Neuromedin U signaling regulates memory retrieval of learned salt avoidance in a C. elegans gustatory circuit

Learning and memory are regulated by neuromodulatory pathways, but the contribution and temporal requirement of most neuromodulators in a learning circuit are unknown. Here we identify the evolutionarily conserved neuromedin U (NMU) neuropeptide family as a regulator of memory retrieval in C. elegans gustatory aversive learning. The NMU homolog CAPA-1 and its receptor NMUR-1 are required for the expression of learned salt avoidance. Aversive learning depends on the release of CAPA-1 neuropeptides from sensory ASG neurons that respond to salt stimuli in an experience-dependent manner. Optogenetic silencing of CAPA-1 neurons blocks the immediate retrieval, but not the acquisition, of learned salt avoidance. CAPA-1 subsequently signals through NMUR-1 in AFD sensory neurons to modulate two navigational strategies for salt chemotaxis. Aversive conditioning thus recruits NMU signaling to eventually modulate locomotor programs for expressing learned avoidance behavior. Because NMU signaling is conserved across bilaterian animals, our findings incite further research into its function in other memory and decision-making circuits. Graphical Abstract


26
Associative learning and memories are crucial for animals to anticipate events and adapt behavioral 27 choices, increasing their chance of survival in a dynamic environment. The experience-dependent 28 association of two stimuli, known as Pavlovian conditioning, creates memory traces that can 29 profoundly influence future behavioral responses 1 . Early studies in invertebrate model organisms, 30 such as Aplysia, have been key to provide insights in neural mechanisms underlying learning and 31 memory 2,3 . Since then it has been suggested in diverse model systems that molecular components of 32 associative learning seem to be evolutionarily conserved from worms to humans 4 . Since memories, 33 in particular of negative experiences, have a large impact on decision-making as well as cognitive 34 and emotional states, mechanisms underlying these functions may share common elements. 35 Modifications in neural signaling mediate and maintain experience-driven changes in brain circuits 36 for memory 5 . The hypothesis that neuropeptides are central modulators of such learning circuits is 37 currently emerging, and neuropeptide receptors have gained significant interest as potential key 38 targets for treatment of cognitive disorders [6][7][8][9][10] . Neuropeptides make up an evolutionary ancient and 39 diverse class of neural messengers that serve a broad range of functions [11][12][13][14][15] . They modulate neural 40 signaling mainly by binding G protein-coupled receptors (GPCRs) 16,17 and are widely distributed in 41 brain areas responsible for learning and memory [18][19][20][21] . 42 A role in experience-dependent plasticity has been established for some neuropeptides, but many 43 questions remain concerning the actions of neuromodulators in learning circuits. Behaviorally, it is 44 unclear whether they regulate specific aspects of a learned behavior or serve as general modulators 45 that coordinately control distinct conditioned responses to a stimulus. Also, the temporal 46 requirement of peptidergic neurons in learning circuits is largely unexplored, in part because most 47 experimental studies examined the effects of peptide injections into the brain, which is limited in 48 temporal and spatial resolution for studying neuropeptidergic modulation 18,20,21 . 49 With a nervous system of 302 neurons, Caenorhabditis elegans provides an excellent model system 50 for dissecting learning and neuropeptidergic circuits [22][23][24][25][26] . The physical connectivity of neurons has 51 been mapped and the nematode shows different kinds of plasticity in gustatory, olfactory, thermo-52 and mechanosensory behaviors. These include associative learning and memory, which require 53 molecular mechanisms similar to those in vertebrates 11,27,28 . We previously found that gustatory 54 associative learning in C. elegans is regulated by nematocin, a neuropeptide of the 55 vasopressin/oxytocin family that promotes salt aversion when worms experienced salt in the 56 absence of food 10 . The C. elegans genome counts at least 153 neuropeptide-encoding genes and 150 57 peptide GPCRs 29-31 , many of which are evolutionarily conserved between worms and humans 32-34 . 58 Learning is likely regulated by multiple neuromodulators, yet our understanding of how individual 59 modulators contribute to a learned behavior in a defined circuit is limited. 60 To uncover neuropeptidergic pathways that regulate learning circuits, we tested C. elegans mutants 61 of neuropeptide receptors for their performance in gustatory associative learning. We focused on 62 evolutionarily conserved neuropeptide systems that are expressed in brain regions for learning and 63 memory. Such a neuropeptide system is the neuromedin U (NMU) pathway that is present in most 64 bilaterian animals 32,35 . We hypothesized that NMU-like signaling regulates learning because NMU 65 peptides and their receptors are widely expressed in memory centers, such as in the vertebrate 66 hippocampus and the insect mushroom bodies [36][37][38] . In mammals, intranasal administration or 67 injection of NMU peptides into the brain affects memory and reward-related behaviors, further 68 suggesting a role for NMU in experience-dependent plasticity 39-43 , although mechanisms underlying 69 these effects remain unknown. 70 Here, we demonstrate that NMU neurons are specifically required for the retrieval of learned salt 71 aversion in C. elegans. In addition, we show that NMU neuropeptides released from a pair of sensory 72 neurons, i.e. the ASG cells that show experience-dependent responses to salt, signal via their 73 receptor NMUR-1 on the AFD sensory neurons to coordinately regulate distinct navigational 74 decisions for adapting salt chemotaxis behavior. While it is widely assumed in many fields, including 75 neuropsychology and neuroeconomics, that decision-making involves learning and memory, the 76 underlying molecular underpinnings and neural mechanisms are still not well understood. Our 77 findings indicate a role for the NMU neuropeptide family as a modulator of the expression of learned 78 aversion, and lay a foundation for further research examining the functions of the conserved NMU 79 system in other learning circuits underlying decision-making tasks. 80

81
A C. elegans neuromedin receptor ortholog, NMUR-1, regulates gustatory aversive learning 82 C. elegans has three predicted NMU receptor orthologs, named NMUR-1 to NMUR-3 32,34 (Figure 1a). 83 We hypothesized that this receptor family is involved in experience-dependent plasticity, because 84 NMU receptors are widely expressed in memory centers of the brain [36][37][38]44 . To test this hypothesis, 85 we analyzed the performance of nmur loss-of-function mutants (Figure 1b) in an established 86 paradigm for gustatory associative learning 10,45 . Similar to classical aversive conditioning, we trained 87 worms by pairing a conditioned salt stimulus with the absence of food during a 15-minute training 88 period ( Figure 1c). We then tested salt chemotaxis behavior by putting worms in the center of a 89 quadrant plate and calculated a chemotaxis index based on the number of animals that migrated 90 towards or away from NaCl-containing quadrants (Figure 1c). Mock-conditioned worms, which have 91 not been pre-exposed to salt, were strongly attracted to NaCl and wild-type animals learned to avoid 92 salt when they previously experienced it in the absence of food ( Figure 1d). Consistent with our 93 hypothesis, nmur-1 mutants were defective in learning salt aversion and were still attracted to salt 94 after aversive NaCl-conditioning ( Figure 1d). Mutants for the NMU receptor orthologs nmur-2 and 95 nmur-3 had no plasticity defects (Figure 1d), suggesting a role specific for NMUR-1 in gustatory 96 aversive learning. 97 The learning defect of nmur-1 animals could be caused by a failure to associate salt with the lack of 98 food, or may alternatively result from general defects in locomotion or salt chemotaxis behavior. To 99 address this, we tested nmur-1 mutants for their behavioral responses to salt in different training 100 regimes (Supplementary figure 1). We first examined chemotaxis behavior of nmur-1 animals for 101 increasing NaCl concentrations and observed no difference compared to wild-type animals 102 (Supplementary figure 1a-b). To determine if food cues contribute to the plasticity defect, we 103 performed conditioning on agar plates instead of pre-exposing animals to salt in liquid 104 (Supplementary figure 1c). Mutants of nmur-1 showed normal salt chemotaxis behavior when salt 105 was paired with food (Supplementary figure 1d). Mock-conditioned nmur-1 animals that were not 106 pre-exposed to salt also behaved like wild-type worms (Figure 1d and Supplementary figure 1d). 107 These results suggest that salt sensing is not affected in nmur-1 mutants and that NMUR-1 signaling 108 promotes gustatory aversive learning. 109 To answer this question, we set up a tracking platform to monitor trajectories of individual worms 120 and quantify locomotion parameters relative to a linear salt gradient on square chemotaxis plates 121 (Figure 2a). We first examined how gustatory aversive learning in wild-type animals is manifested on 122 these linear salt gradients by measuring the position of worms during a 30-minute time window 123 immediately after training. Similar to the quadrant-based assay (Figure 1c), mock-conditioned wild-124 type animals were attracted to salt and had migrated up the gradient after ten minutes (Figure 2b  125 and c). NaCl-conditioning in the absence of food significantly reduced salt attraction, as conditioned 126 animals avoided high NaCl concentrations (Figure 2b and c). Besides the positions of animals on the 127 gradient, we determined a chemotactic index by dividing the mean velocity of each individual's 128 trajectory relative to the gradient by its mean crawling speed 22 . As expected, mock-conditioned wild-129 type animals had a positive chemotactic index, meaning they are attracted to salt, whereas NaCl-130 conditioned animals showed a negative index due to salt avoidance ( Figure 2d). We also calculated 131

NMUR-1 regulates distinct navigational decisions for the experience-dependent modulation of salt
indices to describe the two behavioral strategies for salt chemotaxis. In a biased random walk index, 132 we determined the fractional difference in relative run duration up or down the gradient of 133 individual trajectories (Figure 2e). The klinotaxis index was calculated as the fractional difference in 134 the relative probability of each animal to reorient up or down the gradient after each turn ( Figure  135 2f). In line with former studies 22,48 , we found that both navigational strategies are modulated by 136 aversive experience, because the biased random walk and klinotaxis indices of wild-type animals 137 reversed after NaCl-conditioning (Figure 2e -f). 138 Using this behavioral tracking platform, we further examined the defect in learned salt avoidance of 139 nmur-1 mutants. Similar to the quadrant-based assay (Figure 1d), mock-conditioned nmur-1 animals 140 displayed normal salt chemotaxis behavior (Figure 2c-f), while NaCl-conditioned mutants showed 141 less aversion for high NaCl concentrations than wild-type animals (Figure 2b-d). Interestingly, we 142 observed significant differences in biased random walk and klinotaxis of nmur-1 mutants after 143 aversive conditioning, suggesting that NMUR-1 signaling is involved in the experience-dependent 144 modulation of both the length of runs and the reorientation on the NaCl gradient (Figure 2f and g). 145 To confirm the role of nmur-1 in gustatory aversive learning, we tested whether restoring the 146 expression of nmur-1 rescued the mutant's phenotypes. Therefore, we expressed wild-type copies of 147 the receptor under the control of its endogenous promoter (Figure 2g), which recapitulated the 148 reported expression pattern for nmur-1 in approximately 15 sensory neurons (ADF, ADL and AFD) 149 and interneurons (AIA, AIZ, among others) 52 . This transgene fully rescued salt chemotaxis behavior 150 of nmur-1 mutants after aversive conditioning, including defects in the modulation of klinotaxis and 151 biased random walk (Figure 2j and Supplementary figure 2a for raw data). Next, we asked in which 152 cell(s) NMUR-1 acts to potentiate salt aversive learning by probing each nmur-1 expressing neuron in 153 cell-specific rescue experiments (Figure 2j and Supplementary figures 2-3 for raw data). Re-154 introducing the wild-type nmur-1 cDNA in AFD sensory neurons using various promoters, including  155  the AFD-specific gcy-8 promoter 53 fully restored learned salt aversion, whereas cell-specific  156  expression in other nmur-1 expressing sensory neurons and interneurons did not restore the mutant  157 phenotype (Figure 2h, i, j and Supplementary figure 3a, b and d). Taken together, the NMU receptor 158 ortholog NMUR-1 is required in the AFD sensory neurons to coordinately modulate distinct 159 navigational decisions for establishing learned salt avoidance. 160 NMU-like neuropeptides of the C. elegans CAPA-1/NLP-44 precursor are ligands of NMUR-1 161 NMUR-1 is an orphan GPCR, meaning that so far no ligands have been identified for this receptor. To 162 further characterize the role of NMUR-1 in gustatory aversive learning, we set out to identify its 163 ligand(s) using reverse pharmacology. Therefore, we expressed NMUR-1 in Chinese Hamster Ovary 164 (CHO) cells that were challenged with a synthetic library of over 300 C. elegans peptides, and we 165 monitored GPCR activation by a calcium reporter assay ( Because CAPA-1 peptides activate NMUR-1 in vitro, we hypothesized that these neuropeptides 180 modulate gustatory aversive learning through NMUR-1 signaling. For this purpose, we examined 181 gustatory learning in two capa-1 mutants with independent loss-of-function alleles ( Figure 3c). As 182 expected, we found that both capa-1 mutants display a learning defect, which mimics that of nmur-1 183 mutants (Figure 3e and g). Impaired CAPA-1 signaling, however, has no effect on NaCl chemotaxis 184 behavior of mock-conditioned animals. Like nmur-1 mutants, worms lacking CAPA-1 signaling show 185 defects in the modulation of biased random walk and klinotaxis after NaCl-conditioning 186 (Supplementary figure 4a). The reduced salt avoidance of capa-1 mutants does not result from 187 general defects in locomotion or salt chemotaxis, because loss of capa-1 function did not cause a 188 detectable change in behavioral responses to a range of NaCl concentrations and mutants of capa-1 189 behaved like wild-type animals after pairing salt with the presence of food (Supplementary figure 5a  190 and b). These results suggest that NMU-like CAPA-1 neuropeptides are also required for gustatory 191 aversive learning. 192 Using a bicistronic GFP reporter transgene, we found that capa-1 is expressed in a single pair of 193 neurons (Figure 3f), which is in accordance with previous whole-mount immunostaining using an 194 antiserum that recognizes CAPA-1-derived neuropeptides 35 . Based on the cellular position and 195 morphology, we identified these cells as the ASG neurons. Their cellular identity was further 196 confirmed by co-localization of the GFP signal with an mCherry protein expressed from the gcy-15 197 promoter ( Figure 4f) that is known to drive expression in ASG only 55 . ASG is a chemosensory neuron 198 pair that is involved in detecting food and water-soluble attractants such as NaCl 56-58 . Restoring

CAPA-1/NMUR-1 signaling is not involved in food searching behaviors 216
Previous work indicates that the ASG neurons, which express CAPA-1, regulate foraging behaviors 61 . 217 NMU-like peptides are potent regulators of feeding in insects and mammals 62,63 , and the C. elegans 218 receptor NMUR-1 has also been shown to modulate lifespan depending on the type of bacterial 219 diet 52,64 . These observations prompted us to investigate if the CAPA-1/NMUR-1 pathway is involved 220 in signaling the environmental food context and if this function could underlie its effect on gustatory 221 aversive learning. 222 Loss of NMUR-1 has been shown to extend lifespan only in worms grown on B-type E. coli strains 223 such as OP50, while it has no effect on K-12 type bacterial strains like HT115. We therefore asked if 224 the effect of NMUR-1 signaling on gustatory aversive learning depends on the bacterial diet, by 225 assaying gustatory plasticity of worms cultured on HT115 E. coli (Supplementary figure 7). Mutants 226 of capa-1 and nmur-1 grown on HT115 showed learning defects similar to worms cultured on OP50 227 ( We also tested whether CAPA-1/NMUR-1 signaling is involved in ASG-dependent foraging behaviors. 230 Upon removal from their E. coli food source, C. elegans display an intensive local search behavior 231 with frequent turning and reorientations, which are reduced in ASG-ablated animals 65 conditioned with protocols similar to those used for behavioral learning assays, i.e. we first pre-258 exposed worms to a buffer with (NaCl-conditioned) or without (mock-conditioned) NaCl for 15 259 minutes before testing ASG activity. We then measured ASG calcium responses in response to either 260 up-or downshifts in NaCl concentrations 73 . ASG showed small calcium responses to NaCl in both 261 mock-and NaCl-conditioned animals. However, the response to NaCl downshifts was higher in 262 worms pre-exposed to salt in the absence of food. After aversive conditioning, ASG responded 263 significantly stronger to NaCl removal in comparison to ASG responses in mock-conditioned worms 264 ( Figure 4a). This result suggests that CAPA-1 neurons respond differently to salt stimuli after aversive 265 conditioning and may facilitate learning through modulation of the salt sensing circuit. 266 Next, we asked if NMU signaling adapts salt-evoked activity of the primary salt sensor, ASE. This left-267 right pair of neurons (ASEL and ASER), exhibits functional asymmetry: ASEL and ASER are activated 268 by an increase and decrease in NaCl concentrations, respectively 70,74 . Consistent with this, mock-269 conditioned wild-type worms expressing the ratiometric YC2.12 indicator in both ASEL and ASER 270 showed robust ASEL responses upon NaCl upshifts, while a downshift in NaCl concentration reliably 271 evoked an increase in calcium in the ASER neuron ( Figure 4b). Following NaCl-conditioning, ASEL 272 responses to a NaCl upshift were significantly reduced and ASER responses to NaCl downshifts were 273 sensitized, which is in agreement with previous studies 73, 75 . ASE calcium responses to salt in mock-274 and NaCl-conditioned animals were not affected in nmur-1 mutants, indicating that NMU signaling is 275 not required for the plasticity of ASE responses during aversive learning. 276

CAPA-1 neurons are required for the retrieval, but not the acquisition, of learned salt avoidance 278
To gain further insight into the mechanisms by which NMU-like neuropeptides regulate aversive  279 learning, we set out to uncover the temporal requirement of CAPA-1 neurons in the learning circuit. 280 Therefore, we first investigated the effect of chemically silencing ASG by expressing tetanus toxin 281 light chain (TeTx) specifically in these neurons. TeTx disrupts chemical signaling as it impedes the 282 release of synaptic vesicles and neuropeptide-containing dense core vesicles 76 . As expected from the 283 role of capa-1 in gustatory plasticity, ASG-specific expression of TeTx impaired aversive learning, 284 whereas it did not affect NaCl chemotaxis of mock-conditioned animals (Figure 5a-d). 285 To further determine when ASG signaling is required for gustatory learning, we optogenetically 286 silenced these neurons during NaCl-conditioning or during the immediate recall of learned 287 avoidance behavior when worms are navigating on a NaCl gradient. For this purpose, we expressed 288 the outward-directed proton pump Arch under the control of the capa-1 promoter 77,78 ( Figure 6a). 289 Upon adding the essential cofactor for opsin activity (all-trans retinal, ATR), Arch can be activated by 290 illuminating transgenic worms with green-red light. The subsequent membrane hyperpolarization 291 silences the host neuron. We varied the timing of illumination so that CAPA-1 neurons were silenced 292 either during the acquisition of salt aversion or during retrieval of the learned behavior ( Figure 6b). 293 Silencing CAPA-1 neurons during the acquisition phase had little effect on learned salt avoidance 294 ( Figure 6c-f). In contrast, silencing during the retrieval phase significantly attenuated salt avoidance 295 ( Figure 6c-f). Optogenetically inhibiting CAPA-1 neurons had no effect on general locomotion in the 296 absence of a NaCl gradient (Supplementary figure 9). These results show that CAPA-1 neurons are 297 required for the retrieval of learned salt avoidance, but not for the acquisition of the conditioned 298 response. 299

300
Although neuropeptides are increasingly recognized as modulators of learning and memory, it is not 301 well understood how they adapt behavioral choices and when specific neuromodulators operate in 302 learning circuits. In this study we demonstrate that neuropeptides of the NMU family promote 303 gustatory aversive learning in C. elegans by modulating distinct navigational decisions according to 304 previous experience. We show that NMU neurons are specifically required for the retrieval of 305 learned avoidance behavior. Since NMU neuropeptides are highly conserved in bilaterian animals, 306 including humans, these findings lay a foundation for further research on the involvement of this 307 ancient neuropeptide family in other learning circuits. 308 In C. elegans NMU-like neuropeptides, encoded by the capa-1/nlp-44 gene, are expressed in a single 309 pair of chemosensory neurons, the ASG cells, which promote gustatory aversive learning through 310 CAPA-1 signaling. To determine when signaling from CAPA-1 neurons is required for learning, we 311 used optogenetics to temporally silence these cells during the conditioning phase or during retrieval 312 of the conditioned response immediately after training. Only the latter affected the experience-313 dependent adaptation of NaCl chemotaxis. CAPA-1 neurons thus control the expression of avoidance 314 behavior, but are not required for acquisition of this response. Evidence for such a temporally 315 defined role of neuropeptide signaling in learning circuits also emanates from Drosophila studies on 316 neuropeptide F and insulin-like peptides, which are large neuropeptides of over 100 amino acids 317 that act via GPCR and receptor tyrosine kinase signaling, respectively 79-81 . Of notice, one C. elegans 318 study characterized the memory phase of insulin receptor signaling in an olfactory learning circuit. 319 Using a conditional daf-2 allele, the authors showed that insulin receptor signaling in the olfactory 320 AWC neurons is only partially involved in memory acquisition, but necessary during memory 321 retrieval 82 . By manipulating neuropeptide signaling at the level of peptidergic neurons, our results 322 indicate that small neuropeptide messengers of the NMU family specifically control memory 323 retrieval. A defined temporal role of neuropeptides in learning circuits may be a general feature of 324 neuropeptide-mediated memory. 325 How does CAPA-1 signaling regulate the expression of learned salt avoidance behavior? We 326 identified the NMU receptor ortholog NMUR-1 as a CAPA-1 receptor that is required for gustatory 327 aversive learning. NMUR-1 is present in several sensory neurons and interneurons and has 328 previously been implicated in the food-dependent modulation of lifespan 52,64 . However, the learning 329 defects of capa-1 and nmur-1 mutant animals are not affected by the worms' bacterial diet and 330 CAPA-1 signaling mutants display normal foraging, salt-sensing and locomotion behaviors. 331 Therefore, a general function in sensing salt or the environmental food context does not seem to 332 underlie the effect of CAPA-1 peptides on aversive learning. Our calcium imaging data suggest a 333 model in which aversive conditioning recruits CAPA-1 neurons to modulate the gustatory circuit. 334 Aversive conditioning sensitizes the response of CAPA-1 neurons to NaCl downshifts, which suggests 335 that ASG neurons contribute to the experience-dependent adaptation of salt chemotaxis behavior 336 when worms are exposed to the aversive condition of salt without food. This function is reminiscent 337 of a previous report showing ASG to be recruited to a circuit for processing NaCl cues under another 338 type of stressful condition, hypoxia 71 . After prolonged exposure to hypoxic stress, ASG neurons 339 enhance NaCl attraction through increased serotonin production. Whether CAPA-1 neuropeptides 340 are involved in this process is not known. 341 At the behavioral level, CAPA-1/NMUR-1 signaling facilitates the adaptation of two elementary 342 behavioral strategies for NaCl chemotaxis, i.e. biased random walk and klinotaxis. NMU-like signaling 343 thus acts as a central regulator of different decision-making tasks for expressing learned avoidance 344 behavior, rather than specifically affecting a neural substructure of the behavioral output. Conform 345 to previous work, we found gustatory aversive conditioning to adapt the activity of the primary salt-346 sensor ASE that mediates the attractive drive towards NaCl 73,75 . This happens in an NMUR-1-347 independent manner, suggesting that NMU signaling does not indirectly attenuate the neural 348 pathway for NaCl attraction. By contrast, we found that CAPA-1 facilitates salt learning by acting on 349 NMUR-1 in AFD sensory neurons, which play a well-characterized role in various sensory modalities 350 such as the sensation of thermal, magnetic and CO 2 cues 83-85 . Interestingly, AFD has recently been 351 shown to respond to NaCl fluctuations 86  precursor gene hugin is part of a sensory pathway that relays between gustatory neurons and higher 368 brain centers. This neural subset transduces aversive gustatory cues, such as bitter tastes, resulting 369 in a brake on feeding through avoidance behavior and decreased pharyngeal pumping 38,89,92,93 . In this 370 context, hugin may play a role similar to capa-1 in C. elegans, signaling aversive feeding conditions 371 and stimulating motor programs that allow the animal to migrate to more favorable food 372 environments. Since both the hugin gene and the capa receptor gene (capaR) are expressed in the 373 mushroom bodies, the fruit fly's brain centers that play an important role in sensory integration and 374 memory formation 36 , the Drosophila NMU homolog might be involved in learning as well. Similarly, 375 the NMU neuropeptides and the neuromedin receptor NMUR2 are expressed in discrete regions of 376 the mammalian brain that are instrumental for memory and processing of aversive experiences, 377 including the amygdala, hippocampus and cerebral cortex 37,94,95 . 378 Taken together, we have uncovered a neural mechanism by which the retrieval of learned avoidance 379 is regulated by the evolutionarily conserved NMU neuropeptide pathway. Given that the NMU 380 system is conserved across bilaterians and that NMU specifically regulates retrieval of learned 381 avoidance, our study encourages further research into the memory functions of this neuropeptide 382 system in other animals and in the context of aversive learning disorders.

401
The authors declare no conflict of interest. 402 in blue, while red markers denote NaCl-conditioned worms. 420 significances are relative to NaCl-conditioned wild-type animals tested on the same day. Raw data 445 are depicted on Supplementary Fig. 3 and 4, for cell-specific rescue in the interneurons or sensory 446 neurons, respectively. n ≥ 50 animals per genotype. 447   neurons are not required for NaCl chemotaxis behavior of mock-conditioned animals, because ATR-510 fed animals display normal chemotactic, biased random walk and klinotaxis behavioral indices. 511 Optogenetic silencing of CAPA-1 neurons after NaCl-conditioning during the retrieval phase disrupts 512 the learned salt avoidance response, while silencing during acquisition has no effect. 513 514

732
Further information and requests for resources and reagents should be directed to and will be 733 fulfilled by Liliane Schoofs (liliane.schoofs@kuleuven.be) and Isabel Beets 734 (isabel.beets@kuleuven.be). 735

C. elegans strains 736
C. elegans was cultured using standard methods at 20°C on agar with Nematode Growth Medium 737 (NGM) seeded with OP50 or HT115 E. coli bacteria as a food source. Wild-type worms are of the 738 Bristol variety N2. N2 wild type and RB1288 nmur-1 (ok1387), RB2526 nmur-2 (ok3502), VC1974 739 nmur-3 (ok2295), RB2263 capa-1 (ok3065), CX10 osm-9 (ky10) mutants were acquired from the 740 Caenorhabditis Genetics Center at the University of Minnesota. capa-1 (tm3764) mutants were 741 provided by the National Bioresource Project of Japan. Mutant genotypes were confirmed using PCR 742 and backcrossed to the common wild-type strain (N2) at least 6 times to remove unlinked mutations 743 prior to analysis. All mutant and transgenic strains used in this study are listed in the additional 744 resource table. 745

Molecular biology 759
All oligonucleotide primers used in this study are listed in the additional resource table. 760 For heterologous expression of NMUR-1 in Chinese hamster ovary (CHO) cells, nmur-1 cDNA was 761 amplified by PCR from cDNA of mixed-stage wild-type C. elegans and directionally cloned into the 762 eukaryotic expression vector pcDNA3.1 (Invitrogen). 763 GFP reporter and rescue constructs were generated from a modified pSM vector carrying a GFP 764 reporter sequence preceded by an SL2 trans-splicing sequence (kindly provided by C. Bargmann,765 Rockefeller University, New York, USA). For nmur-1, the nmur-1 cDNA was first amplified from 766 mixed-stage wild-type C. elegans template and inserted into a pSM backbone (linearized by BamHI 767 and NheI restriction digest) using Gibson assembly. The 1kb nmur-1 3'UTR sequence was then 768 inserted after the GFP coding sequence by Gibson assembly into the backbone digested by EcoRI and 769 SpeI, which replaced the unc-54 3'UTR sequence native to the original pSM plasmid. Finally, a 2 kb 770 nmur-1 promoter fragment was inserted prior to the gene's cDNA sequence after SphI and XbaI 771 digestion. 772 For cell-specific rescue experiments, the nmur-1::SL2::gfp::nmur-1 3'UTR pSM backbone was first 773 linearized by PCR amplification, yielding a 5900 bp long linear sequence in which promoters driving 774 expression in selected neurons were inserted using Gibson Assembly. The transgenes for ASG-specific capa-1 knockdown were constructed as previously described 4 . 789 Briefly, the ops-1 promoter (1.9 kb) was fused to a 1132 bp genomic capa-1 fragment in sense and 790 antisense orientation. The genomic capa-1 fragment is identical to that of the clone taken up in the 791 Ahringer feeding RNAi library 5 . 792 The transgene for ASG-specific gcy-15p::mCherry expression was constructed through fusion PCR. 793 First, the 995 bp mCherry sequence was amplified from pGH8, the 1908 bp promoter and the 883 bp 794 3'UTR of the gcy-15 gene amplified. All three fragments were sequentially fused through fusion PCR. 795 In vitro GPCR activation assay 796 The GPCR activation assay was performed as previously described 6,7  of readout, 0.1 % triton X-100 was added to lyse the cells, resulting in a maximal calcium response 806 that was measured for 10 s. After initial screening, concentration-response curves were constructed 807 for HPLC-purified CAPA-1 peptides by subjecting the transfected cells to each peptide in a 808 concentration range from 0.1 nM to 100 µM. Cells transfected with an empty vector were used as a 809 negative control. Assays were performed in triplicate on at least two independent days. 810 Concentration-response curves were fitted using GraphPad Prism 5 (nonlinear regression analysis  811 with a sigmoidal concentration-response equation). 812 A peptide library of over 300 synthetic C. elegans peptides that was used to challenge the NMUR-1 813 expressing CHO cells was compiled based on in silico predictions and peptidomics data 8,9 . Peptides 814 were synthesized by Thermo Scientific and GL Biochem Ltd. 815

Transgenesis and expression pattern analysis 816
To generate transgenic C. elegans, constructs were injected into the syncytial gonad of young adult 817 worms at concentrations ranging from 5 to 50 ng/µL with 50 ng/µL of the coinjection marker elt-818 2p::GFP or unc-122p::dsRED and 17 ng/µL of a 1-kb DNA ladder (Thermo Scientific) as carrier DNA. 819 Expression patterns of reporter transgenes were visualized by an inverted Zeiss AxioObserver Z1 820 microscope fitted with a 40X oil objective, an ORCA-Flash4.0 V2 camera (Hamamatsu) and W-View 821 GEMINI image splitting optics (Hamamatsu). Image acquisition was performed using Metamorph 822 (Molecular Devices) software. For imaging, hermaphrodite animals were immobilized on 10% 823 agarose pads using polystyrene beads between pad and coverslip to restrain the animal's movement 824 (Polybead® Microspheres 0.10μm, Polyscience). Expression patterns were confirmed in at least two 825 independent transgenic strains. 826

Salt chemotaxis assays 827
Salt chemotaxis and gustatory plasticity behavior were assessed as described previously 10,11 ( Figure  828 1). All behavioral assays were performed in a climate-controlled room set at 20°C and 40% relative 829 humidity. Plates were then closed and used on the same day. After animals were allowed to crawl for 10 838 minutes on the quadrant plate, a chemotaxis index was calculated as (n(A) -n(C)) / (n(A) + n(C)) 839 where n(A) is the number of worms within the quadrants containing NaCl and n(C) is the number of 840 worms within the control quadrants without NaCl. 841 For gustatory plasticity, the well-fed synchronized adult worm population is washed in CTX buffer 842 with or without 100 mM NaCl for NaCl-conditioned and mock-conditioned worms, respectively. The 843 response to 25 mM NaCl is then assayed on four-quadrant plates to which 25 mM NaCl has been 844 added to one pair of opposing quadrants. After 10 min, the distribution of worms over the quadrants 845 is determined and a chemotaxis index calculated as described above. 846 For tracking analysis, conditioned behavior is assessed on 0 to 100 mM linear NaCl gradients 12 . These 847 are generated in square Petri dishes (Greiner, 120 x 120 x 17mm, vented), by elevating one side of 848 the plate and filling half of the plate with 50 mL of buffered agar supplemented with 100 mM NaCl. 849 Allowing the agar to solidify for 30 min creates a triangle wedge, after which the plate is laid flat and 850 the other half of the plate is filled with 50 mL buffered agar without NaCl. Behavioral assays were 851 performed 24 hr after the gradients were established. In each assay, 15-20 young adult worms were 852 mock-or NaCl-conditioned as described above, and pipetted to the middle of the square plate 853 corresponding to approximately 50 mM NaCl. 854 For pre-exposure to NaCl in the presence of food 13 , adult animals were washed off culture plates 855 using CTX buffer with or without 100 mM NaCl and allowed to sediment for a few minutes. was calculated per individual, as well as the fraction of time spent in each behavior as determined by 920 the segmentation algorithm explained above. 921

Imaging of neuronal activity in microfluidic chip 922
Transgenic worms expressing GCaMP3 in ASG 20 or YC2.12 in ASE neurons 21 were imaged in a 923 microfluidic polydimethylsiloxane (PDMS) chip 22 . This PDMS chip is attached to inlet tubes, 924 connecting the worm-loading channel to a reservoir of CTX buffer (with or without 100 mM NaCl). 925 Well-fed transgenic worms were first picked in the reservoir, after which they were exposed to the 926 CTX buffer in the absence of food for approximately 15 minutes. NaCl-responses were then assayed 927 by positioning the worm in the channel and challenging the worm to either 0 to 50 mM NaCl upshift 928 or 50 to 0 mM NaCl downshift. The composition of the buffers for imaging is the same as the buffers 929 used in the behavioral assay, except for the addition of sucrose to balance osmolarity (5 mM 930 KH 2 PO 4 /K 2 HPO 4 pH 6.6, 1 mM MgSO 4 , and 1 mM CaCl 2 and 0, 50 or 100 mM NaCl and adjusted to 931 350 mOsmol/kgH 2 O by sucrose). Imaging was conducted on an inverted Zeiss AxioObserver Z1 932 microscope. Fluorescent images were captured using a 40X objective and an ORCA-Flash4.0 V2 933 camera (Hamamatsu) driven by Metamorph software (Molecular Devices). Images were analyzed 934 using custom-written Mathematica (Wolfram) code setting a region of interest around the cell body. 935 For GCaMP3 imaging, the adjacent background was subtracted on each image and the fluorescence 936 intensity during the 10 second period before stimulus delivery was averaged and defined as F 0 . 937 ∆F/F0 (%) is calculated as 100*(background corrected fluorescence -F 0 ) / F 0 . For YC2.12 imaging, 938 ∆R/R (%) is calculated as the ratio between CFP and YFP fluorescent emissions. 939

Optogenetic silencing of ASG 940
Transgenic worms expressing the outward-directed proton pump Arch in ASG were cultured at 25°C 941 on culture plates seeded with sufficient OP50 E. coli. One day before the assay, transgenic worms 942 were transferred to OP50 culture plates to which all-trans retinal (ATR, Sigma Aldrich) has been 943 added (300 µM final concentration in fresh OP50 liquid culture). This cofactor is needed for Arch 944 activity, and allows transgenic worms transferred to culture plates without ATR to serve as negative 945 control. A 0 to 50 mM linear gradient was generated in a 55 mm Petri plate conform to the protocol 946 for linear gradients in square plates (15 mL buffered agar supplemented with 50 mM NaCl is poured 947 in a 55 mm diameter Petri dish elevated to one side. An additional 15.5 mL buffered agar without 948 NaCl is poured on top of the first triangular wedge once it is hardened. NaCl is allowed to diffuse at 949 least 24 h prior the assay). Individual worms were washed for 15 min in a small volume of CTX buffer 950 with or without 100 mM NaCl (NaCl-conditioning or mock-conditioning  respectively. The whiskers extend to the most extreme data points not considered as outlier. For the 974 fraction of time in each behavioral state, the relative fraction was binned over individual worms and 975 analyzed for statistical significance by Chi-squared analysis. The time spent in roaming or dwelling 976 states while foraging on a bacterial lawn was analyzed using a Kruskal-Wallis test. For chemotaxis 977 studies in which worms were pre-exposed to NaCl in the presence or absence of food, a two-way 978 ANOVA was used in combination with a Tukey post-hoc test. Data from optogenetic experiments 979 was analyzed using a three-way ANOVA followed by a Tukey post-hoc test. 980 All behavioral assays were conducted at least four times on at least two separate days. For 981 transgenic experiments, animals with clear target fluorescence emanating from the pSM backbone, 982 TeTx::mCherry or Arch::YFP fusions were selected as transgenic animals. Siblings without this 983 fluorescence (or not having fluorescence from the co-injection marker in the case of nmur-1 rescue 984 and ASG-specific capa-1 RNAi experiments) were used as control and showed behavior similar to 985 non-transgenic mutant animals. 986