Activity manipulation of an excitatory interneuron, during an embryonic critical period, alters network tuning of the Drosophila larval locomotor circuit

As nervous systems develop, activity perturbations during critical periods can lead to permanently altered network function. However, how activity perturbation influences individual synapses, the network response and the underlying signalling mechanisms are not well understood. Here, we exploit a recently identified critical period in the development of the Drosophila larval locomotor circuit to show that activity perturbation differentially affects individual and identified synaptic pairings. Remarkably, we further show that activity-manipulation of a selective excitatory interneuron is sufficient to fully recapitulate the effects induced by network-wide activity disturbance; indicative that some neurons make a greater contribution to network tuning. We identify nitric oxide (NO)-signalling as a potential mediator of activity-dependent network tuning during the critical period. Significantly, the effect of NO-signalling to network tuning is dictated by the prior activity state of the network. Thus, this study provides mechanistic insight that is currently lacking into how activity during a critical period tunes a developing network.


Introduction 38
Neural circuit development requires activity to refine connectivity, modify signalling properties and, 39 ultimately, to ensure emergence of appropriate network output. As networks develop, sets of neurons 40 become spontaneously active. Initial activity shows little coordination across the network. However, 41 as circuits mature and, particularly, as sensory afferents become functional there is a shift towards 42 increasingly coordinated patterned activity (1, 2). This transition is often marked by a critical period: a 43 defined window of heightened plasticity during which activity exerts maximal influence to network 44 properties (3). It is notable that errors made during a critical period can become 'locked-in' such that 45 subsequently the network is unable to correct mistakes made during this period. A well described 46 example is activity manipulation of visual inputs in developing mammals, which studies have shown 47 have permanent and maladaptive effects to the formation of ocular dominance and can lead to clinical 48 conditions such as amblyopia (4). While the existence of critical periods in developing circuits is well 49 described and has revealed a number of important mechanisms, notably a role for a subset of 50 GABAergic inhibitory neurons and the formation of perineuronal nets (3), a reliance on complex 51 mammalian sensory circuits has precluded a cell-specific analysis of how activity, during a critical 52 period, affects cell type-specific synaptic pairings and how such changes combine to impact network 53

tuning. 54
We recently reported a critical period in the development of the Drosophila larval locomotor circuit, 55 indicative that critical periods are a universal phenomenon. Manipulating activity late in 56 embryogenesis (85-95% / 17 to 19h after egg laying) is sufficient to permanently alter the 57 developmental trajectory of the locomotor network, leading to sub-optimal adjustment and instability. 58 This manifests subsequently as a susceptibility to electroshock-induced seizures in larval stages, 59 measured 5 days later at the end of larval life (5). Interestingly, the critical period coincides precisely 60 with the emergence of patterned peristaltic contractions of body wall muscles in the developing 61 embryo (6, 7). Appropriately patterned network activity is necessary for coordinated locomotor 62 movements to emerge, indicative of this phase being essential for network tuning (8). Our previous 63 work demonstrated that the overall level of network activity, during the critical period, is instructive 64 for appropriate adjustment. For example, a gain-of function mutation in a voltage-gated Na + channel 65 (para bss ) leads to increased activity levels and makes animals seizure prone. However, transient 66 optogenetic reduction of activity during the critical period, in these mutants, is sufficient to allow 67 appropriate tuning and entirely suppresses seizure susceptibility (5). At the single cell level, analysis 68 of excitatory synaptic drive to motoneurons, both in para bss or control animals made seizure prone 69 following activity manipulation during the critical period, reveals that excitatory currents have gained 70 a significantly extended duration, suggesting a change to the excitation:inhibition balance (5). 71 The identification of a critical period in the Drosophila larval locomotor circuit presents a new 72 opportunity to determine how synaptic transmission changes between specific, identified components 73 of a network during normal tuning; and how these processes are impacted by transient activity 74 manipulation during the critical period. One of the unique strengths of this locomotor network is that 75 the upstream circuitry of segmentally re-iterated motoneuron-muscle pairs (~30 per half-segment (9)) 76 has recently been characterised by tracing connectivity at the level of single synapses and functionally 77 testing predicted connections (10-13). Thus, we are now able to identify and manipulate defined 78 synaptic pairs and can ask, for the first time, how activity perturbation during a critical period alters 79 signalling at identified synapses. 80

Results 110
Pan-neuronal activity perturbation during the critical period alters network excitation 111 We previously showed that transient perturbation of network activity during late embryonic 112 development (17-19h after egg laying) leads to enduring network instability, measurable some five 113 days later in third instar larvae (L3), as an increased seizure response following electroshock ( Fig 1A). 114 At the level of identified neurons, such transient activity manipulations during late embryogenesis, 115 produces a characteristic broadening of excitatory cholinergic synaptic inputs (termed spontaneous 116 rhythmic currents, SRCs) to identified motoneurons; in this instance the aCC motoneuron ( Fig 1B)  117 (5). SRCs are the product of the simultaneous activity of multiple premotor cholinergic interneurons 118 (17). The same outcomes, to both increased seizure recovery time and SRC duration, are also caused 119 by promoting hyperactivity in developing embryos by other means, e.g. exposure to the proconvulsant 120 picrotoxin (PTX, a chloride channel blocker), or by introduction of the para bss mutation (a Na v 121 hypermorph) (5). Equally, reducing synaptic excitation by inhibition during the critical period, e.g., by 122 pan-neuronal activation of eNpHR, also leads to this effect. Therefore, changing activity levels in the 123 developing network, to deviate from normal, irrespective of the nature or polarity of change, seems 124 deterministic for, and detrimental to, post-embryonic network stability (5). Under conditions where 125 the SRC inputs are broadened, we find the number of action potentials fired per SRC significantly 126 increased (23.04 ± 3.32 vs. 12.08 ± 1.75 APs per bout, para bss vs. wild-type, p = 0.004, unpaired t-test, 127

Neuron-specific activity perturbation during the critical period alters network excitation 131
The cellular mechanisms that ensure stable networks form and that might be affected by transient 132 activity perturbation during a critical period, remain poorly understood. To study these, we took 133 advantage of the fact that the larval Drosophila CNS is composed of identified neurons whose 134 connectivity has been largely characterised (10-13). Specifically, we asked how activity 135 manipulations of identified premotor interneurons, during the critical period, influence network 136 function in later larvae. To test this idea, we focused on A27h: a cholinergic premotor interneuron that 137 provides direct synaptic excitatory input to motoneurons, including the aCC motoneuron (Fig 2A)  138 (12). We selectively manipulated activity of A27h, during the critical period, using A27h-139 Gal4>Chronos (100 ms/1Hz): a variant of ChR which exhibits a green-shifted excitation (λ565 nm, 140 (19), see Methods for full rationale). Five days later, at the L3 stage, spontaneous SRC inputs to aCC 141 (i.e. derived from multiple cholinergic pre-motor interneurons, not just A27h), showed an increased 142 duration (0.98 ± 0.07 s vs. 0.46 ± 0.03 s, -LED vs. +LED, p < 0.0001, unpaired t-test, Fig 2B-C). In 143 addition, electroshock of these L3 larvae showed a statistically significant increase in recovery time to 144 277 ± 16 s (+LED) from 83 ± 4 s (-LED, p < 0.0001, unpaired t-test, Fig 2D) 2E) or activation (233 ± 12 vs. 144 ± 7 s, +LED 100ms vs. -LED, respectively, p < 0.0001). We also 154 observed that transient optogenetic activation of A27h (100 ms) potentiates the seizure severity 155 caused by the para bss mutation (303 ± 12 vs. 258 ± 13 s, +LED 100ms vs. -LED, respectively, p = 0.035, 156 Fig 2F). We previously showed that pan-neuronal inhibition during the critical period is sufficient to 157 rescue the para bss seizure phenotype (5). However, NpHR-mediated inhibition of A27h (600 ms light 158 pulses) was not sufficient to compensate for the para bss -induced network instability (233 ± 11 vs. 258 159 ± 13 s, +LED 600ms vs. -LED, respectively, p = 0.44). Thus, activity manipulation of A27h interneurons 160 is sufficient to destabilise the locomotor network, similar in phenotype to global network activity 161 manipulations; though in the context of an over-excitable network (e.g. in para bss mutants) selective 162 inhibition of A27h activity is insufficient to prevent the network from becoming unstable. 163 Next, we asked whether manipulating the activity of any neuron during the critical period would be 164 sufficient to cause a subsequent seizure recovery phenotype. We tested this idea by identical activity 165 manipulations of the GABAergic interneuron A31k (20A03-AD;87H09-DBD split Gal4 line), which 166 delivers proprioceptive feedback to motoneurons (11). We found that these did not produce any 167 change in recovery time to electroshock treatment at L3 (Fig 2G). We confirmed that all lines express 168

Activity-manipulation differentially influences A27h synaptic drive 176
A key question is how critical period activity manipulations impact on synaptic transmission between 177 partner neurons. We took advantage of having access to known partner neurons in this network and 178 measured the output of A27h→aCC pairing, using optogenetics to selectively activate the A27h 179 interneuron when recording its synaptic drive to the aCC motoneuron. 180 In controls, not exposed to embryonic manipulation, optogenetic stimulation of A27h (Chronos, λ565 181 nm, 1 s) produced an inward current in aCC ( Fig 3A). Current density reached a maximum of 5.07 ± 182 0.66 pA/pF (n = 10, Fig 3B). Selective activity perturbation of A27h (A27h>Chronos, 100 ms/1Hz), 183 during the critical period, resulted in a markedly increased A27h→aCC synaptic current amplitude 184 recorded at L3 (11.27 ± 0.92 pA/pF, n = 10, p < 0.0001, Fig 3A-B). We then compared this with 185 transient pan-neuronal activity manipulations during the critical period. For this we exposed 186 embryonic networks to the proconvulsant picrotoxin (PTX) by feeding it to gravid females (using the 187 GAL4 expression system to selectively stimulate A27h for recording at L3). We previously 188 demonstrated that feeding drugs to gravid females leads to significant transfer of drug to the embryo. 189 This is transient because by late larval stages, when our measurements are made, radio-tracing 190 showed the drug is eliminated (18). Surprisingly, although embryonic exposure to PTX leads to an 191 increase in overall synaptic drive to the aCC motoneuron (cf. Fig 1B), the specific A27h→aCC 192 pairing synaptic transmission is significantly reduced (2.05 ± 0.22 pA/pF, n = 10, p = 0.0048, Fig 3A- 193 B). We find the same reduction in A27h→aCC synaptic drive following a different global activity 194 manipulation that similarly causes network instability, namely the para bss mutation (1.61 ± 0.19 195 pA/pF, n = 10, p = 0.001, Fig 3A- Fig 3C-D). This suggests that both 200 pan-neuronal and A27h-specific activity perturbations lead to excessive excitation of motoneurons, by 201 increasing spontaneous SRC duration (see above) which, in turn, compromises network stability. In 202 order to evaluate the specific contribution of A27h activation to aCC firing, we recorded aCC activity 203 in loose patch mode while simultaneously optogenetically stimulating A27h (A27h>Chronos, λ565 204 nm 1 s pulses, Fig 3E). We observed that the increased synaptic drive resulting from the cell-selective  Thus, we show that both global and cell-specific activity manipulations have comparable effects on 211 network stability, evidenced by increased recovery times following electroshock-induced seizures, as 212 well as endogenous aCC spiking activity. Using a cell-selective A27h manipulation, we show that at 213 least for this excitatory interneuron, increased activation during the critical period leads to 214 strengthened synaptic transmission for both this and other premotor cholinergic interneurons. In 215 contrast, when network activity is globally manipulated during the critical period, transmission at the 216 A27h→aCC pairing is reduced, but the excitatory drive to aCC from other premotor interneurons is 217 increased which results is a similar outcome: network instability to electroshock. Thus, any deviation 218 away from the normal strength of the A27h→aCC coupling, during the critical period, is sufficient to 219 alter network tuning. 220 221

Nitric oxide mediates activity perturbation during the critical period 222
The mechanisms underlying network tuning during a critical period are not well understood. Taking a 223 best-candidate approach, we focussed on nitric oxide (NO)-signalling, since this messenger has been 224 shown to be regulated by activity and able to alter synaptic drive (20). In a first set of experiments, we 225 pharmacologically manipulated NO synthase (NOS) activity in embryos by feeding gravid females 226 either the NOS inhibitor, N(G)-nitro-L-arginine methyl ester (L-NAME, 0.1 M), or the NO donor, 227 sodium nitroprusside (SNP, 1.5 mM). On their own, these manipulations did not lead to changes in 228 SRCs recorded from L3 aCC motoneurons or recovery times following electroshock-induced seizures 229 ( Fig. 4). We then asked whether SRC and seizure recovery phenotypes caused by activity 230 manipulations during the critical period were mediated by NO-signalling. To test this hypothesis, we 231 conducted the same pharmacological NO synthase manipulations in embryos, but this time 232 additionally carried out optogenetic pan-neuronal activity manipulation (elav C155 >ChR) during the 233 critical period. Recordings from aCC in control L3 (absence of NO manipulation) showed an expected 234 increase in SRC duration following optogenetic manipulation (0.52 ± 0.04 vs. 1.25 ± 0.13 s, -LED vs. 235 +LED, respectively, p = 0.003, Fig 4A-B). The presence of the NOS inhibitor L-NAME (0.1 M) 236 abolished SRC broadening (0.57 ± 0.03 vs. 0.61 ± 0.05 s, -LED vs. +LED, respectively, p > 0.9). 237 Conversely, embryos from females fed the NO donor, SNP (1.5 mM), exhibited a significantly larger 238 increase in SRC duration (0.49 ± 0.05 to 2.01 ± 0.27 s, -LED vs. +LED, respectively, p < 0.001). A 239 two-way ANOVA showed both drug treatments produced significant changes in the optogenetically 240 manipulated groups (p = 0.0148 and p = 0.0016, L-NAME and SNP vs. CTRL, respectively), whilst 241 drug treatment per se had no impact on SRCs in non-stimulated (-LED) animals. Throughout, SRC 242 frequency and amplitude were unaffected. 243 Embryonic exposure to the NOS-inhibitor, L-NAME (0.1 M), also counteracted electroshock-induced 244 network instability by preventing the increased recovery time normally associated with pan-neuronal 245 optogenetic manipulation (96 ± 6 vs. 115 ± 6 s, -LED vs. +LED, respectively, p > 0.9, Fig 4C). 246 Conversely, embryonic exposure to the NO donor, SNP (1.5 mM), significantly potentiated the effects 247 of optogenetic activity-perturbation, leading to a further increase in recovery time (from 107 ± 6 s to 248 230 ± 11 s, p < 0.001). A two-way ANOVA showed embryonic manipulation of NO levels 249 significantly affected only the activity manipulated (+LED) groups (p < 0.0001 and p = 0.0156, L-250 NAME and SNP vs. CTRL, respectively), indicating that the concentrations of NOS modifiers used 251 did not by themselves alter motor circuitry development. To further confirm the role of NO-signalling 252 downstream of neuronal activity, we also tested the ability of additional drugs, known to target NOS 253 and other members of the canonical NO-signalling pathway, including the soluble guanylyl cyclase 254 receptor (sGC) and PKG kinase. Consistently, inhibitors of NO-signalling prevented increases in 255 recovery time, while activators caused further potentiation ( Fig S1). 256 We also targeted genetic manipulations of NO-signalling to developing neurons to validate 257 pharmacology. Mirroring the effect of the NOS-inhibitor, L-NAME, pan-neuronal expression of 258 NOS RNAi was sufficient to block an increase of recovery time normally caused by optogenetic network 259 activation during the critical period (elav C155 >ChR;NOS RNAi : 145 ± 9 s, p = 0.0312, Fig 4D). showed no change compared to CTRL (194 ± 12 vs. 189 ± 11 s, GFP RNAi vs. CTRL, respectively, p > 267 0.9). 268 Whilst our previous work has shown that drugs fed to gravid females enter the embryo, perdurance 269 through to final third larval instar does not occur (Marley and Baines, 2011). However, a caveat to 270 results obtained using a Gal4 driver is its expression throughout both embryonic and larval periods. 271 To show that embryonic, but not larval, manipulation of NO-signalling is sufficient to alter synaptic 272 drive we manipulated Gal4 activity with a temperature sensitive Gal4 inhibitor, Gal80 ts (21). 273 Restricting expression of macNOS to motoneurons only during embryogenesis 274 (OK6;Gal80 ts >macNOS) resulted in a significant increase in SRC duration in aCC recorded later in 275 L3 (from 0.50 ± 0.01 s to 0.74 ± 0.03 s, n = 12, p < 0.001, Fig 4E). By contrast, no change in SRC 276 duration was observed when expression of macNOS was restricted to postembryonic larval stages 277 (0.57 ± 0.04 s, n = 10, p = 0.31). To verify that the temperature changes required for GAL80 ts did not 278 affect SRC kinetics, the same temperature shifts were repeated using a control genotype 279 (OK6;Gal80 ts >GFP RNAi ), which showed no significant effects to SRC kinetics ( Fig 4F). 280

Altered NO-signalling in A27h recapitulates changes caused by activity manipulations 281
Our results support the hypothesis that activity regulated changes, during network tuning, are 282 mediated, at least in part, through NO-signalling. If this is indeed the case, and if NO-signalling plays 283 a major role in network adjustment during the critical period, then one might expect manipulations of 284 NO-signalling restricted to neurons such as A27h to mimic effects of cell-specific activity 285 manipulations. To test this, we knocked down NOS transcript (A27h>NOS RNAi ) or increased NO-286 signalling (A27h> UAS-macNOS) in A27h. As predicted, this led to a significantly increased recovery 287 time to electroshock in both conditions (A27h>NOS RNAi : 172 ± 10 s, p < 0.0001; A27h>macNOS: 188 288 ± 10 s, p < 0.0001, vs. A27h>GFP RNAi control: 103 ± 6 s, One-way ANOVA, Fig 5A). As a further 289 control, we expressed the same constructs in the GABAergic interneuron A31k, where activity 290 manipulation was ineffective (cf. Fig 2G). No significant change was detectable following NOS 291 manipulation (A31k>NOS RNAi : 122 ± 5 s, p = 0.1744; A31k>macNOS: 94 ± 5 s, p = 0.1007, vs. 292 control: 109 ± 5 s, One-way ANOVA, n = 30 in each group), suggesting a crucial role for A27h in 293 network stability. 294 Further supporting the involvement of NO in network tuning, embryonic exposure to the NOS-295 inhibitor, L-NAME (0.1 M) increased the recovery time to electroshock caused by selective inhibition 296 (eNpHR +LED 600ms ) of A27h during the critical period (137 ± 6 vs. 331 ± 19 s, -LED vs. +LED, 297 respectively, p < 0.0001, Fig 5B). Conversely, the presence of the NO-donor, SNP (1.5 mM), blocked 298 the effect of this activity manipulation (147 ± 6 to 154 ± 7 s, -LED vs. +LED, respectively, p > 0.9). A 299 two-way ANOVA showed that the embryonic manipulation of NO levels significantly affected the 300 optogenetically stimulated (+LED) groups (p = 0.0007 and p < 0.0001, L-NAME and SNP vs. CTRL, 301 respectively), while leaving unstimulated (-LED) animals unchanged, as also shown above. In a 302 complementary set of experiments, inducing hyperactivity in A27h (eNpHR +LED 100ms ) we observed 303 that the same NO-signalling manipulations caused the opposite changes ( Fig 5C). Specifically, 304 inhibition of NOS prevented A27h hyperactivity-induced increases in recovery time (0.1 M L-NAME, 305 137 ± 6 vs. 148 ± 7 s, -LED vs. +LED, respectively, p > 0.9), while exposure to SNP potentiated the 306 increase in recovery time (1.5 mM, 147 ± 6 to 317 ± 15 s, p < 0.0001). A two-way ANOVA showed 307 that manipulating NO levels significantly affected the +LED groups only (p < 0.0001, both L-NAME 308 and SNP vs. CTRL, respectively). 309

NO-signalling is sufficient to alter network tuning 310
Our observation that NO-signalling is necessary for the effect of activity perturbation during the 311 critical period on SRC duration and recovery time in response to electroshock, raises the important 312 question of whether altering NO alone, during the same embryonic period, is sufficient to induce 313 network instability. To this end, we manipulated NO-signalling in all neurons (elav C155 -Gal4) in the 314 absence of other activity manipulations; expressing UAS-NOS RNAi to reduce, or the constitutively 315 active NOS transgene UAS-macNOS (22) to increase NO-signalling, respectively. Expression of an 316 RNAi transgene against GFP (elav C155 >GFP RNAi ) was used as control. Both manipulations of NOS 317 were sufficient to significantly increase SRC duration in L3 aCC (NOS RNAi : 1.05 ± 0.13 s, n = 10, p < 318 0.0001; macNOS: 1.09 ± 0.06 s, n = 10, p < 0.0001 vs. GFP RNAi : 0.45 ± 0.02 s, n = 13, Fig 6A-B). 319 This mirrors the outcomes achieved by optogenetically increasing or decreasing activity during 320 embryogenesis in otherwise wildtype embryos (5). Genetic manipulation of NO-signalling alone was 321 also sufficient to increase recovery time to electroshock; e.g., pan-neuronal RNAi-mediated knock-322 down of NOS (elav C155 >NOS RNAi : 154 ± 8 s, p < 0.0001, Fig 6C) or expression of macNOS 323 (elav C155 >macNOS: 166 ± 8 s vs. CTRL: 101 ± 8 s, p < 0.0001). 324 Finally, we observed that exposure of developing embryos to higher concentrations of SNP (5 mM 325 rather than 1.5 mM used previously) is sufficient to induce network instability in L3. Thus, we tested 326 whether activity-manipulation of A27h was sufficient to prevent this. Exposure of gravid females 327 SNP (5 mM) produced L3 with a significantly increased recovery time to electroshock (A27h>eNpHR 328 + 5 mM SNP -LED: 182 ± 7 s). However, simultaneous inhibition of A27h during the critical period 329 (A27h>eNpHR + 5 mM SNP +LED 600ms , 17-19hrs AEL) blocked the effect of SNP (124 ± 5 s, p < 330 0.0001, unpaired t-test). This result is consistent with A27h having a significant contribution to 331 network tuning during the critical period. 332

Discussion 333
It is established that activity perturbation during a critical period can induce permanent changes to 334 neural circuit function (3, 4, 23). However, how activity perturbation changes individual synaptic 335 connections across a network and how affected networks respond to such changes is not well 336 understood. In part, this is because our knowledge of critical periods has been largely derived from 337 complex mammalian neural circuits (e.g. vision) that deter single cell resolution. In this respect, it is 338 significant that we have previously identified a critical period during the development of the 339 Drosophila larval locomotor circuitry, during which activity perturbation is sufficient to permanently 340 alter network stability (5). Moreover, we show here that regardless of the nature of the activity 341 manipulation; being inhibitory, excitatory, pharmacological or genetic, the outcome is identical. This 342 predicts that the wiring of this network canalises the impact of activity perturbations, such that they 343 lead to reproducible rather than variable or chaotic outcomes. Moreover, these same outcomes arise 344 regardless of the manipulations having been network-wide or targeted to specific cells. The most 345 parsimonious interpretation for which is that specific components of the motor network (e.g. the pre-346 motor cholinergic interneuron A27h) have a greater influence on network tuning than others (e.g. the 347 GABAergic A31k interneuron), both when selectively manipulated and also during more global 348 manipulations. Selective A27h activity manipulation is also sufficient to counteract global 349 pharmacological manipulations of NO-signalling which, again, is consistent with a stronger 350 contribution for this neuron to network tuning. The perturbations that we observe are evidently 351 destabilising to mature network function increasing both duration of synaptic currents and severity of 352 induced-seizure activity. Activity disturbance during mammalian critical periods has similarly been 353 linked to detrimental effects: e.g. shift in ocular dominance (e.g. amblyopia), language deficits and 354 autism (4, 24). It has, moreover, been shown that therapeutic intervention to normalise activity during 355 a critical period, in an otherwise activity-disturbed network, can have significant beneficial effects for 356 not only ocular dominance but also epilepsy (4, 5, 25). 357 Network tuning during a critical period has been studied most extensively in the vertebrate visual Nevertheless, the critical period in the Drosophila locomotor circuit coincides with the onset of 362 GABA synthesis (34) and the ingrowth of astrocyte-like glia (35, 36). Whether or not, in Drosophila, 363 these events act as instructive cues for the opening and subsequent closure of the critical period 364 remains to be determined. From a circuit perspective in particular, the establishment of an 365 experimental model system such as this, which has an explicit mammalian-like critical period, will 366 greatly facilitate understanding of these enigmatic periods in neural circuit development. 367 Because neural networks are most likely organised with a hierarchical structure, we reasoned that 368 some constituent neurons may have a greater propensity than others for orchestrating activity-369 dependent tuning. To validate this hypothesis, we focussed attention on a defined synaptic pairing 370 formed by the excitatory cholinergic premotor interneuron A27h (12) and its postsynaptic motoneuron 371 aCC. Electrophysiology showed that, while pan-neuronal activity manipulations during the critical 372 period lead to overall increased SRCs in aCC motoneurons, a reduction in synaptic strength 373 unexpectedly occurs at the level of this specific pairing. This outcome clearly demonstrates that 374 defined synaptic pairings respond differently to activity perturbation during a critical period. It further 375 demonstrates that neurons that receive multi-neuron synaptic inputs (e.g. aCC) can specifically, and 376 differentially, modify subsets of these inputs. In this example, the A27h premotor IN is differentially 377 affected compared to other excitatory premotor INs to aCC following pan-neuronal activity 378 manipulations during the critical period. Nevertheless, both network-wide (which include A27h) and 379 A27h-specific activity manipulations during the critical period evoke CNS instability. It is significant 380 that the strength of A27h→aCC synaptic transmission is differentially affected by pan-neuronal vs. 381 A27h-specific activity manipulation (see Fig 7 for a summary of the effects observed in this study). 382 We cannot, at present, differentiate between whether the changes to network tuning we observe are 383 due to alteration of developmental rules governing the formation of the locomotor circuit or are the 384 consequence of homeostatic changes attempting to maintain stability (37). This discrimination will 385 require more detailed recordings from other neurons of the locomotor circuit, including those that 386 synaptically drive A27h. 387 We utilise response to electroshock as a proxy for network stability. A large number of seizure 388 mutations (so-called bang-sensitive mutations) have been identified in Drosophila which show 389 seizure-like activity following exposure to a strong stimulus (electroshock or violent shaking) (38). 390 Similar epilepsy syndromes exist in mammals where exposure to either a loud sound or flashing lights 391 is sufficient to induce a seizure state (39-41). That seizures occur is indicative of underlying networks 392 lacking robustness and which cannot compensate for extremes of activity. At least part of the lack of 393 robustness in Drosophila seizure mutants may be due to the increased synchrony seen in activity of 394 motoneurons between adjacent segments (42). Indeed, increased synchrony within neuron 395 subpopulations is a hallmark of mammalian epilepsy (43-45). Thus, whilst we are unable at present to 396 describe the precise effects to global network tuning which results in increased recovery time to 397 electroshock, this measure is an adequate proxy for network robustness which emerges through 398 appropriate network tuning during development. 399 That single or small subsets of neurons can disproportionately impact the functionality of neural 400 networks, as described in more complex systems, has been described in the mammalian hippocampus 401 and visual cortex (46,47). In this study, we demonstrate that embryonic manipulation of A27h is 402 sufficient to trigger network instability. Our results also suggest that different interneurons (e.g. A31k 403 vs. A27h) may contribute unequally to network tuning, perhaps indicating that their local synaptic 404 connectivity determines how discharges spread from the activated cell to other constituent neurons. 405 Equally, contribution may also be dictated by ability to influence the excitation:inhibition balance of a 406 developing network: in this instance manipulation of excitatory cholinergic neurons (A27h) being 407 more effective than inhibitory GABAergic cells (A31k). Such changes might be expected to 408 differentially perturb the excitation:inhibition balance across a network which may, in turn, contribute 409 to observed postembryonic network stability. 410 Network tuning requires constituent neurons in any given network to be able to "sense" both their 411 own activity and, importantly, the activity of other neurons within the circuit, regardless of whether 412 they are directly connected. NO, acting as a diffusible signal, is an obvious candidate to fulfil this 413 important signalling role (48-51). Our results support this hypothesis. It is notable that, in this context, 414 we find that either increasing or decreasing NO-signalling generates the same outcome, indicating that 415 change in NO levels, rather than the direction of change, is the key determinant that is crucial for 416 network tuning. The Janus nature of NO has been highlighted in different systems, from cell survival 417 (52), to nociceptive transmission (53) and, particularly, epileptogenesis where its role is highly 418 contradictory with significant evidence supporting both proconvulsive, as well as anticonvulsive 419 activity (54). Thus far, the diversity of the downstream target molecules of NO-signalling, coupled to 420 the lack of homogeneity among these different studies (differing drugs, dose and route of 421 administration applied to diverse seizure models), make it difficult to identify a mechanism. Indeed, it 422 has been reported that NO can differentially modulate excitatory (glutamatergic) and inhibitory 423

(glycinergic) synapses in dorsal horn neurons (53). Our results showing that NO manipulation can 424
produce disparate effects dependent on the endogenous activity level present in the developing circuit 425 may also go some way to explaining these contradictory studies. For example, we show that inhibiting 426 NO-signalling is beneficial to a hyperactive network, but increases network instability (i.e. a 427 proconvulsant action) in a physiologically normal network. 428 In summary, although a universal phenomenon, how activity tunes a developing network remains 429 poorly understood. We show here that activity perturbation during a defined critical period in the 430 development of the Drosophila larval locomotor circuit is sufficient to induce significant and 431 permanent change to network stability. We further show that the effect of activity perturbation is 432 mediated by change to NO-signalling sufficient to induce change to specific synaptic pairings 433 between constituent neurons. The degree of understanding, coupled to an ability to identify and 434 selectively manipulate, the individual neurons that form this circuit offer the prospect of exploiting 435 Drosophila to advance understanding of the role of activity in tuning a developing network and, 436 specifically, the role of a critical period in this process.

Optogenetic manipulation of neuronal activity 460
Mated adult females were allowed to lay eggs on grape agar (Dutscher, Essex, UK) plates at 25°C 461 supplemented with a small amount of live yeast paste. To ensure that embryos received enough 462 retinal, adults were fed with 4 mM all-trans-retinal (Sigma-Aldrich, Poole, UK) dissolved in yeast 463 paste twice a day for three days prior to collection. Embryos were collected within a 4 h time range 464 (time 0±2 h after egg laying) and then transferred to a fresh grape agar plate. Plates were placed in a 465 humidified atmosphere inside a 25°C incubator and exposed to collimated light from an overhead 466 LED, positioned 17 cm from the embryos. LEDs had peak emission at λ470 nm (bandwidth 25 nm, 467 irradiance 466 ± 14 nW·cm -2 ; OptoLED, Cairn Instruments, Kent, UK) or λ565 nm (bandwidth 80 468 nm, 250 ± 10 μW·cm -2 ; M565L2, Thorlabs, Newton, NJ). Embryonic exposure to λ470 nm, but not 469 λ565 nm is sufficient to induce a seizure-phenotype at L3 (5) through the activation of cryptochrome-470 expressing neurons (58-60). Therefore, experiments involving blue light required a cry-null (cry 03 ) 471 background, to avoid unspecific effects. Alternatively, we used either eNphR or Chronos, a green-472 shifted variant of ChR, which allowed us to activate specific subpopulations of neurons during 473 embryogenesis without simultaneous activation of the endogenous blue-light sensitive cry neurons. 474 Light was pulsed at 1 Hz using a Grass S48 stimulator (Grass instruments, Quincy, MA, USA). ChR 475 and Chronos were activated with short duration pulses (100 ms). In neurons overexpressing eNpHR, 476 the duration of light pulses varied from 100 ms to 600 ms in order to induce rebound firing activity 477 (hyperactivity) or preventing spike firing (inhibition), as previously described (5). 478 Embryos were optically treated for a pre-determined time period during embryogenesis (between 17-479 19 th ± 2 h AEL), which corresponds to the critical period (5). After manipulation, embryos were 480 transferred into food bottles and maintained at 25°C in complete darkness until ~4 days later when 481 wall-climbing L3 were collected and then electrophysiologically or behaviourally tested. 482

Drug Treatments 483
Embryonic exposure to Picrotoxin (PTX, Sigma-Aldrich) was achieved by feeding gravid females to 484 live yeast paste supplemented with 0.25 mg/ml PTX for three days, prior to embryo collection. 485 Embryos were collected as previously described and transferred to nondrug-containing vials (5). 486 Chemical manipulation of the NO pathway was performed using the same procedure. Feeding gravid 487 females low doses of L-NAME (0.1 M, Sigma-Aldrich), or SNP (1.5 mM, Sigma-Aldrich), was 488 sufficient to affect the outcome of the optogenetic stimulation (increased recovery time and prolonged 489 SRC duration) whilst leaving the control group (-LED) unaltered. Conversely, higher doses (0.5 M L-490 NAME or 5 mM SNP), although more effective, were sufficient to alone mimic the effect of optical 491 stimulation (ChR -LED or in a WT strain, Canton S) thus affecting control levels. We obtained 492 identical results by testing other inhibitors/activators targeting the NO pathway. Hence, for all the 493 experiments shown in this paper, we first pre-determined an optimal concentration for each compound 494 by testing a wide range of doses (data not shown). 495

Electrophysiology 522
Electrophysiological recordings were performed in L3 as previously described (6, 18 an Olympus BX51WI microscope. During recording, light was pulsed onto the sample for 1 s 537 triggered by TTL signals from pClamp (Molecular Devices) to the LED controller. The same 538 stimulation protocol was applied five times to each neuron and the recordings averaged. Current 539 amplitudes were measured as the maximal current evoked and normalised for cell capacitance. 540 Spontaneous rhythmic currents (SRCs) were recorded from L3 aCC motoneurons for 3 minutes. 541 Traces were sampled at 20 kHz and filtered at 0.2 kHz low pass. Cells with input resistance <0.5 GΩ 542 were not considered for analysis. Synaptic current parameters were examined for each recorded cell 543 using Clampfit (version 10.4). To measure the amplitude of SRCs, the change from baseline to peak 544 current amplitude was determined (63). Currents shown were normalized for cell capacitance 545 (determined by integrating the area under the capacity transient resulting from a step protocol from -546 60 to -90 mV). The duration of each synaptic event was defined as the time from current initiation 547 until the return to baseline, as depicted in Fig 1B. 548 Loose patch recordings were performed on L3 aCC motoneurons. Thin-wall borosilicate glass 549 capillaries (GC100TF-10, Harvard Apparatus) were used to pull recording electrodes (unpolished) 550 with resistances between 1.5-2.5 MΩ. Data were acquired with a sampling rate of 20 kHz, filtered 551 with a low-pass filter of 0.2 kHz and analysed in Clampfit 10.4 (Molecular Devices). Excitation of 552 A27h expressing Chronos was achieved using an OptoLED system (Cairn Research) with a 565 nm 553 LED. For each cell, the contribution of A27h to aCC spiking activity was quantified by delivering 554 light pulses of 1 second at 0.2 Hz and averaging the number of action potentials fired in the first 10 555 pulses. Bouts overlapping with endogenous activity were rejected. 556

Electroshock assay 557
Electroshock assay was performed as previously described (18). Briefly, wall-climbing L3 were 558 transferred to a plastic dish after washing to remove food residue and gently dried using paper tissue. 559 Once normal crawling behaviour resumed, a conductive probe, composed of two tungsten wires (0.1 560 mm diameter, ~1-2 mm apart) was positioned over the approximate position of the CNS, on the 561 anterior-dorsal cuticle of the animal. A 2.3 V DC pulse for 2 s, created by a constant voltage generator 562 (DS2A-mkII, Digitimer Ltd., Welwyn Garden City, Hertfordshire, UK), was applied. In response to 563 the stimulus, we observed a transitory paralysis in which larvae were tonically contracted and, 564 occasionally, exhibited spasms. The time to resumption of normal crawling behaviour was measured 565 as recovery time. Normal crawling was defined as a whole body peristaltic wave resulting in forward 566 movement. 567