron type

139

connection (off-target) relative to the largest DN connection (on-target).

140

141

For each presynaptic neuron, we first quantified the percentage of strength of its outgoing synapses

142

that were made onto its most strongly targeted DN. Neurons were required to exceed a threshold

143

value and the percentage of strength of the dominant DN was permitted to exceed the threshold,

144

ensuring a uniquely dominant target. Initially, we assumed threshold values that separate two

145

populations: multi-target vs single-type biases. The final threshold was determined by optimizing

146

the minimization of an error parameter. A statistical comparison of the distributions using the

147

Kolmogorov-Smirnov test (with a significant p-value < 0.001) revealed two distinct distributions

148

of presynaptic neurons, which we defined as single-type biased and multi-target, with threshold

149

values of 0.27 in MANC v1.2.1 and 0.1 in male-cns v0.9. In addition, single-biased cannot send

150

more than 15 connections to other off-target DN to exclude external influence

151

152

In both connectomes, we identified several single-type biased axo-axonic connections (Fig. 3

) to

153

DNs in both MANC v1.2.1 and male-cns v0.9. DNp01 receives the highest number of single-target

154

AACs in MANC (n = 16) and among the highest for male-cns (n = 11). Among these neurons, five

155

belong to the Axc type, and the remaining are labeled as “unknown”.

156

157

Compared with other highly connected DNs, the GFs attain an equivalent synaptic strength with a

158

simple morphology, further demonstrating how morphological complexity alone does not predict

159

axo-axonic input. These results identify the GFs as exceptional hubs for axo-axonic modulation

160

due to their significant connectivity despite their relatively simple morphology compared to other

161

highly connected DNs. However, these findings raise several new questions: How are AAC

162

populations connected, and how do they act on GFs? Is the GF an outlier, or do similar axo-axonic

163

motifs govern other rapid motor pathways in the fly? How do these contacts alter GF excitability?

164

Journal Pre-proof

We address these questions by investigating the strongest single-target AAC population

165

presynaptic to the GFs, the Axc type neurons.

166

167

Sparse yet Globally Connected Axo-Axonic Network

168

To address how AAC populations are connected, and their influence on other DNs, we extracted

169

every DN-to-DN contact from the updated MANC dataset. The resulting adjacency matrix is

170

strikingly sparse: only about 1% of the possible ordered pairs carry even a single synapse (Fig.

171

). Moreover, partner choice is not random: homotypic pairs (same DN type) and heterotypic

172

pairs (different types) show distinct transmitter profiles, as shown for acetylcholine, GABA, and

173

glutamate, implying that neurotransmitter identity contributes to the selectivity of these long-range

174

axo-axonic contacts (Fig. 4

). In homotypic DN connections, ~90% of the connections are

175

cholinergic (excitatory), while heterotypic DN connections are ~60% cholinergic [the remaining

176

are GABA/glutamate (inhibitory)]. This suggests that AACs between DNs, while rare, are

177

predominantly excitatory. connections.

178

179

We then asked whether the same wiring principle applies to non-descending inputs, namely

180

ascending neurons (ANs) and intrinsic interneurons (INs). Repeating the analysis with ANs and

181

INs as presynaptic partners to DNs again produced very sparse matrices: about 1.2% of all possible

182

AN-to-DN pairs and 0.7% of the IN-to-DN pairs contain at least one synapse (Fig. 4

). Figure

183

shows the neurotransmitter profile for AN-to-DN and IN-to-DN pairs. Both pair profiles were

184

similar, with nearly half of the neurons classified as cholinergic and the other half

185

GABA/glutamate. This suggests that AACs between AN-to-DN and IN-to-DN follow the same

186

rules for neurotransmitter identity as DN-to-DN AACs.

187

188

Lastly, we show the distribution of the number of pre-synaptic DN, AN, or IN that project axo-

189

axonic synapses to a post-synaptic DN (Fig. 4

top row) as well as the number of postsynaptic

190

DN that are innervated axo-axonically by the same DN, AN, or IN (Fig. 4

bottom row). The

191

distributions are similar across neuron types, showing a decay pattern resembling an exponential

192

(Poisson) distribution, where a few neurons are highly connected, and a few neurons are sparsely

193

connected. When we merge all three presynaptic classes into a single connectivity matrix (Fig.

194

,) the same pattern from Figure 4

emerges: extreme sparseness (see dots where our

195

exemplary neurons are located; color scheme as in Fig. 1

.). Note that in Fig. 4

there is a clear

196

diagonal line which indicates DNs synapsing onto their contralateral counterparts. We also

197

highlight the degree of convergence across presynaptic classes and identify connectivity hubs

198

based on their total input. Taken together, regardless of presynaptic identity, the findings

199

demonstrate that the network's organization combines stringent partner selectivity with complete

200

coverage of the descending system.

201

202

Global wiring principles were identified to explore more refined AAC spatial connectivity

203

preferences and determine whether they are repeated across DNs, ANs, and INs. Figure 4 shows

204

examples of the contralateral axo-axonic connections between DN-to-DN (Fig. 4

K-L

), AN-to-DN

205

Journal Pre-proof

(Fig. 4

M-N

), and IN-to-DN (Fig. 4

O-P

). Figure 4

shows the percentage of contralateral and

206

ipsilateral connections. AACs between contralateral DNs are rare. However, ANs and INs

207

projecting to contralateral DNs are frequent. Interestingly, the neurotransmitter profiles from our

208

earlier findings are preserved regardless of the type of presynaptic neuron, where approximately

209

half of the neurons are cholinergic (excitatory), and the other half is GABAergic/glutamatergic

210

(inhibitory) (Fig. 4

211

212

Collectively, these findings suggest a coherent narrative. Whether the presynaptic axon descends,

213

ascends, or remains intrinsic to the cord, axo-axonic wiring obeys the same two rules: partner

214

selectivity is extremely high, and midline crossing is common only for neurons that share strong

215

reciprocal ties. The result is a VNC in which local subtypes can interact tightly with their mirror

216

images, yet the great majority of neurons remain insulated from direct axo-axonic influence.

217

218

Graph-Theoretic Architecture of the Axo-Axonic Network

219

To address if GF is an outlier or do similar axo-axonic motifs govern other rapid motor pathways

220

in the fly, we study connectivity matrices that are derived employed as a directed graph to examine

221

the system-wide arrangement of sparse axo-axonic interactions among DNs (Fig. 4

). The graph

222

contains 884 nodes and 2,757 edges, giving a density of 0.007. Despite this low density, the

223

average shortest-path length is only 4.03, indicating that any two DNs can be reached through a

224

small number of intermediate axonal contacts. In fact, we can compare with the chance level: the

225

formula

⟨𝐿⟩ = ln(𝑛) /ln⁡(𝑘̅)

is a well-known and widely used approximation for the average

226

shortest-path length in large random networks, such as the Erdős-Rényi model. Here,

𝑛

is the

227

number of vertices in the graph,

𝑘̅

is the mean degree, which can be obtained by the ratio

𝑚/𝑛

228

with

𝑚

being the edge count. In our problem,

𝑛 = 884

DNs and

𝑚 = 2,757

axo-axonic contacts

229

leave us with

⟨𝐿⟩ ≈ 5.9

. Therefore, the average shortest path length is indeed shorter than you

230

would expect by chance.

The clustering coefficient is likewise low at 0.013, indicating limited

231

local redundancy. Community detection identifies 13 separate modules, consistent with the

232

network's functional subdivision. Node degree ranges from 1 to 29 with a mean of 6.24, reflecting

233

a modest hierarchical spread.

234

235

The rich-club coefficient Φ(k) measures the density of connections among nodes whose total

236

degree (in-degree + out-degree) equals or exceeds a threshold k. Φ(k) = 1 indicates that the high-

237

degree subgraph is fully connected, whereas Φ(k) = 0 means that those nodes share no edges

238

beyond chance level. The rich-club coefficient in our data peaks at 0.076 for nodes with a degree

239

of at least 18, so high-degree hubs do not preferentially connect to each other; integration is

240

therefore distributed rather than centralized (Fig. 4

). The modest peak value and the absence of

241

a sustained plateau indicate that high-degree DNs do not preferentially interconnect to form a rich

242

club; instead, network integration is distributed rather than hub-dominated.

243

244

A slightly positive degree assortativity of 0.029, a measure that suggests a tendency of nodes to

245

connect with other nodes in a similar way, suggests a weak tendency for neurons of similar degree

246

to link. Finally, the maximum betweenness centrality of 0.049, a measure indicating how often a

247

Journal Pre-proof

node lies on the shortest path between all pairs of nodes in a network, highlights a small set of

248

bridge neurons that channel information between modules. Together, these metrics portray a

249

network that combines efficient global communication with modular specialization, an

250

architecture well-suited for coordinating diverse motor outputs.

251

252

Axo-axonic modulators of the Giant Fiber escape circuit.

253

The MANC connectome introduces several neuron types to the Giant Fiber escape circuit. To

254

understand and underscore the significance of AAC modulation of the circuit, it is important to

255

know which individual cells can exert a strong, targeted influence on a system such as the GF

256

system

18,19

. As shown in Figure 3, the GFs are among a small population of DNs that receive

257

single-target connections. A focused search within MANC revealed the Axc type neurons, a cluster

258

of eight neurons whose axo-axonic output is both nearly exclusive and among the strongest to the

259

GFs (Figs. 3-4). These Axc-type neurons are predicted to be cholinergic by the MANC

260

connectome. The BANC connectome demonstrates that these neurons do not connect to the GF in

261

the brain (see Fig. S3).

262

263

To test whether the connectome-predicted Axc type neurons truly form chemical AACs with the

264

GFs, we used split-Gal4 drivers to direct membrane-bound GFP to Axc axons, back-filled the GFs

265

with tetramethyl-rhodamine, and labelled presynaptic active zones with anti-Bruchpilot (nc82).

266

High-resolution confocal stacks revealed discrete GFP-positive boutons that colocalize with Brp

267

puncta on the descending GF tract in the first thoracic neuromere (Fig. 5

), confirming axo-

268

axonic contacts between Axc type neurons and GF. The split-Gal4 line being used labels

269

AN08B098. However, there are a few extra neurons labeled that can come from AN08B099 and

270

make weak connections to the GF. We have also provided (i) single-section confocal images with

271

optical enhancement showing the synaptic connectivity and (ii) similar EM reconstructions from

272

Neuprint to demonstrate that synapse locations along the axon correspond to synapse locations

273

identified in the MANC connectome (Fig. 5

G-O

). Immunostaining for Choline Acetyltransferase

274

(ChAT) revealed that Axc somata, as labelled through genetic expression of mCD8::GFP, are

275

ChAT-positive (Fig. 5

), establishing this cohort as likely cholinergic and therefore excitatory.

276

277

To address how these contacts alter GF excitability, and following this imaging confirmation, we

278

investigated whether these AACs could measurably influence GF activity in a simulation of the

279

complete VNC connectome. To test this, we stimulated all VNC neurons with a Poisson

280

distribution of excitatory synaptic currents at 30 Hz and 5 nS. Additionally, one GF was stimulated

281

more strongly at 10 Hz and 40 nS to establish a baseline level of excitability (Fig. 6

A, where

spikes

282

are indicated by large black dots at the bottom of the raster plot and arrows underneath). When the

283

eight cholinergic Axc neurons were activated, the GF crossed threshold more often, indicating that

284

the axo-axonic inputs are net excitatory (Fig. 6

). Additional simulations with different scenarios

285

where stimulation and deletion of non-Axc (nAxc) populations demonstrate that the Axc cells are

286

paramount for increased excitability (Figs. 6

). GF subthreshold voltage histograms quantify

287

Journal Pre-proof

the increased excitability (Fig. 6

), supported by statistical significance (Fig. 6

). These

288

simulations highlight a previously unrecognized cohort of axo-axonic synaptic modulation,

289

providing a rapid mechanism for amplifying the escape-circuit output.

290

291

As the control experiment, we also tested the possibility of silencing nAxc cells, which are defined

292

here as presynaptic cells to the GF that contribute to polysynaptic pathways, and Axc cells

293

selectively (Figs. 6

). The results from these figures demonstrate that when nAxc connections

294

are silenced, the increased excitability persists, and when the Axc monosynaptic connections are

295

silenced, but nAxc is retained, the excitability is decreased. Fig. 6

shows additional

296

quantification of this Axc to GF excitability, with the only non-significant comparison being Axc

297

stim vs. silencing nAxc + Axc stim, demonstrating that the possibility that Axc acts

298

polysynaptically could be neglected.

299

300

Testing connectome predictions in vivo via optogenetic stimulation of axo-axonic neurons.

301

302

To assess the predictions of these connectome simulations, we tested the ideas

in vivo

using classic

303

electrophysiology and optogenetics. We used the split-Gal4 line that labels the Axc neurons

304

(BDSC #89945) to drive a UAS construct containing a mVenus-tagged red-shifted

305

channelrhodopsin, CsChrimson (BDSC #55136). Expression of CsChrimson was verified with

306

immunohistochemistry (Fig. S2A). We electrically stimulated the GFs at 24 Hz in flies expressing

307

CsChrimson in Axc type neurons and recorded from the jump muscle (TTM) and flight muscle

308

(DLM) as previously described

, (Fig. 7

A-B

). We then set the electrical stimulation of the GF just

309

below threshold. This near-threshold stimulation resulted in occasional (~20%) responses in both

310

muscles (n = 5 specimens, Fig. 7

and 7

). However, when the same preparation was illuminated

311

(590 nm, 0.37mW) during the stimulus train, the sub-threshold electrical stimulus elicited

312

responses from both muscles nearly 100% of the time (n = 5 specimens, Fig. 7

and 7F). A one-

313

way ANOVA found significant differences between responses before, during, and after light

314

application (

p <

0.001), but there was no significant difference for the control

group (

p =

0.53)

315

(Fig.

). Post-hoc multiple comparisons revealed significant differences between responses at

316

baseline (before light application) and during and after light application. We also found significant

317

differences using a paired t-test for the same responses during and after light application

318

(

0.001), but not before (

0.26).

319

320

To verify that light stimulation does not elicit effects in the absence of CsChrimson expression or

321

function, we performed two control experiments. In the first experiment, flies expressed GFP

322

instead of CsChrimson (Fig. S2A); in the second, CsChrimson-expressing flies were raised without

323

dietary cis-trans-retinal, thereby preventing optogenetic activation (Fig. S2

). A one-way

324

ANOVA revealed no significant differences in stimulus-evoked responses across conditions (

325

1). Post-hoc comparisons revealed no significant differences between the baseline, light, and post-

326

light periods. Paired t-tests also found no significant differences between light and no-light

327

Journal Pre-proof

conditions during and after stimulation. These results confirm that the observed effects are

328

specifically attributable to CsChrimson activation.

329

330

How the Axc alters excitability depends in part on the site of spike initiation for the GF. In the

331

present experiments, the stimulating electrodes are in the brain where there are no monosynaptic

332

connections from Axc to GF

. The only monosynaptic Axc-to-GF connections are in the VNC

333

(Fig. 5, S3). How can the optical excitation of the Axcs in the VNC, where they make synaptic

334

contact, have a modulatory effect on the GF spike initiation zone (SIZ) in the brain, where they do

335

not make contact? We have now demonstrated that the GF also possesses a spike-initiating

336

region(s) in the VNC, where the Axc could influence spike initiation. We demonstrated this

337

alternative SIZ by driving CsChrimson in the GF, removing the head to remove the SIZ in the

338

brain, and testing headless animals with optical stimulation to reveal GF activity generated in the

339

VNC (Fig. 7J-

). GF activity driven by the light stimulation is revealed by the response of both

340

TTM and DLM with a constant latency between them indicative of GF action potentials (Fig. 7

341

Therefore, the Axcs can directly influence this SIZ in the VNC.

342

343

Unexpectedly, we also note a prolonged response of the GF after the light was turned off (Fig. 7D,

344

7F and 7G). Before light application, the TTMn response probability was low (~20%, black trace,

345

Fig. 7

), during light application the probability increased to almost 100% (blue trace, Fig. 7

)

346

and after light-off activity returned to baseline very slowly (>100 msec). This prolonged response

347

after light-off has been observed in several other situations for CsChrimson. For instance, at the

348

larval neuromuscular junction (NMJ), CsChrimson drove spiking in motor neurons (monitored by

349

muscle synaptic potentials) and exhibited a post-light response hundreds of milliseconds in

350

duration

. In a second example, descending neurons DNp07 and DNp10, when expressing

351

CsChrimson, exhibit a 100-400 msec response after light-off

. Finally, we tested this idea in the

352

GF by expressing CsChrimson in the GF and illuminating the preparation at 590 nM and observed

353

a long tail of TTM response of approximately 100 msec (Fig. 7

blue trace). In brief, a variety of

354

examples show that CsChrimson when activated produces a strong response with a long slow

355

decay after the light is turned off (Fig 7

F-G)

356

357

To confirm the parallel between the simulation and the

in vivo

experiment, we also developed a

358

simulation that more closely matched the

in vivo

experiments (Fig. 7

). A large-scale network of

359

the VNC, composed of leaky integrate-and-fire neuron models, was used to simulate this

360

experiment. The GF was electrically stimulated just below threshold at 24 Hz. In these simulations,

361

the GF only spikes in the presence of the light. Both the simulation and the optogenetic results on

362

the living animal suggest that the Axc-to-GF connection is excitatory, consistent with the

363

connectome and chemical predictions that the synapse is cholinergic. These results agree with our

364

simulation’s predictions that the main effect of the optogenetic activation of the cholinergic axo-

365

axonic neurons (Axcs) is to increase the excitability of the GF (Fig. 7

366

367

Journal Pre-proof

Discussion

373

By analyzing all 1,314 brain-originating descending neurons, we move beyond scattered

374

descriptions of axo-axonic contacts and derive circuit-level design rules for presynaptic

375

modulation. The quantitative principles that emerged within this work are: (i) extreme sparseness

376

showing that between 0.7 to 1.2% of possible DN-to-DN, AN-to-DN, and IN-to-DN pairs form

377

axo-axonic synapses; (ii) a tight linear relation between synaptic strength and partner multiplicity;

378

and (iii) a small-world architecture that distributes integration rather than concentrating it in a rich-

379

club core. Together, these features constitute a wiring grammar for axo-axonic control in the adult

380

Drosophila

motor system.

381

382

Axo-axonic synapses have previously been identified in the

Drosophila

brain connectome

34–37

383

Excitatory cholinergic synapses, like those investigated here, have been observed in the fly

384

olfactory system, where projecting neurons (PN) form lateral PN-to-PN connections, similar to the

385

DN-to-DN connections we identified

. Axo-axonic connections have also been reported between

386

olfactory sensory neurons (OSNs)

. Bristle mechanosensory neurons (BMNs) are also shown to

387

be axo-axonically connected

. These similarities suggest that axo-axonic connectivity principles

388

might be conserved across the VNC and brains in flies, although the functional role of AACs

389

remains unexplored.

390

391

Notably, axo-axonic wiring has revealed an extraordinarily selective, yet globally percolating

392

schema, in contrast to, for example, the serotonin axo-axonic configuration in the hippocampus,

393

which comprises 30%

10,39

or in the cortex, which comprises 19%

10,40

. In the fly VNC, only ~1 %

394

of the possible DN-to-DN pairs form axo-axonic synapses, an order of magnitude sparser than

395

axo-dendritic contacts in either the MANC or FAFB brain data sets

14,41

. Despite their scarcity,

396

these connections are sufficient to incorporate every DN into a single, strongly connected network,

397

yielding short average path lengths (~4 hops) and thus preserving the ability for any axon to gain

398

rapid influence over remote peers. For comparison, the entire network has an average path length

399

of 2.8 (~3 hops). This combination of stringent partner choice with network-wide reach resembles

400

presynaptic inhibition motifs in mammalian spinal cord and chandelier-cell control of cortical

401

pyramidal neurons, suggesting convergent architectural solutions across phyla

42,43

. What accounts

402

for the sparsity of axo-axonic connections? One possibility is wiring economy: boutons on sister

403

axons consume precious membrane and cytoplasmic resources

. Another is developmental

404

lineage. Many high-degree bridge neurons derive from hemilineages that extend across

405

neuromeres

, hinting that growth-cone timing and Slit-Robo–mediated midline crossing

pre-

406

pattern the opportunities for axo-axonic contact. Comparative EM work in other insects and in

407

zebrafish larval spinal cord will clarify whether these archetypes represent a universal motor

408

blueprint.

409

410

As previously found, rich clubs with dense interlinking of high-degree hubs are present in the fruit

411

Drosophila

connectome

. They are also widely present in vertebrates, such as in the cerebral

412

Journal Pre-proof

cortex

47,48

. In contrast, the axo-axonic fly DN network revealed only a weak, transient rich-club

413

coefficient, suggesting that presynaptic modulation in the nerve cord is decentralized. Control must

414

therefore emerge from many mid-degree “broker” neurons (with fewer connections) and with high

415

betweenness rather than from a minority of super-hubs. Such a layout guards against single-point

416

failure and allows different motor modules to be gated independently, consistent with behavioral

417

data showing that flies can blend local reflexes with whole-body maneuvers

418

419

In terms of functional implications, axo-axonic contacts positioned directly on descending axons

420

are ideally placed to veto, amplify, or synchronize commands emanating from the brain before

421

they fan out onto premotor interneurons and motoneurons. We have shown that Axc is amplifying

422

GF activity. Experiments with headless animals demonstrate the existence of a second spike

423

initiation zone (SIZ) in the VNC. We believe this is the target of the Axc synaptic modulation.

424

Choline acetyltransferase (ChAT) immunostaining confirms that Axc type neurons are

425

cholinergic, and optogenetic activation of these cells reliably enhances the GF’s excitability. Our

426

simulations also demonstrate this principle: activating a cohort of eight cholinergic Axc type

427

neurons, which exclusively target the Giant Fibers (GFs), enhances their firing.

428

429

Crucially, however, the GF is not unique; 17 other DNs receive >50 presynaptic partners that

430

occupy bridge positions between functional modules. These neurons may be promising targets for

431

future studies investigating the role of axo-axonic connections in precise behavior modulation.

432

Future opto-physiological experiments probing these DNs will test whether distributed control is

433

the norm rather than the exception.

434

435

Excitatory axo-axonic connections, exemplified by the Axc to GF projection described here, are

436

not unique to this system and have been reported in other organisms, including the well-

437

characterized Mauthner cells in zebrafish

and dopamine cells in mice

. In these examples, the

438

primary function of the excitatory axo-axonic synapse is to increase the release probability of the

439

axon terminal they innervate. This mechanism can even bypass upstream action potentials,

440

representing a hijacking mechanism that may have physiological roles. The findings presented in

441

this study may contribute to a broader understanding of axo-axonic connectivity principles across

442

species

51–55

443

444

In conclusion, our study presents the first comprehensive blueprint of axo-axonic connectivity

445

within a fully reconstructed adult ventral nerve cord system. The network revealed here offers

446

concrete parameters for next-generation models of rapid motor decision making in both

447

invertebrates and vertebrates. By formalizing direct axon-to-axon communications, the work

448

bridges a critical gap between connectome structure and motor function and opens new avenues

449

for dissecting the fastest circuits in the animal kingdom.

450

451

452

Journal Pre-proof

453

RESOURCE AVAILABILITY

454

Lead contact

455

Further information and requests for resources and reagents should be directed to and will be

456

fulfilled by the Lead Contact, Rodrigo Pena (penar@fau.edu).

457

458

Materials Availability

459

This study did not generate new unique reagents.

460

461

Data and Code Availability

462



Electrophysiology recordings have been deposited as .abf files at Mendeley Data: De

463

Oliveira Pena, Rodrigo (2026), “ceballos2026_ephys_data”, Mendeley Data, V1

464

https://data.mendeley.com/datasets/dh4vghvtpm/1

465



All original code used to create the connectome simulation is publicly available at GitHub:

466

https://github.com/pena-rodrigo/ceballos_et_al2026_iscience

467

468

Limitations of the study

469

One limitation of our study is that the connectome used in the analysis was derived from a single

470

male specimen. New datasets are now being released that are sex-specific or can address

471

developmental differences, such as the Female Adult Nerve Cord Connectome (FANC) and the

472

Brain and Nerve Cord Connectome (BANC). All functional whole-network tests are

in silico

, and

473

although experimental verification is provided through optogenetics, the large-scale network based

474

on leaky integrate-and-fire neuron model respects synapse counts but ignores conductance

475

dynamics and neuromodulators influences. Furthermore, EM of the MANC connectome (v1.2.1)

476

did not detect gap-junctional coupling, suggesting that the present map under-reports electrical

477

axo-axonic contacts that could further reshape information flow

51–55

478

479

Lastly, dendro-axonic synapses account for only 2% of all synapses

25,56

. Thus, we cannot rule out

480

the possibility that some of the synapses investigated in this study are dendro-axonic, although the

481

vast majority are axo-axonic. Nevertheless, we curated a large number of EM images (Fig. S1) to

482

demonstrate the presence of mitochondria in axo-axonic synapses of descending neurons,

483

suggesting that this structure can be used as a marker of axo-axonic synapses. Based on this, we

484

found that synapses of AN08B098 onto Giant Fiber contain presynaptic mitochondria, suggesting

485

that these synapses are axo-axonic.

486

487

We found that the number of neurons in the Gal4 line was larger than the number of neurons found

488

in the connectome that belong to the type AN08B098. We cannot establish the source of this

489

difference. Although we characterized two other Gal4 lines (Bloomington stock #87537 and

490

#87664) and we found similar counts.

491

492

Journal Pre-proof

The long tail of response following activation of Axc in Fig 7 could be explained by multiple

493

underlying mechanisms; local recurrent excitatory loop between GF and Axc, unknown

494

descending neurons that feedforward to this circuit, but we are unable to identify the source of

495

these inputs.

496

497

Acknowledgments

498

R.P. was funded by an IBRO Collaborative Research Grants. The authors thank the FAU Jupiter

499

Life Sciences Initiative (JLSI) for funding. This material is also based upon work supported by

500

the U.S. Department of Education under Research and Development Infrastructure Grant No.

501

P116H230018. Any opinions, findings, and conclusions or recommendations expressed in this

502

material are those of the author(s) and do not necessarily reflect the views of the U.S. Department

503

of Education.

504

505

Author contributions

506

C.C.C. and R.P. were responsible for the conceptualization of the work. C.C.C., J.L., T.R., D.S.,

507

S.J., and C.S. analyzed the data and made the figures. R.P. worked on graph-theoretical measures.

508

D.S. performed and analyzed the electrophysiology recordings. S.J. and C.S. performed

509

immunofluorescence. K.R. maintained fly stocks and set up experiment crosses. T.R. built and ran

510

the network simulations. R.P., C.S., and R.M. supervised the research and secured funding. All

511

authors participated in the writing, reviewed, and approved the manuscript.

512

513

Declaration of interests

514

The authors declare no competing interests.

515

516

Figure 1. Descending neurons (DNs) and their axo-axonic inputs in the VNC.

)-(

) Exemplary Neuroglancer

517

reconstructions of three DNs of interest (scale bars indicate 500

𝜇

m). (

)

(

) EM images illustrate axo-axonic

518

connections formed by the three reconstructed neurons: (

) shows DNp01 to DNp01 connectivity, (

) shows DNd03

519

connectivity with DNp42, and

) shows DNg02 to DNg02 connectivity (same colors as in A-C; scale bars indicate

520

500 nm). Red circles identify the locations of presynaptic T-bars; blue circles show postsynaptic densities. Circles

521

were enlarged for visibility. (

) Histogram of the number of axo-axonic neurons innervating one DN. Presynaptic

522

neurons can be DN, AN, or IN. (

) Histogram of the cumulative strength of the synaptic connections for each DN.

523

Presynaptic neurons can be DN, AN, or IN. (

) Scatter plot and linear regression of the number of axo-axonic neurons

524

vs their synaptic strength. R

= 0.92 and

p <

0.001.

525

526

Figure 2. Analysis of morphology and synaptic strength between DN-to-DN axo-axonic connections.

(

)

527

Examples of DNs include (legend colors indicate pre- and postsynaptic in the figure): (A) low connectivity with simple

528

morphology, (B) moderate connectivity and morphology, and (C) high connectivity with complex morphology.

529

Additionally, (D) low connectivity with complex morphology and (E) high connectivity with simple morphology

530

represent exceptions. These neurons generally align with the overall trend—a linear correlation between

531

morphological complexity and synaptic strength—while (D) and (E) are outliers. Dots indicate synapse locations. (

)

532

Kernel density estimation (KDE) of the scatter data of total input strength vs cable length, (

) Scatter plot of total

533

input strength vs cable length and the corresponding linear regression, polynomial regression of a quadratic function,

534

and symbolic regression.

535

Journal Pre-proof

Figure 3. Neurons that provide exclusive axo-axonic connections to DNs.

(

) Sketch of single type biased vs (

)

536

multi-target neuron. Single-type biased neurons project a significant proportion of their outgoing connections to a

537

single DN, while unbiased neurons form several strong connections among partners. (

) Number of exclusive neurons

538

per DN for both MANC v1.2.1 (blue histogram) and Male-cns v0.9 (orange histogram). The Giant Fibers (DNp01;

539

marked by an asterisk) are among a small population of DNs that receive most single-type biased connections, i.e., as

540

seen in (D), there are approximately 16 neurons that innervate DNp01 in MANC and 11 in Male-cns. (

) For each

541

dataset, the fraction of outputs made from neurons onto their most strongly targeted DN type was computed per

542

neuron, and a dataset-specific cutoff was identified from the distribution as the point separating the population of

543

unbiased neurons and single-type biased neurons (MANC 0.27, male-cns 0.1). Single-type biased neurons across both

544

datasets had similar properties, including high target preference, reduced target entropy, and reduced off-target

545

connection strength (Kolmogorov-Smirnov test, p < 0.001). We further refined this by removing neurons whose off-

546

target connections had 15 or greater synapses, suggesting non-influence over their synaptic partners

547

548

Figure 4. Axo-axonic connectivity analysis.

(

) Connectivity matrix of pairs of DN-to-DN pairs. Only 1% of all

549

possible connections were identified, suggesting that DN-to-DN axo-axonic connections are extremely rare. (

)

550

Connectivity matrix of pairs of AN-to-DN pairs. Only 1.2% of all possible connections were found, which suggests

551

that AN-to-DN axo-axonic connections are also exceptionally rare. (

) Connectivity matrix of pairs of IN-to-DN

552

pairs. Only 0.7% of all possible connections were observed, indicating that IN-to-DN axo-axonic connections are

553

uncommon. (

) Pie chart of predicted neurotransmitter of DN of the different type (left, e.g., DNp01 -> DNp02) and

554

same type (right, e.g., DNp01 ->DNp01) that make pairs of axo-axonic connections. (

) Pie chart of predicted

555

neurotransmitter of AN-to-DN and IN-to-DN axo-axonic connections. (

) Histograms of the number of (F1)

556

presynaptic DN (in-degree) and (F2) postsynaptic DN (out-degree) for DN-to-DN pairs. (

) Histograms of the

557

number of (G1) presynaptic AN and (G2) postsynaptic DN for AN-to-DN pairs. (

) Histograms of the number of

558

(H1) presynaptic IN and (H2) postsynaptic DN for IN-to-DN pairs. (

) Rich-club analysis of the DN-to-DN axo-

559

axonic network. (

) Connectivity matrix of pairs of AN/DN/IN-to-DN connections. Filled circles follow the same

560

color scheme as in Fig. 1

. Cyan square identifies Axc-type neurons. Dashed green lines and gray tones for dots

561

separate ANs, DNs, and INs. (

) Examples of each category from DN-to-DN, AN-to-DN, and IN-to-DN. (From

562

left to right: (

) DNx01-DNx01, (

) DNb04-DNb04, (

) AN08B099-DNp01, (

) AN10B019-DNp01, (

)

563

IN11A001-DNp01 and (

) IN06B059-DNp01). (

) Percentage of contralateral and ipsilateral connections between

564

DN-to-DN, AN-to-DN, and IN-to-DN. (

) Pie chart of predicted neurotransmitter of DN to contralateral DN, AN

565

to contralateral DN, and IN to contralateral DN of axo-axonic connections.

566

Figure 5. Axcs make cholinergic axo-axonic synapses on the GFs as predicted by the MANC connectome.

567

(

) AN08B098-type neurons (green) form axo-axonic connections at distinct locations along Giant Fiber (red, GF)

568

axons. The scale bar shown is 20 µm. (

) The AN08B098-GF connections reconstructed in Neuroglancer

569

(Neuroglancer scale 530.45 µm/vh) mirror the morphology seen with fluorescence in

. (

) We validated

570

AN08B098-to-GF connectivity using anti-bruchpilot to stain for T-bars in active zones (white) along AN08B098-

571

type neurons and colocalized the resulting fluorescence with the GFs (red), which were filled with

572

tetramethylrhodamine. The scale bar shown is 20 µm. (

) Neuroglancer reveals presynaptic sites at similar locations

573

seen in colocalized fluorescent image (Neuroglancer scale 530.45 µm/vh). (

E-F

) XY and XZ plane views of the

574

preparation shown in

and

. The zoomed-in 6 µm (scale bars 3 µm) inlays show AN08B098 forming a single

575

synapse with the GF in E. The yellow arrows detail the precise location AN08B098 forms the Brp-positive chemical

576

synapse (white) to the GFs in F. Scale bars shown are 5 µm. (

) EM images showing monosynaptic connections

577

between single GF (green) and AN08B098 neurons identified by the following MANC id: (G-H) 21041, (I) 21589,

578

(J) 23949, (K) 152261, (L) 16900, (M) 20444, (N) 22275, and (O) 24038. Pre-and postsynaptic sites are detected in

579

EM slices using a 3D convolutional neural network to identify T-bars (cyan dots) and postsynaptic densities

580

(PSDs, magenta dots). (

) We expressed anti-choline acetyltransferase (anti-ChAT) and anti-GFP in AN08B098

581

neurons. Anti-ChAT colocalizes to AN08B098 cells, with particularly bright staining in the cell bodies. This finding

582

Journal Pre-proof

suggests acetylcholine synthesis is present within these cells, validating connectome transmitter predictions for

583

AN08B098. Scale bars shown are (P) 20 µm and (Q-S) 5 µm.

584

Figure 6.

Simulations using the whole VNC network.

(

) Stimulation of the GF and corresponding raster plot of

585

all the neurons in the VNC and voltage trace of GF response (spikes are shown as large black dots at the bottom of

586

the raster plot, also marked with an arrow). (

) Schematic of the GF stimulation by Axc type neurons and

587

corresponding raster plot of all the neurons in the VNC and voltage trace of GF response (GF spikes are represented

588

by large black dots at the bottom of the raster plot, while Axc spikes are represented by large red dots). (

) Similar to

589

previous plots, but now the stimulation is driving non-Axc (nAxc), and the Axc is removed from the simulation. (

)

590

Similar to previous plots, but now the stimulation is driving Axc cells, and nAxc are deleted. (

) Similar to

, but now

591

Axc cells are deleted. (

) Histogram comparison of subthreshold voltages during GF-only and GF+Axc stimulation.

592

(

) Similar to

, but now the histogram shows the difference between silencing nAxc and stimulation in Axc vs.

593

silencing Axc and stimulation in nAxc. (

) Violin plots showing statistical analysis of some of the comparisons. The

594

overall excitability of the GF increases when the Axc neurons are activated, leading to higher firing probability for

595

the GF.

596

597

Figure 7.

Optogenetic activation of the Giant Fiber system.

(

) Sketch of the electrophysiological recordings with

598

optogenetic activation of axo-axonic cells (Axc). (

) Example of raw traces for control (no light application; scale bar

599

is 5mV, time is shown on the x-axis). A zoomed-in panel is presented, showing a constant difference in latencies of

600

~0.6 ms between TTM and DLM indicative of GF spiking (scale bar is 5 mV and 1 ms). (

) Simulation of a

601

computational model of the VNC connectome emulating the experimental results under subthreshold conditions and

602

light stimulation (for the voltage, scale bar is 10mV, time is shown on the x-axis). Note that GF only spikes when the

603

light is applied and is time-locked with the electrical pulses. Action potentials are indicated with arrows. (

) In vivo

604

example of raw traces when light was applied (scale bar is 10mV, time is shown on the x-axis). A zoomed-in panel is

605

presented, showing difference in latencies of ~0.6 ms between TTM and DLM responses (scale bar is 10 mV and 1

606

ms). (

) Response probability per fly (n = 5) under no-light conditions. (

) response probability per fly (n = 5) under

607

light conditions (light was applied during the blue shaded region). (G) Average and SEM of response probability

608

(paired t-test, * =

<0.001).

Spike initiation in the VNC

. (

) Sketch of the electrophysiological recordings with

609

optogenetic activation of the GF before and after decapitation. (

) Recordings from both TTM and DLM before

610

decapitation (scale bar is 20 mV and 100 ms). The constant latency between TTM and DLM indicates the presence of

611

GF action potentials. A zoomed-in panel shows the constant latency between TTM and DLM ~0.77 ms (scale bar is

612

20 mV and 1 ms). (

Response of the TTM to optical stimulation in a specimen driving CsChrimson in the GF as

613

above. The blue curve represents the response of the TTM in animals with the head on, exhibiting a long tail of post-

614

light response. The black curve is after the head was removed to remove the spike-initiating zone in the brain. (

)

615

Recordings from both TTM and DLM after decapitation (scale bar is 50 mV and 100 ms). The constant latency

616

between TTM and DLM indicates the presence of GF action potentials. A zoomed-in panel shows the constant latency

617

between TTM and DLM ~0.76 ms (scale bar is 50 mV and 5 ms).

618

619

Journal Pre-proof

STAR Methods

620

621

Detailed methods include the following:

622

• KEY RESOURCES TABLE

623

• RESOURCE AVAILABILITY

624

• EXPERIMENTAL MODELS AND SUBJECT DETAILS

625

• METHODS DETAILS

626



Electrophysiology

627



Solution preparations

628



Optogenetics

629



Dye-injection

630



Immunostaining

631



Confocal Microscopy

632



Image Analysis

633



Computational model

634



Neuron model

635



Synapse model

636

• QUANTIFICATION AND STATISTICAL ANALYSIS

637



Data acquisition and analysis

638



Connectivity analysis

639



Network analysis

640



Statistical and visual analysis

641



Software and libraries

642

643

Journal Pre-proof

EXPERIMENTAL MODELS AND SUBJECT DETAILS

644

Fly Genetics

645

Flies were raised at 25°C with a natural (12-hr) light cycle, and data were collected from both male

646

and female adults 2-4 days old. The Gal4 drivers used in our study were selected through

647

NeuronBridge

. Neurons of interest were identified through NeuPrint+, then identified using

648

NeuronBridge software v3.2.6, and finally chosen based on similarity score and visual accuracy.

649

A split-Gal4 driver specific to AN08B098 (Axc) was identified using NeuronBridge software

650

v3.2.6 and chosen based on similarity score. SS25451 contains w[1118]; VT063643-p65.AD;

651

VT059224-GAL4.DBD (BDSC #89495) (see Fig S4). Constructs used were 20XUAS-IVS-

652

CsChrimson.mVenus (#55136), and 20XUAS-IVS-mCD8::GFP (BDSC #32194).

653

654

METHOD DETAILS

655

Electrophysiology

656

Flies were mounted dorsal side up in dental wax after CO

anesthesia. Giant Fibers were activated

657

by extracellular stimuli applied using Grass stimulators with tungsten wire electrodes placed in the

658

brain through the eyes, while a tungsten wire electrode was placed in the fly's abdomen to ground

659

the circuit

. Muscle recordings were obtained using glass microelectrodes placed directly into

660

both jump (TTM) and flight (DLM) muscles

18,30,55,58,73

. Recording microelectrodes were filled with

661

O'Dowd's Drosophila saline

We do not directly monitor GF spikes but instead use TTM

662

responses as a reporter. Extracellular stimulation of the GF pathway activates both the TTM and

663

the DLM with a short and constant latency between them (~0.6 ms) as described in the early

664

recordings of this type in the 1980s

37,59

. Action potentials in the GF are indicated by synaptic

665

potentials in both the TTM and the DLM with the constant latency between them (see Figure 7) as

666

described in a more recent protocol for the preparation

. It has long been assumed that the GF has

667

the lowest threshold for descending neurons using this method of brain stimulation due to its large

668

size and the presence of a spike initiation region in the brain. Recently, a spike initiation zone

669

(SIZ) has been confirmed to be located in the brain at the dendrite-axon bifurcation

. The Axon

670

Digidata 1440A Data Acquisition System was used to digitize recordings which are then visualized

671

using Clampex software (version 10 & 11, Molecular Devices). Muscle responses were obtained

672

by stimulating at 24 Hz.

673

674

Solution preparations

675

Phosphate buffered saline was made from tablets purchased from Fisher BioReagents (BP2944-

676

100). One tablet was dissolved in double distilled water to yield a solution of 10 mM Phosphate

677

Buffer, 2.7 mM KCl, and 137 mM NaCl. Solutions were stored at the recommended 4



678

679

All trans retinal (500 mg), which was purchased from Sigma-Aldrich, is dissolved in 17.6 ml

680

absolute ethanol to make 100 mM stock solutions. 100 µl of 100 mM stock solution is

681

transferred to small tubes and wrapped in aluminum foil and kept in -20°C freezer (ATR should

682

be kept away from light since it is sensitive to light and it would be degraded, thus becoming

683

Journal Pre-proof

ineffective if it is exposed to light for a long time). To prepare fly food supplemented with 400

684

µM ATR, 5 mL fly food is dissolved in microwave. The food is left to cool down, then 20 µl of

685

400 µM ATR is mixed well with fly food or 20 µl of absolute ethanol is mixed with food as a

686

control. The food vial is wrapped in aluminum foil then the food is left until it was solidified

687

well; otherwise, the flies get stuck in the wet food. The flies are transferred from its vial to ATR

688

containing vial and kept in a dark place (to keep the ATR from degradation) at room temperature

689

22-23°C.

690

691

O’Dowd saline was made by dissolving: NaCl (0.59 g); NaHCO

(0.174 g); KCl (0.0224 g); MgCl

692

(0.0814 g); CaCl

(0.0148 g); NaH

(0.0138 g); Trehalose (0.1892 g); and Sucrose (1.54 g) in

693

100 mL double distilled water. Saline was kept in 4



C for storage or in room temperature as

694

needed.

695

696

Optogenetics

697

We drove channelrhodopsin expression in neurons of interest using the UAS-Gal4 system. Flies

698

expressing csChrimson (BDSC #55136) were crossed with our driver line to produce: w1118;

699

VT063643-p65/+; 20XUAS-IVS-csChrimson.mVenus/VT059224-GAL4.DBD. Adults were

700

placed in the dark at 50% relative humidity on standard cornmeal-molasses food, supplemented

701

with all-trans-retinal (0.4 mM post-eclosion) for two days before testing. Light stimulation was

702

performed for 100 ms while flies are mounted in dental wax as described above. We used a 590

703

nm high-power fiber-coupled LED (1.5 mW maximum light intensity, Thorlabs M590F3) with an

704

optic fiber of diameter 200 µm and N.A. 0.22 to activate the mVenus csChrimson (M118L02). We

705

tested various placements for the optic fiber and found the strongest activation of csChrimson to

706

occur when the fiber was less than 1 cm from the sample and aimed at the neck. Light intensity

707

was controlled via T-Cube LED Driver (Thorlabs LEDD1B) and kept at maximum power during

708

testing. Each sample received electrical-only, optogenetic-only or paired opto-electrical

709

stimulation. If Axc is stimulated only, no response is recorded, meaning that csChrimson alone is

710

unable to elicit spikes in the GF.

711

712

We also conducted two control experiments for the optogenetic protocol (Fig. S2). The first control

713

used flies of the same genotype as the experimental group but were not raised on food

714

supplemented with all-trans-retinal. The second control involved flies that were raised on retinal-

715

supplemented food but lacked the UAS-csChrimson.mVenus transgene (w1118; VT063643/+;

716

VT059224-GAL4/VT059224-GAL4). The results are shown in Figure S2 B-C.

717

718

Dye-injection

719

The VNC and connective of young adult male flies were dissected and kept in cold O’Dowd’s

720

saline within a Sylgard 184-filled dissection dish. Samples were immediately transferred to a

721

Nikon Eclipse E600FN fluorescent microscope. A tetramethylrhodamine-labeled dextran dye

722

(10,000MW, Invitrogen, D1817) solution was backfilled into a pulled glass electrode tip (~30 M



723

Journal Pre-proof

resistance tip) and filled with O’Dowd’s saline. Electrodes were placed into a 3D hydraulic

724

micromanipulator. The electrode was lowered into place, penetrating the GF axons near the neck

725

connective. A Gettings Model 5A amplifier was used to inject positive current pulses (between 1

726

nA and 5 nA), with a reference electrode placed into the saline-filled dish. GFs were filled with

727

dye for approximately 5 minutes each.

728

729

Immunostaining

730

A mouse primary antibody against ChAT (1:20, DSHB, 4B1) was used to confirm cholinergic

731

cells and coupled to secondary Goat anti-mouse AlexaFluor 647 (1:500, 115-605-003). A mouse

732

primary antibody against bruchpilot (1:50, DSHB, NC82) was used to label presynaptic chemical

733

active zones and coupled to secondary Goat anti-mouse AlexaFluor 647 (1:500, 115-605-003).

734

Primary rabbit anti-GFP (1:500, Invitrogen A11122) and secondary goat anti-rabbit (1:500,

735

Jackson Immuno, 111-225-144) were used to amplify the fluorescence of mCD8::GFP. mVenus-

736

tagged csChrimson. Fluorescence was amplified using a conjugated antibody (1:500, Millipore,

737

MABE1906). Primary antibody solutions contained 3% BSA and 0.3% Triton X-100 in PBS. The

738

secondary antibody solutions contained PBS.

739

740

Samples were dissected and processed according to the dye-injection methods above. Samples

741

were then immediately fixed in PBS + 4% PFA for 25 minutes at room temperature (22°C).

742

Samples were washed with PBS and placed on a rotator (60RPM) for 30 minutes. Samples were

743

permeabilized with PBS + 0.5% Triton X-100 for 45 minutes at 22°C. Primary antibodies were

744

added, and samples were stored at 4°C for 48 hours. Samples were then washed on a rotator with

745

PBS for 4 hours at 22°C, changing the solution every 30 minutes. Secondary antibodies were

746

added, and samples were stored at 4°C for 24 hours. Samples were then washed on a rotator with

747

PBS for 4 hours at 22°C, changing the solution every 30 minutes. Samples were dehydrated with

748

a series of ethanol steps, 50%, 70%, 90%, and 100% for 8 minutes each step. Samples were

749

trimmed and embedded in 100% methylsalycylate, covered with cover slips, sealed, and imaged

750

within 24 hours.

751

752

Confocal Microscopy

753

Samples were imaged on a Nikon AXR confocal using an NSPARC detector and a 60x oil

754

immersion objective. Z-dimension step size was 194



m for all results displayed. Laser channels

755

used were 488 nm, 561 nm, and 640 nm. Channels were imaged in series using filters to avoid

756

cross-excitation of close wavelengths, and 2x integration was used.

757

758

Image Analysis

759

Z-stack files were imported into Imaris software v10.2.0, and the fluorescence intensity of each

760

channel was adjusted for best fit. GF and AN08B098 (Axc) channels were surfaced using a

761

thresholding method that matched fluorescence in each Z dimension. Masks of each channel were

762

Journal Pre-proof

created for each surface to detect colocalization of Brp or ChAT. Images were acquired in volume

763

view and exported.

764

765

Computational model

766

To study the impact of axo-axonic projections onto DNp01 (GF), we constructed an

in-silico

model

767

of the ventral nerve cord (VNC) by utilizing the entire MANC v1.2.1 connectome. First, we

768

scrubbed the connectivity data by omitting connections with a total strength less than 6, then we

769

excluded all connectivity data where the neurotransmitter type, neuron type, or neuron ID was

770

unknown or invalid. This resulted in a total of 23,437 valid neurons and 1,152,548 connections

771

between them, with 619,684 excitatory (cholinergic) and 532,864 inhibitory (glutamatergic or

772

GABAergic) connections. In the fly, cholinergic receptors are excitatory, whereas glutamatergic

773

and GABAergic receptors are inhibitory

774

775

Neuron model

776

To achieve biological plausibility while avoiding computational intractability, leaky integrate-and-

777

fire (LIF) neuron models were used:

778

𝐶

𝑚

𝑑𝑣

𝑑𝑡

− 𝑔

𝐿

(

𝑣 − 𝑉

𝑟𝑒𝑠𝑡

)

−

𝑔

𝑎𝑐ℎ

( 𝑣 −

𝐸

𝑎𝑐ℎ

)

−

𝑔

𝑔𝑎𝑏𝑎

(

𝑣 −

𝐸

𝑔𝑎𝑏𝑎

)

− 𝑔

𝑔𝑙𝑢𝑡

(

𝑣 −

𝐸

𝑔𝑙𝑢𝑡

(1)

779

780

Unlike other models

, we employ distinct parameters for the GF compared to the rest of the

781

neurons in the connectome, i.e., the membrane resistance. In Eq. (1),

𝑣

is the neuron membrane

782

potential,

⁡𝑔

𝐿

𝑔

𝑎𝑐ℎ

, 𝑔

𝐺𝐴𝐵𝐴

, and

𝑔

𝑔𝑙𝑢𝑡

are the conductances for leak channels and acetylcholine

783

(nicotinic), GABA (GABA

), and glutamate ionotropic receptors (nAChR, GABA

R, and GluCl,

784

respectively). Similarly,

𝑉

𝑟𝑒𝑠𝑡

𝐸

𝑎𝑐ℎ

, ⁡𝐸

𝐺𝐴𝐵𝐴

, and

𝐸

𝑔𝑙𝑢𝑡

are their respective reversal potentials.

785

Since GABA

R and GluR are chloride-permeable channels, their reversal potentials were obtained

786

from chloride reversal potential

𝐶

𝑚

is the membrane capacitance. These are parameterized with

787

values obtained from the literature, which investigates the biophysical properties of fruit fly

788

synapses. Furthermore, we collected physiologically relevant values for the membrane

789

capacitance, resistance, and leak reversal potential

Simulations of these neurons were made using

790

the BRIAN2 simulator

791

792

Synapse model

793

For each of the synaptic currents (

𝑥 = 𝑎𝑐ℎ, 𝑔𝑎𝑏𝑎, 𝑔𝑙𝑢𝑡

), the conductance obeys

794

795

𝑑𝑔

𝑥

𝜏

𝑥

= −𝑔

𝑥

(2)

796

which increases by

𝐽

𝑥

upon a presynaptic spike. For parameters, see table.

797

798

Journal Pre-proof

Parameter

Symbol

Value

Ref.

Acetylcholine receptor (nAChR)

conductance

𝐽

𝑎𝑐ℎ

0.27 nS

nAChR decay time

𝜏

𝑎𝑐ℎ

1.1 ms

nAChR reversal potential

𝐸

𝑎𝑐ℎ

0 mV

GABA receptor (GABA

conductance

𝐽

𝐺𝐴𝐵𝐴

0.8 nS

Estimated

GABA

R decay time

𝜏

𝐺𝐴𝐵𝐴

5.4 ms

GABA

R reversal potential

𝐸

𝐺𝐴𝐵𝐴

-60 mV

60,62

Glutamate receptor (GluCl) conductance

𝐽

𝑔𝑙𝑢𝑡

0.8 nS

Estimated

GluCl decay time

𝜏

𝑔𝑙𝑢𝑡

5 ms

GluCl reversal potential

𝐸

𝑔𝑙𝑢𝑡

-60 mV

Membrane Resistance (GF)

𝑅

𝑚

25 MOhm

Membrane Resistance (other neurons)

𝑅

𝑚

300 MOhm

Resting membrane potential

𝑉

𝑟𝑒𝑠𝑡

-55 mV

61,68

Action potential threshold

𝑉

𝑡ℎ

-40 mV

Reset potential

𝑉

𝑟𝑒𝑠𝑒𝑡

-55 mV

Estimated

Refractory period

𝜏

𝑟𝑒𝑓

2 ms

Estimated

Membrane capacitance

𝐶

𝑚

72 pF

List of parameters used in the connectome simulation of the VNC in this paper. For each of our simulations, we ran

799

the entire VNC network with a step-size of 0.1 ms during 2000 ms. Furthermore, for each simulation, each neuron in

800

the VNC network was provided with an excitatory Poisson input of 30 Hz at a synaptic conductance of 5 nS to simulate

801

typical endogenous activity driven by neurons in the fly brain. During stimulation experiments, neurons DNp01,

802

DNg02, and/or DNd03 were stimulated with additional Poisson input at 10 Hz at a synaptic conductance of 40 nS. In

803

Figure 8, the same network was used, but the GF was electrically stimulated just below threshold with a Poisson-

804

distributed stimulation at 24 Hz.

805

806

807

QUANTIFICATION AND STATISTICAL ANALYSIS

808

809

Data acquisition and analysis

810

Data for VNC neurons used in the study came from the publicly available

Drosophila

male adult

811

nerve cord connectome

. NeuPrint+ API

26,28

access is granted to registered users, and Python

812

accessibility is maintained through the neuPrint-python package

. The latest

Drosophila

MANC

813

data can be accessed here: https://www.janelia.org/project-team/flyem/manc-connectome.

814

815

Journal Pre-proof

Connectivity analysis

816

We retrieved all simple synaptic connections using neuprint.fetch_simple_connections

26–28

817

separately for: outputs (presynaptic AN, IN, and DN as presynaptic) and Inputs (postsynaptic DN).

818

Connections were summed by partner neuron and filtered to those with total strength ≥ 6. For each

819

DN, we then counted the number of distinct AN, IN, and DN partners in both the output and input

820

directions.

821

822

Network analysis

823

Network analysis was performed using Python with the NetworkX

library to extract quantitative

824

measures of connectivity. From the structural connectivity matrix, we calculated a comprehensive

825

set of graph-theoretical metrics to characterize the network topology. These included basic

826

properties (number of nodes and edges, network density), integration measures (average path

827

length, diameter), segregation indices (clustering coefficient, modularity with community

828

detection using the Louvain algorithm), centrality metrics (degree, betweenness, closeness, and

829

eigenvector centrality), and organizational principles (degree assortativity, rich club coefficient).

830

831

The rich club analysis was specifically conducted across multiple degree thresholds to identify

832

potential hierarchical organization among high-degree hub regions. All weighted graph measures

833

incorporated the synaptic strengths from the original connectivity matrix. This multimodal graph-

834

theoretical approach allowed for a detailed characterization of both local and global properties of

835

the network's structural connectome, revealing important aspects of neural communication

836

efficiency, functional segregation, and network resilience.

837

838

Statistical and visual analysis

839

SWC analysis was conducted in Python using the Navis client

. We stored morphology and

840

input/output counts using the Pandas library. Relationships between morphology metrics (cable

841

length, branch points, node count) and partner counts were assessed with Pearson correlation and

842

linear regression (scipy.stats.pearsonr, linregress). Results were visualized with Matplotlib

843

Network‐style connection diagrams were rendered interactively with ipycytoscape.

844

845

Software and libraries

846

NeuPrint+ analyses were performed in Python 3.12.4 using the neuprint-python API. We used the

847

Janelia NeuPrint client (neuprint), and Navis (v1.10.0) for data access and network building; data

848

manipulation and grouping were done with pandas v2.2.2 and NumPy v1.26.4; graph‐based

849

operations with NetworkX v3.2.1; interactive visualization with ipycytoscape v1.3.3 and

850

ipywidgets v8.1.5; plotting and PDF export with Matplotlib v3.10.1; and statistical analyses

851

(Pearson’s r, linear regression) with SciPy

v1.13.1. Auxiliary parsing and utility functions relied

852

on Python’s built-in json, re, itertools, and warnings modules.

853

854

Journal Pre-proof

References

859

Brette, R. Theory of axo-axonic inhibition.

PLoS Comput. Biol.

860

e1013047 (2025).

861

Schneider-Mizell, C. M.

et al.

Structure and function of axo-axonic

862

inhibition.

Elife

, (2021).

863

Stretton, A. O. W.

et al.

Structure and physiological activity of the

864

motoneurons of the nematode Ascaris.

Proc. Natl. Acad. Sci. U. S. A.

865

(1978).

866

Pearson, K. G. & Goodman, C. S. Presynaptic inhibition of transmission

867

from identified interneurons in locust central nervous system.

868

Neurophysiol.

, (1981).

869

Atwood, H. L. & Kwan, I. Reciprocal axo-axonal synapses between

870

excitatory and inhibitory neurons in crustaceans.

Brain Res.

174

, (1979).

871

Altrup, U., Speckmann, E. J. & Caspers, H. Differentiation between anti- and

872

orthodromic responses to nerve stimulation in neurons with axo-axonal

873

synapses (Helix pomatia).

Neurosci. Lett.

, (1979).

874

Andersen, P., Eccles, J. C., Schmidt, R. F. & Yokota, T. Depolarization of

875

presynaptic fibers in the cuneate nucleus.

J. Neurophysiol.

, (1964).

876

Wendell-Smith, C. P., Blunt, M. J., Baldwin, F. & Paisley, P. B. Neurone -

877

Satellite cell relationship.

Nature

205

, (1965).

878

Gobel, S. & Dubner, R. Fine structural studies of the main sensory trigeminal

879

nucleus in the cat and rat.

Journal of Comparative Neurology

137

, (1969).

880

10.

Cover, K. K. & Mathur, B. N. Axo‐axonic synapses: Diversity in neural

881

circuit function.

Journal of Comparative Neurology

529

, 2391–2401 (2021).

882

11.

Dudok, B.

et al.

Recruitment and inhibitory action of hippocampal axo-

883

axonic cells during behavior.

Neuron

109

, (2021).

884

12.

Lacoste, A. M. B.

et al.

A Convergent and Essential Interneuron Pathway for

885

Mauthner-Cell-Mediated Escapes.

Current Biology

, 1526–1534 (2015).

886

13.

Takemura, S.-Y.

et al.

A Connectome of the Male \textit{Drosophila}

887

Ventral Nerve Cord.

bioRxiv

(2023).

888

14.

Cheong, H. S. J.

et al.

Transforming descending input into behavior: The

889

organization of premotor circuits in the Drosophila Male Adult Nerve Cord

890

connectome.

bioRxiv

(2023).

891

15.

Marin, E. C.

et al.

Systematic annotation of a complete adult male

892

Drosophila nerve cord connectome reveals principles of functional

893

organisation.

Elife

, (2024).

894

16.

Lin, A.

et al.

Network statistics of the whole-brain connectome of

895

Drosophila.

Nature 2024 634:8032

634

, 153–165 (2024).

896

17.

Clements, J.

et al.

neuPrint: Analysis Tools for EM Connectomics.

bioRxiv

897

2020.01.16.909465 (2020) doi:10.1101/2020.01.16.909465.

898

18.

Allen, M. J., Godenschwege, T. A., Tanouye, M. A. & Phelan, P. Making an

899

escape: Development and function of the Drosophila giant fibre system.

900

Seminars in Cell and Developmental Biology

vol. 17 Preprint at

901

https://doi.org/10.1016/j.semcdb.2005.11.011 (2006).

902

19.

Asinof, S. K. & Card, G. M. Neural Control of Naturalistic Behavior

903

Choices.

Annu. Rev. Neurosci.

, 369–388 (2024).

904

Journal Pre-proof

20.

Cheong, H. S. J.

et al.

Transforming descending input into behavior: The

905

organization of premotor circuits in the Drosophila Male Adult Nerve Cord

906

connectome.

bioRxiv

(2023).

907

21.

Stürner, T.

et al.

Comparative connectomics of Drosophila descending and

908

ascending neurons.

Nature 2025

1–15 (2025) doi:10.1038/s41586-025-

909

08925-z.

910

22.

Takemura, S.

et al.

A Connectome of the Male Drosophila Ventral Nerve

911

Cord.

Elife

, (2024).

912

23.

Marin, E. C.

et al.

Systematic annotation of a complete adult male

913

Drosophila nerve cord connectome reveals principles of functional

914

organisation.

Elife

, (2024).

915

24.

Schneider-Mizell, C. M.

et al.

Quantitative neuroanatomy for connectomics

916

in Drosophila.

Elife

, (2016).

917

25.

Plaza, S. M.

et al.

neuPrint: An open access tool for EM connectomics.

918

Front. Neuroinform.

, (2022).

919

26.

Berg, S. & Schlegel, P. neuprint-python. Preprint at

920

https://doi.org/10.5281/zenodo.7592899 (2017).

921

27.

Clements, J.

et al.

neuPrint: Analysis Tools for EM Connectomics.

bioRxiv

922

2020.01.16.909465-2020.01.16.909465 (2020)

923

doi:10.1101/2020.01.16.909465.

924

28.

Bates, A. S.

et al.

Distributed control circuits across a brain-and-cord

925

connectome. Preprint at https://doi.org/10.1101/2025.07.31.667571 (2025).

926

29.

Allen, M. J. & Godenschwege, T. A. Electrophysiological recordings from

927

the Drosophila giant fiber system (GFS).

Cold Spring Harb. Protoc.

928

(2010).

929

30.

Klapoetke, N. C.

et al.

Independent optical excitation of distinct neural

930

populations.

Nat. Methods

, (2014).

931

31.

Ache, J. M.

et al.

Neural Basis for Looming Size and Velocity Encoding in

932

the Drosophila Giant Fiber Escape Pathway.

Current Biology

, (2019).

933

32.

Scheffer, L. K.

et al.

A connectome and analysis of the adult drosophila

934

central brain.

Elife

, (2020).

935

33.

Manoim, J. E., Davidson, A. M., Weiss, S., Hige, T. & Parnas, M. Lateral

936

axonal modulation is required for stimulus-specific olfactory conditioning in

937

Drosophila.

Current Biology

, 4438-4450.e5 (2022).

938

34.

Gruber, L.

et al.

The unique synaptic circuitry of specialized olfactory

939

glomeruli in Drosophila melanogaster. Preprint at

940

https://doi.org/10.7554/eLife.88824.1 (2023).

941

35.

Calle-Schuler, S. A., Santana-Cruz, A. E., Kmecová, L., Hampel, S. &

942

Seeds, A. M. A comprehensive mechanosensory connectome reveals a

943

somatotopically organized neural circuit architecture controlling stimulus-

944

aimed grooming of the

Drosophila

head. Preprint at

945

https://doi.org/10.1101/2025.05.19.654894 (2025).

946

36.

Oleskevich, S., Descarries, L., Watkins, K. C., Se´gue´la, P. & Daszuta, A.

947

Ultrastructural features of the serotonin innervation in adult rat

948

hippocampus: An immunocytochemical description in single and serial thin

949

sections.

Neuroscience

, 777–791 (1991).

950

Journal Pre-proof

37.

Séguéla, P., Watkins, K. C. & Descarries, L. Ultrastructural relationships of

951

serotonin axon terminals in the cerebral cortex of the adult rat.

Journal of

952

Comparative Neurology

289

, 129–142 (1989).

953

38.

Lin, A.

et al.

Network statistics of the whole-brain connectome of

954

Drosophila.

Nature 2024 634:8032

634

, 153–165 (2024).

955

39.

Lu, J.

et al.

Selective inhibitory control of pyramidal neuron ensembles and

956

cortical subnetworks by chandelier cells.

Nat. Neurosci.

, (2017).

957

40.

Woodruff, A. R., Anderson, S. A. & Yuste, R. The Enigmatic Function of

958

Chandelier Cells.

Front. Neurosci.

, (2010).

959

41.

Cuntz, H., Forstner, F., Borst, A. & Häusser, M. One rule to grow them all:

960

A general theory of neuronal branching and its practical application.

PLoS

961

Comput. Biol.

, (2010).

962

42.

Lacin, H. & Truman, J. W. Lineage mapping identifies molecular and

963

architectural similarities between the larval and adult Drosophila central

964

nervous system.

Elife

, (2016).

965

43.

Rajagopalan, S., Nicolas, E., Vivancos, V., Berger, J. & Dickson, B. J.

966

Crossing the midline: Roles and regulation of Robo receptors.

Neuron

967

(2000).

968

44.

Harriger, L., van den Heuvel, M. P. & Sporns, O. Rich Club Organization of

969

Macaque Cerebral Cortex and Its Role in Network Communication.

PLoS

970

One

, e46497 (2012).

971

45.

Liang, X.

et al.

The Rich-Club Organization in Rat Functional Brain

972

Network to Balance Between Communication Cost and Efficiency.

Cerebral

973

Cortex

, 924–935 (2018).

974

46.

Card, G. & Dickinson, M. H. Visually Mediated Motor Planning in the

975

Escape Response of Drosophila.

Current Biology

, (2008).

976

47.

Kramer, P. F.

et al.

Synaptic-like axo-axonal transmission from striatal

977

cholinergic interneurons onto dopaminergic fibers.

Neuron

110

, 2949-

978

2960.e4 (2022).

979

48.

Ammer, G., Vieira, R. M., Fendl, S. & Borst, A. Anatomical distribution and

980

functional roles of electrical synapses in Drosophila.

Current Biology

981

(2022).

982

49.

Lopez, J., Boerner, J., Robbins, K., Pena, R. F. O. & Murphey, R.

983

Frazzled/DCC regulates gap junction formation at a Drosophila giant

984

synapse through transcription. Preprint at

985

https://doi.org/10.1101/2025.04.08.647628 (2025).

986

50.

Holcroft, C. E.

et al.

Innexins Ogre and Inx2 are required in glial cells for

987

normal postembryonic development of the Drosophila central nervous

988

system.

J. Cell Sci.

126

, 3823–3834 (2013).

989

51.

Phelan, P.

et al.

Molecular mechanism of rectification at identified electrical

990

synapses in the Drosophila giant fiber system.

Curr. Biol.

, 1955–1960

991

(2008).

992

52.

Orr, B. O., Borgen, M. A., Caruccio, P. M. & Murphey, R. K. Netrin and

993

frazzled regulate presynaptic gap junctions at a Drosophila giant synapse.

994

Journal of Neuroscience

, (2014).

995

Journal Pre-proof

53.

Winding, M.

et al.

The connectome of an insect brain.

Science (1979).

379

996

(2023).

997

54.

Gu, H. & O’Dowd, D. K. Cholinergic Synaptic Transmission in Adult

998

Drosophila

Kenyon Cells

In Situ

The Journal of Neuroscience

, 265–272

999

(2006).

1000

55.

Clements, J.

et al.

NeuronBridge: an intuitive web application for neuronal

1001

morphology search across large data sets.

BMC Bioinformatics

, (2024).

1002

56.

McCarthy, E. V

et al.

Synchronized bilateral synaptic inputs to Drosophila

1003

melanogaster neuropeptidergic rest/arousal neurons.

Journal of Neuroscience

1004

, (2011).

1005

57.

Shiu, P. K.

et al.

A Drosophila computational brain model reveals

1006

sensorimotor processing.

Nature

634

, (2024).

1007

58.

Schnizler, K.

et al.

A novel chloride channel in Drosophila melanogaster is

1008

inhibited by protons.

Journal of Biological Chemistry

280

, (2005).

1009

59.

Stimberg, M., Brette, R. & Goodman, D. F. M. Brian 2, an intuitive and

1010

efficient neural simulator.

Elife

, (2019).

1011

60.

Moreno-Sanchez, A.

et al.

Morphology and synapse topography optimize

1012

linear encoding of synapse numbers in Drosophila looming responsive

1013

descending neurons. Preprint at https://doi.org/10.7554/eLife.99277.1

1014

(2024).

1015

61.

Lee, D. & O’Dowd, D. K. Fast excitatory synaptic transmission mediated by

1016

nicotinic acetylcholine receptors in Drosophila neurons.

Journal of

1017

Neuroscience

, (1999).

1018

62.

Lee, D., Su, H. & O’Dowd, D. K. GABA receptors containing Rdl subunits

1019

mediate fast inhibitory synaptic transmission in Drosophila neurons.

Journal

1020

of Neuroscience

, (2003).

1021

63.

Saumweber, T.

et al.

Behavioral and synaptic plasticity are impaired upon

1022

lack of the synaptic protein SAP47.

Journal of Neuroscience

, (2011).

1023

64.

Churgin, M. A.

et al.

A neural correlate of individual odor preference in

1024

Drosophila.

Elife

, (2025).

1025

65.

Hagberg, A. A., Schult, D. A. & Swart, P. J. Exploring Network Structure,

1026

Dynamics, and Function using NetworkX. in

Proceedings of the 7th Python

1027

in Science Conference

(2008). doi:10.25080/tcwv9851.

1028

66.

Schlegel, P.

et al.

navis-org/navis: Version 1.10.0. Preprint at (2025).

1029

67.

Hunter, J. D. Matplotlib: A 2D graphics environment.

Comput. Sci. Eng.

1030

(2007).

1031

68.

Virtanen, P.

et al.

SciPy 1.0: fundamental algorithms for scientific

1032

computing in Python.

Nat. Methods

, (2020).

1033

1034

1035

Journal Pre-proof

KEY RESOURCES TABLE

REAGENT or RESOURCE

SOURCE

IDENTIFIER

Antibodies

Mouse Anti-Drosophila choline acetyltransferase

DSHB

RRID:AB_528122

Mouse anti-Bruchpilot (nc82)

DSHB

RRID:AB_2314866

Rabbit anti-GFP

Invitrogen

RRID:AB_221569

Goat anti-rabbit secondary

Jackson
ImmunoResearch

RRID:AB_2338021

Goat anti-mouse Alexa Fluor 647 secondary

Jackson
ImmunoResearch

RRID:AB_2338902

Anti-mVenus antibody (conjugated)

MilliporeSigma

MABE1906

Bacterial and virus strains

N/A

Biological samples

N/A

Chemicals, peptides, and recombinant proteins

All trans retinal 500mg

MilliporeSigma

R2500-500MG

Tetramethylrhodamine-labeled dextran (10,000 MW)

Invitrogen

D1817

O'Dowd's Drosophila saline

N/A

Formaldehyde, Para

Fisher Scientific

T353-500

Triton X-100

MilliporeSigma

T8787-100ML

Critical commercial assays

N/A

Deposited data

N/A

Experimental models: Cell lines

N/A

Experimental models: Organisms/strains

SS25451: w[1118]; VT063643-p65.AD; VT059224-
GAL4.DBD

Bloomington
Drosophila Stock
Center

RRID: BDSC_89495

w[1118]; P{y[+t7.7] w[+mC]=20XUAS-IVS-
CsChrimson.mVenus}attP2

Bloomington
Drosophila Stock
Center

RRID: BDSC_55136

w[*]; P{y[+t7.7] w[+mC]=20XUAS-IVS-mCD8::GFP}attP2 Bloomington

Drosophila Stock
Center

RRID: BDSC_32194

Control-

Bloomington
Drosophila Stock
Center

RRID:BDSC_

Oligonucleotides

N/A

Recombinant DNA

N/A

Software and algorithms

Python

http://www.python.org/

RRID:SCR_008394

Anaconda

https://docs.anaconda.
com/

RRID:SCR_025572

Journal Pre-proof

Jupyter Notebook

https://jupyter.org/

RRID:SCR_018315

neuprint-python API

N/A

NetworkX

https://networkx.org/

RRID:SCR_016864

MATLAB

https://www.mathwork
s.com/products/matlab
.html

RRID:SCR_001622

MANC

N/A

Inkscape

https://inkscape.org/en
/

RRID:SCR_014479

Brian2

http://briansimulator.or
g/

RRID:SCR_002998

NeuronBridge v3.2.6

N/A

Other

Axon Digidata 1440A Data Acquisition System

Molecular Devices

Model 1440A

Clampex software (version 10 and 11)

http://www.moleculard
evices.com/products/s
oftware/pclamp.html

RRID:SCR_011323

Nikon AXR confocal microscope with NSPARC detector

Nikon

N/A

Nikon Eclipse E600FN fluorescence microscope

Nikon

N/A

590 nm optogenetic LED setup

Thorlabs

M590F3; M118L02;
LEDD1B

Journal Pre-proof