ROS and cGMP signaling modulate persistent escape from hypoxia in Caenorhabditis elegans

The ability to detect and respond to acute oxygen (O2) shortages is indispensable to aerobic life. The molecular mechanisms and circuits underlying this capacity are poorly understood. Here, we characterize the behavioral responses of feeding Caenorhabditis elegans to approximately 1% O2. Acute hypoxia triggers a bout of turning maneuvers followed by a persistent switch to rapid forward movement as animals seek to avoid and escape hypoxia. While the behavioral responses to 1% O2 closely resemble those evoked by 21% O2, they have distinct molecular and circuit underpinnings. Disrupting phosphodiesterases (PDEs), specific G proteins, or BBSome function inhibits escape from 1% O2 due to increased cGMP signaling. A primary source of cGMP is GCY-28, the ortholog of the atrial natriuretic peptide (ANP) receptor. cGMP activates the protein kinase G EGL-4 and enhances neuroendocrine secretion to inhibit acute responses to 1% O2. Triggering a rise in cGMP optogenetically in multiple neurons, including AIA interneurons, rapidly and reversibly inhibits escape from 1% O2. Ca2+ imaging reveals that a 7% to 1% O2 stimulus evokes a Ca2+ decrease in several neurons. Defects in mitochondrial complex I (MCI) and mitochondrial complex I (MCIII), which lead to persistently high reactive oxygen species (ROS), abrogate acute hypoxia responses. In particular, repressing the expression of isp-1, which encodes the iron sulfur protein of MCIII, inhibits escape from 1% O2 without affecting responses to 21% O2. Both genetic and pharmacological up-regulation of mitochondrial ROS increase cGMP levels, which contribute to the reduced hypoxia responses. Our results implicate ROS and precise regulation of intracellular cGMP in the modulation of acute responses to hypoxia by C. elegans.


Introduction
Animals have evolved sophisticated mechanisms to maintain cellular homeostasis when encountering changes in oxygen ( carotid body, vascular smooth muscle cells, and chromaffin cells of fetal adrenal medulla, react acutely to hypoxia to ensure sufficient oxygenation of critical organs [1][2][3][4][5]. O 2 -sensitive cells respond to hypoxia by closure of K + channels and subsequent opening of voltage-gated Ca 2+ channels. A variety of models have been proposed for how a drop in O 2 levels is detected and transmitted to the downstream K + channels. Accumulating evidence suggests that the rapid decrease of O 2 is detected by the mitochondria, which rapidly generate reactive oxygen species (ROS) to inhibit K + channels on the plasma membrane [1,[6][7][8]. Perturbation of mitochondrial complex I (MCI) or mitochondrial complex III (MCIII) leads to continuous ROS production, which subsequently prevents acute response to hypoxia [1,[6][7][8]. However, it remains unclear whether mitochondrial ROS directly modifies K + channels on the plasma membrane. Several studies implicate cGMP signaling in acute hypoxia sensing, with cGMP inhibiting the O 2 sensitivity of chemosensory organs [9][10][11][12]. For example, atrial natriuretic peptide (ANP), which enhances cGMP production, dampens acute vasoconstriction of the pulmonary artery in response to hypoxia [9,12]. Similarly, nitric oxide (NO) stimulated cGMP production represses the sensory activity of the carotid body [11]. How a rise in cGMP blocks the sensitivity of O 2 sensors is unclear. By contrast, it is well established in the nematode Caenorhabditis elegans that cGMP is the primary second messenger mediating responses to acute hyperoxia (21% O 2 ). C. elegans avoids both atmospheric (21%) and hypoxic (1%) O 2 concentrations [13]. When O 2 levels rise from 7% to 21%, atypical soluble guanylate cyclases in the O 2 sensing neurons URX, AQR, and PQR that directly bind O 2 become progressively activated, elevating cGMP levels, which opens cyclic nucleotide gated (CNG) channels and causes neuron depolarization [13][14][15][16][17][18][19].
The remarkable ability of C. elegans to tolerate extreme O 2 concentrations, from near anoxia to 100% O 2 [20], makes it a good model to study O 2 responses in an intact, behaving, animal. By contrast to our detailed understanding of avoidance and escape from hyperoxia, little is known about how C. elegans responds to acute hypoxia [20,21]. Here, we characterize changes in C. elegans' locomotory behavior when O 2 levels drop abruptly to hypoxic levels and delineate the importance of intracellular cGMP and mitochondrial ROS in the regulation of these responses.

Hypoxia evokes immediate and sustained changes in C. elegans behavior
We set out to examine the effect of acute hypoxia on C. elegans locomotory behavior. A sudden decrease from 7% to 1% O 2 elicited a bout of turning maneuvers (omega turns and reversals, S1A and S1B Fig) followed by a switch to rapid forward movement (S1C Fig). Strikingly, while the increased turning behavior was transient, the locomotory arousal elicited by hypoxia was sustained for as long as O 2 levels remained low and showed no sign of adapting to repeated stimulation (S1D and S1E Fig). Acute hypoxia thus evokes both transient and sustained changes in C. elegans locomotory behavior, implying the existence of fast and slowly adapting neuronal mechanisms.
To explore behavioral responses to more severe hypoxia, we shifted animals from 7% O 2 to pure nitrogen. The microfluidic chamber arena we used for our behavioral assays was not completely gastight, so pumping pure nitrogen into the chamber resulted in measured O 2 levels of approximately 0.2% (henceforth "near-anoxia"), whereas we measured approximately 1.2% O 2 (hypoxia) when we pumped 1% O 2 into the chamber. A shift to near-anoxia elicited a sharp increase in speed of locomotion that peaked soon after stimulation and then decreased slowly (S1F Fig). The speeds reached by animals in near-anoxia were lower than those of animals in hypoxia (S1F Fig). When switched back from near-anoxia to 7% O 2 , animals did not reduce speed immediately and often transiently increased their locomotory activity before slowing down (S1F Fig). These observations, together with earlier work [22,23], suggest acute behavioral responses to hypoxia and anoxia differ both quantitatively and qualitatively [20].

Mutants lacking the BBSome show disrupted responses to acute hypoxia
To identify molecules that mediate or regulate acute responses to hypoxia, we performed a candidate screen for mutants that failed to increase their locomotory speed when O 2 levels acutely dropped from 7% to 1%. Our screen included mutants defective in guanylate cyclases, globins, neuropeptides, potassium channels, G protein signaling, cilia intraflagellar transport (IFT) and mitochondrial electron transport (S1 Table). This screen yielded multiple strains with strong defects in their responses to acute hypoxia. A subset of these strains contain lossof-function mutations in bbs genes, encoding the subunits of Bardet-Biedel syndrome (BBS) protein complex BBSome [26], and in the dyf-3 gene, whose product plays a role in IFT and is the ortholog of vertebrate qilin/cluap1 [27][28][29][30] (S1 Table). CLUAP1 is the only member of the IFT-B complex that reliably co-immunoprecipitates with BBSome [26], which traffics structural and signaling components into and out of cilia [31][32][33][34]. dyf-3 animals moved more slowly at 7% O 2 and showed a greatly reduced behavioral response to hypoxia ( Fig 1D). This phenotype could be rescued by a wild-type copy of dyf-3 ( Fig 1D). bbs mutants, including bbs-1, bbs-2, bbs-4, bbs-7, bbs-8, and bbs-9, each displayed low basal movement at 7% O 2 and strong suppression of hypoxia-evoked behavioral responses (Figs 1E, S2A and S2B and S1 Table). Expressing bbs-9 cDNA in ciliated neurons fully restored the hypoxia response to bbs-9 mutants (Fig 1E). Mutants defective in other components of the IFT machinery, or in ciliogenesis, also showed defects in response to acute hypoxia, to varying degrees (S1 Table). These data implicate cilia in the hypoxia response.
Attenuated behavioral responses to 1% O 2 in cilia-defective mutants could reflect more general deficiencies and reduced locomotion in mutant animals, rather than a specific defect in the ability to sense hypoxia. To probe this, we examined whether mutations in bbs-9 and dyf-3 impaired the very similar behavioral responses to 21% O 2 , using an npr-1 null background. Both dyf-3; npr-1 and bbs-9; npr-1 mutants were strongly aroused by 21% O 2 but not by 1% O 2 (S2C and S2D Fig). Thus, the reduced behavioral response of dyf-3 and bbs-9 mutants to hypoxia cannot be explained by a general locomotory defect, but reflects an inability of C. elegans to respond appropriately to a drop in ambient O 2 .
We next sought to identify the neurons where DYF-3 and the BBSome act to promote acute hypoxia responses. Expressing bbs-9 or dyf-3 cDNAs in all ciliated neurons, using the osm-6 promoter, fully restored acute hypoxia responses to the respective mutants ( Fig 1F). Selective expression in the ADL and ASH neuron pairs using the gpa-11 promoter gave partial albeit significant rescue (Fig 1G and 1H). Selective expression in either ADL (sra-6p) or ASH (srh-

PLOS BIOLOGY
220p) neurons alone gave weak but still significant rescue (Fig 1G and 1H). Cell-specific inactivation of bbs-9 by sense and antisense RNA interference (RNAi) constructs [35] in ADL and ASH neurons partially inhibited acute response to hypoxia (S2E Fig). These data suggest that DYF-3 and BBSome act in ADL and ASH, as well as other ciliated neurons, to promote acute responses to hypoxia.

Increased cGMP signaling inhibits acute hypoxia responses
Earlier studies have shown that defects in IFT lead to increased cGMP/PKG signaling [36,37] and that loss-of-function mutations in genes encoding guanylate cyclases, or in the cGMPdependent protein kinase (PKG) egl-4, can suppress a subset of phenotypes associated with IFT defects [36][37][38][39]. We therefore tested whether disrupting egl-4 suppressed the hypoxia response defects of dyf-3 and bbs-9 mutants. A loss-of-function mutation in egl-4 restored the hypoxia responses of both dyf-3 and bbs-9 mutants to wild-type levels (Fig 2A and 2B). One possible mechanism underlying suppression of the bbs-9 hypoxia phenotype is restoration of cilium function to egl-4; bbs-9 double mutants; however, the dye filling defects of bbs-9 mutants were not suppressed by loss of egl-4 (S3A and S3B Fig). EGL-4 contains a low and a high affinity cGMP binding site. Mutating either site abolishes EGL-4 nuclear localization [40,41]. The hypoxia response defect of dyf-3 and bbs-9 mutants was partially suppressed when the low affinity cGMP binding site of the endogenous egl-4 locus was edited using CRISPR/ Cas-9 (Fig 2C and 2D). In addition, an egl-4 gain-of-function mutation, which enhances PKG signaling [42], potently inhibited acute responses to 1% O 2 ( Fig 2E). We conclude that constitutive high PKG activity inhibits acute hypoxia responses.
Our data are consistent with dyf-3 and bbs mutants having elevated intracellular cGMP, leading to high EGL-4/PKG activity. To quantify total cGMP levels in these animals, we used ELISA of whole worm lysates [43]. As a positive control, we measured cGMP in mutants defective in the 4 C. elegans phosphodiesterases (PDEs) that hydrolyse cGMP, PDE-1, 2, 3, and 5. As expected, we observed a substantial increase in cGMP in pde-1,2,3,5 quadruple mutants. We also observed a small but significant cGMP rise in bbs-9 and dyf-3 mutants (Fig 2F). pde-1,2,3,5 quadruple mutants were completely defective in hypoxia-evoked locomotory responses, supporting the notion that high intracellular cGMP levels inhibit arousal in response to hypoxia ( Fig 2G). To analyze this further, we assessed pde single and all possible double mutants and found that pde-2; pde-3 animals were as defective as pde quadruple mutants (S4A-S4D CNG channels are another major effector of cGMP besides PKG. However, mutations in the C. elegans CNG channel genes tax-4, cng-1, cng-3, and cng-4 did not significantly alter acute hypoxia responses, and only cng-1; cng-3 weakly suppressed the hypoxia response defects of bbs-9 mutants (S4F- S4H Fig). Activation of CNG channels is therefore unlikely to be required either for acute hypoxia sensing or for suppressing acute hypoxia responses in bbs-9 mutants.
Disrupting GCY-28 suppresses the hypoxia response defects of bbs mutants cGMP is generated by guanylate cyclases. To identify guanylate cyclases whose activity suppresses hypoxia responses in bbs-9 mutants, we studied hypoxia-evoked locomotory responses in guanylate cyclase mutants. None of the guanylate cyclase mutants was defective in their response to acute hypoxia, suggesting guanylate cyclases are dispensable for hypoxia sensing (S1 Table). Disrupting all 7 soluble guanylate cyclases encoded by the C. elegans genome increased the baseline speed of bbs-9 mutants, but failed to restore their hypoxia-evoked behavioral response (S5A Fig). A survey of mutants defective in each of the 27 predicted receptor guanylate cyclases encoded in the C. elegans genome [44,45] identified several that partially suppressed the defective hypoxia response of bbs-9 animals (S5B Fig). Strikingly, disrupting gcy-28 alone was sufficient to restore hypoxia-evoked escape fully to bbs-9 mutants (Figs 3A and S5B). Deleting gcy-28 also robustly suppressed the hypoxia response defects of pde-1,2,3,5 mutant animals ( Fig 3B). Conversely, overexpressing either the gcy-28.c or the gcy-28.d splice isoforms from the gcy-28.c promoter effectively eliminated locomotory responses to hypoxia in wild-type animals ( Fig 3C). By contrast, we did not observe any inhibitory effects when 2 other guanylate cyclases, gcy-35 or gcy-15, were overexpressed (S5C and S5D Fig). Taken together, these data suggest that GCY-28 has a modulatory role in hypoxia-evoked behavioral responses.
We next asked whether the increased cGMP levels in bbs-9 mutants were dependent on GCY-28. Total cGMP levels were markedly decreased in gcy-28 mutants compared to wild type, and the elevated cGMP in bbs-9 mutants was significantly reduced by removing gcy-28 ( Fig 3D). Our ELISA data suggest that GCY-28 is the major source of total cGMP in C. elegans, despite the 33 other guanylate cyclases encoded in the genome. This is consistent with the broad expression pattern of GCY-28; other guanylate cyclases are only expressed in a small number of neurons [45]. To visualize how cGMP levels in individual neurons were altered by loss of bbs-9, we used the FlincG3 cGMP sensor [46]. To help us avoid movement artifacts, we used pseudo-ratiometric imaging by co-expressing FlincG3 in the ASK, AFD, and AIA neurons with mCherry. The FlincG3 signal in each of these neurons was significantly higher in bbs-9 mutants compared to wild type, indicating increased cGMP levels (Figs 3E, 3F, and S5E-S5H). These data suggest that bbs mutants inhibit the acute hypoxia response by promoting cGMP production.
To further investigate critical sites of cGMP signaling, we optogenetically manipulated intracellular cGMP levels using a guanylate cyclase rhodopsin from Blastocladiella emersonii, BeCyclOp, that enables rapid light-triggered increases in cGMP production [43]. Without blue light, and in the presence of the cofactor all trans-retinal (ATR), transgenic animals expressing BeCyclOp from the gcy-28.c (Fig 4C), or gpa-3 ( Fig 4D) promoters responded strongly to 1% O 2 , although a weak suppression was observed. By contrast, exposing these animals to blue light and 1% O 2 in the presence of ATR effectively suppressed acute hypoxia responses (Fig 4C  and 4D). To probe the temporal dynamics and reversibility of this manipulation, we kept animals in 1% O 2 and waited until the animals' speed plateaued, at which point light stimulation of BeCyclOp triggered an abrupt slowing followed by resumption of rapid movement when light was switched off again ( Fig 4E). In the absence of ATR, responses to 1% O 2 were unaffected in animals expressing BeCyclOp (S7A and S7B Fig), supporting the notion that heightened intracellular cGMP inhibits acute hypoxia responses. Intriguingly, BeCyclOp-mediated cGMP production in AIA neurons, but not in ADL and ASH neurons, significantly inhibited locomotory responses to acute hypoxia (Figs 4F and S7C), consistent with the gcy-28 overexpression data ( Fig 4B). To confirm the role of AIA neurons in acute hypoxia responses, we inhibited AIA neurons using a histamine-gated chloride channel (HisCl1) [48]. Inhibition of AIA neurons increased the basal speed at 7% O 2 while completely eliminating responses to 1% O 2 (Fig 4G), suggesting that AIA neurons are an important element of the circuit mediating responses to acute hypoxia.

BBSome acts in ciliated neurons to indirectly modulate cGMP production in AIA interneurons
The fact that cGMP levels are elevated in bbs-9 mutants, and that GCY-28 is a major contributor to cGMP production, prompted us to speculate that GCY-28 expression or localization might be altered in bbs-9 mutants. However, we observed reduced instead of increased expression of GFP-tagged GCY-28.C and GCY-28.D in bbs-9 mutants, and no obvious change in the subcellular localization of GCY-28-GFP when expressed either from a single copy MosSCI insertion (GCY-28.C) or from extrachromosomal arrays (GCY-28.D) (S8A and S8B Fig). It is possible that GCY-28 activity, not expression, is affected by loss of the BBSome. However, since GCY-28 is not localized to the cilia [47] and its main action site does not overlap with BBSome expressing neurons, its activity is unlikely to be directly targeted by the BBSome. Moreover, bbs-9 mutants showed increased cGMP production even in the absence of GCY-28 ( Fig 3D, columns 3 and 4), suggesting that other guanylate cyclases may be activated in a bbs-9 mutant background. Supporting this, mutations in several other guanylate cyclase genes besides gcy-28 partially suppressed the hypoxia response defects of bbs-9 mutants (S5B Fig). Taken together, we suspect that GCY-28 may not be directly involved in abnormal cGMP generation in bbs-9 mutants. However, because GCY-28 is a major guanylate cyclase in cGMP production ( Fig 3D), loss of GCY-28 could effectively neutralize any increases in cGMP generated by other guanylate cyclases in bbs-9 mutants, thereby reversing their inhibition of acute responses to hypoxia.

PLOS BIOLOGY
How does the BBSome, which acts in ciliated neurons, modulate cGMP production in downstream circuits such as the AIA neurons? Disrupting daf-19, which encodes an RFX transcription factor essential for ciliogenesis [49], attenuated but did not abolish hypoxia responses (S1 Table). This makes it unlikely that loss of acute responses to hypoxia in bbs-9 mutants reflects loss of cilia function. An alternative hypothesis, supported by previous studies, is that bbs mutants have acquired ectopic signaling capability, perhaps because they accumulate signaling components in their cilia due to defective retrograde traffic [50][51][52][53][54]. To test this, we asked if blocking accumulation of signaling molecules in cilia rescues the hypoxia response defects of bbs mutants. Indeed, eliminating the core IFT component OSM-6 suppressed the hypoxia response defects of bbs-9 to the levels found in osm-6 single mutants (Fig 5A).
Since bbs-9 hypoxia response defects were rescued by expressing bbs-9 cDNA in ADL and ASH neurons, we surmised that signals from these 2 neurons could be heightened in bbs-9 mutants. If this was the case, silencing ADL and ASH neurons should suppress the defects of bbs-9 mutants. Indeed, simultaneously but not individually inhibiting ADL and ASH neurons using the histamine-gated chloride channel HisCl1 partially restored locomotory responses to bbs-9 mutants (Fig 5B). By contrast, inhibiting or ablating these 2 neurons in the wild-type background did not affect responses to acute hypoxia (S8C and S8D Fig). These data suggest that abnormal signaling from ADL and ASH in bbs-9 mutants is at least partially responsible for loss of acute hypoxia responses in these animals.

BBSome modulates cGMP production through GPCR signaling
We next probed the nature of the ectopic signaling in bbs mutants in ADL and ASH neurons. Mutants lacking the Gi/Go-like protein GPA-3, similar to bbs mutants, have high cGMP levels [55]. Disrupting gpa-3 weakly reduced the locomotory response to hypoxia (Fig 5C). By assaying mutants in other Gα subunits, we found that simultaneously disrupting gpa-3 and odr-3 strongly attenuated responses to acute hypoxia ( Fig 5C). Mutations in gcy-28 or egl-4 suppressed this defect (Fig 5D and 5E) and, consistent with this, gpa-3 odr-3 double mutants showed significantly elevated cGMP (Fig 5F), suggesting that GPA-3 and ODR-3 redundantly inhibit cGMP production. We examined whether the BBSome and G proteins act in parallel or in the same pathway to modulate GCY-28 signaling. We found that simultaneously overexpressing gpa-3 and odr-3 restored hypoxia-evoked behavioral responses to bbs-9 mutants, suggesting that in genetic terms the BBSome acts upstream of G protein signaling to promote acute hypoxia responses ( Fig 5G). We next examined if GPA-3 and ODR-3 act in the same neurons as BBSome. The behavioral defects of gpa-3 odr-3 double mutants were partially rescued by expressing gpa-3 under the gpa-11 promoter (Fig 5H). These observations suggest that G proteins, similar to the BBSome, act in ADL and ASH neurons to modulate locomotory responses to hypoxia. Since both GPA-3 and ODR-3 are localized to sensory cilia, we tested whether this localization was affected in bbs-9 mutants. The expression of both G proteins was significantly decreased in bbs-9 mutants compared to wild type, both in neuronal cell bodies and in cilia of ADL and ASH neurons (Fig 5I and 5J). ODR-3 expression in the cilia of AWA neurons showed the same tendency (S8E Fig). By contrast, and consistent with previous observations, levels of the ODR-10 olfactory GPCR in cilia increased >2-fold in bbs-9 mutants ( Fig  5K) [53]. Together, these data suggest that disrupting the BBSome alters G protein localization and signaling in ciliated neurons to modulate cGMP production in downstream circuits.

Abnormal neuroendocrine secretion disrupts acute hypoxia responses
BBSome mutants have been shown to enhance neuropeptide secretion from ADL neurons [51]. This increased neurosecretion can be abrogated by disrupting aex-6, the C. elegans ortholog of human RAB27, or by mutations in the RAB27 regulator RBF-1. These findings prompted us to ask if increased cGMP levels inhibit acute responses to hypoxia by promoting neurotransmission. Consistent with this, deleting aex-6 or rbf-1 restored acute hypoxia responses to bbs-9 mutants (Fig 6A and 6B). Moreover, enhanced secretion of the insulin DAF-28 and the FMRF-like peptide FLP-21 in bbs-9 mutants was suppressed by a mutation in gcy-28, although loss of gcy-28 did not alter release of these peptides in WT (Fig 6C-6F). In addition, the defective acute hypoxia responses in animals overexpressing gcy-28.c under its endogenous promoter were suppressed by mutations in rbf-1 (Fig 6G). These observations suggest that increased cGMP signaling enhances neuroendocrine secretion, which, in turn, inhibits behavioral responses to 1% O 2 .

Hypoxia-induced neuronal inhibition
Our data suggest that despite the similarity of the behavioral responses, the neural circuits mediating escape from 1% and 21% O 2 are distinct. Whereas 21% O 2 evokes a tonic rise in Ca 2+ in RMG interneurons, which is sufficient to drive rapid movement [14,16], 1% O 2 evoked a decrease of Ca 2+ levels in RMG (Fig 7A). We sought neurons that respond to hypoxia, using the ratiometric Ca 2+ sensor YC3.60. None of the amphid neurons we imaged showed a hypoxia-evoked rise in Ca 2+ . However, we observed stimulus-evoked decreases of Ca 2+ in ASK, AFD, and ADL neurons, suggesting neuronal hyperpolarization, which was frequently followed by a poststimulus Ca 2+ "rebound" (Fig 7B-7D). We observed a rapid bleaching of the calcium sensor when imaging ADL neurons, which made the hypoxia-evoked Ca 2+ decrease less obvious in these neurons. However, a clear rise of Ca 2+ at 7% O 2 suggests that there is an intrinsic hypoxia response in ADL neurons. Ca 2+ transients evoked by hypoxia were not observed in ASH neurons (Fig 7E), suggesting that hypoxia did not trigger a general decrease of Ca 2+ in neurons. This is consistent with our cell ablation data, suggesting that ASH neurons are not involved in the sensing step of hypoxia. To probe the circuitry underlying hypoxia responses further, we monitored Ca 2+ responses evoked in the amphid interneurons AIB, AIY, and AIA, by shifting ambient O 2 from 7% to 1%. We detected clear responses only in the AIA neurons, which also appeared to be inhibited by a 1% O 2 stimulus (Fig 7F). This observation is consistent with our finding that stimulating cGMP production specifically in AIA neurons inhibits hypoxia responses. However, we do not exclude the possibility that the Ca 2+ responses are indirect consequences of hypoxia such as altered pH values and ATP production.

Defects in the mitochondrial electron transport chain disrupt acute hypoxia responses
Assaying mutants with compromised mitochondrial function identified 2 additional components, GAS-1 and ISP-1, required for acute responses to hypoxia (S1 Table). gas-1 is the

PLOS BIOLOGY
Acute hypoxia responses in C. elegans C. elegans ortholog of Ndufs2, which contributes to the ubiquinone binding site of MCI. Previous work has suggested that NDUFS2 is required for acute hypoxia sensing in the carotid body [6]. In C. elegans, mutations in gas-1 entirely eliminated acute locomotory responses to hypoxia, and this defect could be rescued by expressing gas-1 cDNA under its endogenous promoter (Fig 8A). isp-1 encodes C. elegans ortholog of Rieske protein, the iron sulfur protein of MCIII. Disrupting Rieske protein in pulmonary vascular smooth muscle cells inhibits acute responses to hypoxia [56]. We found that isp-1 mutants in C. elegans were strongly defective in locomotory responses to acute hypoxia ( Fig 8B). However, repairing the point mutation of isp-1(qm150) by CRISPR, or adding fosmid transgenes bearing a wild-type isp-1 gene, only partially rescued the hypoxia response phenotype, suggesting that a background mutation enhanced the hypoxia response defect in this isp-1 mutant strain (S9A and S9B Fig). To further evaluate the role of ISP-1 in behavioral responses to hypoxia, we expressed sense and antisense RNAi constructs to silence isp-1 expression [35]. RNAi knockdown of isp-1 using the isp-1 or myo-2 promoters led to lethality, showing the efficacy of the RNAi strategy. RNAi knockdown of isp-1 in the nervous system, but not in other tissues, inhibited escape from 1% O 2 without disrupting responses to 21% O 2 (Figs 8C and S9C-S9E). Thus, ISP-1 activity in neurons specifically promotes the acute hypoxia response.

PLOS BIOLOGY
MCI and MCIII are major production sites of ROS. The primary ROS generated in mitochondria is superoxide, which is detoxified to H 2 O 2 by mitochondrial superoxide dismutase (SOD). We measured in vivo ROS production by staining worms with dihydroethidium (DHE). Both gas-1 and isp-1 mutants, but not bbs-9, dyf-3 and pde-1,2,3,5 quadruple mutants, had significantly increased ROS levels (S9F and S9G Fig). Mitochondrial superoxide is selectively increased in isp-1 mutants [57]. To test whether superoxide levels help signal acute hypoxia, we examined acute hypoxia responses in mitochondrial SOD mutants. Loss of sod-2 partially inhibited acute responses to hypoxia, and a stronger inhibition was observed when all SOD genes were disrupted (S9H Fig). These data prompted us to test if increasing mitochondrial ROS production using pro-oxidant drugs altered acute hypoxia responses. Treatment with paraquat and juglone, which induce mitochondrial superoxide generation, blocked locomotory speed increase to 1% O 2 while responses to 21% O 2 were affected to a much lesser extent (Figs 8D and S9I). As paraquat and juglone may have other effects, we explored other superoxide inducing drugs. Rotenone is a strong inducer of superoxide [58][59][60][61] and is known to block acute hypoxia sensing by peripheral chemoreceptors in mammals [62][63][64]. Treatment with rotenone abolished the hypoxia-evoked increase in speed but had only a mild effect on the response to 21% O 2 (Fig 8E). These observations suggest persistently high ROS levels inhibit C. elegans' response to acute hypoxia.
The inhibition of hypoxia-evoked locomotory responses by high cGMP promoted us to ask if high mitochondrial ROS levels stimulate cGMP production. We found that both gas-1 and isp-1 mutants showed increased total cGMP (S10A Fig). Moreover, treatment with paraquat rapidly stimulated cGMP production in wild-type animals (S10B Fig). Pro-oxidant induced cGMP production was blocked by including the antioxidant N-acetyl cysteine (NAC) during paraquat exposure and was absent in GCY-28 mutants (S10C Fig), suggesting the ROS induced cGMP generation is GCY-28 dependent. The behavioral defects of gas-1 mutants were partially suppressed in a double mutant with gcy-28 (Fig 8F), suggesting that increased cGMP contributes to the phenotype of mitochondrial mutants. To explore if ROS acts in the same neurons as cGMP, we selectively inactivated ISP-1 by RNAi in different neurons. Expressing RNAi constructs from the gpa-3 promoter robustly inhibited behavioral responses to 1% but not to 21% O 2 (Figs 8G and S10D). Potent inhibition was also observed by using the gcy-28.c promoter to drive the RNAi constructs ( Fig 8H). In addition, silencing isp-1 in AIA neurons, but not in ADL and ASH, significantly reduced the locomotory arousal evoked by 1% O 2 , suggesting that ROS and cGMP act in the same neurons (Fig 8G and 8H). These results suggest mitochondrial ROS can inhibit acute hypoxia responses by stimulating cGMP signaling in a GCY-28-dependent manner.

Discussion
Here, we characterize the acute locomotory responses of C. elegans to a hypoxic (1% O 2 ) environment. A rapid decrease in O 2 to hypoxic (1%) levels triggers a bout of reorientation events, as animals seek to avoid hypoxia, followed by a large increase in locomotory activity that is sustained as long as animals remain in hypoxia. These behavioral responses to hypoxia occur within seconds and are remarkably similar to those evoked by 21% O 2 , a signal of surface exposure that is also a noxious cue for C. elegans. Notwithstanding the similar responses, the known O 2 -sensing neurons AQR, PQR, URX, and BAG, and the downstream interneurons RMG, which mediate escape from 21% O 2 , are not required for escape from hypoxia. Moreover, disrupting all 7 C. elegans soluble guanylate cyclases, GCY-31-37, which are thought to bind O 2 at their heme co-factor, or the cGMP-gated channels TAX-4/TAX-2 that transduce O 2 -evoked cGMP signals to drive escape from 21% O 2 , does not disrupt hypoxia-evoked escape responses. These results suggest the neuronal and molecular substrates underlying responses to 1% and 21% O 2 are different. They also suggest that the hypoxia-sensing mechanism mediating C. elegans escape from 1% O 2 differs from that used by Drosophila larvae, which depends on soluble guanylate cyclases [65].
C. elegans' acute responses to hypoxia are, however, inhibited by increased cGMP signaling through PKG. Most notably, rapid generation of cGMP using optogenetics is sufficient to prevent C. elegans arousal in response to 1% O 2 . These observations raise the possibility that C. elegans responds to acute hypoxia by directly and rapidly inhibiting cGMP production. However, other observations argue against this hypothesis. If the locomotory responses to hypoxia were triggered by a decrease in cGMP, we might expect that mutants defective in cGMP signaling would display constitutive high locomotory activity at preferred O 2 levels, as animals tonically signal that they are in constant hypoxia. However, both egl-4 and gcy-28 mutants show normal locomotory activity at 7% O 2 , suggesting that reduced cGMP signaling is not sufficient to evoke escape behavior. Moreover, the fact that none of the guanylate cyclase mutants show defective responses to hypoxia suggests GCYs are not essential for sensing acute hypoxia, although we cannot rule out redundancy between cyclases. Thus, we favor a model in which cGMP is a modulator rather than the primary signal of acute hypoxia responses. Our observation is consistent with the inhibitory role of cGMP in acute hypoxia responses in mammalian chemosensory organs [9][10][11][12]. It also echoes a gasotransmitter model, in which high cGMP and high PKG activity inhibit the synthesis of H 2 S to prevent acute responses to hypoxia [66]. All these suggest a conserved role of cGMP/PKG in modulating acute hypoxia sensing.
In C. elegans many cGMP signaling molecules, e.g., most guanylate cyclases, are concentrated in sensory cilia. However, the BBSome does not appear to directly regulate traffic of GCY-28, a guanylate cyclase that contributes disproportionately to cGMP production in both wild-type and bbs mutants. Unlike many other guanylate cylases, GCY-28 does not localize to cilia [47] and its expression is not increased, nor its localization altered, in bbs mutants (S8A and S8B Fig). The defects of bbs-9 mutants are rescued by expressing bbs-9 cDNA in ADL and ASH neurons that do not reliably express GCY-28. Moreover, increasing cGMP production selectively in ADL and ASH, either optogenetically or by overexpressing gcy-28, is not sufficient to inhibit hypoxia-evoked locomotory responses. Together, these data suggest that GCY-28 and the BBSome act in different neurons to regulate hypoxia responses. The BBSome may instead modulate acute responses to hypoxia by regulating the function of the GPA-3 and ODR-3 G proteins. Supporting this, GPA-3 and ODR-3 act in the same neurons as BBSome components to promote escape from hypoxia. Moreover, the expression of GPA-3 and ODR-3 is down-regulated in ADL and ASH neurons in bbs-9 mutants (Fig 5I and 5J), and overexpressing GPA-3 and ODR-3 partly restores hypoxia responses to bbs-9 mutants. The phenotypic suppression of G protein mutants by disrupting cGMP signaling places cGMP further downstream of, or possibly in parallel to, ODR-3 and GPA-3 G protein signaling. Consistent with this, cGMP production in several neurons, including AIA interneurons, can inhibit acute hypoxia responses. How signals from ADL and ASH neurons interact with signals from cGMP producing neurons to regulate hypoxia responses is unclear.
Mitochondria have historically been regarded as good candidates for harboring the mechanism for acute hypoxia sensing. In glomus cells of the carotid body, increased ROS generation is thought to regulate opening of adjacent K + channels in the plasma membrane, thereby signaling drops in O 2 tension and establishing an integrative mitochondria-to-membrane signaling model of acute hypoxia sensing [67][68][69]. Precisely how mitochondrial ROS alters the open probability of K + channels in the plasma membrane remains to be elucidated. In both carotid body and smooth muscle cells of pulmonary arteries, high ROS levels in MCI and MCIII mutants prevent further ROS increase upon acute hypoxia, thereby inhibiting the responses to the drop of O 2 tension [1,6,8]. Our work shows that similar manipulations in C. elegans eliminate locomotory responses to hypoxia, suggesting that molecular mechanism of acute hypoxia sensing may be conserved from C. elegans to mammals. We also provide insight into how increased ROS levels inhibit the responses to acute hypoxia. Both genetic and pharmacological perturbations of ROS production increase intracellular cGMP levels, raising the possibility that ROS inhibits acute hypoxia sensing by stimulating cGMP production. This is supported by the fact that ROS act in the same cells as cGMP, and the defects of gas-1 mutants in acute responses to hypoxia are partially suppressed by gcy-28 mutants. These data suggest a mechanism by which increased ROS can inhibit acute hypoxia sensing.
Our work begins to shed light on acute hypoxia responses in C. elegans. A rapid drop in O 2 is detected by multiple sensory neurons, which transmit the signals to the downstream circuits including the AIA interneurons to generate escape responses (Fig 9A). The hypoxia sensing circuit can be potently dampened by increased cGMP production (Fig 9B). Even though ADL and ASH neurons are not directly involved in sensing hypoxia, disruption of the BBSome generates or enhances an unknown signal from these 2 neurons to stimulate GCY-28-mediated cGMP production in the hypoxia sensing circuit, thereby inhibiting acute responses to hypoxia (Fig 9B). Many questions still remain to be addressed. We implicate mitochondria and ROS in the sensing mechanism, but have not demonstrated that ROS increases robustly and persistently upon hypoxia exposure, to explain the persistent behavioral arousal evoked in C. elegans by hypoxia. While we show that hypoxia elicits a decrease in neuronal Ca 2+ in several neurons, we have not identified the ion channels involved. Nevertheless, the robust behavioral response we describe now allows us to bring to bear the powerful genetics available in C. elegans to delineate the molecular details of acute hypoxia sensing in this animal.

Strains
C. elegans were maintained using standard protocols [70]. Strains used are listed in S2 Table.

Behavioral analysis
C. elegans locomotion assays were performed as described previously [16]. Briefly, assay plates were seeded with OP50 and grown overnight. To prepare the bacterial lawn for our assays we used a PDMS stamp that had an approximately 0.8 cm × 0.8 cm square cut away from its center. PDMS stamp was made by mixing silicone elastomer base with the curing agent at the ratio of 10:1 (Sylgard 184 silicone elastomer kit) and cured for >1 day in a 15-cm petri dish. Pressing the flat side of this stamp against the bacterial lawn removed the border, leaving an approximately 0.8 cm × 0.8 cm bacterial patch that perfectly fitted our microfluidic chambers. The chamber was placed on top of 25 to 30 young adults and defined gases mixtures delivered into the chamber at a flow rate of 3 ml/min using a PHD 2000 Infusion syringe pump (Harvard Apparatus, United States). Telfon valves (AutoMate Scientific, United States) were used to rapidly switch between 2 gas mixtures (7% and 0%, 7% and 1%, or 7% and 21% O 2 ) from the syringe pump. The closing and opening of valve channels were operated by ValveBank Perfusion Controller (AutoMate Scientific). The O 2 concentration in the chamber was measured using the Fibox sensor system (Pre-Sens, Germany). We placed an oxygen sensor spot (PreSens) inside the chamber and monitored the O 2 levels as we pumped in gases. To track animals' locomotory responses to O 2 stimulation, a Grasshopper camera (FLIR, United States) was mounted on a Zeiss Stemi 508 microscope. Videos were acquired at 2 frames/second for 2 minutes at each O 2 concentration and analyzed using Zentracker, a custom-written MATLAB program (https://github.com/wormtracker/zentracker). Each data point represents at least 3 assays for each genotype.
To examine the effects of paraquat, rotenone and juglone on acute responses to hypoxia, day 1 adult animals were exposed to 200 mM paraquat for 10 minutes, 10 μM rotenone for 1 hour, and 1 mM juglone for 2 minutes, on drug-containing plates in the presence of food; treated animals were then assayed on regular assay plates.
Optogenetic experiments followed published protocols [16,43]. Transgenic L4 animals were picked to plates either with or without 100 μM ATR (R2500, Sigma, United States), and kept in the dark for 16 to 20 hours. Blue light illumination was from an ultra-high-power LED lamp (UHP--MIC-LED-460, Prizmatix, Israel), attenuated to 70 μW/mm 2 . Light intensity was measured by a PM50 Optical Power Meter (ThorLabs, United States). The transmitted light used for worm picking and video recording was filtered through a long-pass optical filter that blocked light with wavelengths shorter than 595 nm. Videos were captured and analyzed as for regular behavioral assays.
Neuronal inhibition using the histamine gated chloride channel HisCl1 was performed as described [48]. All HisCl1 constructs were injected at 10 ng/μl. To inhibit AIA interneurons, animals at L4 stage were picked to fresh plates containing 10 mM histamine (Sigma, H7250) and grown for 10 minutes. To inhibit the other neurons, animals were grown on histamine containing plates for both 10 minutes and 24 hours. Assays were performed using day 1 adults, on assay plates containing 10 mM histamine. Videos were acquired and analyzed as for regular behavioral assays.

CRISPR/Cas-9-based genome editing
Genome editing with CRISPR/Cas-9 was performed as described [71]. To generate gene knockouts, we used a single-strand DNA oligo (ssODN) repair template that targeted sequences early in the coding region of the gene. The ssODN contained 2 homology arms, each of 35 bases that flanked either side of the targeted PAM site while deleting 16 bases of coding sequence to generate a frameshift mutation. In addition, we introduced 2 stop codons and a unique restriction site for genotyping between the 2 flanking sequences. To generate point mutations, we chose a PAM site close to the codon of interest and used a ssODN donor that included 35 bases homologous to sequences on either side of the targeted codon. If the PAM site was not close to the codon, the homology arms were extended accordingly. GFP knockins were generated by using double strand DNA repair templates [72]. GFP specific oligos tailed with 35bp homology arms were used to amplify gfp coding sequence. The template was melted as described to improve editing efficiency. Moreover, 0.5 μl of Cas-9 protein (IDT) was mixed with 5 μl of 0.4μg/μl tracrRNA (IDT, United States) and 2.8 μl of 0.4 μg/μl crRNA (IDT). The mix was incubated at 37˚C for at least 10 minutes before adding 2.2 μl of 1μg/μl ssODN (or 500 ng dsDNA) and 2 μl of 0.6 μg/μl rol-6 co-injection marker. Nuclease-free water was used to bring the final volume to 20 μl. The injection mixture was centrifuged for at least 2 minutes before use. Large deletions were generated with an injection mix that included 2 crRNA guides targeting the start and the end of the gene. Moreover, 0.5 μl of Cas-9 protein, 5 μl of 0.4μg/μl tracrRNA, and 1.4 μl of 0.4 μg/μl each of 2 crRNA guides were mixed and incubated at 37˚C. Furthermore, 2 μl of 0.6 μg/μl rol-6 co-injection marker and nuclease-free water was added after incubation to bring the final volume to 20 μl.

Microscopy
FlincG3 was used to report neuronal cGMP levels; in these experiments, we performed pseudoratiometric imaging by collecting both GFP (from FlincG3) and free mCherry signals. Transgenic animals at L4 stage were picked and grown overnight under standard laboratory conditions. Day 1 adult animals were mounted on 2% agarose pads containing 5 mM levamisole. Z-stack confocal Images of 20 to 50 animals were collected using a 63× oil objective N.A 1.3 on a Nikon A1 confocal microscope with Nikon NIS elements software. A similar procedure was followed to quantify the neuropeptide secretion. Only the signals from posterior pair of coelomocytes were collected. The mCherry signal of neuropeptides and GFP signals from Punc-122::gfp were captured simultaneously for pseudoratiometric comparison. Signals from 20 to 50 animals were collected. For GPCR, G protein and gcy-28 expression, the absolute GFP intensity was calculated and compared among different genotypes. All images were processed in Fiji by subtracting background, adjusting threshold, and analyzing the particles in the region of interest.

Dye filling
Synchronized 1-day adults were soaked for 1 hour in 1 ml of DiO or DiI (Molecular Probes, United States), diluted 1:200 in M9 from a 2 mg/ml stock solution in DMF. The worms were washed 3× with 1 ml M9 and destained on seeded plates for > 2 hours. DiO and DiI uptake was monitored by mounting live animals for microscopy and imaging as described above.

DHE staining
Synchronized 1-day adults were washed 3× in PBS and stained in 3 μM of DHE (diluted from a 30 mM DHE stock solution in DMSO) for 30 minutes. Worms were washed twice with 1ml PBS and immobilized on a 2% agarose pad for confocal imaging. Images were acquired and processed as described above.

cGMP quantification in worm lysates
Total cGMP in C. elegans lysates was measured using a direct cyclic GMP enzyme immunoassay kit (Arbor Assays, United States). Moreover, 3 to 4 plates of synchronized young adult animals were harvested and washed 3× with PBS. The worm pellet was washed again with 1 ml 0.1 M HCl. A total of 50 μl of worm pellet was resuspended in 400 μl 0.1M HCl and frozen in liquid nitrogen. For worms treated with paraquat, animals were exposed in the paraquat solution after PBS washing. Paraquat was immediately replaced with PBS, and the worms washed quickly 3× with PBS before resuspension in 0.1 M HCl. Worms were disrupted using a beadbeater with 0.5-mm glass beads. The worm lysate was centrifugated at 2,500 rpm for 2 minutes. Protein concentration was determined using the Quick start Bradford dye reagent (Bio-Rad, United States). cGMP in 20 μl of lysate was quantified by following the acetylation protocol from Arbor Assays. The optical density at 450 nm was read using a ClarioStar 96-well microplate reader (BMG Labtech, Germany). The cGMP concentration was calculated using a curve generated from cGMP standards in the kit and normalized to sample protein concentration.

Ca 2+ imaging
Ca 2+ imaging of immobilized animals was performed using YC2.60 (RMG neurons) and YC3.60 (ASK, AFD, and AIA neurons) on an inverted microscope (Axiovert, Zeiss, Germany), with a 40× 1.2 NA C-Apochromat lens, and MetaMorph acquisition software (Molecular Devices, United States). Worms were glued to agarose pads (2% in M9 buffer, 1 mM CaCl 2 ) using Dermabond tissue adhesive with their nose and tail immersed in a mix of OP50 and M9 buffer. The glued animals were sealed in an airtight microfluidic chamber, and the O 2 was delivered and measured as described in the behavioral analysis. Recordings were carried out at 2 frames/sec with an exposure time of 100 ms using MetaMorph software. Photobleaching was minimized using 2.0 or 1.5 optical density filters. An excitation filter (Chroma, United States) restricted illumination to the cyan channel, and a beam splitter (Optical Insights, UK) was used to separate the cyan and yellow emission light. Neuron Analyzer, a custom-written MATLAB program, was used to analyze the resulting image stacks (https://github.com/ neuronanalyser/neuronanalyser) [16].