A carbon dioxide avoidance behavior is integrated with responses to ambient oxygen and food in Caenorhabditis elegans

Abstract

Homeostasis of internal carbon dioxide (CO₂) and oxygen (O₂) levels is fundamental to all animals. Here we examine the CO₂ response of the nematode Caenorhabditis elegans. This species inhabits rotting material, which typically has a broad CO₂ concentration range. We show that well fed C. elegans avoid CO₂ levels above 0.5%. Animals can respond to both absolute CO₂ concentrations and changes in CO₂ levels within seconds. Responses to CO₂ do not reflect avoidance of acid pH but appear to define a new sensory response. Sensation of CO₂ is promoted by the cGMP-gated ion channel subunits TAX-2 and TAX-4, but other pathways are also important. Robust CO₂ avoidance in well fed animals requires inhibition of the DAF-16 forkhead transcription factor by the insulin-like receptor DAF-2. Starvation, which activates DAF-16, strongly suppresses CO₂ avoidance. Exposure to hypoxia (<1% O₂) also suppresses CO₂ avoidance via activation of the hypoxia-inducible transcription factor HIF-1. The npr-1 215V allele of the naturally polymorphic neuropeptide receptor npr-1, besides inhibiting avoidance of high ambient O₂ in feeding C. elegans, also promotes avoidance of high CO₂. C. elegans integrates competing O₂ and CO₂ sensory inputs so that one response dominates. Food and allelic variation at NPR-1 regulate which response prevails. Our results suggest that multiple sensory inputs are coordinated by C. elegans to generate different coherent foraging strategies.

CO₂ is an important sensory cue for many organisms. Insects can use elevated CO₂ as part of an alarm signal or to find food (1–3). In fungi, high CO₂ can induce filamentation (4) and regulate sporulation (5). Nematode parasites of plants and animals can follow CO₂ gradients to locate their hosts (6, 7). Internal CO₂ levels also provide important signals. For example, insects and mammals monitor internal CO₂ to modulate respiratory exchange (8–10). This homeostatic function prevents respiratory poisoning and pH changes in body fluids, which can occur if CO₂ levels rise above 5% (11).

Several mechanisms have been implicated in sensing CO₂. In Drosophila, avoidance of high CO₂ is mediated by a pair of odorant receptors (2, 12, 13). Artificially activating neurons expressing these receptors elicits the escape response (14). Less is known about how insects monitor internal CO₂ to control opening of spiracles (15). In mammals internal CO₂ levels regulate breathing, diuresis, blood pH, and blood flow (8). In most cases the molecular sensors involved are unclear although pH changes associated with hydration of CO₂ are thought to be important. Carbonic anhydrases, which catalyze the hydration of CO₂ to produce H⁺ and HCO₃⁻, are widely expressed in mammals. HCO₃⁻ has been shown to regulate the activity of a family of adenylate cyclases that is conserved from bacteria to man (16). However, the role of these enzymes in CO₂ signaling in animals is unclear. In fungi an HCO₃⁻-regulated adenylate cyclase modulates development in response to elevated CO₂ (4).

Caenorhabditis elegans belongs to the Nematoda, one of the largest phyla. Little is known, at a mechanistic level, about how these animals respond to CO₂. Nematodes lack specialized respiratory structures, and gaseous exchange is thought to occur through their cuticle. Previous studies have described C. elegans chemotaxis to HCO₃⁻ but have not examined responses to gradients of CO₂ (17, 18). C. elegans thrives in compost, mushroom beds, and decaying fruit, where it feeds on bacteria (19, 20). Broad ranges in O₂ and CO₂ concentrations exist in such environments depending on microbial growth, temperature, aeration, and moisture, and CO₂ levels can rise to 10% (21, 22). Here we investigate how C. elegans responds to CO₂.

Results

C. elegans Avoids Elevated CO₂.

To investigate how C. elegans responds to CO₂, we first exposed N2 (Bristol) wild-type animals to spatial CO₂ gradients. Gas gradients were set up over worms on agar surfaces using microfluidic chambers connected to defined gas mixtures (Fig. 1 A and B and ref. 23; see Methods). Within these chambers laminar flow operates such that a linear gas gradient is generated by simple diffusion between the two ends of the chamber. Unless otherwise indicated, O₂ was kept at 21% in these mixtures: CO₂ was increased at the expense of N₂. When only air was pumped into the chamber, N2 animals distributed equally to both sides of the chamber space (Fig. 1A). However, on introduction of a 5% to 0% CO₂ gradient, animals rapidly (<10 min) vacated areas of the chamber where CO₂ levels were high (Fig. 1B). To examine the concentration dependence of C. elegans CO₂ avoidance, we also assayed animals in gradients of 0.25% to 0%, 0.5% to 0%, 1% to 0%, and 3% to 0% CO₂. Avoidance of CO₂ was concentration-dependent, and animals avoided high CO₂ both in the presence and in the absence of a lawn of Escherichia coli food [Fig. 1 C and D and supporting information (SI) Fig. S1]. However, bacteria slightly but significantly reduced the strength of the avoidance response (Fig. 1 C and D and Fig. S1). The significance threshold for C. elegans CO₂ response was 1% CO₂ on food and 0.5% CO₂ off food at the 0.01% significance level (Fig. 1 C and D and Fig. S1). Thus, CO₂ is a potent repellent for N2 animals.

Fig. 1.

C. elegans avoids elevated levels of CO₂. (A and B) Distribution of N2 animals in microfluidic devices after 10 min without a CO₂ gradient (A) or with a 5% to 0% CO₂ gradient (B). Assays are in the absence of food. Gases pumped into the chamber are indicated at the top. (C and D) Distribution of N2 animals in CO₂ gradients in the absence (C and Table S1) or presence (D) of E. coli food (see also Fig. S1). Bin numbers refer to different portions of the microfluidic chamber. High CO₂ is to the left, as indicated by the wedge. Distribution of animals in all CO₂ gradients shown was significantly different from 0–0% CO₂ (P < 0.0001). Distribution of animals in all CO₂ gradients shown on food was significantly different from that off food (P < 0.0001). In this and all subsequent figures measurements were taken 10 min after the assay began.

To provide a simple measure for the CO₂ response we calculated a chemotaxis index by subtracting the number of animals in the low CO₂ half of the chamber from the number in the high CO₂ half and dividing by the total number of animals in the assay. Chemotaxis indices of +1, 0, and −1 indicate perfect attraction, indifference, and perfect avoidance of CO₂, respectively. The chemotaxis indices for CO₂ gradients of 1% to 0%, 3% to 0%, and 5% to 0% were −0.28, −0.66, and −0.80, respectively (see Fig. S1C, Assays without food).

To examine how C. elegans avoids CO₂, we exposed N2 animals to temporal CO₂ gradients by pumping defined gas mixtures at set rates into a behavioral arena. We subjected animals to both increases (0% to 3% or 5%) and decreases (from 3% or 5% to 0%) in CO₂. Animals subjected to a 0–3% rise in CO₂ responded within 10 s of the gas switch; forward movement briefly ceased, animals reversed, and by ≈25 s after the switch most animals had committed a near to 180° turn (Movie S1, Animals on food). To quantify this response we took a reversal to be backward movement of an animal by greater than one-quarter of its body length and a turn to be when an animal brings its head close to its tail to create the shape of the Greek letter omega (Ω). Raising CO₂ levels transiently stimulated reversals and turns both in the presence and absence of bacterial food (Fig. 2 A–D). Reversals were sustained for longer in the presence of food, suggesting that bacterial signals modify C. elegans CO₂ response pathways (Fig. 2 A–D). Because the responses do not persist after CO₂ levels plateau at 3%, they are likely evoked by a neural circuit that responds to changes in CO₂ concentration rather than absolute concentrations.

Fig. 2.

Behavioral mechanisms involved in avoidance of CO₂. (A–D) Fraction of animals reversing (A and C) or executing a turn (B and D) after a switch in CO₂ concentration. A and B show responses on food, and C and D show responses off food. Events are binned into 15-s time intervals. Gas switches (indicated by an arrow) occur at time 0. Blue bars represent animals subjected to an increase in CO₂, from 0% to 3%; red bars represent animals subjected to a decrease in CO₂ from 3% to 0%. “pre” indicates responses in a 15-s interval immediately before the gas switch. Asterisks indicate significances compared with responses before the gas switch (pre). In this and all subsequent figures, *** or +++ indicates P < 0.001, ** or ++ indicates P < 0.01, and * or + indicates P < 0.05. (E) Feeding N2 animals respond to high CO₂ by increasing their movement. Animals were subjected to a rise in CO₂ (indicated by the first arrow) from 0% to 5% followed by a fall in CO₂ (indicated by the second arrow) from 5% to 0%. “pre” refers to speed before the first gas switch. The gas stimulus regime is indicated below the graph. Speed was measured for each animal every second and then binned into 50-s intervals. Asterisks indicate the significance compared with speed before the up step (“pre”). + indicates significance compared with the 50-s interval before the down step. (F) The average speed of feeding N2 animals exposed to 5% CO₂ remains elevated as long as CO₂ levels are high. Animals were exposed to 0% CO₂ for 4 min, switched to 5% CO₂ for 30 min, and then returned to 0% CO₂ for 4 min. Bars represent the average speed of animals during 50-s intervals just before increasing CO₂ levels, just before decreasing CO₂ levels, and 3 min after return of CO₂ levels to 0%. Fifty-second intervals are indicated by shaded boxes in the gas stimulus regime displayed below the graph. Asterisks indicate significance compared with speed at 0% CO₂. (G) In the absence of food, N2 animals respond to a rise in CO₂ by reducing their speed. Speeds were averaged over the 50-s intervals indicated by shaded boxes in the gas stimulus regime displayed below the graph. (H) CO₂ is potentially a complex stimulus. Aqueous CO₂ as well H⁺ and HCO₃⁻ could be sensory cues for the nematode. Because nematodes are gas-permeable, CO₂ detection could involve both external and internal sensors. Double-headed arrows indicate equilibration of CO₂ among gas, liquid, worm, and agar phases. (I) Avoidance of 5% CO₂ persists with little or no change in magnitude across a broad range of external pH. All pairwise comparisons of chemotaxis indices at different pH values are not significantly different.

CO₂ Stimulates C. elegans Locomotory Activity on Food but Not off Food.

The speed an animal moves at influences how rapidly it can escape an aversive cue. This led us to examine whether elevated CO₂ stimulated movement in C. elegans. Raising CO₂ from 0% to 5% led to a doubling of the average speed of feeding N2 animals, from 46 to 92 μm/s (Fig. 2E). Unlike the increase in reversals and turns, which lasted for only 1–2 min (Fig. 2 A–D), the increased rate of movement was sustained as long as CO₂ levels remained high (>30 min; Fig. 2F). This perdurance suggests that absolute levels of CO₂, rather than change in its concentration, can signal to control speed of movement.

When returned from 5% CO₂ to 0%, feeding animals showed a further transient increase in speed before slowing down to the speed they exhibited before the CO₂ rise (Fig. 2E). In contrast to our observations in the presence of food, raising CO₂ levels from 0% to 5% in the absence of food caused a decrease in the average speed of movement, from 235 to 183 μm/s (Fig. 2G). Returning animals to atmospheric CO₂ levels reversed this inhibition. In summary, our data suggest that C. elegans can respond both to absolute levels of CO₂, which can regulate speed, and to changes in CO₂ levels, which modulate reversals and turns and, to some extent, speed too.

CO₂ Avoidance Is Distinct from Avoidance of Acid pH.

CO₂ is potentially a complex sensory stimulus. C. elegans lives in aqueous films and responds to chemical stimuli dissolved in these films. CO₂ is highly soluble in water, reacting to form carbonic acid that dissociates to yield H⁺ and HCO₃⁻ (Fig. 2H). HCO₃⁻ can dissociate further to yield H⁺ and CO₃²⁻, but CO₃²⁻ concentrations are negligible at physiological pH. Thus at the air–water interface an equilibrium is set up between gaseous CO₂ and its solvation products (Fig. 2H).

Previous studies have indicated that C. elegans avoids acid pH (24). This raised the possibility that CO₂ avoidance reflects escape from acid pH. We therefore examined how a 5% to 0% CO₂ gradient changed agar pH across the microfluidic chamber (Fig. S2). We observed a pH change of <0.1 pH units across the chamber, from pH 6.22 to pH 6.29. The small size of the pH change was expected because the agar substrate is buffered (see Methods). This small pH change and the previous observation that C. elegans avoids acid only below pH 4 (24) suggest that changes in external pH are unlikely to explain CO₂ avoidance.

C. elegans could also avoid CO₂ by responding to changes in HCO₃⁻ levels in the medium. To test this we examined CO₂ responses on agars buffered at different pH values, from 4.9 to 7.1. The concentrations of HCO₃⁻ generated by any given partial pressure of CO₂ should vary 100-fold across this pH range. We saw no substantial differences in avoidance of 5% CO₂ at different pH values (Fig. 2I). These data suggest that changes in external H⁺ and HCO₃⁻ are unlikely to be the sensory stimuli that trigger CO₂ avoidance. However, the permeability of CO₂ across lipid bilayers is high (≈0.35 cm s⁻¹) (25), and the C. elegans genome encodes several genes with homology to carbonic anhydrases, the enzymes that catalyze hydration of CO₂ (www.wormbase.org). C. elegans could therefore sense CO₂ fluctuations by monitoring internal (extracellular or intracellular) H⁺ or HCO₃⁻ levels. Alternatively, C. elegans could respond to molecular CO₂.

Signaling Through cGMP-Gated Ion Channels Contributes to CO₂ Avoidance.

Two major chemosensory pathways have been defined in C. elegans. One is mediated by a cGMP-gated ion channel encoded by the tax-2 and tax-4 genes (26, 27). A second is mediated by transient receptor potential V-like (TRPV-like) ion channels encoded by osm-9 and its associated subunits encoded by ocr genes (28, 29). We tested whether mutations in these genes disrupted CO₂ avoidance. Loss of osm-9 did not cause a carbon dioxide avoidance defective (Cdad) phenotype in the presence or absence of food (Fig. 3A). In contrast, mutations in tax-2 or tax-4 completely disrupted CO₂ avoidance on food but only partially disrupted avoidance off food (Fig. 3A). Thus, cGMP pathways contribute to CO₂ avoidance, but other signal transduction pathways may also be important.

Fig. 3.

Multiple signal transduction pathways contribute to CO₂ avoidance. (A) Disrupting the cGMP-gated ion channel encoding genes tax-2 and tax-4 reduces avoidance of 5% CO₂ both on and off food, whereas loss of osm-9, which encodes a TRPV-related channel, does not. In this and all other panels of this figure, a 5% to 0% CO₂ gradient was used to test behavior. * and + indicate significance compared with N2. Alleles used were tax-4(p678), tax-2(p691), and osm-9(ky10). “N2 (air)” represents a negative control with no CO₂ gradient. (B) N2 animals deprived of food gradually reduce CO₂ avoidance. Asterisks indicate significance compared with unstarved N2. (C) Disrupting DAF-2 insulin-like receptor signaling results in strong defects in CO₂ avoidance. Reduced DAF-2 signaling inhibits CO₂ responses by activating the DAF-16 Forkhead transcription factor. Alleles used were daf-2(e1370) and daf-16(mgDf47). Because daf-2(e1370) is a temperature-sensitive allele, animals were grown at 15°C and assayed at 22°C. *, significance compared with N2; ns, not significant compared with N2; +, significance compared with daf-2. (D) Mutations that disrupt pdk-1 3-phosphoinositide-dependent protein kinase 1 or akt-1 protein kinase B also disrupt avoidance of CO₂. PDK-1 and AKT-1 link activation of DAF-2 to inhibition of DAF-16. Alleles used were pdk-1(sa709) and akt-1(mg306). *, significance compared with N2. (E) egl-9 mutants grown in 21% O₂ exhibit attraction to high CO₂. This switch in CO₂ response requires HIF-1. ns, not significant compared with N2; *, significance compared with N2; +, significance compared with egl-9 (sa307). (F) Exposing feeding N2 animals to hypoxia (1% O₂) for 1 h inhibits CO₂ avoidance in a HIF-1-dependent manner. ns, not significant compared with N2; *, significance compared with N2; +, significance compared with N2 conditioned in 1% O₂ for 1 h on food. (G) Genetic pathways contributing to CO₂ avoidance and its modulation.

Starvation Suppresses CO₂ Avoidance.

C. elegans thrives in decaying organic matter where microbial activity can significantly raise local CO₂ levels (21, 22). It was therefore surprising that N2 animals avoided CO₂. Studies of other nematodes, both free-living bacteriophagous species (e.g., Panagrellus silusiae) and plant (e.g., Meloidogyne incognita) and animal (e.g., Steinernema sp.) parasites, have reported chemoattraction not chemorepulsion to CO₂ (6, 30, 31). This led us to examine whether C. elegans avoidance of CO₂ is context-dependent. We began by asking whether starvation alters CO₂ avoidance. We removed N2 animals from food for 1, 3, or 5 h and then tested their responses in a 5% to 0% CO₂ gradient off food. Food deprivation suppressed CO₂ avoidance: N2 animals showed no significant CO₂ avoidance after 3 h without food and weak attraction toward CO₂ after 5 h without food (Fig. 3B). Thus, whereas well fed or feeding animals strongly avoid CO₂, starved animals do not.

Insulin-Like Signaling Sustains CO₂ Avoidance.

Several neuroendocrine pathways signal feeding state in C. elegans (32–35). These include the daf-2 insulin-like receptor pathway: high DAF-2 signaling is associated with the well fed state, whereas low signaling is associated with food deprivation. We speculated that starvation might suppress CO₂ avoidance by inhibiting DAF-2 signaling. This hypothesis predicts that mutants in this pathway would behave like starved wild-type animals even when they are well fed. Consistent with this, mutants in the insulin-like signaling pathway, including the daf-2 insulin-like receptor, the 3-phosphoinositide-dependent kinase pdk-1, and the protein kinase B serine/threonine kinase akt-1 showed reduced CO₂ avoidance or even weak attraction (Fig. 3 C and D). Insulin-like signaling thus sustains avoidance of high CO₂.

The effects of food deprivation on CO₂ responses occurred over several hours (Fig. 3B), a timescale consistent with a transcriptional reconfiguration of CO₂-sensing circuits. Reduced DAF-2 signaling activates the DAF-16 Forkhead transcription factor (32, 36). We therefore asked whether DAF-16 was responsible for suppressing CO₂ avoidance in daf-2 mutants. Consistent with such a scenario, daf-2; daf-16 double mutants strongly avoided high CO₂ and behaved indistinguishably from N2 animals (Fig. 3C). Together these data are consistent with a model in which starvation reconfigures CO₂ responses, at least in part, by down-regulating insulin-like signaling and activating the DAF-16 forkhead transcription factor.

Hypoxia Suppresses CO₂ Avoidance via Activation of HIF-1.

Because CO₂ is the by-product of aerobic respiration, we speculated that O₂-sensing pathways might regulate CO₂ responses. One pathway regulated by O₂ is the hypoxia-inducible pathway. In both C. elegans and mammals, severe hypoxia (<1% O₂) induces hypoxia-inducible factor (HIF) transcription factors. In high O₂ HIFs are targeted for degradation by prolyl hydroxylases. These enzymes use molecular O₂ as a cosubstrate and are active in high, but not low, O₂. C. elegans encodes a single HIF, called HIF-1 (37), which is targeted for degradation by the prolyl hydroxylase EGL-9 (38). Loss of egl-9 leads to high levels of HIF-1 irrespective of ambient O₂. egl-9 mutants were attracted to CO₂ (Fig. 3E). To investigate whether this reversal of CO₂ chemotaxis was due to high HIF-1 activity, we examined the behavior of egl-9; hif-1 double mutants. Loss of hif-1 restored strong CO₂ avoidance to egl-9 mutant animals (Fig. 3E). Finally, we asked whether wild-type animals suppress CO₂ avoidance after experiencing hypoxia. After 1 h in 1% O₂, N2 animals, but not hif-1 mutant animals, suppressed CO₂ avoidance (Fig. 3F). Taken together, these data suggest that hypoxia signals through HIF-1 to reconfigure CO₂-sensing circuits, leading to indifference or even attraction to high CO₂.

The NPR-1 Neuropeptide Receptor Promotes CO₂ Avoidance.

We chose to extend our studies on the interplay between O₂ and CO₂ sensing. Previous work has shown that natural variation in the neuropeptide receptor npr-1 alters C. elegans foraging behavior (39). Strains expressing the less active NPR-1 215F isoform avoid high ambient O₂ and burrow as they feed. Strains bearing the more active NPR-1 215V isoform do not avoid high O₂ as they feed and do not burrow (23, 39). We asked whether npr-1 regulated not only O₂ but also CO₂ responses. Consistent with this hypothesis, npr-1 loss-of-function mutants showed striking defects in CO₂ avoidance both on and off food (Fig. 4A). This CO₂ avoidance defect could be rescued by an npr-1 215V transgene (Fig. 4A).

Fig. 4.

The naturally polymorphic NPR-1 receptor promotes CO₂ avoidance. (A) Mutations in npr-1 reduce avoidance of 5% CO₂ both in the presence and in the absence of E. coli; this phenotype is rescued by an npr-1 215V transgene. N2 animals carrying the npr-1 215F natural allele also exhibit reduced avoidance of CO₂ in the presence of food but maintain avoidance in its absence. ns, not significantly different compared with N2; nd, not determined. (B) The CO₂ avoidance defect in npr-1 mutants is not a consequence of their aggregation behavior. npr-1 animals grown in isolation (GII) retain a strong defect in avoidance of 5% CO₂ compared with similarly reared N2 animals. The weighted chemotaxis index was calculated by recording the position of each animal in a CO₂ gradient at 1-s intervals for 5 min and weighting this according to location in the CO₂ gradient (see Methods). “N2 air” represents a negative control with no CO₂ gradient. *, significance for comparisons between N2 and npr-1; +, significance between N2 GII and npr-1 GII.

To test whether the natural npr-1 215F allele also modified CO₂ avoidance, we compared the behavior of N2 animals to a near isogenic strain, AX613, which bears the npr-1 215F allele from the German wild strain RC301 backcrossed 20 times into N2. Animals bearing npr-1 215F showed a significant reduction in CO₂ avoidance compared with N2 on food but not off food (Fig. 4A). Thus, animals having high npr-1 activity strongly avoid CO₂ whereas animals with low npr-1 activity exhibit weaker avoidance.

Under normal cultivation conditions npr-1 mutant animals aggregate strongly. One explanation for their reduced CO₂ avoidance is a difference in experience compared with N2. To explore this possibility we grew N2 and npr-1 animals in isolation. npr-1(ad609) animals grown in isolation retained a strong defect in CO₂ avoidance (Fig. 4B). Together, these data suggest that signaling from the NPR-1 neuropeptide receptor promotes CO₂ avoidance, particularly when food is present.

Sensory Integration of CO₂ and O₂ Signals in C. elegans.

In our previous experiments we exposed animals to gradients of CO₂ in a background of 21% O₂. However, in nature C. elegans is likely to encounter simultaneous gradients of O₂ and CO₂. To explore how C. elegans navigates these more complex situations, we placed animals in combined gradients of O₂ and CO₂ with 11% O₂ and 5% CO₂ at one end of the chamber and 21% O₂ and 0% CO₂ at the other. As controls we tested animals in identical gradients of only CO₂ or O₂. Integration of CO₂ and O₂ stimuli was particularly interesting in the context of different alleles of npr-1 because natural variation at this receptor modifies both CO₂ and O₂ responses. We therefore tested strains carrying npr-1 215V (which occurs in N2 and all dispersing wild isolates), npr-1 215F (which occurs in all aggregating wild isolates), and the loss-of-function mutant npr-1(ad609).

The response of N2 animals in the crossed gradient was dominated by CO₂ avoidance: both on and off food animals accumulated at 21% O₂/0% CO₂ (Fig. 5). Thus, the avoidance of high O₂ by N2 animals when food is absent was suppressed by avoidance of high CO₂. By contrast, the response of npr-1(ad609) and npr-1 215F animals in the crossed gradient depended on context (Fig. 5 and Fig. S3). On food, the behavior of these animals was dominated by the O₂ response: animals ignored high CO₂ to accumulate at low O₂. Conversely, off food it was the response to CO₂ that dominated: animals behaved as if they were in a gradient that consisted only of CO₂ (compare Fig. 5 D–F).

Fig. 5.

C. elegans integrates antagonistic gradients of O₂ and CO₂ according to food availability and genotype at the npr-1 locus. Data show distribution of N2 and npr-1(ad609) animals in simple and mixed gradients of O₂ and CO₂ when food is present (A–C) or absent (D–F). The gas gradients are indicated below each set of panels: 5% to 0% CO₂ in A and D; 11% to 21% O₂ in B and E; and a combined gradient of 5% to 0% CO₂ and 11–21% O₂ in C and F. N2 animals strongly avoid CO₂ both on and off food, even if this requires migration to high-O₂ environments. In contrast, the behavior of npr-1 mutants and animals bearing the npr-1 215F allele (see Fig. S3) depends on context. These animals accumulate at low O₂/high CO₂ if food is present (C): an adverse CO₂ gradient does not appear to affect their avoidance of high O₂. Conversely, if food is absent, they tend to migrate to high O₂/low CO₂.

Thus, C. elegans integrates antagonistic inputs from CO₂- and O₂-sensing pathways to generate a coherent behavioral response in which one input dominates. The activity of the NPR-1 receptor reconfigures which of the two sensory responses dominates within the context of food availability.

Discussion

Well fed C. elegans avoid elevated CO₂, even though they seek environments where O₂ levels are between 11% and 7% (23, 40). The threshold we observed for CO₂ response is ≈0.5%. This is >10-fold higher than atmospheric CO₂ levels, but decaying organic matter can have much higher CO₂ concentrations, of 10% or more. C. elegans can respond both to absolute levels of CO₂, by modifying speed, and to change in CO₂ concentration, by altering direction of movement. Interestingly, C. elegans responses to O₂ are also coupled to changes in both concentration and absolute levels (40).

Behavioral and genetic dissection of the C. elegans CO₂ response reveals surprising complexity. Several observations are most easily explained if C. elegans has several pathways that respond to changes in CO₂. First, single mutations in known sensory transduction pathways are not sufficient to abolish CO₂ avoidance under all feeding conditions. Second, CO₂ responses are switched from repulsion to attraction by mutations in some genes. Third, the effects of CO₂ on speed of movement are complex. Although we have not identified CO₂-responsive sensory neurons in this study, one set of candidate neurons is those expressing the TAX-2/TAX-4 cGMP-gated ion channel.

Avoidance of CO₂ is modulated by contextual cues such as feeding state, exposure to hypoxia, and bacteria (Fig. 3G). Starvation completely suppresses CO₂ avoidance. This may represent a tradeoff in which food-deprived animals ignore an aversive cue to explore a wider range of environments. Previous work has shown that starvation inhibits signaling from the insulin-like receptor daf-2 and promotes entry of the DAF-16 forkhead transcription factor into the nucleus (32). Our data are consistent with high DAF-2 signaling in well fed animals sustaining avoidance of high CO₂ and low DAF-2 signaling in starved animals reducing CO₂ avoidance by activating DAF-16. DAF-2 has been implicated in modulating behavior previously, notably in studies of salt chemotaxis and thermotaxis (33, 35, 41). The daf-2 pathway may therefore act globally to reset behavioral state according to feeding conditions. Suppression of CO₂ avoidance in hypoxia may enable animals to migrate through CO₂-rich environments to reach more aerobic environments. Suggestions for how HIF-1 might alter CO₂ responses come from microarray studies. In both mammals and C. elegans, HIF regulates expression of carbonic anhydrases (42).

Bacterial signals also modulate CO₂ sensing: the CO₂ responses of well fed animals, both wild type and mutant, differ depending on whether food is present or not. Perhaps different combinations of sensory neurons mediate responses to CO₂ on and off food. Such a scenario has been described for the response of C. elegans to the aversive odorant octanol (43).

Sensory responses to CO₂ and O₂ are integrated by the worm in ways that depend on context and genotype at the naturally varying npr-1 locus. Previous data have shown that NPR-1 215V suppresses avoidance of high O₂ in feeding animals. Here we show that NPR-1 215V also promotes CO₂ avoidance. By coordinately stimulating avoidance of high CO₂ and inhibiting avoidance of high O₂, npr-1 215V is likely to promote migration to surface environments. In contrast, the npr-1 215F allele permits strong avoidance of high O₂ and weak avoidance of CO₂, promoting migration to subsurface environments. We speculate that these niche preferences may favor speciation.

Why does C. elegans avoid CO₂? One reason may be that high external CO₂ can acidify the body fluid of C. elegans. However, there are other possibilities. Comparison of local O₂ and CO₂ levels may allow the animal to monitor aeration and escape from an environment before it becomes anaerobic.

In summary, C. elegans CO₂ avoidance defines a novel behavior. CO₂ avoidance is highly integrated with other sensory cues of natural importance to the worm, such as food and ambient O₂. One exciting challenge for the future will be to identify the neuronal substrates of CO₂ avoidance in C. elegans and to examine how contextual changes alter cellular behavior, leading to the alterations in organismal behavior patterns that we have observed in this study.

Methods

Strains.

Strains were maintained at 22°C by using standard methods unless otherwise indicated (44). Strains used in this study are listed in SI Materials and Methods.

Behavioral Assays.

Spatial CO₂ gradients were generated by using custom-made 33 × 15 × 0.4-mm microfluidic devices fabricated from polydimethylsiloxane (PDMS). Design was modified from ref. 23. Devices were placed over 50–150 nematodes on nematode growth medium (NGM) agar. CO₂ gradients were formed by pumping a high percentage of CO₂ at one end of the chamber and 0% CO₂ at the other end with a syringe pump (PhD 2000; Harvard Apparatus). Flow rate through each inlet was 2 ml/min. A 5% to 0% CO₂ gradient was used in most assays; the background O₂ level was 21%. Assays were run for 10 min. The distribution of nematodes was recorded by counting animals in each of nine equal divisions of the chamber as well as in the two spaces at either end of the chamber (Fig. 1A). For assays in the absence of food, animals were washed with M9 Buffer before assay. Details of the wash method, which was designed to avoid giving animals a hypoxic shock, are in SI Materials and Methods. Assays in the presence of food were performed on NGM plates on lawns seeded 2 days earlier with OP50 (44). Defined CO₂::O₂::N₂ gas mixtures were obtained from The BOC Group.

Measurements of speed were performed by using the Digital Image Analysis System (DIAS) software as described previously (40). Each data point represents at least six assays. In all bar graphs, statistical significance was determined by using the two-tailed t test. In all worm distribution plots, significance was determined by pairwise comparison between different strains and conditions using Pearson's χ² test at the P < 0.0001 level. In all figures, error bars denote SEM.

Environmental Manipulations.

In Fig. 2I, the pH of the nematode substrate was varied by using different buffers as follows: pH 4.9 (40 mM sodium acetate, pH 4.75), pH 5.7 (40 mM malate, pH 5.33), and pH 7.1 (40 mM phosphate, pH 7.2).

In starvation experiments (Fig. 3B), two culture plates of N2 animals were washed three times in M9 before transfer to conditioning plates (6 or 9 cm of unseeded NGM). Animals were left for 0, 1, 3, or 5 h and then washed once before being assayed off food for CO₂ avoidance.

In the hypoxia conditioning experiments (Fig. 3F), C. elegans cultures were placed in a glove box (Coy Laboratory Products) at 1% O₂ for 1 h before being assayed off food for CO₂ avoidance.

In Fig. 4B three animals per plate were grown from the L2/L3 larval stage to adulthood. Pools of 25 animals were then assayed in CO₂ gradients in the presence of food. The position of each worm in the PDMS chamber was recorded over a 5-min period, beginning 10 min after the onset of the assay, with a CCD camera mounted on a dissecting microscope. Resulting films were analyzed, and the positions of the worms in the chamber were determined with DIAS (Soll Technologies). See SI Materials and Methods for further details.

pH Measurements.

We measured CO₂-induced pH changes using NGM containing 500 μM pH-sensitive chromophore 8-hydroxypyrene-1,3,6-trisulphonic acid (HPTS; Sigma). For the HPTS fluorescence (F) measurement method, see SI Materials and Methods.