Hypoxia and the HIF-1 transcriptional pathway reorganize a neuronal circuit for oxygen-dependent behavior in Caenorhabditis elegans

Abstract

Rapid behavioral responses to oxygen are generated by specialized sensory neurons that sense hypoxia and hyperoxia. On a slower time scale, many cells respond to oxygen through the activity of the hypoxia-inducible transcription factor HIF-1. Here, we show that in the nematode Caenorhabditis elegans, prolonged growth in hypoxia alters the neuronal circuit for oxygen preference by activating the hif-1 pathway. Activation of hif-1 by hypoxia or by mutations in its negative regulator egl-9/prolyl hydroxylase shifts behavioral oxygen preferences to lower concentrations and eliminates a regulatory input from food. At a neuronal level, hif-1 activation transforms a distributed, regulated neuronal network for oxygen preference into a smaller, fixed network that is constitutively active. The hif-1 pathway acts both in neurons and in gonadal endocrine cells to regulate oxygen preference. These results suggest that physiological detection of hypoxia by multiple tissues provides adaptive information to neuronal circuits to modify behavior.

Oxygen availability is a powerful motivator of behavior. Many animals sense internal oxygen levels (1), and organisms that live in soil and water environments with variable oxygen levels can detect external oxygen as well (2, 3). In the soil nematode Caenorhabditis elegans, specialized chemosensory neurons sense environmental oxygen gradients to guide migration to intermediate oxygen levels (7–14% O₂). The oxygen sensors in these neurons have been proposed to be soluble guanylate cyclase homologs (sGCs) that bind directly to molecular oxygen (4, 5).

Avoidance of hyperoxia (>14% O₂) is mediated by a distributed network of C. elegans sensory neurons. Two sets of neurons that express sGCs are largely redundant for hyperoxia avoidance (6). In addition to the sGC neurons, hyperoxia avoidance depends on the activity of nociceptive neurons and serotonergic neurons that sense stress (6, 7). The distributed and partly redundant function of these neurons generates subtle gradations in oxygen preference.

C. elegans oxygen preference is regulated by food context and by naturally-occurring genetic variation. The standard wild-type strain, N2, does not avoid hyperoxia in the presence of bacterial food (4). Food suppresses hyperoxia avoidance by signaling through the G protein-coupled neuropeptide receptor NPR-1 and the TGF-β homolog DAF-7 (4, 6). Natural isolates of C. elegans with low npr-1 activity display hyperoxia avoidance on food, accumulate at the border of a bacterial lawn, and aggregate into social feeding groups (4, 8). Hyperoxia avoidance is a component of all of these behaviors; lawn borders and aggregates are attractive in part because they have lower, more favorable oxygen concentrations (4, 7). As expected from these linked activities, common genes are required for hyperoxia avoidance on food and for social feeding behavior.

Along with the rapid neuronal pathways for oxygen detection, the slower hypoxia-inducible factor-1 (HIF-1) transcriptional pathway monitors internal oxygen levels to regulate metazoan metabolism, vascular biology, cell survival, and cell proliferation (9). In normoxia, where HIF-1 is inactive, the HIF-1α subunit of the HIF-1 transcription factor is modified by prolyl hydroxylases (PHDs) that use oxygen as a cofactor (reviewed in ref. 10). Modified HIF-1α is recognized by the von Hippel-Lindau tumor suppressor protein (VHL), which targets HIF-1α for degradation. In hypoxia, PHD activity is reduced, and unmodified HIF-1α is stable and transcriptionally active. The HIF-1 pathway is conserved in C. elegans, where the genes encoding homologs of PHDs, VHL, and HIF-1α are named egl-9, vhl-1, and hif-1, respectively (11, 12). As in other animals, C. elegans HIF-1 protein accumulates to high levels after 4–8 h of hypoxia (1% or 0.1% O₂) but disappears within minutes in normoxia because of the actions of EGL-9 and VHL-1 (11, 12). Unlike Hif-1α knockout mice, which die embryonically (13, 14), C. elegans hif-1 mutants are viable in normoxia, and embryonic lethal only in hypoxia (1% O₂) (11).

How do physiological and behavioral pathways for oxygen homeostasis interact? A study in ref. 5 showed that C. elegans strains with low npr-1 activity shift their oxygen preferences after a few hours in hypoxia, but the wild-type N2 strain does not. Here, we show that longer-term exposure to hypoxia changes the behavioral response of wild-type C. elegans, causing it to prefer lower oxygen levels and avoid hyperoxia in the presence of food. This modified preference results from activation of the hif-1 pathway in both neurons and nonneuronal cells, and leads to a striking reorganization of the neuronal circuit for oxygen preference.

Results

Cultivation in Hypoxia Modifies Oxygen Preference Through the egl-9/hif-1 Pathway.

To examine long-term consequences of growth in hypoxia, wild-type animals were cultivated for 2 days at 1% O₂, where C. elegans metabolic rates are approximately half of normal (11, 15), and then tested for their oxygen preference in a gas-phase linear gradient of 0–21% O₂. The behavior of animals in these gradients is called aerotaxis and reflects both hyperoxia and hypoxia avoidance (4). Wild-type animals grown in standard laboratory conditions (normoxia) distributed themselves through the chamber with a median preferred oxygen concentration ≈10%, whereas animals cultivated in hypoxia shifted their oxygen preference to a lower median preferred oxygen concentration near 8% (Fig. 1 A and B). This shift in oxygen preference was similar to, but weaker than, the shift in npr-1 mutants grown in hypoxia (5) [supporting information (SI) Fig. S1].

Fig. 1.

Cultivation in hypoxia modifies oxygen preference through the egl-9/hif-1 pathway. (A) Diagram showing typical distribution of C. elegans (dots) in an aerotaxis assay. Preferred area is shaded. (B–H) Aerotaxis of wild-type animals and egl-9/hif-1 pathway mutants grown in normoxia or in 1% O₂ for 2 days as adults. Asterisks denote distributions different at P < 0.01 by χ² analysis. (I) The hyperoxia avoidance index, defined as [avg(fraction in 5–14% O₂) − avg(fraction in 14–21% O₂)]/[avg(fraction in 5–14% O₂) + avg(fraction in 14–21% O₂)]. Asterisks indicate values different from the same genotype and condition without food at P < 0.05 by t test. Crosses indicate values different from normoxia-grown wild-type on food at P < 0.01 by Dunnett test. In all panels, n ≥ 3 assays per genotype, 80–100 animals/assay; error bars denote standard error of mean (SEM).

The addition of bacterial food to the assay uncovered a second difference between wild-type animals grown in normoxia and hypoxia. Animals cultivated in normoxia did not avoid high oxygen levels in the presence of food (Fig. 1C), but animals cultivated in hypoxia had similar oxygen preferences whether food was present or absent (Fig. 1D). We conclude that cultivation in hypoxia eliminates the food regulation of aerotaxis.

Because of its known role in physiological adaptation to hypoxia, the hif-1 transcriptional pathway was considered as a possible mediator of the altered oxygen response. Animals mutant for hif-1 were cultivated in normoxia or hypoxia and tested for oxygen preferences in the absence or presence of food. Whether cultivated in normoxia or hypoxia, hif-1 null mutants behaved like wild-type animals in normoxia: They avoided hyperoxia in the absence of food but not in its presence (Fig. 1 E and F). Null mutants in egl-9/PHD, the negative regulator of hif-1, had a complementary defect. These mutants with constitutively high HIF-1 levels behaved like wild-type animals grown in hypoxia, and avoided hyperoxia in the presence or absence of food (Fig. 1G). egl-9 hif-1 double mutants closely resembled hif-1 single mutants, indicating that egl-9 acts through hif-1 to change behavioral responses to oxygen (Fig. 1H).

These results demonstrate that hypoxia regulates aerotaxis through the hif-1 pathway. When the hif-1 pathway is active (in chronic hypoxia or in egl-9 mutants), animals avoid hyperoxia in the presence of food. When the hif-1 pathway is inactive (in normoxia or in hif-1 mutants), food suppresses hyperoxia avoidance. As expected for a slow transcriptional pathway, hif-1 activity was required for changes in oxygen preference after 2 days of hypoxia, but it was not required for changes in wild-type or npr-1 mutants after 6 h of hypoxia (Fig. S1). Like cultivation in hypoxia, hif-1 activation in egl-9 mutants also caused a small but significant shift in the median preferred oxygen concentration (Fig. S1), but, for further analysis, we focused on its strong effect on regulation by food. This effect can be summarized in a hyperoxia avoidance index that compares the relative accumulation of animals at preferred oxygen concentrations (5–14% O₂) versus hyperoxia (14–21% O₂) (Fig. 1I; see also Methods). The 5–14% preferred interval was chosen to capture both the normal preference for 10% O₂ and the shift to 8% O₂ observed in hypoxic animals.

egl-9 Activity in Neurons and Nonneuronal Cells Regulates Oxygen Preference.

hif-1 and egl-9 are coexpressed in virtually all cells in C. elegans (11, 16) (Fig. 2A) and are expected to act in the same cells. To determine where the hif-1 pathway acts to regulate oxygen preference, we performed tissue-specific rescue of egl-9 mutants, monitoring hyperoxia avoidance on food. We chose to rescue egl-9 because egl-9 mutants were defective when grown in normoxia, whereas hif-1 defects were only apparent after cultivation in hypoxia (Fig. 1).

Fig. 2.

egl-9/PHD activity in neurons and nonneuronal secretory cells regulates hyperoxia avoidance on food. (A) Expression patterns of egl-9 transgenes under the listed promoters and rescue of aerotaxis and egg-laying. Colored boxes indicate expression of the transgene or rescue of behavior. (See also SI Methods.) (B) Hyperoxia avoidance index. ND, not determined. Asterisks indicate values different from the same genotype without food at P < 0.05 by t test. Crosses indicate rescued lines with values different from egl-9 in the presence of food at P < 0.01 by Dunnett test. Aerotaxis assays shown in Fig. S2.

Expression of an egl-9 cDNA from its endogenous promoter rescued food regulation of hyperoxia avoidance in egl-9 mutants, as did expression of the egl-9 cDNA from the unc-31 promoter (Fig. 2 A and B and Fig. S2). The unc-31 promoter is expressed in all neurons and in secretory cells of the somatic gonad: the spermatheca, vulval muscles, and uv1 uterine-vulval cells (17). egl-9 aerotaxis behavior was not rescued when the egl-9 cDNA was expressed only in all neurons or in sGC or serotonergic subsets of neurons that regulate oxygen preferences (Fig. 2 A and B). Negative results were also obtained with egl-9 expression in body wall and vulval muscles, uterine cells, uv1 uterine-vulval cells, pharyngeal muscles, pharyngeal marginal cells, pharyngeal gland cells, XXX endocrine cells, and hypodermis (Fig. 2 A and B). However, expression of egl-9 both in neurons and in the uv1 cells (H20+tdc-1 promoters) rescued food regulation of hyperoxia avoidance, unlike either H20::egl-9 or tdc-1::egl-9 alone (Fig. 2 A and B).

egl-9 mutants were first isolated based on defective egg-laying behavior and retention of late-stage eggs in the uterus (18). Egg-laying in egl-9 mutants was rescued by the same transgenes that rescued oxygen-dependent behaviors: egl-9::egl-9, unc-31::egl-9, and H20::egl-9+tdc-1::egl-9 transgenes (Fig. 2A). An involvement of uv1 gonadal cells in egg laying was further supported by genetic mosaic analysis of egl-9 expressed from its own promoter (SI Methods). These rescue experiments indicate that egl-9/hif-1 act in at least two sets of cells to regulate egg laying and hyperoxia avoidance: neurons and nonneuronal uv1 cells of the somatic gonad.

The egl-9/hif-1 Pathway Alters Neuronal Requirements for Hyperoxia Avoidance.

Studies have defined four groups of sensory neurons that promote hyperoxia avoidance (4, 6, 7) (Fig. 3A). The four groups are (i) URX, AQR, and/or PQR, which express sGCs (the URX set); (ii) SDQ, ALN, and/or PLN, which express sGCs (the SDQ set); (iii) ASH, which expresses transient receptor potential vanilloid (TRPV) channels; and (iv) ADF, which expresses TRPV channels and synthesizes serotonin. Hyperoxia avoidance requires at least one set of sGC-expressing neurons and at least one class of the TRPV-expressing neurons.

Fig. 3.

Induction of the egl-9/hif-1 pathway alters neuronal requirements for hyperoxia avoidance. (A) A distributed network of sensory neurons generates hyperoxia avoidance in wild-type N2 animals (6). URX, AQR, and PQR neurons (URX set) and SDQ, ALN, and BDU neurons (SDQ set) express sGC homologs and are likely oxygen sensors. ASH and ADF neurons express the TRPV channels osm-9 and ocr-2; ADF produces serotonin. In the presence of food, hyperoxia avoidance is suppressed by the activity of the neuropeptide receptor NPR-1 and TGF-β homolog DAF-7. (B) Summary of results in C–H. Aerotaxis of egl-9/PHD mutants requires URX, AQR, and PQR sensory neurons but not TRPV channel-expressing neurons. The NSM neurons are a more important source of serotonin than ADF neurons. (C and D) Effects of qaIs2241 and osm-9 on egl-9 aerotaxis off food. The qaIs2241 transgene kills URX, AQR, and PQR neurons (6). Asterisks denote distributions different from the double mutant at P < 0.01 by χ² analysis. (E) Hyperoxia avoidance index off food. Asterisks indicate double mutants different from egl-9 controls at P < 0.01 by Dunnett test. Double crosses indicate values different from the designated single mutant at p < 0.05 by Bonferroni t test. Data for some single mutants from ref. 6. (F and G) Effects of tph-1 on egl-9 aerotaxis off food and rescue of tph-1. ADF promoter was srh-142; NSM promoter was ceh-2; and serotonergic neurons other than ADF and NSM were not tested. Asterisks denote distributions different from tph-1; egl-9 at P < 0.01 by χ² analysis. (H) Hyperoxia avoidance of tph-1 strains. Asterisks indicate values different from egl-9 control at P < 0.01 by Dunnett test. Crosses and double-crosses indicate values different from nonrescued strain at P < 0.05 by Bonferroni t test. Additional aerotaxis assays shown in Fig. S3.

To determine which neurons were affected by the egl-9/hif-1 pathway, we systematically crossed egl-9 into mutants that affect these sensory neurons, and tested the double mutants for their aerotaxis behavior on and off food. egl-9 mutants were expected to resemble animals grown in hypoxia (12) (Fig. 1) and therefore could also shed light on the effects of environmental hypoxia on wild-type behaviors.

To our surprise, the neuronal requirements for hyperoxia avoidance in egl-9 mutants were very different from those of wild-type animals (Fig. 3B and Fig. S3). Only the URX set of neurons was important for hyperoxia avoidance, with other sensory neurons making negligible contributions. Notably, a transgene that kills the URX set of neurons disrupted hyperoxia avoidance in egl-9 mutants, but not in wild-type animals (Fig. 3 C and E). egl-9 mutants required the sGC gcy-36, which acts in the URX set, but were only slightly affected by gcy-35, which acts in both the URX set and the SDQ set (4–6) (Fig. 3E). These sGC requirements are consistent with a greater role for the URX set, and a diminished role for the SDQ set, in egl-9 mutants.

The TRPV channel genes osm-9 and ocr-2 are required for ASH and ADF function and for hyperoxia avoidance in wild-type animals (6, 7). By contrast, osm-9 or ocr-2 were not required in an egl-9 background (Fig. 3 D and E), suggesting that ASH and ADF are unimportant in this context.

Hyperoxia avoidance in wild-type N2 animals is potentiated by the neurotransmitter serotonin, which is produced by the ADF neurons (6). Mutants in the tryptophan hydroxylase enzyme gene tph-1 are defective in serotonin synthesis in ADF neurons and four other classes of C. elegans serotonergic neurons (19). Like N2 animals, egl-9 mutants required tph-1 for hyperoxia avoidance (6) (Fig. 3 F and H). However, unlike tph-1 single mutants, tph-1;egl-9 double mutants were not rescued by expression of tph-1 in ADF neurons (6) (Fig. 3 G and H). Instead, transgenic expression of tph-1 in only the NSM neurons resulted in small improvements in hyperoxia avoidance that were increased by expression in both NSM and ADF neurons together (Fig. 3 G and H). These results suggest that the essential serotonin sources differ in wild-type and egl-9 backgrounds (Fig. 3 A and B).

In summary, induction of the hif-1 pathway in egl-9 mutants changes the neuronal requirement for hyperoxia avoidance from a multicellular network of sGC-expressing and TRPV-expressing neurons to a smaller network of neurons. In egl-9 mutants, the essential elements are sGC-expressing URX, AQR, and/or PQR neurons and serotonin signaling from NSM neurons (Fig. 3B).

Cultivation in Hypoxia Alters the Neural Circuit for Oxygen Preference.

The initial analysis of egl-9 mutants suggested that they resembled wild-type animals after long-term cultivation in hypoxia (Fig. 1). To strengthen the connection between egl-9 mutants and the hypoxia response, we asked whether cultivation in hypoxia could shift the genetic requirements for aerotaxis in the same way as an egl-9 mutation. These experiments focused on the most striking differences between wild-type and egl-9, which were the differential requirements for the URX set of neurons (required in egl-9 but not in wild-type) and for osm-9/TRPV neurons (required in wild-type but not in egl-9). After cultivation in hypoxia, wild-type animals lacking the URX set of neurons were profoundly defective in hyperoxia avoidance (Fig. 4 A and C). Conversely, cultivation in hypoxia suppressed the hyperoxia avoidance defect of osm-9 mutants (Fig. 4 B and C). The altered behaviors in these strains suggests that growth in hypoxia changes the neuronal requirements for hyperoxia avoidance, increasing the contribution of the URX set of neurons and diminishing the requirement for TRPV neurons. These changes are similar to the changes in egl-9 mutants, supporting the hypothesis that chronic hypoxia changes the aerotaxis circuit by activating hif-1.

Fig. 4.

Cultivation in hypoxia alters the neural circuit for oxygen preference. (A and B) Aerotaxis of qaIs2241 (URX, AQR, and PQR killed) (A) and osm-9 (B) animals grown in 1% O₂ for 2 days as adults. Asterisks denote distributions different at P < 0.01 by χ² analysis. (C) Hyperoxia avoidance index. Asterisks indicate values different from qaIs2241 at P < 0.05 by Bonferroni t test. Crosses indicate values different from osm-9 at P < 0.05 by Bonferroni t test.

egl-9/hif-1, npr-1, and daf-7/TGF-β Pathways Regulate Oxygen Preferences Independently.

Like egl-9 mutants, animal with mutations in the neuropeptide receptor npr-1 or the neuronal TGF-β homolog daf-7 avoid hyperoxia in the presence of food (4, 6). This similarity motivated us to examine possible relationships among the egl-9/hif-1 pathway and the npr-1 and daf-7 pathways.

In npr-1 mutants, deleting either the URX set of neurons or the TRPV channels restores food regulation of hyperoxia avoidance (6). However, the same genetic manipulations did not restore food regulation to egl-9 mutants (Fig. 5A and Fig. S4). Conversely, a hif-1 mutation that restored food regulation to egl-9 did not restore food regulation to npr-1 (Fig. 5B). These results suggest that egl-9 and npr-1 mutants use different mechanisms to maintain hyperoxia avoidance on food.

Fig. 5.

egl-9/hif-1, npr-1, and daf-7/TGF-β pathways act independently to modulate oxygen preference. (A, B, and D) Hyperoxia avoidance index and food regulation in egl-9/sensory mutants (A), egl-9/npr-1 mutants (B), and egl-9/daf-7 pathway mutants (D). Asterisks indicate values different from the same genotype without food at P < 0.05 by t test. Some data published in ref. 6 are included for clarity. (C) Aerotaxis of egl-9;npr-1 double mutants. Asterisks indicate distributions different from double mutants at P < 0.01 by χ² analysis. Additional aerotaxis assays shown in Fig. S4.

Additional results supported the conclusion that egl-9/hif-1 regulation of oxygen responses is distinct from and parallel to regulation by npr-1. Like the respective single mutants, egl-9;npr-1 double mutants exhibited strong hyperoxia avoidance that was not regulated by food (4, 6) (Fig. 5B). However, egl-9;npr-1 double mutants preferred even lower oxygen concentrations than either single mutant, consistent with parallel functions (Fig. 5C). They also showed enhanced aggregation on food compared to egl-9 and npr-1 single mutants (data not shown).

Another gene that regulates oxygen responses on food is the TGF-β homolog daf-7 (6). In daf-7 mutants, food regulation of hyperoxia avoidance can be restored by a mutation in the downstream SMAD transcription factor daf-3 (6, 20). daf-3 did not affect egl-9 hyperoxia avoidance on food (Fig. 5D), indicating that egl-9 does not function through the daf-7/TGF-β pathway. We also examined a known transcriptional target of daf-7/TGF-β pathway, the serotonin biosynthetic enzyme tph-1. In daf-7 mutants, tph-1 expression in ADF neurons is increased, and this increase stimulates hyperoxia avoidance in the presence of food (6). By contrast, no changes in tph-1::GFP expression were observed in ADF or NSM neurons of egl-9 mutants, hif-1 mutants, or wild-type worms grown in hypoxia (Fig. S5). Thus, the egl-9/hif-1 and daf-7/daf-3 transcriptional pathways regulate hyperoxia avoidance by different mechanisms.

Transcriptional targets of the hif-1 pathway have been identified through microarray analysis (21, 22), but none have known effects on oxygen preference. By screening additional transcriptional reporters in egl-9 mutants, we identified several regulated genes in relevant cell types. Decreased expression of sGC::GFP transcriptional reporters in URX neurons was reported in animals grown at 1% O₂ (5); we found that expression of four sGC::GFP transcriptional reporters was decreased in URX neurons of egl-9 mutants (Fig. S5). egl-9 mutants also had reduced daf-7::GFP fluorescence in ASI neurons and reduced expression of the catecholaminergic biosynthetic enzyme tdc-1 in uv1 gonadal cells (Fig. S5). Although these molecular changes could contribute to the behavior, neither sGCs, daf-7, or tdc-1 are sufficient to explain the effects of hypoxia and hif-1 (Fig. 3, Fig. 5, and data not shown).

Discussion

After long-term cultivation in hypoxia, C. elegans prefers lower oxygen concentrations and avoids hyperoxia in the presence of food. Our work provides three insights into this process. First, these changes occur through activation of the hif-1 transcriptional pathway, which couples physiological and behavioral adaptations to environmental oxygen. Second, the hif-1 pathway acts both in neurons and in uv1 cells of the somatic gonad to modify oxygen preferences, suggesting that uv1 senses hypoxia through egl-9/hif-1 and signals to the nervous system, which also senses hypoxia. Third, the neuronal control of oxygen preference changes dramatically after cultivation in hypoxia. Even in the absence of food, where the behavioral responses appear similar, neuronal contributions are different.

The use of HIF-1 to modify oxygen preferences provides an elegant integrated solution to match physiological needs and behavioral priorities. C. elegans adapts to a range of environmental conditions by deploying HIF-1 in hypoxia and antioxidative enzymes in hyperoxia (11, 23). Antagonistic physiological adaptations to hypoxia and hyperoxia provide a rationale for the enhanced avoidance of hyperoxia after cultivation in hypoxia. Normoxia-grown animals can tolerate high oxygen; hypoxia-grown animals may be more sensitive to high oxygen due to HIF-1-dependent physiological changes. We speculate that other physiological pathways will drive specialized behavioral adaptations by tuning the relevant neural circuits.

Transcriptional modulation by the egl-9/hif-1 pathway provides adaptive information to mammalian sensory behaviors as well. In mice, reduced HIF-1 activity causes abnormal ventilatory adaption to chronic hypoxia, in part because of defects in the oxygen-sensing carotid body (24). Acute ventilatory responses to hypoxia appear normal in HIF-1-deficient mice, but the primary regulators of the hypoxia response in HIF-1 mutants are vagal afferents instead of the carotid body (24). Thus, HIF-1-dependent transcriptional regulation of circuits may be a general feature of oxygen sensation.

In C. elegans, egl-9 acts both in neurons and in uv1 cells of the somatic gonad to detect hypoxia. The uv1 cells express neuropeptides, catecholamines, and the secretory machinery for their release and therefore could generate long-range signals to the nervous system (17, 25–27). Oxygen diffusion through C. elegans tissues is expected to be rapid, but the gonad may have special needs to monitor hypoxia. Reproductive activity is associated with the highest oxygen consumption in the C. elegans life cycle (28, 29), and embryos are preferentially susceptible to hypoxia-induced lethality (11). The action of the hif-1 pathway in the somatic gonad may drive appropriate behavior at the crucial period during which eggs are produced and laid.

The molecular targets of egl-9/hif-1 that regulate oxygen preference are unknown. The cell fates of the oxygen-sensing neurons appear unaffected by egl-9/hif-1 genes (Fig. S5 and data not shown). Because the transcriptional pathway acts in several cell types, it is likely to have multiple relevant targets, perhaps including sGCs, daf-7, and tdc-1. At least one target in uv1 is predicted to affect intercellular signaling after chronic hypoxia.

The most striking effect of the egl-9/hif-1 pathway is the reorganization of neuronal contributions to oxygen preference. Hypoxia simplifies the circuit, eliminating food regulation and the contributions of several sensory neurons. These changes may increase the reliability of the behavior but decrease its flexibility. Even for elements of the circuit that seem similar, closer examination reveals differences in normoxia and hypoxia. Serotonin is important for hyperoxia avoidance in both contexts, but its neural source is different. In egl-9 mutants, NSM pharyngeal neurons, rather than ADF neurons, are the most important source of serotonin. NSM neurons integrate feeding state or starvation into egg-laying, locomotion, feeding, and male mating behaviors (30–32). Similarly, NSM may monitor food intake and other signals in the pharynx to modulate hyperoxia avoidance.

egl-9, npr-1, and daf-7 mutants avoid hyperoxia in the presence of bacterial food, and all exhibit the oxygen-dependent behaviors of aggregation and bordering on bacterial lawns (4, 6, 8, 33). However, each strain achieves hyperoxia avoidance on food by regulating a different combination of neurons. Analogous studies have demonstrated that different combinations of neurons mediate locomotion and olfactory avoidance behaviors in C. elegans depending on whether animals are starved or well fed (31, 34). It is generally assumed that outwardly similar behaviors have a common neural basis. Our results lead to the surprising conclusion that this is not always true, even in the simple nervous system of C. elegans. A change in the environment can uncover alternative circuits within the apparently fixed anatomy of the nervous system.

Materials and Methods

Nematode Growth and Strains.

Strains were cultured under standard conditions (35) with Escherichia coli HB101; under these conditions, they are exposed to ≈12–21% O₂ (4). For growth in hypoxia, animals were transferred onto fresh culture plates in a gas-tight hypoxic chamber at 1% O₂ (Coy Laboratory Products). Wild-type animals were C. elegans Bristol strain N2. Mutants in the hif-1 pathway were hif-1(ia4) and egl-9(sa307) (11, 16). The qaIs2241 transgene kills URX, AQR, and PQR neurons (6). A complete strain list is included in SI Methods. In all cases, null alleles or strong loss-of-function alleles were used.

Behavioral Assays.

Aerotaxis assays were performed by placing animals on nematode growth medium (NGM) agar in custom-made microfluidic devices fabricated from polydimethylsiloxane with a gas-phase linear gradient from 0 to 21% O₂, and monitoring animals' accumulation in nine bins across the gradient after 25 min (4, 6) (see SI Methods). For all genotypes and conditions, at least three independent assays of 80–100 animals each were performed on different days. Primary data are in Dataset S1.

For assays on food, thin bacterial lawns of E. coli HB101 were made by seeding NGM plates for overnight growth at 37°C and returning them to room temperature for at least 1 h before assay. Animals grown in 1% O₂ were assayed immediately upon removal from the hypoxic chamber.

The hyperoxia avoidance index used in this study differs from that used in ref. 6 to normalize the index for the preference shift toward lower oxygen in hypoxia-reared wild-type animals and egl-9 mutants (Fig. 1), but is the same index used in ref. 4. It varies from −1.0 (all animals in hyperoxic regions, >14% O₂) to + 1.0 (all animals in normoxic regions, 5–14% O₂) with 0 representing no preference between normoxia and hyperoxia (equal densities of animals in hyperoxic and normoxic regions). Some data published in ref. 6 were recalculated by using the modified hyperoxia avoidance index for presentation here.

Statistical Analysis.

Statistical analysis of aerotaxis was conducted as described in ref. 6. The overall distributions of animals in the aerotaxis assay were first compared by χ² analysis (9 bins per aerotaxis assay × 2 strains, yielding 8 degrees of freedom). For controls, such as N2 or egl-9, that were compared to multiple mutants, P < 0.01 was used as the level of significance. Most results were significant at P < 0.001 (Dataset S2). Any stated differences in a secondary measure, such as hyperoxia avoidance index or median preferred oxygen concentration, also passed this first test. In the second stage of analysis, unpaired t tests were used to compare hyperoxia avoidance indices or median preferred oxygen concentrations. ANOVA plus Bonferroni t tests or Dunnett tests were used to correct for multiple comparisons (Statview). Further analysis of sources of variance was conducted by principal components analysis (SI Methods and Fig. S6).