Insulin/IGF Signaling Regulates Presynaptic Glutamate Release in Aversive Olfactory Learning

Information flow through neural circuits is continuously modified by context-dependent learning. In the nematode Caenorhabditis elegans, pairing specific odors with food deprivation results in aversion to the odor. Here we identify cell-specific mechanisms of insulin/IGF receptor signaling that integrate sensory information with food context during aversive olfactory learning. Using FLP::FRT recombination of an endogenously targeted locus, we show that aversive learning to butanone, an odor sensed only by the AWCON olfactory neuron, requires the insulin/IGF receptor DAF-2 in AWCON. Learning requires an axonally-localized DAF-2c isoform and the insulin receptor substrate (IRS) protein IST-1, and is partly independent of the FoxO transcription factor DAF-16. Food deprivation, the unconditioned stimulus for learning, increases DAF-2 expression post-transcriptionally through an insulin- and ist-1-dependent process. Aversive learning suppresses odor-regulated glutamate release from the AWCON axon in wild-type animals but not in ist-1 mutants, suggesting that insulin signaling drives presynaptic depression to generate an aversive memory.


INTRODUCTION
Insulin and Insulin-like Growth Factors (IGFs) are central regulators of metabolism and growth.
In mammals, they function as endocrine or local signals to stimulate glucose uptake, cell and organ growth, cell proliferation, and reproduction (Tokarz, et al., 2018;Yakar & Adamo, 2012).
Mammalian insulin/IGF receptors are transmembrane tyrosine kinases that initiate signaling after ligand binding by cross-phosphorylation to form docking sites for Insulin Receptor Substrate (IRS) proteins (Kavran, et al., 2014). IRS proteins are phosphorylated, and then recruit PI3K  to the cell membrane, where it converts the lipid PIP2 (Phosphatidylinositol 4,5-bisphosphate) to PIP3 (Phosphatidylinositol (3,4,5)trisphosphate). PIP3 in turn activates PDK and AKT kinases, resulting in phosphorylation of multiple downstream targets. For IGF receptor signaling, important targets include FoxO, a transcription factor that is phosphorylated and retained in the cytoplasm in an inactive form by Insulin/IGF signaling. For insulin receptor signaling, critical downstream targets include a cytoplasmic signaling pathway that regulates trafficking of the plasma membrane glucose transporter GLUT-4 to support glucose uptake. The insulin/IGF pathway is negatively regulated by PTEN, a phosphatase that dephosphorylates PIP3 to form PIP2 and inorganic phosphate.
Insulin/IGF signaling also functions in learning and memory. In the nematode worm Caenorhabditis elegans, it regulates aversive learning and memory in contexts where food deprivation acts as an unconditioned stimulus that drives learning (Cho, et al., 2016;Tomioka, et al., 2016;Lin, et al., 2010;Tomioka, et al., 2006). In Drosophila, insulin/IGF signaling acts at multiple times and places to promote olfactory associative learning: it supports development of the mushroom bodies, increases learning efficiency, and drives glucose mobilization for longterm memory formation (de Tredern, et al., 2021;Chambers, et al., 2015;Naganos, et al., 2012).
Mammalian insulin/IGF signaling has been implicated in CNS plasticity and learning in the hypothalamus, hippocampus, olfactory bulb, nucleus accumbens, ventral tegmental area, and infralimbic cortex (Reviewed in (Ferrario & Reagan, 2018;Dyera, et al., 2016)). Despite the robust circuit-level effects of insulin/IGF in mammals, the relevant cellular mechanisms of insulin/IGF action are only partly understood.
In C. elegans, pairing odors, salts, or temperature with differing food contexts can result in behavioral changes that are highly specific to the conditioned stimulus. For example, pairing the attractive odor butanone with food deprivation decreases attraction to butanone, without suppressing chemotaxis to other attractive odors (Colbert & Bargmann, 1995), whereas pairing butanone with food increases attraction to butanone (Torayama, et al., 2007). In each case, butanone attraction behavior relies upon a single olfactory neuron, AWC ON (Torayama, et al., 2007;Wes & Bargmann, 2001). Similarly, pairing salt with food deprivation suppresses salt attraction, and pairing salt with food increases salt attraction (Kunitomo, et al., 2013), in both cases through a single sensory neuron, ASER. The insulin-like neuropeptide INS-1 is a candidate to convey the unconditioned stimulus of food deprivation to these sensory neurons during learning. Aversive learning after pairing food deprivation with odors or salt is lost in ins-1 mutants (Tomioka, et al., 2016;Lin, et al., 2010;Tomioka, et al., 2006), but their appetitive learning after pairing butanone with food is normal (Lin, et al., 2010). INS-1 is released from AIA integrating neurons, and possibly other cells, and acts on sensory neurons to support aversive learning and memory. Both INS-1 and AIA are required in adults near the time of memory formation and retrieval (Cho, et al., 2016;Lin, et al., 2010;Tomioka, et al., 2006).
In addition to learning, C. elegans insulin/IGF signaling regulates development, metabolism, fertility, and lifespan (Murphy & Hu, 2013). Its function has been best characterized in dauer formation, a stress-induced developmental arrest. A single gene, daf-2, is the ortholog of both insulin receptor and IGF receptors in mammals. Like those receptors, it is predicted to be a receptor tyrosine kinase that is regulated by insulin-like peptides, which can be either agonists or antagonists of daf-2. daf-2 acts through age-1, which encodes a PI3K homolog, and the protein kinase homologs encoded by pdk-1, akt-1, and akt-2. The primary target of this signaling pathway is the forkhead FoxO transcription factor encoded by daf-16, which is sequestered in the cytoplasm by daf-2 signaling, preventing access to its transcriptional targets. Surprisingly, the daf-2/daf-16 pathway is only modestly affected by inactivating the sole C. elegans IRS homolog, ist-1 (Wolkow, et al., 2002).
Studies of salt chemotaxis learning in ASER have revealed a second daf-2 signaling pathway that is different from the daf-2/daf-16 transcriptional pathway (Tomioka, et al., 2016;Ohno, et al., 2014). daf-2 encodes at least seven alternative protein isoforms, with DAF-2a being the most widely expressed and best characterized. In ASER, however, aversive learning requires a DAF-2c isoform that includes 82 additional amino acids encoded by an alternative exon (Ohno, et al., 2014). While the canonical DAF-2a protein is limited to the ASER cell body, the DAF-2c isoform required for learning is also present in the ASER axon, where its expression is increased by food deprivation (Ohno, et al., 2014). This localization suggests that axonal DAF-2c signaling may contribute to learning, and indeed, acute local activation of PI3K in the ASER axon can suppress salt attraction.
The insulin/IGF receptor daf-2 is broadly expressed in neurons and non-neuronal tissues (Kimura, et al., 2011;Hunt-Newbury, et al., 2007). Null mutants in daf-2 are developmentally arrested or sterile, with widespread metabolic changes, creating challenges for studying the insulin receptor pathway at cellular resolution (Murphy & Hu, 2013). Here we demonstrate that C. elegans ist-1 mutants lacking Insulin Receptor Substrate (IRS) activity are highly defective in aversive olfactory learning. By characterizing ist-1, which is expressed only in a small number of neurons, along with single-cell knockouts of the endogenous daf-2 locus, we identify an insulin/IGF signaling pathway required for aversive learning and show that it is acutely modulated by food deprivation, ins-1, and ist-1. The insulin/IGF receptor pathway suppresses odor-regulated glutamate release at olfactory synapses, providing a cellular mechanism for aversive memory.

The insulin receptor substrate mutant ist-1 is defective in aversive learning.
During a targeted mutant screen, we made the serendipitous finding that an ist-1 mutation caused profound defects in aversive olfactory learning (see Methods). As learning and memory can engage different molecular mechanisms depending on training and testing protocols, we characterized behavioral responses across a range of conditions designed to distinguish between aversive learning, appetitive learning, and non-associative desensitization. In aversive learning, an Unconditional Stimulus (US), food deprivation, is paired with a Conditional Stimulus (CS), butanone exposure, resulting in reduced chemotaxis to butanone (tested at a 1:1000 dilution) ( Figure 1A-B). In appetitive learning, the US is bacterial food, paired with the same CS of butanone exposure; this training results in enhanced attraction to butanone (tested at a 1:10 dilution) (Arey, et al., 2018;Torayama, et al., 2007). Finally, after exposure to very high butanone concentrations, the AWC ON neuron becomes desensitized, regardless of food context, also resulting in reduced chemotaxis to 1:1000 butanone ( Figure 1A-B) (Cho, et al., 2016;Pereira & Van der Kooy, 2012). By varying the amount of butanone used in training, we found that desensitization required 100-fold higher butanone training concentrations than aversive learning, and >1000-fold higher butanone training concentrations than appetitive learning ( Figure 1A-B).
Using these assays, we found that ist-1 null mutants had normal chemotaxis to butanone at baseline, but were highly defective in aversive learning across a 100-fold range of butanone training conditions ( Figure 1C). They had a partial defect in desensitization and were normal in appetitive learning. We further examined their behavior by using computational video tracking to examine wild-type and mutant chemotaxis in butanone gradients. C. elegans approaches attractants using a biased random walk, in which it makes sustained forward runs when odor increases, but reorients its movement when odor decreases ( Figure S1A) (Levy & Bargmann, 2020;Pierce-Shimomura, et al., 1999). These reorientation behaviors were similar in naïve wildtype and ist-1 animals ( Figure S1B). After training, reorientations were no longer correlated with the direction of the odor source in wild-type animals, but ist-1 mutants retained a substantial odor-oriented reorientation bias consistent with chemotaxis ( Figure S1B-C).
Expression of an ist-1b cDNA under a 14.8 kb genomic fragment encompassing 9.3 kb upstream of the first initiation site, the first exon, and the 5.3 kb intron before the second initiation site resulted in full rescue of learning ( Figure 2C). The same 14.8 kb fragment drove reliable GFP expression in eight pairs of head neurons (AWC, ASE, ASG, ASI, and BAG sensory neurons and RIM, AIM, and RIG interneurons ( Figure 2D) and in some tail neurons ( Figure S2C), overlapping the ist-1 expression pattern obtained from single-cell RNA sequencing data (Taylor, et al., 2021). Notably, ist-1 expression is absent in many neuronal and non-neuronal cells that express the insulin receptor daf-2 (Kimura, et al., 2011;Hunt-Newbury, et al., 2007). To define cells in which ist-1 acts in aversive learning, ist-1b was expressed under shorter regions of the endogenous ist-1 promoter and under cell-specific promoters that overlapped with its expression pattern ( Figure 2C, Figure S2). Expression of ist-1 in the AWC ON neuron, which is the primary butanone-sensing neuron, fully rescued aversive learning in ist-1 null mutants, whereas expression in other neurons or in the intestine did not ( Figure 2C, Figure S2). A GFP-tagged IST-1 protein that rescued the ist-1 aversive learning defect was broadly distributed in the axon, dendrite and cell body of AWC ON and appeared to be excluded from the nucleus ( Figure 2E).
These results suggest that ist-1 functions autonomously in the AWC ON sensory neuron to drive aversive learning.

ist-1 interacts with insulin pathway genes in aversive learning
Aversive learning to butanone requires the insulin-related gene ins-1, which is expressed in the AIA interneurons, other neurons, and intestine, and the daf-2-regulated PI3 kinase age-1, which is required in AWC ON neurons (Takeishi, et al., 2020;Cho, et al., 2016) (Figure 3A-B). We found that several additional genes in the insulin pathway had either butanone chemotaxis defects or aversive learning defects: loss of function mutants in the akt-1 kinase had a deficit in aversive learning comparable to that of ins-1, ist-1 and age-1 mutants, and daf-18 PTEN phosphatase mutants failed to chemotax to butanone before or after conditioning ( Figure 3A-B).
To further determine the relationship between ist-1 and other genes in the pathway, we examined double mutants. Double mutants between ist-1 and the PTEN phosphatase daf-18 had mixed phenotypes: the daf-18 butanone chemotaxis defect was partially restored in daf-18; ist-1 double mutants, and aversive learning was restored as well ( Figure 3C). These results are consistent with a model based on the mammalian insulin pathway, in which the insulin receptor DAF-2 activates the PI3 kinase AGE-1, the adaptor protein IST-1 potentiates the DAF-2/AGE-1 interaction, the DAF-18 phosphatase antagonizes AGE-1, and simultaneous loss of both IST-1 and DAF-18 permits residual AGE-1 activation by DAF-2 ( Figure 3A). In agreement with this model focused on AGE-1 activation, age-1; daf-18 double mutants resembled age-1, with robust chemotaxis and no aversive learning ( Figure 3C). However, the daf-18 mutation used in this experiment is not a null allele, which limits the interpretation of these results.
daf-16 FoxO mutants do not chemotax to butanone (Daniels, et al., 2000). In addition, their behavior was not modified by aversive conditioning ( Figure 3C). daf-16; ist-1 double mutants had a mixed phenotype: they had intermediate levels of chemotaxis to butanone, with diminished levels of aversive learning ( Figure 3C). Since both ist-1 and daf-16 alleles were null alleles, this epistasis analysis indicates that each of these genes has some functions that are independent of the other. The daf-16 null phenotype was also partly suppressed in daf-16; age-1 double mutants ( Figure 3C).
The role of ist-1 in dauer development, the target of the canonical C. elegans insulin/IGF pathway, has been studied only through RNA interference (Wolkow, et al., 2002). Using the ist-1 deletion mutant ok2706, we confirmed that ist-1 mutants did not form dauers under well-fed conditions (data not shown). They were also similar to wild-type in mildly dauer-inducing conditions with limited food at high temperatures ( Figure 3D), and were only slightly more likely than wild-type to form dauers when exposed to synthetic pheromones that resulted in substantial dauer formation ( Figure 3D).
In summary, ist-1 (insulin receptor substrate), ins-1 (an insulin), age-1 (PI3K), and akt-1 (AKT) are all required for aversive butanone learning, and daf-18 (PTEN) antagonizes ist-1 in a way that is consistent with ist-1 and daf-18 action upstream of age-1. However, there are distinctions between the aversive learning pathway and the canonical C. elegans insulin/IGF pathway (Murphy & Hu, 2013). ins-1 acts as an inhibitor of insulin signaling in dauer formation, but apparently as an activator (with a similar phenotype to age-1) in learning. ist-1 and akt-1 are essential for aversive learning, but not for dauer formation, where akt-1 is redundant with akt-2.
daf-16 is the final output of the daf-2 pathway in dauer formation, and is epistatic to all other genes, but ist-1 function in aversive learning is partially independent of daf-16.

The DAF-2c insulin receptor isoform supports aversive learning in AWC ON
One concern in studying insulin pathway mutants, particularly the essential gene daf-2, is that systemic physiological abnormalities in mutants might indirectly affect learning. To begin to address this point, we examined two daf-2 mutants that do not have overt developmental defects, but have taste avoidance defects when salt is paired with food deprivation (Nagashima, et al., 2019;Ohno, et al., 2014) ( Figure 4A). The daf-2(pe1230) missense mutation affecting the extracellular domain was highly defective for butanone aversive learning ( Figure 4B). A partial aversive learning defect was caused by the DAF-2c-specific allele daf-2c(pe2722) ( Figure 4B), which selectively disrupts isoform-specific exon 11.5 that adds 82 amino acids to DAF-2c (Nagashima, et al., 2019;Ohno, et al., 2014).
Next, we tested full-length daf-2a and daf-2c cDNAs expressed under the AWC ON -selective str-2 promoter for rescue of daf-2 mutant phenotypes. daf-2c fully rescued the aversive learning defects of both daf-2(pe1230) and daf-2c(pe2722) mutants. daf-2a did not significantly rescue aversive learning in daf-2c(pe2722) or daf-2(pe1230) mutants, although there was a trend toward rescue in daf-2(pe1230) ( Figure 4C). The mutant and cell-specific rescue results suggest that the daf-2c isoform has a primary role in aversive learning in AWC ON , although other isoforms such as daf-2a may also contribute.
To further refine the site of daf-2 action, we generated a conditional allele of the endogenous daf-2 locus susceptible to flippase-FRT inactivation. Two FRT sites flanking the second half of the daf-2 gene were inserted by CRISPR-Cas9 recombination, and flippase was expressed under cell-selective promoters to inactivate daf-2 only in selected cells ( Figure 4A). Expression of flippase under an AWC ON -selective promoter resulted in a partial aversive learning defect after butanone conditioning ( Figure 4D). As a flippase activity reporter indicated that this transgene was also active in ASI and RIA neurons (Muñoz-Jiménez, et al., 2017) (Figure 4D), we used other promoters to confirm AWC ON as a relevant site of daf-2 action. An ASI-specific daf-2 knockout did not affect aversive learning, while a sensory-specific daf-2 knockout that affected AWC ON but not RIA disrupted aversive learning. Together, these results support a role for daf-2 in AWC ON in aversive learning. The incomplete effect of daf-2 inactivation in AWC ON in may be due to daf-2 activity in other cells, incomplete flippase recombination, or pre-existing daf-2 mRNA or protein that persists in AWC ON after recombination.

IST-1 has a minor role in the DAF-2c pathway for taste avoidance learning
The involvement of the daf-2c isoform in aversive olfactory learning suggested an analogy with taste avoidance learning, the associative learning behavior in which daf-2c was first characterized (Ohno, et al., 2014). Taste avoidance learning has similar properties to olfactory avoidance learning: when high salt is paired with food deprivation, animals lose their attraction to high salt ( Figure 5A). ist-1 mutants had a partial decrease in taste avoidance learning ( Figure   5B), while retaining normal preferences when high or low salt was paired with food ( Figure S3). This defect was rescued by expressing ist-1 in ASER, the sensory neuron required for taste avoidance learning ( Figure 5B). The ist-1 mutant defect in taste avoidance learning was milder than the defect in daf-2c(pe2722) mutants. daf-2c; ist-1 double mutants had strong defects resembling those of daf-2c ( Figure 5B). daf-16 mutants are also defective in taste avoidance learning, resembling daf-2c(pe2722) ( Figure   5B). Double mutants between the daf-16 and ist-1 mutants had a stronger defect than either single mutant ( Figure 5B). A similar enhanced defect was observed in daf-16; daf-2c double mutants (Nagashima, et al., 2019) and in daf-16; daf-2c; ist-1 triple mutants ( Figure 5B). These results are consistent with overlapping functions for daf-2c and ist-1, and distinct functions for daf-16, in taste avoidance learning.
In summary, ist-1, daf-2c, and daf-16 all affect both aversive olfactory learning and taste avoidance learning. For salt chemotaxis learning, the daf-2c(pe2722) mutant defect was more severe than that of ist-1, whereas for aversive olfactory learning the ist-1 defect was more severe than that of daf-2c(pe2722). For both olfactory learning and taste learning, daf-16; ist-1 double mutants were different from either single mutant, indicating that daf-16 and ist-1 function at least partly in parallel ( Figure 3B, 5B).

Food deprivation and insulin signaling regulate DAF-2 levels in AWC ON
In the ASER neurons that mediate taste avoidance learning, mVenus-tagged DAF-2a protein is restricted to the cell body, while mVenus-tagged DAF-2c protein is present in the cell body, axons, and synaptic regions (Ohno, et al., 2014). We observed similar subcellular localization patterns of mVenus-tagged DAF-2a and DAF-2c in AWC ON . DAF-2a was present in the AWC ON dendrite and cell body, with a somatic pattern suggesting nuclear exclusion, but was undetectable in the AWC ON axon ( Figure 6A-C), whereas DAF-2c was present in the dendrite, cell body and axon of AWC ON ( Figure 6D-F).
The levels of tagged DAF-2a and DAF-2c protein in AWC ON increased after 90 minutes of food deprivation, the unconditioned stimulus used for aversive conditioning, but their subcellular localization patterns were unchanged ( Figure 6A-F). In control experiments, expression of GFP under the same AWC ON -selective str-2 promoter was not affected by food deprivation, suggesting that DAF-2 translation, trafficking, or stability is altered post-transcriptionally ( Figure   6G).
In mammals, insulin signaling results in internalization of insulin receptors, which may subsequently return to the cell surface or be degraded (Chen, et al., 2019). With that in mind, we asked whether insulin signaling itself might contribute to the observed changes in DAF-2 protein expression. Indeed, food deprivation did not affect AWC ON DAF-2 levels in ist-1 or ins-1 mutants ( Figure 6B-C, Figure 6E-F). Since ist-1 has a minor role in salt learning, we asked whether ist-1 also plays a role in the localization of DAF-2c in the salt-sensing ASER neuron.
The levels of mVenus-tagged DAF-2c in the ASER axon increased after 90 minutes of food deprivation in wild-type animals, but not in ist-1 mutants ( Figure 6H). These results suggest that food deprivation upregulates DAF-2 protein levels in sensory neurons via ins-1 and ist-1.

Aversive learning suppresses odor-dependent glutamate release from AWC ON
To ask how ist-1-dependent insulin signaling affects AWC ON responses to odor in naïve and trained animals, we employed the genetically-encoded calcium indicator GCaMP5. AWC ON calcium responses to butanone pulses were normal in ist-1 mutants across a 100,000-fold range of butanone concentrations, consistent with their normal chemotaxis ( Figure 7A-B). Sustained exposure to high butanone, with or without food, causes a ten-fold reduction in AWC ON butanone sensitivity across its dynamic range that persists for at least 30 minutes (Cho, et al., 2016). This nonassociative AWC ON desensitization was similar in wild-type and ist-1 animals ( Figure 7B).
We conclude that the aversive learning defect in ist-1 mutants arises downstream of the AWC ON calcium signal, as previously proposed for ins-1 and age-1 insulin/IGF pathway genes (Cho, et al., 2016).
The AWC ON neuron releases the classical neurotransmitter glutamate and a number of neuropeptides (Chalasani, et al., 2010;Chalasani, et al., 2007). An AWC-selective knockout of eat-4, the vesicular glutamate transporter, was generated using FLP::FRT recombination targeted to the endogenous eat-4 locus (López-Cruz, et al., 2019). AWC knockout of eat-4 resulted in a butanone chemotaxis defect comparable to that of the eat-4 null mutant ( Figure 7C). These animals had a slight reduction in chemotaxis after pairing butanone and food deprivation, as did eat-4 null mutants. Thus, AWC glutamate release is important for butanone chemotaxis, but likely functions together with additional AWC neurotransmitters.
To investigate the synaptic properties of AWC ON , we used a reporter strain in which the  transporter was fused to pHluorin, a pH-sensitive version of GFP whose fluorescence increases when acidic glutamatergic synaptic vesicles fuse with the plasma membrane (Ventimiglia & Bargmann, 2017). Naïve wild-type animals and ist-1(ok2706) mutants had similar AWC ON responses to butanone addition and removal using this synaptic reporter ( Figure 7D, E). After pairing butanone with food deprivation, EAT-4::pHluorin responses to odor pulses were reduced in wild-type AWC ON neurons for at least 30 minutes, consistent with a dampened synaptic response to odor ( Figure 7D, E). Importantly, this synaptic effect was observed at butanone concentrations that elicited strong GCaMP responses in both naïve and trained animals ( Figure   7B). By contrast, AWC ON neurons in ist-1 mutants retained substantial odor-regulated glutamate release after training, providing a synaptic correlate of their defect in aversive learning ( Figure   7D, E). These results suggest that aversive learning reduces odor-regulated glutamate release from AWC ON via ist-1-dependent insulin/IGF signaling.
To further explore the relationship between synaptic transmission and aversive learning, we examined PKC-1, a presynaptic protein kinase C homolog that stimulates glutamate release from AWC ON neurons ( Figure 3A) (Ventimiglia & Bargmann, 2017) and neuropeptide secretion (Sieburth, et al., 2007). PKC-1 is activated by Gq-PLC signaling; in AWC ON , pkc-1 and Gq are key effectors of appetitive learning after pairing butanone with food (Arey, et al., 2018). pkc-1 mutants were repelled by butanone, and only slightly affected by pairing butanone with food deprivation ( Figure 7F). Their chemotaxis defect was rescued by expressing pkc-1 in AWC ON , and interestingly, the rescued strain was defective in aversive learning ( Figure 7F). The learning defect might be due to pkc-1 overexpression in AWC ON or the absence of pkc-1 in another cell type, so we next examined a strain in which a pkc-1(gf) cDNA was specifically expressed in AWC ON in wild-type animals. The AWC ON ::pkc-1(gf) strain was highly defective in aversive learning, consistent with a role for AWC ON presynaptic regulation in learning and memory ( Figure 7F).

DISCUSSION
These results provide three new insights into insulin/IGF signaling in olfactory learning. First, aversive learning in AWC ON engages a cell type-specific insulin-daf-2 signaling pathway that includes the adaptor protein IST-1. Second, insulin/IGF signaling mobilizes the DAF-2c insulin receptor protein during food deprivation, suggesting an acute function that sets the stage for learning. Third, aversive training inhibits odor-regulated glutamate release from AWC ON via the insulin/IGF signaling pathway. Through cell-specific inactivation of the endogenous daf-2 locus, and through the analysis of ist-1 mutants that have restricted defects in insulin/IGF signaling, we were able to localize ist-1 and daf-2 functions in aversive learning to AWC ON sensory neurons.
These genetic tools allowed us to avoid systemic developmental and metabolic effects of this critical endocrine pathway.

Aversive learning uses a cell type-specific insulin/IGF signaling pathway
The insulin-related molecule encoded by ins-1 and the insulin receptor encoded by daf-2 are both required for aversive learning. By contrast, ins-1 and daf-2 have opposite effects on dauer formation (Kodama, et al., 2006). C. elegans has 40 genes that encode insulin-like peptides, and the daf-2 gene encodes at least seven different protein isoforms with different functions (Ohno, et al., 2014;Martinez, et al., 2020). Structural studies of mammalian insulin and IGF-1 receptors predict that the DAF-2c-specific exon is adjacent to the primary ligand-receptor interface, and therefore could alter interactions between DAF-2 and insulin ligands (Gutmann, et al., 2019;Li, et al., 2019;Scapin, et al., 2018). We speculate that INS-1 activates DAF-2c, the isoform implicated in aversive learning, but inhibits the DAF-2a isoform used in dauer development.
In alternative models, INS-1 might not act directly on AWC ON , but indirectly via other cells.
Insulin-to-insulin relays have been observed in several settings in C. elegans: for example, in pathogen-related learning, INS-7 from the URX sensory neuron antagonizes the insulin/IGF receptor DAF-2 in RIA interneurons, to prevent learning, while INS-6 secreted from the ASI sensory neurons suppresses INS-7 production from URX neurons, thereby disinhibiting RIA and allowing learning (Chen, et al., 2013). Several other neuronal insulins participate in long-term activity-dependent regulation that results in compensation across circuits (Yeon, et al., 2021).
Results like these underscore the value of the cell type-specific daf-2 knockout used here, which demonstrates a direct requirement for daf-2 in AWC ON , although it leaves open the possibility that daf-2 acts in other cell types as well.
Insulin and insulin-like signaling are usually associated with nutrients and growth in both vertebrates and invertebrates. However, the aversive learning paradigms for odor and salt in C. elegans require ins-1, daf-2c, and insulin signaling in the opposite context, food deprivation. ins-1 signaling from the intestine drives food deprivation-induced thermosensory plasticity, consistent with a role for ins-1 in encoding the low-food context (Takeishi, et al., 2020).
Moreover, ins-1 expression is upregulated by starvation both in the intestine and in the AIA neurons that regulate AWC plasticity (Takeishi, et al., 2020). We observed that DAF-2 levels in AWC ON increase upon food deprivation through a post-transcriptional mechanism that requires ist-1 and ins-1. These results suggest that acute insulin signaling during food deprivation results in altered translation, stabilization, or trafficking of DAF-2 receptors, and sets the stage for learning. There is precedent for such post-transcriptional regulation in mammals, where insulin/IGF receptors are internalized and may be degraded after ligand binding (Chen, et al., 2019), but IRS (IST-1) can promote retention of IGF-IR complexes on the plasma membrane instead of degradation by blocking an AP2 binding site (Yoneyama, et al., 2018).
IRS proteins are required for insulin/IGF signaling in mammals and in Drosophila, but were not previously thought to be important in C. elegans (Wolkow, et al., 2002). We found that ist-1 has a prominent role in aversive olfactory learning. Expression of ist-1 is highly enriched in AWC, ASE, AFD, and BAG sensory neurons, all of which show robust plasticity upon food deprivation. One possibility is that ist-1 preferentially contributes to neuronal and behavioral daf-2 signaling, and less to endocrine and developmental daf-2 signaling in C. elegans; a second possibility is that IST-1 potentiates INS-1/DAF-2c interactions through inside-out signaling.
Olfactory learning also has a strict requirement for the AKT-related kinase akt-1, whereas in development akt-1 is largely redundant with akt-2. akt-1 expression is enriched in many of the same neurons that express ist-1, including AWC, ASE, and BAG (Taylor, et al., 2021). Whether akt-1 and akt-2 differ only in their expression, or also in function, remains to be determined. Mammalian Akt1, Akt2, and Akt3 have different cellular and subcellular distributions (Santi & Lee, 2010), and distinct roles in synaptic plasticity in the hippocampus (Levenga, et al., 2017).
The analysis of double mutants suggests that insulin/IGF receptor signaling has two targets in AWC ON : one that depends on the daf-16 transcriptional regulatory pathway, and one that does not. Similarly, in salt chemotaxis, insulin signaling occurs through one pathway that requires daf-2a and daf-16, and a second pathway that is daf-2c-dependent and daf-16-independent (Nagashima, et al., 2019). Using temperature-sensitive mutations in daf-2, Lin et al. (2010) showed that insulin/IGF signaling is required both for the formation and for the retrieval of an aversive olfactory memory. It is possible that these temporally distinct functions are associated with molecularly distinct insulin/IGF signaling pathways.

Aversive learning suppresses presynaptic glutamate release from AWC ON
Odor responses in AWC ON are regulated by olfactory experience and/or insulin/IGF signaling across multiple timescales. Sustained butanone exposure with food results in non-associative desensitization of AWC ON calcium responses to butanone, but this is insufficient for aversive learning (Cho, et al., 2016). Insulin/IGF signaling transiently suppresses AWC ON calcium responses through a rapid feedback loop, but this functional plasticity is short-lived compared to aversive learning (Cho, et al., 2016;Chalasani, et al., 2010). Here, we show that sustained pairing of butanone odor with food deprivation results in suppression of odor-regulated glutamate release from AWC ON . This presynaptic depression requires association of food deprivation with odor, as expected for aversive learning, and is retained for at least 30 minutes, consistent with the hour-long chemotaxis assay. As glutamate release from AWC is required for efficient chemotaxis and the odor-regulated turning behavior used in chemotaxis, this functional change is likely to contribute to behavioral changes after learning (Tsunozaki, et al., 2008;Chalasani, et al., 2007). A similar mechanism has been proposed to play a role in aversive taste learning, where calcium responses to salt in ASER sensory neurons are preserved but synaptic pHluorin signals are decreased after training (Oda, et al., 2011). The detailed mechanism by which insulin signaling suppresses glutamate release remains to be determined, and might involve AGE-1 mobilization of lipids that regulate synaptic proteins such as protein kinase C (PKC-1) (Ohno et al., 2014), AKT-1-dependent protein phosphorylation, or regulation of gene expression.
How is food deprivation associated with odor cues? Olfactory signal transduction in AWC ON is mediated by cGMP, and one site of odor-food convergence is the cGMP-dependent protein kinase EGL-4. When butanone is present for an hour or more, and DAF-2/AGE-1 signaling is active, EGL-4 translocates from the AWC ON cytoplasm to the nucleus (Cho, et al., 2016;Lee, et al., 2010;O'Halloran, et al., 2009;Torayama, et al., 2007;L'Etoile, et al., 2002). Like INS-1, EGL-4 is required for aversive but not appetitive learning (Cho, et al., 2016;Torayama, et al., 2007). The sustained coincidence of EGL-4 and insulin/IGF activity when odor is paired with food deprivation could represent a signal for associative learning. Proteomic studies indicate that EGL-4 is enriched in presynaptic regions (Artan, et al., 2021), where it could interact with DAF-2 signaling components. As mammalian EGL-4 homologs potentiate synaptic release in nociceptive neurons (Luo, et al., 2012), it is possible that depletion of presynaptic EGL-4 could contribute to synaptic depression in AWC ON , along with EGL-4-dependent transcriptional changes in the nucleus.
Olfactory neurons in all animals consist of many classes with distinct odor responses. Recent results have shown that mouse olfactory sensory neurons regulate dozens of activity-sensitive genes within a few hours in a novel olfactory environment, in a cell type-specific fashion, resulting in altered odor responses (Tsukahara, et al., 2021). Food deprivation is a powerful motivating stimulus in all animals (Smith & Grueter, 2021), and insulin receptors are prominently expressed in the rodent olfactory bulb (Kuboki, et al., 2021;Aimé, et al., 2012). It will be interesting to ask whether insulin signaling can modify mammalian olfactory neurons to provide associative input to an odor context. More broadly, insulin resistance in type-2 diabetes is associated with increased risk of Alzheimer's disease, whose early symptoms include defects in olfactory processing (Talbot, et al., 2012). We speculate that exploring the relationship between insulin/IGF and learning in mammals could lead to a deeper understanding of these clinical findings.  (total # of animals in assay) after one hour. White bars, Control: animals were either deprived of food or fed for 90 min, without exposing to odor, then tested for chemotaxis to 1:1000 butanone or 1:10 butanone. Pink bars, Aversive learning: animals were deprived of food while exposed to varying concentrations of butanone odor for 90 min, then tested for chemotaxis to 1:1000 butanone. Grey bars, Desensitization: animals were fed while exposed to varying concentrations of butanone odor for 90 min, then tested for chemotaxis to 1:1000 butanone. Blue bars, Appetitive learning: animals were fed while exposed to varying concentrations of butanone odor for 90 min, then tested for chemotaxis to 1:10 butanone.  Table S2. Statistical tests are described in Table S2.
Statistical tests are described in Table S2.  B) Taste avoidance learning in insulin pathway mutants including ist-1, daf-2c, and daf-16, and rescue of taste avoidance learning by ist-1 expression under the ASER-specific gcy-5 promoter.
Animals were food-deprived with 100 mM NaCl for 5 hours (trained, blue bars) and tested for chemotaxis in a salt gradient as above.
Error bars indicate S.E.M. *, values differ at p ≤ 0.05. **, values differ at p ≤ 0.01. ****, values differ at p ≤ 0.0001. ns, values are not significantly different. Statistical tests are described in Table S2. Additional salt training and chemotaxis results are in Figure S3.  F) Chemotaxis and aversive learning in pkc-1 mutants (see Fig 3A), pkc-1 rescued in AWC ON , and wild-type animals expression pkc-1(gf) in AWC ON . Animals were trained and tested as above.
Statistical tests are described in Table S2.

CONTACT FOR REAGENT AND RESOURCES SHARING
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dr. Cornelia Bargmann (cori@rockefeller.edu).

C. elegans strains
All new CRISPR-generated strains and transgenic strains are listed in Table 1.

Nematode maintenance and molecular biology
All strains were maintained at 20°C on Nematode Growth Medium (NGM) plates, seeded with E. coli OP50 bacteria as a food source (Brenner, 1974). Wild-type animals were the genetically defined strain PD1074 derived from the Bristol strain N2 (Yoshimura, et al., 2019).

Chemotaxis assays
Animals were grown at 20°C on 10 cm round NGM plates spotted with 900 µl of E. coli OP50.
Chemotaxis plates are square plates containing 12 ml of modified NGM agar (without peptone and cholesterol, 1.7% Granulated, purified agar (Agar-agar, Millipore), 50 mM NaCl, 25 mM KHPO4 (pH6), 1 mM CaCl2, 1 mM MgSO4). The chemotaxis plates are allowed to air-dry without lids for 15 minutes before use, or are stored at 4°C for up to 2 weeks before use. Fifteen minutes before the start of the assay, one 1 µl spot of NaN 3 was spotted at the midline of each

Training for aversive learning
The training protocol for aversive learning is based on the assay developed by (Colbert & Bargmann, 1995) with modifications. One-day old adults were washed three times with NGM buffer and placed on conditioning plates consisting of NGM agar without peptone and cholesterol (same as chemotaxis plates). For the trained group, 20 µl of butanone were distributed on four small pieces of agar on the inside of the lid of the plate, while the control group did not have butanone on the lid. The plates were then sealed with parafilm for 90 minutes before animals were washed 2-3 times with NGM buffer and tested in chemotaxis assays according to the previous section.

Training for desensitization and appetitive learning
The training protocol for desensitization and appetitive learning (or enhancement) is identical; the two assays are distinguished by test conditions (Torayama, et al., 2007). One-day old adults were washed three times with NGM buffer and placed on conditioning plates consisting of NGM agar seeded with 900 ul of OP50. For the trained group, 20 µl of butanone were distributed on four small pieces of agar on the inside of the lid of the plate, while the control did not have butanone on the lid. The plates were then sealed with parafilm and incubated for 90 minutes before animals were washed and tested in chemotaxis assays according to the section "Chemotaxis assays" with 1:1000 butanone for desensitization, and 1:10 butanone for appetitive learning.

Salt concentration learning assay
Salt concentration learning assays were performed according to our previously reported procedure (Tomioka, et al., 2016). For conditioning on agar plates, adult worms were transferred to NGM plates with 25-or 100-mM NaCl in the absence or presence of a bacterial food (NA22) for 4-5 h. After conditioning, the worms were placed at the center of a test plate with an NaCl gradient and allowed to crawl for 45 min. A chemotaxis index was determined according to the following equation: NaCl Preference Index = (NA-NB) / (Nall-Nc), where NA and NB are the number of worms in the high-and low-salt areas, respectively, Nall is the total number of worms on a test plate, and Nc is the number of worms in the area around the starting position.

Chemotaxis recordings and analyses
40-60 animals were placed on a square plate with modified NGM agar. A 1-hour movie was recorded at 3 frames per second with Streampix software and a 6.6 MP PL-B781F CMOS camera (Pixelink). Animals' trajectories were extracted by a custom MatLab script. A longreversal-omega turn (pirouette) event was defined as a reversal coupled with an omega turn, and identified as a sharp change in angular speed (≥75 degree/s), followed by a sharp reorientation (body-enclosing ellipse eccentricity ≤0.875, filtered by an angular speed threshold ≥60 degree/s), as previously described (Pokala, et al., 2014). Long-reversal-omega frequency was calculated by binning the events with respect to the angle between each worm's heading and the direction of the odor source before the reorientation, at 30° intervals. The normalized long-reversal-omega frequency was calculated by binning the event/worm/minute with respect to angle between each worm's heading and the direction of the odor source before the reorientation, at 90° intervals.

Dauer formation assays
Dauer formation assays were performed using synthetic dauer-inducing pheromones with modifications to previously described methods (Neal, et al., 2013;Butcher et al., 2008;). Assay plates were prepared 24 hours before each experiment by mixing 1 ml NGM-agar with 100 µl synthetic pheromone mixture (ascarosides C3 and C6, 73 nM final concentration each) or ethanol control. On the day of each experiment, 20 μl heat-killed OP50 (8 mg/ml in NGM buffer) was added and dried onto each plate. 14 adult animals were picked onto each assay plate and allowed to lay eggs for 3-4 hours before being removed. Plates were sealed with parafilm and incubated at 25 o C with wet paper towels to prevent desiccation. After 72 hours of incubation, the proportion of dauers (including partial dauers) in each assay plate was counted based on their distinguishing characteristics: radial constriction, posture, gut refraction, locomotion, and cessation of pumping (Neal, et al., 2013).

Calcium imaging in the arena chip
Multiple animals expressing GCaMP5 in AWC ON were imaged simultaneously as described (Larsch, et al., 2013;Cho, et al., 2016) with slight modifications. PDMS chips with two arenas were made using a custom mold. The chips were flooded with NGM buffer containing 10 mM tetramisole hydrochloride, and 10-15 adult animals were loaded into each arena. Animals were trained in the chip for 90 minutes with 10 mM butanone (or NGM buffer for control groups) followed by a 15-minute wash with NGM buffer, then imaged on a Zeiss AxioObserver A1 microscope with a 2.5X objective and Metamorph software for synchronized image capture with pulsed illumination. A three-way valve was used to switch between buffer and odor flow into the chip, and a Hamilton valve was used to switch between different concentrations of odor. The stimulation protocol was three 30-second alternations between odor and buffer followed by 90 seconds of buffer, and then the sequence was repeated at 10-fold higher odor concentration for a total of six butanone concentrations, ranging from 10 -8 M to 10 -3 M, plus a buffer-to-buffer control. Fluorescence was measured using a custom ImageJ script (Larsch, et al., 2013). Calcium imaging data analysis was identical to that used in (Cho, et al., 2016) using custom Matlab scripts to analyze calcium imaging data generated by ImageJ tracking programs.

pHluorin imaging in the single-animal trapped chip
The pHluorin imaging method is similar to that developed by (Ventimiglia & Bargmann, 2017). imaging chips were fabricated as described in    (Ohno, et al., 2014), and more similar to that described in (Nagashima, et al., 2019). This decision was based on the suggestion from the authors of the articles. A rectangular region of interest (ROI) was drawn manually inside the area of the cell body or a piece of axon before using ImageJ to calculate the average intensity of fluorescence in the ROI. Average fluorescence intensity from at least five ROIs was used to quantify axon fluorescence of a given image stack.

1-day old adults expressing an AWC
Background fluorescence intensity was averaged from at least five ROIs adjacent to the cell body or axon. The normalized fluorescent was calculated by subtracting the average fluorescence intensity of the background from the average fluorescence intensity of the cell body or axon.
For ASER imaging, fluorescence ratio was calculated based on the relative signal intensities of DAF-2c::mVenus and mCherry co-expressed in ASER.
Whole genome sequencing of these samples was performed by the Rockefeller Genomic Resource Center using an Illumina HiSeq 2500 System. Sequence reads were analyzed by Patrick McGrath (Georgia Tech) by comparing the sequence of the backcrossed strains to the wild-type strains using a custom program. A T->G mutation was found in the splicing motif for ist-1 after the end of the 3 rd exon and verified using Sanger sequencing. The ist-1(ky1071) point mutation, which is predicted to disrupt a splice donor, is: atgtttatttagaaattatgtgcacaattcaattttttgagcagttgtcgatcaaaaagaaaat tctttcagatttatttttaattatgcataaaatcttcatattcc (site of point mutation underlined in red, mutation in square bracket, 3 rd exon in uppercase, introns in lowercase)

CRISPR/Cas9-generated mutant strains
New mutations in ist-1 were generated using a co-CRISPR protocol (Arribere, et al., 2014), with a template-guided dominant mutation on the rol-6 gene to induce a roller phenotype to aid selection of the gene-edited progeny. Young hermaphrodites with only a few eggs in the gonad were injected with a mix of plasmids encoding Cas9, a gRNA targeting rol-6, a gRNA targeting the gene of interest (such as ist-1), and a ssDNA repair template that induces the rol-6(su1006) mutation. F1 animals carrying roller phenotypes were isolated onto individual plates, allowed to lay eggs, and then screened for a target mutation by Sanger sequencing. non-Roller F2 animals with frameshift mutations in ist-1 were chosen to generate homozygous strains. The homozygous strains were then backcrossed twice to wild-type strains before being used to test behavior.
The sequence of the ist-1(ky1075) mutation (resulting in a frameshift in the third shared exon, affecting all isoforms) is: GCTAATGTTTGTGACACTCACGGAACGATGTTTAGAACTTCACGAGTCGGAGAAATC CTATCGTGCG (Deletion underlined).
The sequence of the ist-1(ky1076) mutation (resulting in a frameshift in exon 2 affecting all isoforms) is: GCGCCAGGAAGAAACAGGAAGAAACTGTAAAGAAATgtaagtgtcgttctgttatggttgttaaatt (Deletion underlined).

Generating a conditional daf-2 allele
The flippase-FRT-modified daf-2 strain, including an FRT site sequence, the let-858 3'UTR, and the mCherry coding sequence before the 3' UTR of the daf-2 gene, was made using a protocol developed by the Mello lab (Dokshin, et al., 2018). Briefly, S. pyogenes Cas9 protein and synthesized tracrRNA and crRNA were used to generate the cut required for recombination. The Although the mechanism is unknown, the Mello lab found these heating-cooling steps are critical to achieve high knock-in efficiencies. The repair template, tracrRNA, crRNA and rol-6(su1006) plasmid co-injection marker were then injected into young hermaphrodite gonads, and mutants were selected as described above. (See section "CRISPR/Cas9-generated mutant strains"). The upstream FRT site was inserted in the middle of the daf-2 gene such that excision would delete exons 10 through 17 which encode all intracellular domains including the kinase domain. We note that the daf-2b isoform, which encodes a secreted protein that antagonizes insulin signaling, might be expressed in a shortened form (exons 1-9, instead of its normal sequence from exons 1-11) (Martinez, et al., 2020). For the second site, we inserted a let-858 3'UTR-FRT-mCherry coding sequence between the stop codon and the 3' UTR of daf-2, with the intention that mCherry should be expressed after successful knockout of daf-2 in cells of interest. Due to oversight in design, the mCherry reporter could not be visualized after FRT recombination.
Instead, the BN294 bqSi294 strain, which carries an integrated transgene that expresses mCherry in every cell, with a masked GFP, was used to report cell-specific flippase expression (Muñoz-excised while unmasking GFP expression. The sequence surrounding the 5' insertion of the FRT site in daf-2 (CX1095) conditional knockout strain is:

DATA AND SOFTWARE AVALIABILITY
All datasets and scripts generated during the current study will be made available on Mendeley Data and Github, respectively. Figure S1. ist-1 regulates biased random walk chemotaxis during aversive learning A) C. elegans utilizes a biased-random walk strategy to navigate toward an odor source, in which turning increases as odor decreases. For panels (B) and (C), the angle between the animals' heading direction and the direction of the odor source was determined shortly before each reversal-turn behavior during chemotaxis assays.

SUPPLEMENTARY FIGURE LEGENDS
B) Event count of reversal-turns for 50 wild-type or 50 ist-1 mutant animals during one chemotaxis session. The angle between animals' heading direction and odor source is binned by an increment of 30°. Animals were food-deprived (control, white bars), or food-deprived with butanone for 90 mins (trained, pink bars).
C) Average frequency of reversal-turns for wild-type (black) or ist-1 mutant (red) animals when they are traveling towards the odor (white bars with black and red borders), or traveling away from the odor (solid black and red bars).
In (B) and (C), wild-type and ist-1 control animals initiated more reversal-turn reorientations when traveling away from the odors, an act of course-correction for chemotaxis. Wild-type animals that were trained with butanone for 90 mins initiated similar frequencies of reversalturns regardless of their heading direction, a finding that correlates with decreased chemotaxis after training. ist-1 animals that were trained with butanone for 90 mins continued to initiate more reversal-turn reorientations when traveling away from the odors (*, values differ at p ≤ 0.05). Statistical tests are described in Table S2.

Figure S2. ist-1 promoters and isoforms
A) Genomic structure of the C. elegans insulin receptor substrate-1 (ist-1) gene and flanking regions. White blocks indicate noncoding regions; black boxes indicate ist-1 exons.
B) The 9.3 kb genomic region between the first exon and the next gene upstream from ist-1 drives GFP in the intestine and pharynx, and does not rescue ist-1 aversive learning when driving an ist-1b cDNA.
C) The 5.3kb genomic region between the first exon and second exon of ist-1 gene drives expression in the AWC, ASE, AUA and BAG neurons, and rescues ist-1 aversive learning when driving an ist-1b cDNA.