A single set of interneurons drives opposite behaviors in C. elegans

Summary

Many chemosensory stimuli evoke innate behavioral responses that can be either appetitive or aversive depending on an animal’s age, prior experience, nutritional status, and environment [1–9]. However, the circuit mechanisms that enable these valence changes are poorly understood. Here, we show that Caenorhabditis elegans can alternate between attractive or aversive responses to carbon dioxide (CO₂) depending on its recently experienced CO₂ environment. Both responses are mediated by a single pathway of interneurons. The CO₂-evoked activity of these interneurons is subject to extreme experience-dependent modulation, enabling them to drive opposite behavioral responses to CO₂. Other interneurons in the circuit regulate behavioral sensitivity to CO₂ independent of valence. A combinatorial code of neuropeptides acts on the circuit to regulate both valence and sensitivity. Chemosensory valence-encoding interneurons exist across phyla, and valence is typically determined by whether appetitive or aversive interneuron populations are activated. Our results reveal an alternative mechanism of valence determination in which the same interneurons contribute to both attractive and aversive responses through modulation of sensory neuron to interneuron synapses. This circuit design represents a previously unrecognized mechanism for generating rapid changes in innate chemosensory valence.

Results and Discussion

Here we investigated the molecular, cellular, and circuit mechanisms that determine CO₂ response in the free-living roundworm C. elegans. CO₂ is a byproduct of cellular respiration that can signal the presence of food, mates, predators, or pathogens [10–13]. We found that CO₂ can be attractive or repulsive for C. elegans adults depending on their recently experienced environmental CO₂ levels. We raised animals at either ambient or high (2.5%) CO₂ for one generation, and tested their response to CO₂ in a chemotaxis assay (Figure S1A). Although the level of atmospheric CO₂ is only 0.038% [10], wild C. elegans inhabit environments rich in rotting organic matter, where CO₂ levels can rise above 10% [14]. As previously reported, animals raised at ambient CO₂ avoided CO₂ (Figure 1A) [15, 16]. In contrast, animals raised at high CO₂ showed robust CO₂ attraction (Figure 1A). In both cases, response valence was consistent across concentrations (Figure S1B–C). Thus, C. elegans can show attraction or repulsion to CO₂ depending on its prior cultivation conditions.

Figure 1. C. elegans shows both attractive and aversive responses to CO₂.

(A) Animals raised at ambient CO₂ (0.038%) avoid 5% CO₂, while animals raised at high (2.5%) CO₂ are attracted to 5% CO₂. ***p<0.001, Mann-Whitney test. n=12–14 trials per condition.

(B) Adults raised at ambient CO₂ were incubated at high (2.5%) CO₂ for 1, 3, 6, or 9 h and then assayed for their response to 2.5% CO₂ (blue), while adults raised at high CO₂ were put at ambient CO₂ for 3, 6, or 9 h and then assayed for their response to 2.5% CO₂ (orange). A switch in CO₂ environment triggers a rapid change in CO₂ response valence. n=8–24 trials per condition.

(C) BAG sensory neurons are required for CO₂ avoidance and attraction. Wild-type animals raised at ambient CO₂ avoid 2.5% CO₂, while wild-type animals raised at high (2.5%) CO₂ are attracted to 2.5% CO₂. BAG-ablated animals (BAG-) do not respond to CO₂ under either condition. ***p<0.001, two-way ANOVA with Sidak’s post-test. n=8–16 trials per condition.

(D) BAG neurons of animals raised at high (2.5%) CO₂ respond more robustly to CO₂ than BAG neurons of animals raised at ambient CO₂. Graph shows the calcium responses of BAG to 15%CO₂, for animals raised at ambient CO₂ (blue) or high CO₂ (orange), measured using the ratiometric calcium indicator yellow cameleon YC3.60. Solid lines indicate average calcium responses; shading represents SEM. Black line indicates the CO₂ pulse. Animals raised at high CO₂ show an increased BAG response relative to animals raised at ambient CO₂ (***p<0.001, unpaired t test). n=10–15 animals per condition.

(E-F) eat-4 and flp-17 are required for normal CO₂ response. (E) Mutation of eat-4 or flp-17 abolishes CO₂ avoidance in animals raised at ambient CO₂. Responses shown are to 5% CO₂. ***p<0.001, Kruskal-Wallis test with Dunn’s post-test. n=8–14 trials per genotype and condition.

(F) Mutation of either eat-4 or flp-17 reduces CO₂ attraction, and mutation of both genes abolishes CO₂ attraction, in animals raised at high (2.5%) CO₂. Responses shown are to 1% CO₂. ***p<0.001, one-way ANOVA with Dunnett’s post-test. n=8–16 trials per genotype and condition.

For A-C, E and F, graphs depict medians with interquartile ranges. See also Figures S1 and S2.

We then examined the behavior of animals raised at either ambient or 2.5% CO₂ in three different CO₂ gradients: 0–2.5%, 2.5–10%, or 2.5–40%. We found that animals raised at ambient CO₂ avoided the higher concentration of CO₂ under all three conditions (Figure S1D). By contrast, animals raised at 2.5% CO₂ were attracted to the higher CO₂ concentration in both the 0–2.5% and 2.5–10% CO₂ gradients (Figure S1D). Thus, cultivating animals at high CO₂ results in a drive toward higher CO₂ levels rather than a preference for their previous cultivation condition. This is in contrast to other sensory behaviors in C. elegans, including salt chemotaxis and thermotaxis, where animals are attracted to their prior cultivation condition [17, 18]. Animals raised at high CO₂ were not attracted to 40% CO₂ in the 2.5%−40% CO₂ gradient (Figure S1D), suggesting that they retain the ability to avoid toxic levels of CO₂ [19].

To determine if CO₂ preferences are flexible, we transferred animals raised at ambient CO₂ to high CO₂ and vice versa, and assayed their responses to 2.5% CO₂ over the course of 9 hours. We found that animals displayed rapid adaptation to their new environment. Animals raised at ambient CO₂ showed CO₂ attraction after 1 hour at high CO₂ and exhibited maximum attraction by 6 hours (Figure 1B). Animals raised at high CO₂ displayed decreased attraction after 3 hours at ambient CO₂ and recovery of CO₂ avoidance by 9 hours (Figure 1B). Thus, CO₂ response valence is experience-dependent and flexible.

The same pair of sensory neurons is required for CO₂ attraction and repulsion

We next investigated the mechanisms that drive CO₂-evoked behaviors of opposing valence. We previously showed that the BAG sensory neurons detect CO₂ and are required for CO₂ avoidance [12, 15, 20]. We therefore examined the role of BAG in mediating CO₂ attraction. We found that BAG is required for CO₂ attraction as well as repulsion: animals lacking BAG failed to respond to CO₂ regardless of their prior cultivation conditions (Figures 1C, S1E). We tested animals with increased neurotransmission in BAG due to cell-specific expression of a gain-of-function (gf) allele of the protein kinase C gene pkc-1 [21, 22], and found that BAG::pkc-1(gf) animals raised at ambient CO₂ showed enhanced CO₂ avoidance (Figure S1E). Thus, BAG activity modulates the strength of the behavioral response to CO₂.

To test whether recently experienced CO₂ levels affect the response of BAG to CO₂, we compared the CO₂-evoked activity of BAG in animals raised at ambient vs. high CO₂ using calcium imaging. In animals raised at ambient CO₂, exposure to CO₂ resulted in a rapid depolarization, consistent with previous reports (Figure 1D) [12, 20, 23]. In animals raised at high CO₂, exposure to CO₂ resulted in a 2.5-fold increase in the magnitude of the depolarization (Figure 1D). In addition, the CO₂-evoked responses of BAG in animals raised at ambient CO₂ were previously shown to be concentration-dependent [20], and we observed that the CO₂-evoked responses of BAG in animals raised at high CO₂ are also concentration-dependent (Figure S1F). Since CO₂ response valence is consistent across concentrations in animals raised at ambient or high CO₂ (Figure S1B C), yet the CO₂-evoked activity of BAG is concentration-dependent in both cases, CO₂ response valence is encoded downstream of the calcium response of BAG.

The increased BAG activity in animals raised at high CO₂ correlated with increased expression of the receptor guanylate cyclase gene gcy-9, which encodes a putative CO₂ receptor [20, 23–25] (Figure S2A). These results are consistent with the increased behavioral sensitivity to CO₂ exhibited by animals raised at high CO₂: whereas animals raised at ambient CO₂ are repelled by CO₂ concentrations above 5%, animals raised at high CO₂ are attracted to CO₂ concentrations as low as 0.25% (Figure S1B–C). Thus, cultivation at high CO₂ alters both the valence and sensitivity of CO₂-evoked behavior. The increased CO₂ sensitivity following prolonged exposure to high CO₂ is unusual in that prolonged exposure to a chemosensory stimulus generally results in reduced sensitivity as a result of adaptation [26], and in fact prolonged exposure of insects and fish to high CO₂ environments results in reduced sensitivity to CO₂ [27, 28]. C. elegans responds differently to prolonged CO₂ exposure, perhaps because it often inhabits high CO₂ environments. That the sensitivity of C. elegans to CO₂ may be determined by regulating the level of expression of the CO₂ receptor in BAG rather than by regulating interneuron input to BAG may reflect the fact that C. elegans chemosensory neurons respond to multiple chemosensory stimuli due to the compact nature of the C. elegans nervous system [29]. In particular, BAG responds to both CO₂ and O₂ [30], and therefore regulating the sensitivity of BAG to CO₂ by regulating CO₂ receptor expression may be a mechanism that adjusts CO₂ sensitivity while leaving O₂ sensitivity unaltered.

CO₂ avoidance and attraction require neuropeptide and glutamate signaling

We then investigated the signaling mechanisms used by BAG in animals raised at ambient vs. high CO2 to generate attractive or repulsive responses to CO₂. The FMRFamide-like neuropeptide FLP-17 is expressed specifically in BAG [31, 32]. We found that flp-17 mutants raised at ambient CO₂ did not respond to any concentration of CO₂ (Figures 1E, S2B). By contrast, flp-17 mutants raised at high CO₂ were still attracted to CO₂, but to a lesser extent than wild-type animals (Figures 1F, S2C). Thus, FLP-17 is required for CO₂ avoidance but acts in combination with other signaling mechanisms to mediate CO₂ attraction. A candidate for such an additional signaling mechanism is glutamate, since BAG expresses the vesicular glutamate transporter EAT-4 [33]. We found that eat-4 mutants raised at ambient CO₂ failed to respond to CO₂, while eat-4 mutants raised at high CO₂ showed decreased CO₂ attraction (Figures 1E–F, S2B–C). However, eat-4; flp-17 double mutants raised at high CO₂ did not respond to CO₂, suggesting that FLP-17 and glutamate act partially redundantly to mediate CO₂ attraction (Figures 1F, S2C). Moreover, restoring eat-4 expression specifically in BAG partially restored CO₂ avoidance and attraction (Figure S2D–E). Thus, eat-4 acts in BAG to mediate CO₂ response, although glutamatergic signaling from other neurons may also contribute. These results indicate that BAG mediates both CO₂ avoidance and attraction via neuropeptide and glutamate signaling.

A single pathway of interneurons drives CO₂ avoidance and attraction

To gain insight into the CO₂ circuit that operates downstream of BAG, we examined the requirement for the 8 interneurons postsynaptic to BAG [34, 35] that have been previously implicated in chemosensory behaviors [36–42] (Figure S3A). We first tested whether these interneurons are required for CO₂ avoidance in animals raised at ambient CO₂ using strains in which individual interneurons or subsets of interneurons were genetically ablated or silenced with tetanus toxin [43, 44]. When tested with 1% CO₂, a concentration that is neutral to wild-type animals, AIB- AIZ- and AIY- animals showed avoidance and RIA- animals showed attraction (Figure 2A). AIB- animals responded normally to CO₂, suggesting that the phenotype of the AIBAIZ- animals resulted from the loss of AIZ activity (Figure 2A). When tested with 10% CO₂, RIG-animals showed reduced avoidance relative to wild-type animals (Figure 2B). The increased avoidance of AIB- AIZ- and AIY- animals, and the reduced avoidance of RIG- and RIA- animals, were consistent across concentrations (Figure 2C–F). In contrast, AIY::pkc-1(gf) animals showed weaker avoidance, while RIG::pkc-1(gf) and RIA::pkc-1(gf) animals showed enhanced avoidance relative to wild-type animals (Figure 2C–E). Transiently silencing AIY and RIA in adult animals using the histamine-gated chloride channel HisClI [45, 46] had the same effect on CÜ2-evoked behavior as genetic ablation (Figure 2G). Together, these results suggest that CO₂ avoidance is mediated by four pairs of first-order interneurons - AIY, AIZ, RIA, and RIG - whose real-time activity levels determine behavior.

Figure 2. Distinct interneurons act in opposition to regulate CO₂ avoidance in animals raised at ambient CO₂.

(A-B) In animals raised at ambient CO₂, silencing of AIZ and ablation of AIY enhances CO₂ avoidance, ablation of RIG reduces CO₂ avoidance, and ablation of RIA results in CO₂ attraction. Animals were raised at ambient CO₂ and screened for responses to 1% CO₂ (A) and 10% CO₂

(B). Interneurons were either genetically ablated individually (AIB-, AIY-, RIA-, RIB-, and RIG-); genetically ablated simultaneously (AVA- AVE- AVD-); or silenced with tetanus toxin simultaneously (AIB- AIZ-). *p<0.05, **p<0.01, ***p<0.001, one-way ANOVA with Dunnett’s post­test (A) or Kruskal-Wallis test with Dunn’s post-test (B). n=8–26 trials per genotype and condition.

(C) Ablation of AIY enhances CO₂ avoidance. By contrast, animals with more active AIY neurons due to AIY-specific expression of pkc-1(gf) show reduced CO₂ avoidance. Animals were raised at ambient CO₂. ***p<0.001, two-way ANOVA with Dunnett’s post-test. n=8–26 trials per genotype and condition.

(D) Ablation of RIG reduces CO₂ avoidance. By contrast, animals with more active RIG neurons due to RIG-specific expression of pkc-1(gf) show enhanced CO₂ avoidance. Animals were raised at ambient CO₂. *p<0.05, **p<0.01, two-way ANOVA with Dunnett’s post-test. n=8–30 trials per genotype and condition.

(E) Ablation of RIA reduces CO₂ avoidance. By contrast, animals with more active RIA neurons due to RIA-specific expression of pkc-1(gf) show enhanced CO₂ avoidance. Animals were raised at ambient CO₂. *p<0.05, **p<0.01, ***p<0.001, two-way ANOVA with Dunnett’s post-test. n=8–16 trials per genotype and condition.

(F) Silencing of AIZ enhances CO₂ avoidance. Animals were raised at ambient CO₂. ***p<0.001, two-way ANOVA with Sidak’s post-test. n=8–16 trials per genotype and condition.

(G) Animals with AIY neurons transiently silenced by expression of the histamine-gated chloride channel HisCl1 in AIY show enhanced CO₂ avoidance. By contrast, animals with RIA neurons transiently silenced using the same approach show CO₂ attraction. Responses to 1% CO₂ were tested for wild-type animals and animals expressing HisCl1 in either AIY or RIA without histamine (negative control) and with histamine (neuronal silencing); changes in CO₂ response were observed only in the presence of histamine. Animals were raised at ambient CO₂. *p<0.05, ***p<0.001, Kruskal-Wallis test with Dunn’s post-test. n=8–12 trials per genotype and condition.

For A-G, graphs show medians and interquartile ranges. See also Figure S3.

To identify interneurons that regulate CO₂ attraction, we raised interneuron-ablated or silenced strains at high CO₂ and assayed their responses to 0.1% and 0.25% CO₂. Animals raised at high CO₂ were tested with lower concentrations of CO₂ than animals raised at ambient CO₂ due to their increased CO₂ sensitivity; 0.1% and 0.25% CO₂ were chosen because they elicited non­saturating responses from wild-type animals (Figure S1C). We found that AIB- AIZ- and RIA-animals showed stronger attraction, and AIY- animals showed weaker attraction, than wild-type animals (Figure 3A–E). AIB- animals showed normal CO₂ attraction, suggesting the phenotype of AIB- AIZ- animals is due to the loss of AIZ activity (Figure 3A–B). In contrast, AIY::pkc-1(gf) animals raised at high CO₂ showed stronger attraction, while RIA::pkc-1(gf) showed weaker attraction relative to wild-type animals (Figure 3C–D). Transiently silencing AIY and RIA in animals raised at high CO₂ had the same effect on CO₂-evoked behavior as genetic ablation (Figure 3F–G). Moreover, we found that RIG- animals showed a delayed shift from attraction to avoidance following a transition from high to ambient CO₂ (Figure 3H–I). Whereas wild-type animals recovered 77% of their maximum avoidance after 6 hours, RIG- animals recovered only 11%. Conversely, AIY- animals showed a delayed shift, and RIA- animals showed an accelerated shift, from repulsion to attraction following a transition from ambient to high CO₂ (Figure 3J–K). After 3 hours at high CO₂, wild-type animals reached 81% of their maximum attraction, whereas AIY-animals reached only 23% and RIA- animals reached 99%. Taken together, these results suggest that the same set of interneurons regulates CO₂ attraction and repulsion.

Figure 3. The same set of interneurons contributes to CO2 avoidance and attraction.

(A-B) In animals raised at high (2.5%) CO₂, silencing of AIZ and ablation of RIA enhances CO₂ attraction, while ablation of AIY reduces CO₂ attraction. Graphs show responses to 0.1% CO₂ (A) or 0.25% CO₂ (B). *p<0.05, ***p<0.001, one-way ANOVA with Dunnett’s post-test (A) or Kruskal-Wallis test with Dunn’s post-test (B). n=6–20 trials per genotype and condition.

(C) Ablation of AIY reduces CO₂ attraction. By contrast, animals with more active AIY neurons due to AIY-specific expression of pkc-1(gf) show enhanced CO₂ attraction. Animals were raised at high (2.5%) CO₂. **p<0.01, ***p<0.001, two-way ANOVA with Dunnett’s post-test. n=6–20 trials per genotype and condition.

(D) Ablation of RIA enhances CO₂ attraction. By contrast, animals with more active RIA neurons due to RIA-specific expression of pkc-1(gf) show reduced CO₂ attraction. Animals were raised at high (2.5%) CO₂. **p<0.01, ***p<0.001, two-way ANOVA with Dunnett’s post-test. n=8–24 trials per genotype and condition.

(E) Silencing of AIZ enhances CO₂ attraction. Animals were raised at high (2.5%) CO₂. *p<0.05, two-way ANOVA with Sidak’s post-test. n=8–20 trials per genotype and condition.

(F-G) Animals with RIA neurons transiently silenced by expression of the histamine-gated chloride channel HisCl1 in RIA show enhanced CO₂ attraction (F). By contrast, animals with AIY neurons transiently silenced using the same approach show reduced CO2 attraction (G). Responses to 0.1% CO₂ (F) and 0.25% CO₂ (G) were tested for wild-type animals and animals expressing HisCl1 in either RIA or AIY without histamine (negative control) and with histamine (neuronal silencing); changes in CO₂ response were observed only in the presence of histamine. Animals were raised at high (2.5%) CO₂. *p<0.05, **p<0.01, one-way ANOVA with Sidak’s post-test (F) or Kruskal-Wallis test with Dunn’s post-test (G). n=8–14 trials per genotype and condition.

(H-I) Ablation of RIG delays the shift from CO₂ attraction to repulsion in animals raised at high (2.5%) CO₂ and transferred to ambient CO₂. Animals were tested for their response to 2.5% CO₂ after 3, 6, or 9 h at ambient CO₂. Responses were compared to those of animals of the same genotype raised at ambient CO₂ and high CO₂ to determine the percent change in valence (see Methods). Graphs show the percent change in valence after 6 h (H) or as a function of time (I). **p<0.01, ***p<0.001, Kruskal-Wallis test with Dunn’s post-test (H) or two-way ANOVA with Sidak’s post-test (I). n=8–26 trials per genotype, time point, and condition.

(J-K) Ablation of AIY delays the shift, and ablation of RIA accelerates the shift, from CO₂ repulsion to attraction in animals raised at ambient CO₂ and transferred to high (2.5%) CO₂. Animals were tested for their response to 2.5% CO₂ after 1, 3, 6, or 9 h at high CO₂. Responses were compared to those of animals of the same genotype raised at ambient CO₂ and high CO₂ to determine the percent change in valence (see Methods). Graphs show the percent change in valence after 3 h

(J) or as a function of time (K). *p<0.05, ***p<0.001, Kruskal-Wallis test with Dunn’s post-test (J) or two-way ANOVA with Dunnett’s post-test (K). n=8–16 trials per genotype, time point, and condition.

For A-K, graphs show medians and interquartile ranges. See also Figure S3.

AIY expresses the inhibitory glutamate-gated chloride channel subunit GLC-3 [47]. We therefore tested whether GLC-3 plays a role in suppressing AIY activity to promote CO₂ avoidance. We found that glc-3 mutants grown at ambient CO₂ responded normally to CO₂, but glc-3 mutants raised at high CO₂ and transferred to ambient CO₂ showed a delayed shift from attraction to repulsion (Figure S3B–C). This phenotype was rescued by cell-specific expression of glc-3 in AIY (Figure S3C). Thus, inhibition of AIY via glc-3 is required for animals to adapt normally to a shift from high to ambient CO₂.

We next examined the CO2-evoked activity of AIY, RIA and RIG by calcium imaging to determine how CO₂ response valence arises from the activity of these interneurons. We found that AIY showed a CO₂-evoked depolarization in animals raised at high CO₂, and a small but significant hyperpolarization in animals raised at ambient CO₂ (Figure 4A). In contrast, RIA showed a CO₂-evoked depolarization in animals raised at ambient CO₂, consistent with a previous report [48], and a CO₂-evoked hyperpolarization in animals raised at high CO₂ (Figure 4B). RIG also showed a CO2-evoked depolarization in animals raised at ambient CO2, but showed no response in animals raised at high CO₂ (Figure 4C). The CO₂-evoked responses of AIY, RIA, and RIG were BAG-dependent (Figure 4A–C). Thus, AIY, RIA, and RIG show qualitative differences in their CO₂-evoked activity in animals raised at ambient vs. high CO₂.

Figure 4. First-order interneurons contribute to CO₂ avoidance and attraction through experience-dependent modulation of their CO₂-evoked activity.

(A) AIY is inhibited by CO₂ in animals raised at ambient CO₂ and activated by CO₂ in animals raised at high (2.5%) CO₂. Both responses are BAG-dependent. *p<0.05, **p<0.01, ***p<0.001, Kruskal-Wallis test with Dunn’s post-test. n=8–14 animals per genotype and condition.

(B) RIA is activated by CO₂ in animals raised at ambient CO₂, and inhibited by CO₂ in animals raised at high (2.5%) CO₂. Both responses are BAG-dependent. *p<0.05, ***p<0.001, Kruskal-Wallis test with Dunn’s post-test. n=9–10 animals per genotype and condition.

(C) RIG is activated by CO₂ in animals raised at ambient CO₂, but does not respond to CO₂ in animals raised at high (2.5%) CO₂. The response is BAG-dependent. **p<0.01, ***p<0.001, Kruskal-Wallis test with Dunn’s post-test (raised at ambient CO₂) or p=0.1060, unpaired t test (raised at high CO₂). n=8–14 animals per genotype and condition.

(D) AIZ is activated by CO₂ exposure in animals raised at both ambient and high (2.5%) CO₂. Both responses show BAG-dependent and BAG-independent components. **p<0.01, ***p<0.001, one-way ANOVA with Sidak’s post-test. n=10 animals per genotype and condition.

For A-D, calcium responses were measured using the ratiometric calcium indicators yellow cameleon YC3.60 or YC2.12. Left graphs show composite calcium responses to a 20-s pulse of 15% CO₂. Solid lines indicate average calcium responses; shading represents SEM. Blue lines indicate the response of wild-type animals to CO₂; green lines indicate the response of wild-type animals to an air control; orange lines indicate the response of BAG-ablated animals to CO₂; purple lines indicate the response of BAG-ablated animals to an air control. Right graphs show maximum values (for excitatory or neutral responses) or minimum values (for inhibitory responses) of % ΔR/R0 for each animal. Lines show medians and interquartile ranges.

(E) A model for CO₂ response in C. elegans. Animals raised at ambient CO₂ avoid CO₂, while animals raised at high CO₂ (2.5% CO₂) are attracted to CO₂. CO₂ response valence is determined by the coordinated activity of three interneuron pairs postsynaptic to the CO₂-sensing BAG neurons: AIY, RIA, and RIG. In animals raised at ambient CO₂, activation of RIG and RIA combined with inhibition of AIY results in CO₂ avoidance. In animals raised at high CO₂, activation of AIY, inhibition of RIA, and silencing of RIG results in CO₂ attraction. Activation of a fourth interneuron pair, AIZ, dampens behavioral sensitivity to CO₂ regardless of valence. CO₂ response is regulated by a combinatorial code of neuropeptides: NLP-1 reduces CO₂ avoidance in animals raised at ambient CO₂, FLP-16 reduces CO₂ attraction in animals raised at high CO₂, and FLP-27 enhances CO₂ response under both conditions.

See also Figures S3 and S4.

Together, our behavioral and calcium imaging data demonstrate that CO₂ repulsion and attraction are mediated by the coordinated activity of the same set of first-order interneurons. The CO₂-evoked activity of these interneurons is contextually modulated to drive opposite responses to CO₂. Inhibition of AIY promotes avoidance, while activation of AIY promotes attraction. In contrast, inhibition of RIA and silencing of RIG promotes attraction, while activation of RIA and RIG promotes avoidance. Thus, CO₂ avoidance arises from activation of RIA and RIG, and inhibition of AIY; CO₂ attraction arises from activation of AIY, inhibition of RIA, and silencing of RIG.

Distinct interneurons regulate valence and sensitivity

In contrast to AIY, RIA, and RIG, AIZ showed similar CO₂-evoked activity in animals raised at ambient vs. high CO₂ (Figure 4D). This activity was decreased but not eliminated in BAG- animals, suggesting additional CO2-dependent inputs into AIZ (Figure 4D). The magnitude of the depolarization in AIZ did not differ under ambient vs. high CO₂ conditions despite BAG activity being significantly different (Figures 1D, 4D), suggesting that AIZ activity may be constrained by gain control mechanisms. In addition, the fact that AIZ activity does not differ in animals raised at ambient vs. high CO₂ demonstrates that raising animals at high CO₂ does not result in a general physiological change that alters the CO₂-evoked activity of all interneurons in the circuit; rather, it results in cell-specific changes to the CO₂-evoked activity of RIG, RIA, and AIY. Furthermore, we found that AIB does not show CO₂-evoked activity in animals raised at ambient or high CO₂ (Figure S3D), suggesting that the phenotypes of the AIB- AIZ- animals are attributable to AIZ. However, we cannot exclude the possibility that AIB contributes indirectly to CO₂ response in combination with AIZ. Together with our behavioral data showing that silencing AIZ enhances both CO₂ attraction and repulsion (Figures 2F, 3E), these results demonstrate that AIZ regulates behavioral sensitivity to CO₂ regardless of valence. Thus, distinct interneurons regulate valence and sensitivity.

A combinatorial code of neuropeptides regulates valence and sensitivity

The rapid shift in CO₂ response valence that occurs following a change in environmental CO₂ levels is consistent with a mechanism of valence encoding that involves neuromodulation rather than synaptic rewiring [49–51]. To gain insight into the neuromodulators that regulate CO₂ response valence, we examined the CO₂-evoked behaviors of animals lacking individual neuropeptides that were previously shown to be enriched in BAG [20]. We first examined the CO₂ response of animals raised at ambient CO₂ and found that animals lacking the FMRFamide-like neuropeptide FLP-27 showed reduced avoidance, while animals lacking the neuropeptide-like protein NLP-1 showed enhanced avoidance (Figure S4A–B). We then examined the CO₂ response of animals raised at high CO₂ and found that animals lacking FLP-27 showed reduced attraction, while animals lacking FLP-16 showed enhanced attraction (Figure S4C–D). These data suggest that FLP-16 reduces CO₂ attraction, NLP-1 reduces CO₂ repulsion, and FLP-27 enhances CO₂ response regardless of valence. Thus, different neuropeptides play distinct roles in regulating CO₂ response valence and sensitivity. Our results raise the intriguing possibility that BAG secretes different combinations of neuropeptides in animals raised at ambient vs. high CO₂ to generate experience-appropriate responses to CO₂. Activity-dependent regulation of neuropeptide expression has been demonstrated in BAG, where flp-19 expression is greatly reduced in the absence of the CO₂-sensing pathway [52]. However, these neuropeptides may also act in other neurons to regulate the CO₂ circuit. Alternatively, modulation of the CO₂ circuit could be achieved through changes in receptor expression in the postsynaptic AIY, RIA and RIG interneurons, or modulatory input from other neurons.

A novel mechanism of valence encoding

We have examined the neural basis of CO₂ response and found that both CO₂ attraction and repulsion are mediated by a single microcircuit consisting of the BAG sensory neurons and four postsynaptic interneuron pairs: AIY, AIZ, RIA, and RIG (Figure 4E). CO₂ avoidance results from activation of RIA and RIG, and inhibition of AIY. CO₂ attraction results from activation of AIY, inhibition of RIA, and silencing of RIG. The valence associated with activation of each interneuron does not change as a result of experience: activation of AIY is always correlated with CO₂ attraction, and activation of RIA and RIG is always correlated with CO₂ avoidance. However, our data demonstrate that CO₂ response is determined by the combined activity states of AIY, RIA and RIG, and not solely by the interneuron(s) whose activation is correlated with its valence. The fourth interneuron pair, AIZ, regulates sensitivity regardless of valence. Furthermore, a combinatorial code of neuropeptides acts on the CO₂ circuit to regulate valence and sensitivity. Thus, the specific behavioral response to CO₂ is determined by the coordinated activity of four interneuron types, two of which are capable of showing both CO₂-evoked excitation and CO₂-evoked inhibition.

CO₂-evoked behaviors are also subject to modulation in other organisms. For example, CO₂ avoidance by the fruit fly Drosophila melanogaster and CO₂ attraction by the mosquito Aedes aegypti are reduced in the presence of food odorants through direct inhibition of CO2-detecting sensory neurons [53, 54]. CO₂ avoidance in D. melanogaster can also be attenuated by food odors through a mechanism in which the pathway mediating the response to food odors and the pathway mediating CO₂ response converge in the mushroom body [55, 56]. In addition, CO₂ is aversive to walking flies but attractive to flying flies, and different sets of chemoreceptors are required in the two conditions [57]. Thus, CO₂ response can be modulated by a number of different mechanisms across phyla.

In both invertebrates and vertebrates, a common mechanism for determining chemosensory valence involves two separate pathways, one that promotes attraction and one that promotes repulsion. Valence is then determined by which of these opposing pathways is activated [3, 58]. Here, we describe a novel mechanism of valence determination that instead involves a single pathway of interneurons. The CO₂-evoked activity of these interneurons is subject to extreme experience-dependent modulation based on the animal’s recent exposure to CO₂, allowing them to contribute to both attractive and aversive responses to CO₂ (Figure 4E). Thus, the functional connectivity of sensory neuron to interneuron synapses rapidly changes to reflect the current ethological state of the animal. Valence-encoding interneurons have been identified in mammals [3, 59], but whether their activity is modulated to drive changes in innate valence has not yet been investigated. Our finding that C. elegans can generate opposite behavioral responses to the same chemosensory input as a result of experience-dependent modulation of sensory neuron to interneuron synapses raises the possibility that similar mechanisms operate in mammals to mediate rapid changes in innate valence.

STAR Methods

CONTACT FOR REAGENT AND RESOURCE SHARING

Further information and requests for strains and reagents should be directed to and will be fulfilled by the Lead Contact, Dr. Elissa Hallem (ehallem@ucla.edu).

EXPERIMENTAL MODEL AND SUBJECT DETAILS

The free-living nematode Caenorhabditis elegans was used as the experimental model for this study. C. elegans has two sexes, hermaphrodites and males. Our experiments were carried out with hermaphrodites; males were used only for crosses. Unless otherwise noted, all experiments were done using C. elegans young adults. Strains were maintained at room temperature (RT, ~22°C) and ambient CO₂ (0.038% CO₂) on Nematode Growth Media (NGM) plates containing a thin lawn of Escherichia coli OP50 bacteria, according to standard methods [60]. Strains raised at high CO₂ were placed in a Tritech Research DigiTherm® CO₂ heating/cooling incubator, at 22°C and 2.5% CO₂, for one generation and subsequently tested. Strains transferred from ambient CO₂ to high CO₂CO₂ were maintained at RT and ambient CO₂, and transferred to the CO₂ incubator (22°C; 2.5% CO₂) for the indicated amount of time. Strains transferred from high CO₂ to ambient CO₂ were maintained at RT and ambient CO₂, raised in the CO₂ incubator (22°C; 2.5% CO₂) for one generation, and subsequently transferred to RT and ambient CO₂ for the indicated amount of time. All transgenic strains were made by microinjection of plasmid DNA into N2 hermaphrodites. See Key Resources Table for details.

EAH202 was obtained from Y. Iino (University of Tokyo, Tokyo, Japan) and then given an EAH strain number for identification purposes. EAH284 was generated by microinjecting the following plasmids, obtained from D. Colón-Ramos: DACR335 ttx-3::casp-3(p17) and DACR336 ttx- 3::casp-3(p12). EAH314 was derived from VM4770 [42] by outcrossing to N2 for 3 generations. The following strains were used to confirm results with independent transgenes: TV2217, EAH319, EAH267, and EAH346. TV2217 was obtained from D. Colón-Ramos [61]. The AIY ablation phenotype was confirmed using strain OH8, which contains a ttx-3 mutation. EAH319 was generated by microinjecting the following plasmids, obtained from D. Colón-Ramos: DACR77 glr-3::casp-3(p17) and DACR76 glr-3::casp-3(p12). The following GFP reporter strains were used to confirm cell ablations: OH99, EAH242, EAH269, IK716. EAH242 and EAH269 were generated using pCZGY#1534, obtained from Y. Jin [62]. The following strains were used to confirm results with independent deletion alleles: FX04829, RB2275, RB1902, FX04612, RB1340. The following strains were tested to rule out the possibility that the phenotypes observed with VC2012 were due to a deletion in Y17G7B.22 rather than flp-27: VC2180 and VC2063. The AIB AIZ-silenced strain IV316 was obtained from S. Chalasani [63]. Calcium imaging of AIY and AIZ was performed using strains IK1405 and IK686, respectively, which were obtained from I. Mori [64, 65]. Calcium imaging of RIA was performed using strain AX2361, which was obtained from M. de Bono [48]. Calcium imaging of AIB was performed using EAH259, which was obtained from T. Hirotsu [66].

METHOD DETAILS

CO₂ chemotaxis assays

Chemotaxis assays were performed as previously described (Figure S1A) [12]. Young adult animals were washed off plates using M9 buffer [60] and collected into a 65-mm Syracuse watch glass. Animals were washed 3x with M9 buffer and transferred to a 1-cm × 1-cm square of Whatman paper. Animals were then transferred from the filter paper to the center of a 100-mm NGM or chemotaxis plate [67]. The actual potential crawling distance of the animals is the diameter of the inside base of the plate, which measures 84.4 mm. A CO₂ gradient was generated by delivering gas stimuli to the plate through holes in the plate lids as previously described [12]. Unless otherwise indicated, a 21% O₂, balance N2 air control was delivered through one hole, and a certified mixture containing a designated CO₂ concentration with 21% O₂ and the balance N2 (Airgas) was delivered through the other hole. Gases were pumped through ¼-inch flexible PVC tubing using a syringe pump (PHD 2000, Harvard Apparatus) at a rate of 2 mL/min. The duration of the assay was 20 min. The number of animals in a 20-mm diameter circle centered under each gas inlet was counted and used to determine the chemotaxis index (CI), according to the formula:

$$CI=\frac{\text{\#\hspace{0.17em}}ofanimalsatC{O}_2-\text{\hspace{0.17em}\#\hspace{0.17em}}ofanimalsataircontrol}{\text{\#\hspace{0.17em}}ofanimalsatC{O}_2+aircontrol}$$

To control for directional bias due to subtle room vibrations, two identical assays were always performed simultaneously with the CO₂ gradient in opposite directions. Assays were discarded if the difference in the CI for the two plates was >0.9 or if fewer than 7 worms moved into the scoring regions on either plate. Transgenic strains expressing the histamine-gated chloride channel HisCl1 were incubated on NGM plates containing 10 mM histamine but not E. coli OP50 [45] for 20 min prior to the chemotaxis assay.

For assays where animals were raised at ambient CO₂ and transferred to high (2.5%) CO₂, the extent to which the animals had switched valence to attraction was calculated for each trial by comparing the CI for the current trial to the mean CIs for animals of the same genotype cultivated at either ambient or high CO₂ according to the following formula:

$$\%changeinvalence=\frac{CIforcurrenttrial-meanCIforambientC{O}_{2}trials}{meanCIforhighC{O}_{2}trials-meanCIforambientC{O}_{2}trials}\times 100$$

For assays where the animals were raised at high CO₂ and transferred to ambient CO₂, the extent to which the animals had switched valence to avoidance was calculated for each trial by comparing the CI for the current trial to the mean CIs for animals of the same genotype cultivated at either ambient or high CO₂ according to the following formula:

$$\%changeinvalence=\frac{meanCIforambientC{O}_{2}trials-CIforcurrenttrial}{meanCIforhighC{O}_{2}trials-meanCIforambientC{O}_{2}trials}\times 100$$

Values below 0 were counted as 0. Values greater than 100 were counted as 100.

Calcium imaging

Imaging was performed as previously described [12] using the genetically encoded calcium indicators yellow cameleon YC2.12 and YC3.60 [68]. Young adults were immobilized onto a cover glass containing a 2% agarose pad made with 10 mM HEPES using Butler Schein Animal Health Surgi-lock 2-octyl cyanoacrylate instant tissue adhesive. A custom-made gas delivery chamber was secured over the cover glass. Gases were delivered at a rate of 0.7–0.8 L/min. During the assay, 20 s of 21% O₂ was followed by a 20-s pulse of 15% CO₂, followed by 20 s of 21% O₂. Imaging was performed on a Zeiss AxioObserver A1 inverted microscope using a 40X EC Plan-NEOFLUAR lens, a Hamamatsu C9100 EM-CCD camera, and AxioVision software. The EM gain was set at 30. The emission image was passed through a DV2 beam splitter (Photometrics) as previously described [12]. Image analysis was performed using Zeiss AxioVision Software and Microsoft Excel.

For each recording, the mean pixel value of a background region of interest was subtracted from the mean pixel value of a region of interest containing the neuron soma (RIG, AIZ, AIB) or neuron process (AIY, RIA). When imaging from AIY, we focused on the synapse-rich segment of the process designated zone 2 [69]. When imaging from RIA, we focused primarily on the “loop” segment of the RIA process, and occasionally on nrV, the ventral segment of the process in the nerve ring [70]. Fluorescence values were normalized to the average values obtained 10 s prior to CO₂ delivery. The YFP/CFP ratio was calculated as previously described [20]. Images were baseline-corrected using a linear baseline correction. Traces with unstable baselines prior to the onset of the CO₂ pulse were discarded. To establish appropriate criteria for including traces as either depolarizations or hyperpolarizations, we recorded control traces for each genotype using a 21% O₂, balance N2 air control pulse. We then calculated the standard deviation of the set containing the maximum value of each control trace (max set), and the standard deviation of the set containing the minimum value of each control trace (min set). Traces recorded with a CO₂ pulse where the maximum value exceeded 3 standard deviations from the mean of the air control (max set) were designated depolarizations; traces recorded with a CO₂ pulse where the minimum value exceeded 3 standard deviations from the mean of the air control (min set) were designated hyperpolarizations. For cases where we observed CO₂-evoked depolarizations or hyperpolarizations, traces where the maximum or minimum value, respectively, was within 3 standard deviations of the mean of the max or min set for the air control were discarded.

For imaging animals raised at high CO₂, animals were incubated at 2.5% CO₂ for one generation. Prior to imaging, cameleon-expressing animals were placed on individual plates so they could subsequently be removed from the incubator one at a time to minimize time at ambient CO₂. Immediately prior to recording, individual animals were removed from the incubator and then prepared for imaging, spending approximately 5 min at ambient CO₂ before imaging. We note that for all imaging experiments, the CO₂ concentrations used for calcium imaging cannot be directly compared to the CO₂ concentrations used for behavioral assays due to differences in the setup for CO₂ delivery in the two cases.

Fluorescence microscopy

Images of gcy-9::GFP-expressing animals were acquired as previously described [12]. Animals were selected at the L4 stage using the co-injection marker pax-2::GFP. pax-2 expression is visible in the vulva at the L4 stage [71]. Selection based on pax-2 expression was used to limit bias and obtain a representative sample of gcy-9 expression.

Molecular biology

To achieve BAG-specific expression of pkc-1(gf), a 3-kb sequence upstream of the flp-17 gene [31, 72] was PCR-amplified from genomic DNA using primers 5’-gcggccgcaaaattatctggattcaccaac-3’ and 5’-ggatccggaaaatatttccacacagaat-3’, and used to drive expression of pkc-1(gf) [44]. A plasmid containing the pkc-1(gf) sequence was obtained from C. Bargmann (Rockefeller University, NY). AIY-specific expression of pkc-1(gf) was achieved using a 4-kb region of the ttx-3 gene [69, 73] that was PCR-amplified from genomic DNA using primers 5’-gcggccgcaagcttttttgaaacgatctt-3’ and 5’-ggatccatttgacaccgaagacaatt-3’. RIG-specific expression of pkc-1(gf) was achieved using a 149-bp region of the twk-3 promoter [74]. A plasmid containing the twk-3 promoter sequence was obtained from L. Salkoff (Washington University, MO). RIA-specific expression of pkc-1(gf) was achieved using a 1.2-kb region of the glr-3 gene [70] that was PCR-amplified from genomic DNA using primers 5’-gcatgcatcactgagccagagatgag-3’ and 5’-ggatccatgttaatagcaaatattgaagattc-3’. To generate a BAG-specific rescue of eat-4, we obtained a plasmid from I. Mori (Nagoya University, Japan) containing eat-4 cDNA. eat-4 cDNA was PCR-amplified from the plasmid using the following primers 5’-gctagccatgtcgtcatggaacgag-3’ and 5’-ggtaccagatggcgatctgatgacag-3’. Using our previously generated flp-17::pkc- 1(gf)::SL2::GFP plasmid, we replaced the pkc-1(gf) sequence with the eat-4 cDNA sequence and injected the resulting flp-17::eat-4::SL2::GFP plasmid into the MT6308 eat-4(ky5) strain, at 50 ng/μL. Behavioral results were confirmed with two independently derived rescue lines.

Interneuron ablation strains were generated using the two-component reconstituted caspase system previously described [75]. For genetic ablation of AIY, the following plasmids were obtained from D. Colón-Ramos (Yale University): DACR335 ttx-3::casp-3(p17) and DACR336 ttx- 3::casp-3(p12) [69, 73]. For genetic ablation of RIB, cell-specific expression was achieved using the cex-1 promoter [69]. A 1 -kb sequence upstream of cex-1 was PCR-amplified from genomic DNA using primers 5’-gtcgactttttaaatggaaagtaaaacga-3’ and 5’-ggatccttctgaaagtataagatttgactga-3’. The cex-1 sequence, along with DACR335 and DACR336, were used to generate cex-1::casp- 3(p17) and cex-1::casp-3(p12). For genetic ablation of RIG, cell-specific expression was achieved using the same 149-bp promoter region of twk-3 used to generate RIG::pkc-1(gf) (described above). The twk-3 sequence, along with DACR335 and DACR336, were used to make twk- 3::casp-3(p17) and twk-3::casp-3(p12). For genetic ablation of RIA, cell-specific expression was achieved using the promoter region of the glr-3 gene [70]. The following plasmids were obtained from D. Colón-Ramos (Yale University): DACR77 glr-3::casp-3(p17) and DACR76 glr-3::casp- 3(p12). Plasmids were injected at 50 ng/μL (AIY), 15 ng/μL (RIB), 35 ng/μL (RIG) or 35 ng/μL (RIA), along with the coinjection marker myo-2::dsRed (10 ng/μL), using standard microinjection techniques. Stable transgenic lines expressing myo-2::dsRed were crossed to the following GFP reporter strains to confirm loss of the respective interneurons: OH99 (AIY), EAH243 (RIB), EAH269 (RIG), and IK716 (RIA).

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis was performed using GraphPad Prism 6 using standard significance tests. Significance values were calculated relative to the N2 control, unless otherwise indicated. All statistical details for each experiment can be found in the figure legends. The D’Agostino-Pearson omnibus normality test was used to determine whether values came from a Gaussian distribution; if data were not normally distributed, non-parametric tests were used.