Temperature, Oxygen, and Salt-Sensing Neurons in C. elegans Are Carbon Dioxide Sensors that Control Avoidance Behavior

1As the major by-product of oxidative metabolism, CO2 is ubiquitous in nature. Although CO2 comprises only ∼0.038% of Earth's atmosphere, it can accumulate to higher levels in environments with high respiration rates. Organisms have evolved CO2-sensing mechanisms to monitor both external and internal CO2 concentrations, but how these systems function to control physiology and behavior remain poorly understood.

2Mice can smell environmental CO2 concentrations as low as 0.066% CO2 using specialized olfactory neurons that express carbonic anhydrase II. Carbonic anhydrases catalyze hydration of CO2 to generate H+ and HCO3−. HCO3− is thought to stimulate the mouse olfactory neurons by activating a guanylate cyclase, GC-D. In humans the GC-D homolog is a pseudogene, and we cannot smell CO2. However, we can taste CO2 in carbonated solutions via sour-sensing cells on our tongues. In rodents, CO2 levels of 10% or more elicit an innate fear response in which animals freeze and avoid open spaces. This response requires activation of the acid-sensing ion channel ASIC-1A in cells of the amygdala. High concentrations of inhaled CO2 also modulate wakefulness by stimulating midbrain neurons.

3Insects also sense and respond to environmental CO2. Drosophila adults and larvae avoid CO2 levels as low as 0.1%. Like the CO2-evoked fear behavior in mice, Drosophila CO2 avoidance is innate and may be part of an alarm response: stressed flies release 3- to 4-fold more CO2 than unstressed flies. Drosophila senses gaseous CO2 using two olfactory receptors, Gr21a and Gr63a, which are expressed in antennal sensory neurons. Like other insect olfactory receptors, these do not have homologs in vertebrates or worms. Artificial activation of the Gr21a/Gr63a-expressing neurons elicits an avoidance response. Whether the Gr21a/Gr63a receptor binds molecular CO2 or a CO2 derivative is not known. Interestingly, some food-associated odorants inhibit Gr21a/Gr63a CO2 receptor function, and the presence of food reduces CO2 avoidance. Although Drosophila avoids gaseous CO2, it is attracted to carbonated substrates, a response mediated by HCO3−-sensitive neurons in the proboscis.

4Besides monitoring external CO2, many animals also monitor internal CO2. Internal CO2 levels are regulated by respiratory gas exchange, but when left unregulated can lead to toxic changes in body fluid pH and death. Mammalian respiratory CO2 chemoreception occurs in the brain and carotid bodies. The molecular mechanisms are unclear, but CO2-sensitive cells express carbonic anhydrases, and changes in extracellular or intracellular pH modulate signaling via H+-sensitive ion channels. Insects achieve respiratory gas exchange by opening and closing spiracles, but the control mechanisms involved are not known.

5Many small animals, including the nematode C. elegans, lack a specialized respiratory system and use diffusion for gas exchange. As in other animals, high CO2 levels are toxic. C. elegans appears to control internal CO2 by avoiding environments where this gas exceeds ∼0.5%. Avoidance requires cGMP-gated ion channels containing the TAX-2 and TAX-4 subunits. Also implicated are the BAG sensory neurons, required for acute avoidance of a high CO2 and low O2 mixture. Recent work indicates that the BAG neurons are transiently activated when ambient O2 levels fall below 10%.

6Here, we show that the C. elegans head sensory neurons AFD, BAG, and ASE are primary CO2 sensors. AFD, BAG, and ASE were previously only known to detect changes in temperature, O2, and salt ion levels, respectively. Using Ca2+ imaging, we describe the CO2 responses of these neurons, which include ON, OFF, and perduring responses. We show that some, but not all, of the Ca2+ responses to CO2 depend on a cGMP-gated ion channel. Finally, we dissect how the C. elegans CO2 sensory system regulates CO2-evoked behavior. We find that the contribution of different sensors to behavior varies widely, depending on both context and stimulus dynamics.

7When placed in a 5%-0% CO2 gradient, C. elegans migrate away from high CO2 (Figures 1A and 1B). We used this assay to identify potential CO2-sensing neurons. Mutants defective in either the TAX-4 α or TAX-2 β cGMP-gated ion channel subunits show reduced CO2 avoidance, both in the presence and absence of E. coli food (Figure 1C). The defects of tax-2; tax-4 double mutants recapitulated those of single mutants (Figure 1C), consistent with α and β subunits functioning together. tax-2 and tax-4 are coexpressed in 14 of 40 C. elegans sensory neuron classes, implicating a subset of these neurons in CO2 sensing. A tax-2 promoter mutation, tax-2(p694), also disrupted CO2 avoidance (Figure 1C). Previous work reported that this allele deletes exon 1 and ∼1.6 kb of tax-2 upstream sequences. However, our sequencing data suggest that it removes only 365 bp in this interval (details in Supplemental Experimental Procedures available online). tax-2(p694) mutants have deficits in behaviors mediated by the AFD, BAG, ASE, AQR, PQR, and URX neurons but appear wild-type for responses mediated by other tax-2 expressing neurons. Selectively expressing tax-2 cDNA in AFD, BAG, ASE, AQR, PQR, and URX in tax-2(p694) mutants restored CO2 avoidance to the same extent as a full-length tax-2 genomic fragment (Figures 1C and 1D). We next attempted to rescue the tax-2(p694) defect by expressing tax-2 cDNA from neuron-specific promoters, confirming appropriate expression by polycistronic constructs that coexpress tax-2 and gfp. Expressing tax-2 cDNA in the AFD thermosensory neurons strongly rescued CO2 avoidance, both on and off food (Figure 1D). In contrast, restoring tax-2 to the BAG O2-sensing neurons rescued CO2 avoidance on food, as shown previously, but not off food. Expressing tax-2 cDNA in the ASE taste neurons or in the AQR, PQR, and URX O2-sensing neurons also partially rescued CO2 avoidance, both on food and off food (Figure 1D). These data implicate functionally diverse sensory neurons in CO2 avoidance.

8The AFD neurons are transiently activated when temperatures exceed cultivation levels. To test whether AFD also responds to CO2, we monitored AFD intracellular Ca2+ levels during CO2 exposure using the ratiometric Ca2+ sensor cameleon YC3.60, expressed in AFD under control of the gcy-8 promoter. Animals expressing the Ca2+ sensor retained wild-type CO2 responses (Figure S1A; see Experimental Procedures). To deliver CO2 stimuli, we used a Y-shaped microfluidic chamber that enables the gas phase over an immobilized animal to be switched in less than 3 s. In all experiments, O2 was maintained at 21%, with nitrogen (N2) completing the balance. AFD Left and AFD Right neurons responded equally to CO2 (Figure 2A; data not shown). On CO2 exposure the AFD neurons exhibited a fall in intracellular Ca2+ that slowly reversed to rise above baseline levels (“CO2-ON” response) within 2 min of CO2 coming on (Figures 2A and 2C). Thus, the AFD CO2-ON response has two components to it, an “ON-minimum” and an “ON-maximum.” Strikingly, AFD also responded to removal of CO2 with a fast Ca2+ spike that peaked within 10 s (“CO2-OFF” response, Figures 2A and 2D). The OFF-maximum was the largest feature of the AFD Ca2+ pattern, being on average 3- to 4-fold greater than the ON-maximum (Figure 2B). All three components of the AFD CO2 response were concentration dependent (Figure 2B). To exclude the possibility that the observed activity could be due to AFD temperature sensing, we exposed animals to 0%-0%-0% CO2 mock switches. Under these conditions AFD gave no responses (first 9 min, Figure 2E).

9We next examined whether repeated stimulation altered AFD Ca2+ responses. Some C. elegans sensory neurons, such as the ALM anterior touch neurons, habituate upon repeated stimulation. The AFD OFF response remained undiminished upon repeated exposure to 3% CO2 (Figures 2E, 2F, and S1B). We also asked whether prolonged CO2 exposure affects AFD responses. After a 9 min exposure to 3% CO2, the ON-maximum had decayed to baseline levels, whereas the OFF-maximum was unaltered (Figure 2G).

10CO2-evoked activity in AFD could be due to synaptic input to AFD. To test this, we imaged CO2 responses in unc-13 mutants, which have severe defects in synaptic release. The AFD CO2 responses of unc-13 animals were indistinguishable from wild-type (Figures 2H and S1C). These data suggest that, as well as being a thermosensory neuron, AFD is a CO2 sensor with both ON and OFF responses. The sensory endings of AFD have many finger-like projections, potentially providing a large surface for CO2 and temperature reception.

11AFD only responds to a temperature rise above the cultivation temperature. If AFD temperature and CO2-sensing are distinct, AFD might be expected to respond to CO2 at temperatures below the cultivation temperature. To test this, we built a temperature-controlled stage (see Supplemental Experimental Procedures). In animals grown at 22°C, AFD responded to CO2 both at 15°C and at 22°C (Figures S1E and S1F). The shape of the response was similar at the two temperatures but smaller at 15°C than at 22°C. These data support the idea that AFD CO2 and temperature-sensing pathways are at least partly distinct.

12Recent work has shown that the BAG neurons are transiently activated when O2 levels drop below 10%. showed that feeding animals lacking the BAG neurons have reduced avoidance of a 10% CO2/10% O2 mixture. We have previously shown that O2 responses can modulate CO2 avoidance. These data suggest that either BAG responds exclusively to O2 but modulates neural circuits mediating CO2 responses or that BAG is a primary sensor of both O2 and CO2.

13To test BAG neuron CO2 sensitivity, we created animals expressing cameleon YC3.60 in BAG from a pflp-17::YC3.60 transgene and imaged Ca2+ levels. The BAGL and BAGR neurons were exquisitely sensitive to a rise in CO2 (Figures 3A–3C). Cameleon reported a rise in Ca2+ that peaked after ∼30 s and then decayed (Figures 3A and 3B). The excitability threshold of BAG was below 0.25% CO2. A plot of mean fluorescence ratio change against percent (%) CO2 suggests that BAG reaches half-maximal activity at ∼2.9% CO2 (Figure 3D). Thus, BAG neurons respond to both O2 and CO2.

14Elevated CO2 persistently stimulates locomotory activity in feeding C. elegans, suggesting that some CO2-sensing circuits can signal tonically in high CO2. During prolonged high CO2 the BAG Ca2+ spike decayed to a plateau that persisted until CO2 removal, at which point Ca2+ returned to resting levels (Figure 3E). Thus, BAG exhibits both a transient peak and a perduring Ca2+ plateau in response to elevated CO2. As with AFD, we asked whether BAG neurons habituate. During five stimulus cycles of 3% CO2, BAG showed a decrement in response amplitude after the first CO2 stimulus, but no habituation thereafter (Figures 3F–3H).

15To test if the BAG neurons are primary CO2 sensors, we disrupted synaptic input to BAG using the unc-13 and unc-31 mutations. unc-31 mutants are defective in dense-core vesicle release, but not synaptic vesicle release. Neither the unc-13 nor the unc-31 mutations disrupted BAG Ca2+ responses, suggesting that BAG neurons are intrinsically CO2 sensitive (Figures 3I–3K). However, the magnitude of Ca2+ responses in these mutants was significantly enhanced, particularly in unc-31 animals, suggesting that BAG activity is normally inhibited by neuromodulators.

16We next examined CO2 responses in the ASE neurons that mediate chemotaxis to water-soluble cues, including salt ions such as Na+ and Cl−. ASEL and ASER are functionally asymmetric. ASEL is activated by a rise in the concentration of NaCl, whereas ASER is activated by a drop. For NaCl responses, activation of ASEL inhibits animals from reversing, whereas activation of ASER increases reversal likelihood.

17We imaged ASEL and ASER Ca2+ responses to CO2, using animals expressing the Ca2+ sensor YC2.12 in ASE from a pflp-6::YC2.12 transgene. Both ASEL and ASER were activated by 1%, 3%, and 5% CO2 (Figures 4A–4E), although the responses of ASEL were generally ∼2-fold larger than those of ASER (Figure 4E). ASE responses to CO2 were slow, taking around 2 min for Ca2+ levels to peak (Figure 4F). Sustained elevated CO2 led to sustained increases in Ca2+ (Figure 4F). As for AFD and BAG, ASE neurons appeared to be intrinsically CO2 sensitive because Ca2+ responses were intact in unc-13 mutants (Figures 4G and S1D). In summary, ASEL and ASER both respond to CO2 by a slow rise in Ca2+ that persists while CO2 is high and returns to baseline when CO2 returns to baseline.

18We examined whether the AQR, PQR, and URX O2-sensing neurons respond to CO2 because our tax-2 rescue data indicated that these neurons contribute, albeit weakly, to CO2 avoidance. Average Ca2+ traces indicated that unlike AFD, BAG, and ASE, none of these neurons respond reliably to CO2 (Figures S2A–S2D). URX most consistently showed CO2-evoked activity, and this was retained in unc-13 mutants (Figures S2A, S2E, and S2F). AQR and PQR occasionally showed a Ca2+ rise associated with an increase in CO2 but also showed apparent spontaneous activity that lay out of synchrony with the CO2 stimulus (Figures S2B–S2D). The response of PQR to a 0%-3%-0%-3% CO2 stimulus was dwarfed by its response to a 21%-11%-21%-11% O2 stimulus (Figure S2C).

19Having identified three C. elegans neuron classes that responded strongly to CO2 and a further three that responded weakly to CO2, we considered the possibility that all sensory neurons show some CO2 responsiveness. Therefore, we imaged Ca2+ responses to CO2 in the ASH neurons that respond to various aversive stimuli. ASH showed no response to 3% CO2 (Figure 4H). This suggests that AFD, BAG, and ASE are functionally specialized as CO2 sensors.

20Our tax-2 rescue data suggested that CO2 sensing in BAG and AFD neurons involves cGMP signaling. To examine this further we imaged BAG responses to CO2 in tax-2(p694) and tax-4(null) mutants. Both mutations completely abolished CO2-evoked Ca2+ responses in BAG (Figures 5A and 5C). This suggests that BAG CO2 sensory transduction is mediated by TAX-2/TAX-4 cGMP-gated channels and by extension, upstream guanylate cyclases (gcy).

21The only gcy genes known to be expressed in BAG are the atypical soluble guanylate cyclases gcy-31 and gcy-33. These appear to be O2 regulated because both are required for BAG O2 responses. To examine if GCY-31, GCY-33, or both are required in CO2 sensory transduction, we imaged BAG responses to 3% CO2 in gcy-31; gcy-33 double-deletion mutants. Loss of gcy-31 and gcy-33 reduced the CO2-evoked BAG Ca2+ response (Figures 5B and 5C). This suggests that GCY-31 and/or GCY-33 forms part of the CO2 sensory system in BAG, although other molecules are likely to be involved.

22We next imaged AFD responses in tax-2(null) and tax-2(p694) animals. Expression from the gcy-8 promoter is markedly reduced in tax-2 and tax-4 mutants, and YC3.60 expression was correspondingly low in AFD in tax-2(ot25null) animals. In contrast, expression in tax-2(p694) animals was similar to wild-type (data not shown). Both tax-2 mutations significantly reduced the AFD CO2 response, but neither completely abolished it (Figures 5D–5F). The AFD ON-minimum appeared to be absent in both tax-2 mutants, whereas the AFD ON-maximum was absent in tax-2(null) animals but enhanced in tax-2(p694) animals (Figures 5D–5F). Our data suggest that all three components of the AFD CO2 response involve TAX-2 mediated cGMP pathways but that other pathways also contribute.

23To further investigate molecular mechanisms of CO2 sensing, we asked whether C. elegans CO2 sensors express carbonic anhydrases, hallmarks of CO2-responsive neurons in other animals. Database searches indicate that the C. elegans genome encodes eight predicted carbonic anhydrases. Six, cah-1 to cah-6, belong to the alpha family, and two, bca-1 and bca-2, to the beta family. Because many members of the beta family are mitochondrial, we focused our studies on the alpha family. We fused upstream promoter regions of each gene to gfp and examined the resulting expression patterns. We found that cah-1, 2, 3, and 6 show strong neuronal expression in adults (Figure S3A). cah-4 was primarily expressed in the hypodermis (excluding the seam cells) and in the excretory cell, consistent with a kidney-like function for this cell. cah-3 and cah-5 show expression in intestinal cells, with cah-3 expression being especially strong. Using a pBAG::mCherry marker, we showed that cah-2, but not apparently any of the other five cah genes, was expressed in BAG (Figure S3B). cah-2 was also expressed in a set of four quadrant head neurons, other unidentified head neurons, the canal neurons CANL/R, whose processes run parallel to the tracts of the excretory cell, and a pair of tail neurons (Figure S5). Previous data suggest that cah-2 is also expressed in AFD. These data suggest that BAG and AFD neurons are specialized CO2 sensors that coexpress carbonic anhydrases and CO2-regulated cGMP pathways. They also raise the possibility that other C. elegans neurons and tissues respond to CO2.

24To investigate how CO2 sensors contribute to avoidance in spatial gradients, we genetically ablated neurons. We focused on AFD and BAG neurons because the Ca2+ responses of ASE to CO2 stimuli were slow, and those of AQR, PQR, and URX, weak. Specification of the AFD neurons requires the otd/Otx homeodomain transcription factor ttx-1, which is expressed only in AFD. ttx-1 mutants show thermotactic defects equivalent to those of animals in which AFD has been removed by laser ablation. ttx-1 mutants had a strong CO2 avoidance defect off food, and a weaker defect on food (Figure 5G). Wild-type avoidance was restored to ttx-1 mutants by a transgene containing ttx-1 genomic DNA (Figure 5G). These data suggest that the AFD neurons promote CO2 avoidance in spatial CO2 gradients.

25To ablate BAG we expressed the egl-1 programmed cell death activator from a BAG-specific gcy-33 promoter (we thank M. Beverly and P. Sengupta for this line). Both BAGL and BAGR neurons were absent in greater than 90% of animals bearing this transgene (Table S1 available online). Surprisingly, the CO2 avoidance of BAG-ablated animals was not significantly different from wild-type, both on and off food (Figure 5G). We asked if combined genetic ablation of AFD and BAG causes a synthetic CO2 avoidance phenotype. Ablating the BAG neurons disrupted the residual CO2 avoidance of ttx-1(p767) mutants on food (Figure 5G). However, in the absence of food, ttx-1(p767); pgcy-33::egl-1 animals showed no greater defect than ttx-1(p767) single mutants (Figure 5G). These data show that AFD and BAG promote CO2 avoidance in spatial gradients on food, and that AFD and at least one other neuron that is not BAG promote avoidance when food is absent. Thus, the importance of different sensory neurons for CO2 avoidance in spatial gradients depends on context.

26In 5%-0% CO2 spatial gradients (Figure 1), a C. elegans moving at ∼0.3 mm/s experiences a change of 0%-0.05% CO2/s, depending on bearing relative to the gradient. In our Ca2+-imaging experiments, immobilized animals experienced much sharper temporal gradients of ∼1% CO2/s. In the wild, animals are likely to encounter a variety of CO2 gradients. To analyze behavioral responses to sharp CO2 gradients, we designed a square-shaped microfluidic chamber that enables CO2 levels over freely moving animals to be switched rapidly (Movie S1 available online). We recorded responses and used custom software to extract instantaneous speed, reversal rate, and rate of omega turns, turns in which an animal's head and tail touch to form an “Ω” shape (N2, Figure 6B). In the absence of food, a rise in CO2 from 0% to 5% elicited a brief slowing followed by a transient increase in reversals and omega turns (Figure 6B). A rapid drop in CO2, from 5% to 0%, elicited an acceleration that coincided with suppression of reversals and omega turns.

27The timing of CO2-evoked Ca2+ responses in both AFD and BAG correlated with peaks in locomotory activity (Figure 6A). We investigated these correlations directly by ablating AFD and/or BAG and examining behavioral responses (Figure 6B). For statistical comparison, we chose time intervals before and after gas switches according to the occurrence of peaks in wild-type behavioral rates. In the absence of food, neither AFD nor BAG ablation abolished modulation of speed across shifts in CO2 (Figures 6B and S4). Stronger phenotypes were observed for reversal and omega rates (Figure 6B). Unexpectedly, ablation of AFD increased reversal and omega rates following a sharp CO2 rise (ttx-1, Figures 6B, 7B, 7C, 7H, and 7I) and reduced suppression of omega turns following a CO2 fall (ttx-1, Figures 6B, 7K, and 7L), suggesting that AFD acts to suppress reversals and omega turns at these two time points. Ablation of BAG abolished reversal and omega responses to a rise in CO2 (pBAG::egl-1, Figures 6B, 7B, 7C, 7H, and 7I) and reduced the suppression of omega turns following a CO2 fall (pBAG::egl-1, Figures 6B, 7K, and 7L), consistent with BAG excitation promoting reversals and omega turns. Coablation of AFD and BAG abolished the suppression of reversals and omega turns following a fall in CO2 (ttx-1; pBAG::egl-1, Figures 7F and 7L). This effect was due to reduced reversal and omega rates under prolonged high CO2 (ttx-1; pBAG::egl-1, red bars, Figures 7E and 7K). These data suggest that together BAG and AFD act to suppress reversals and omega turns when CO2 decreases.

28Curiously, AFD-ablated BAG-ablated animals continued to show a transient increase in reversals following a CO2 rise (ttx-1; pBAG::egl-1, Figures 6B, 7B, and 7C). This result suggests that there is at least one other CO2 “ON” sensory neuron, XYZ, that promotes reversals in response to a CO2 rise. It also suggests that after a CO2 rise, AFD acts antagonistically to both BAG and the hypothetical XYZ neuron to inhibit reversals. We investigated whether the ASE or AQR, PQR, URX neurons could be XYZ by ablating them together with AFD and BAG. Ablating ASEL/R had no significant effect on the reversal rate of AFD-ablated BAG-ablated animals immediately following a CO2 rise (che-1; ttx-1; pBAG::egl-1, Figures S5A–S5D) but did alter reversal rates under prolonged high CO2 (Figures S5E and S5F). The ablation of AQR, PQR, URX by an integrated pgcy-36::egl-1 transgene caused an increase in the reversal rate of AFD-ablated BAG-ablated animals in air alone (Figures S5A–S5D). These data suggest that the ASE neurons suppress reversals under prolonged high CO2 and that the AQR, PQR, URX neurons suppress reversals in the absence of CO2. However, even animals defective in AFD, BAG, ASE, AQR, PQR, and URX retained some CO2 responsiveness, suggesting that C. elegans has additional CO2 sensors.

29Wild-type C. elegans (N2) exhibit distinct locomotory patterns in the presence and absence of food. Animals move slowly and reverse frequently on food, whereas in its absence they move rapidly with fewer reversals. The escape mechanisms elicited by a CO2 rise on and off food were correspondingly different (Movies S1 and S2 and Figure S6). Feeding animals still briefly slowed down when CO2 levels rose but then switched to a high locomotory rate as high CO2 persisted (Figure S6). Coupled to the slowing response was a much stronger transient increase in omega turns (Figure S6). Feeding animals also persistently suppressed reversals in high CO2. These mechanisms increased the exploratory behavior of feeding animals, presumably helping them to escape from high CO2.

30To investigate whether AFD and BAG contribute to differences between on- and off-food behavior, we ablated them. AFD ablation abolished the increased speed response to high CO2 and resulted in inappropriately high-reversal and omega rates under high CO2 (ttx-1, Figure S6). In contrast, ablating only BAG had little or no effect (pBAG::egl-1, Figure S6). Ablating neither AFD nor BAG alone abolished the dramatic spike in omega turns following a CO2 rise, but ablating both neurons together nearly did (ttx-1; pBAG::egl-1, Figure S6). As for off food, loss of AFD and BAG did not eliminate CO2 responses, suggesting that other neurons contribute to rapid CO2-evoked behavior on food.

31In summary, genetic ablation suggests that AFD and BAG account for much of the different behavioral strategies employed in CO2 avoidance on and off food. In both contexts one or more other neurons also contribute to CO2 avoidance.

32C. elegans, like mammals, monitors CO2 using multiple neuron types. CO2 sensors include the ASE neurons with sensory endings directly exposed to the external environment and AFD and BAG neurons whose dendrites lie within the animal. All three neuron types are primary CO2 sensors: their CO2 responses are unimpaired in unc-13 mutants defective in synaptic release. Each neuron type has a unique CO2 response. In AFD, a rise in CO2 triggers an initial drop in intracellular Ca2+ levels (AFD ON-minimum), then a rise above baseline (AFD ON-maximum), and when CO2 is removed, a spike (AFD OFF-maximum). This complexity may reflect multiple CO2-transduction mechanisms. In contrast, BAG and ASE neurons are activated by a rise, but not a fall, in CO2. In BAG, Ca2+ peaks within 60 s of a rise in CO2, then decays to a plateau that persists as long as CO2 remains high; Ca2+ drops back to baseline upon CO2 removal. ASE responds slowly to CO2 exposure: Ca2+ takes 2 min to peak but remains elevated while CO2 is high. The tonic activity of BAG and ASE neurons in high CO2 may allow C. elegans to modify responses to other cues, perhaps by affecting sensory pathways or interneuron networks.

33AFD, BAG, and ASE also sense other stimuli. AFD senses temperature, BAG senses ambient O2, and ASE senses salt. This may enable sensory integration within sensory neurons. For each of the three neurons, CO2 and non-CO2 stimuli evoke distinct Ca2+ responses. When temperature rises above the cultivation level, AFD responds with a monophasic Ca2+ spike that lasts a few seconds. The dissimilar CO2 and temperature responses suggest that the two stimuli are sensed differently. Supporting this, AFD responds to CO2 below the cultivation temperature. The Ca2+ responses of BAG to high CO2 and low O2 are more similar in shape (Figure 3). In contrast, the responses of ASE to CO2 and NaCl differ markedly (Figure 4). First, unlike CO2, NaCl evokes an asymmetric response in ASEL and ASER: a rise in NaCl triggers a Ca2+ spike in ASEL but a drop in Ca2+ in ASER. Second, ASEL/R Ca2+ responses to NaCl adapt rapidly, whereas sustained CO2 stimulation leads to sustained high Ca2+ in ASE (Figure 4F). Third, whereas ASE responses to CO2 are slow, taking around 2 min for Ca2+ to peak, responses to NaCl peak within 30 s of stimulus exposure. The slowness of ASE CO2 responses could reflect rate-limiting hydration of environmental CO2.

34CO2 sensing in AFD, BAG, and ASE involves cGMP signaling. Mutating the cGMP-gated channel subunit tax-2 partially abolishes the AFD Ca2+ response to CO2 and completely abolishes CO2-evoked activity in BAG (Figure 5). CO2-evoked Ca2+ responses in ASE likely also depend on cGMP-gated channels because expression of tax-2 cDNA in ASE in tax-2 mutants partially restores CO2 avoidance (Figure 1). In mouse olfactory epithelia, CO2 sensing requires the transmembrane guanylate cyclase GC-D, which is activated by HCO3−. The hallmarks that make GC-D HCO3− regulated are unknown, but the C. elegans genome encodes 27 transmembrane guanylate cyclase (gcy), a subset of which could be similarly regulated. The AFD neurons express gcy-8, gcy-18, gcy-23, and gcy-29. gcy-8 gcy-18 gcy-23 triple mutants have a thermotaxis defect similar to that of the AFD specification mutant ttx-1, but have no defect in CO2 avoidance in a 5%-0% CO2 gradient (data not shown). ASE neurons express 11 transmembrane guanylate cyclases, nine of which are expressed asymmetrically either in ASEL or ASER.

35Transmembrane guanylate cyclase expression has not been reported in BAG. However, BAG expresses the atypical soluble guanylate cyclases GCY-31 and GCY-33. Simultaneously disrupting gcy-31 and gcy-33 reduced the CO2-evoked Ca2+ response amplitudes in BAG, suggesting that GCY-31 and/or GCY-33 contribute to CO2 sensing. GCY-31 and GCY-33 are thought to function as heterodimers that have an O2-binding heme cofactor and are required for BAG O2-evoked Ca2+ responses when O2 drops below 10%. An intriguing possibility is that the GCY-31/GCY-33 heterodimer is inhibited by O2 and activated by CO2, making it a sensory integrator of CO2 and O2 signals in BAG (Figure 8A); however, we cannot rule out the possibility of a linked mutation disrupting BAG responses.

36AFD, BAG, and ASE are unlikely to be the only CO2-responsive neurons in C. elegans. The AQR, PQR, and URX O2-sensing neurons showed sporadic responses to CO2 (Figure S2), and selective expression of tax-2 cDNA in these neurons partially restored CO2 avoidance to tax-2(p694) mutants, suggesting that they are CO2 sensitive. Moreover, more than ten C. elegans neurons express carbonic anhydrases, some of which may be unidentified CO2 sensors.

37Why does C. elegans have multiple CO2 sensors? One reason is that sensors are deployed differently according to the dynamics of the CO2 stimulus. For example, when food is absent, BAG mediates responses to sharp CO2 gradients but is less important for navigating shallow gradients (compare Figures 5G and 6B). A second reason is that context modifies the behavioral changes needed to escape CO2. For example, when food is present, C. elegans move slowly and reverse frequently. To efficiently escape high CO2 in a food-containing environment, C. elegans increase speed and suppress reversals relative to the “on food” ground state. By contrast when food is absent, animals are already moving quickly and reversing less frequently. Correspondingly, the importance of BAG for CO2 avoidance depends on both stimulus shape and food context. Whereas BAG-ablated animals respond poorly to rapid CO2 changes when food is absent, they respond like wild-type animals when food is present (pBAG::egl-1, Figures 6 and S6). Conversely, in shallow gradients BAG acts redundantly with AFD to promote CO2 avoidance when food is present but is not important when food is absent, even when AFD is ablated (Figure 5G).

38How do the Ca2+ responses of CO2 sensory neurons encode behavior? CO2-evoked neuronal events in AFD and BAG correlate with peaks and troughs in locomotory rates (Figure 6A). To investigate these relationships, we ablated CO2 sensors. One caveat of neuronal ablation is that it can only remove a neuron in its entirety, and not individual components of its responses. Ablation of AFD and BAG neurons one at a time and together suggests that: (1) BAG activation and the AFD ON-minimum act antagonistically, promoting and suppressing reversal and omega rates, respectively (Figures 7C and 7I); (2) BAG plateau activity and the AFD ON-maximum both act to promote reversal and omega rates during maintained high CO2 (ttx-1; BAG, Figures 7E and 7K); and (3) decay of BAG activity and the AFD OFF-maximum act together to suppress reversals and omega turns following CO2 removal (ttx-1; BAG, Figures 7F and 7L). Together our data suggest that when an animal is migrating up a CO2 gradient, BAG and AFD trigger turning, whereas when an animal is migrating down a CO2 gradient, AFD and BAG suppress turning (Figure 8B). Therefore, it appears that the three different components of the AFD CO2 response may differentially regulate behavior (1, 2, 3, AFD, Figure 8B). Because AFD(−) BAG(−) animals still respond to CO2, we also infer the existence of an additional sensory neuron, XYZ, that is neither ASE nor AQR, PQR, URX, that promotes turning when CO2 rises (Figure 8B).

39Elevated tissue CO2 is toxic. In C. elegans, CO2 levels exceeding 9% disrupt body muscle organization and general development and reduce fertility. The CO2 responses of AFD, BAG, and ASE neurons do not habituate upon multiple exposures to CO2 (Figures 2 and 3; data not shown). C. elegans CO2 avoidance in spatial gradients is also nonhabituating over a similar period (data not shown). By contrast, C. elegans attraction to benzaldehyde, response to noxious Cu2+ ion stimuli, and response to nose touch all habituate. Moreover, BAG and ASE neurons show tonic signaling while CO2 levels are high, at least over 20 min. We speculate that C. elegans CO2 avoidance habituates slowly and performs a homeostatic function by preventing CO2 poisoning of body tissues. C. elegans CO2 avoidance provides an opportunity for detailed examination of a CO2 homeostatic system with comparative ease relative to the systems of more complex animals.

40Strains were grown at 22°C under standard conditions. Mutant combinations were made by following visible phenotypes or using PCR to confirm genotype. A full list of strains can be found in Supplemental Experimental Procedures.

41Spatial CO2 gradient assays were as described. Briefly, polydimethylsiloxane (PDMS) chambers connected to gas syringe pumps were placed over adult worms on a 9 cm agar plate. After 10 min the distribution of worms was used to calculate a chemotaxis index (Figure 1). Chemotaxis bar graphs represent the average of nine independent assays performed over 3 days.

42For temporal gradient assays a square 11 × 11 × 0.2 mm PDMS chamber was placed over adult worms on 6 cm agar plates. For off-food assays, ∼40 animals were picked after washing in M9 Buffer to remove adhering E. coli. For on-food assays, a 2-day-old 20 μl E. coli lawn was used. Worms were allowed to crawl on food for 1 hr. After placing the chamber, animals were left for 4 min before exposure to a 0%-5%-0% CO2 stimulus. Behavior was captured using a Grasshopper CCD camera (Point Grey Research). A TTL-output from a frame counter (custom built) controlled opening and closing of Teflon™ pinch valves (Automate Scientific) at defined time points, controlling the switching of gases. Worms were tracked using DIAS Software (Solltech), and worm object paths were created. The centroid X and Y coordinates, maximum length, mean width, perimeter, and roundness were extracted for each worm object across frames. From these parameters, speed, omega initiation rate, and reversal initiation rate were calculated using a custom-written program in MATLAB (The MathWorks). Omega turns were detected by circular object topologies. This method gave 90.9% success using the stringent criterion that worm head touches worm tail. Reversal events were defined as forward movement (F), followed by backward movement (B), followed by return to forward movement (F). Using the criterion of an F-B-F event and optimized parameters minimum allowable reversal angle (150°), maximum reversal duration (7.5 s), and minimum reversal distance (0.3 mm, life size), reversal detection success rate ran at 81.25%. Detection parameters were optimized by minimizing the sum of the squared differences between detection outputs of computer and a human observer for Movie S1. Behavior occurring during merger of worm objects was discarded. Temporal gradient assay data represent the average of 16 or more movies for off food and nine or more for on food.

43In all experiments, percent (%) CO2 was balanced by percent (%) N2 while 21% O2 was maintained. In rescue experiments, transgenic animals were preselected by following coinjection markers. In all figures, statistical significance was determined using the two-tailed Student's t test.

44Ca2+ imaging was on an inverted microscope (Axiovert; Zeiss), using a 40× C-Apochromat lens and MetaMorph acquisition software (Molecular Devices). Agarose pads were made in M9 Buffer (pH 6.8) and 1 mM CaCl2, mimicking an NGM substrate. Worms expressing the Ca2+ sensor YC3.60 showed wild-type avoidance in 5%-0% CO2 gradients (Figure S1). Worms were glued to pads using Nexaband glue (WPI Inc.) and placed under the stem of the Y-chamber microfluidic device. Photobleaching was minimized using a 2.0 optical density filter and a shutter to limit exposure time to 100 ms per frame. An excitation filter (Chroma) restricted illumination to the cyan channel. A beam splitter (Optical Insights) was used to separate the cyan and yellow emission light. The ratio of the background-subtracted fluorescence in the YFP and CFP channels was calculated with Jmalyze. Fluorescence ratio (YFP/CFP) plots were made in MATLAB. Movies were captured at 2 fps. Average Ca2+ traces were compiled from at least six recordings made on 2 or more days.