GLOBIN-5-Dependent O2 Responses Are Regulated by PDL-1/PrBP That Targets Prenylated Soluble Guanylate Cyclases to Dendritic Endings

Aerobic animals constantly monitor and adapt to changes in O2 levels. The molecular mechanisms involved in sensing O2 are, however, incompletely understood. Previous studies showed that a hexacoordinated globin called GLB-5 tunes the dynamic range of O2-sensing neurons in natural C. elegans isolates, but is defective in the N2 lab reference strain (McGrath et al., 2009; Persson et al., 2009). GLB-5 enables a sharp behavioral switch when O2 changes between 21 and 17%. Here, we show that GLB-5 also confers rapid behavioral and cellular recovery from exposure to hypoxia. Hypoxia reconfigures O2-evoked Ca2+ responses in the URX O2 sensors, and GLB-5 enables rapid recovery of these responses upon re-oxygenation. Forward genetic screens indicate that GLB-5's effects on O2 sensing require PDL-1, the C. elegans ortholog of mammalian PrBP/PDE6δ protein. In mammals, PDE6δ regulates the traffic and activity of prenylated proteins (Zhang et al., 2004; Norton et al., 2005). PDL-1 promotes localization of GCY-33 and GCY-35, atypical soluble guanylate cyclases that act as O2 sensors, to the dendritic endings of URX and BAG neurons, where they colocalize with GLB-5. Both GCY-33 and GCY-35 are predicted to be prenylated. Dendritic localization is not essential for GCY-35 to function as an O2 sensor, but disrupting pdl-1 alters the URX neuron's O2 response properties. Functional GLB-5 can restore dendritic localization of GCY-33 in pdl-1 mutants, suggesting GCY-33 and GLB-5 are in a complex. Our data suggest GLB-5 and the soluble guanylate cyclases operate in close proximity to sculpt O2 responses.


Introduction
O 2 's importance for respiration, and its destructive potential via ROS, has led animals to evolve behavioral, physiological, and cellular responses to acute and chronic changes in O 2 . The O 2sensing mechanisms mediating these responses are incompletely understood. In mammals changes in O 2 evoke rapid responses in specialized tissues, e.g., the carotid bodies and the pulmonary neuroepithelium (Ward, 2008). Tuning of these O 2 responses varies according to the O 2 tension experienced normally by the tissue. The molecular identity of acute O 2 sensor(s) in different mammalian tissues, however, remains elusive. Whether there is one, or several sensors, and whether different sensors work in concert is uncertain.
O 2 -sensing enables natural C. elegans isolates to escape 21% O 2 , accumulate where bacteria is thickest, usually at the lawn border, and to settle on food when O 2 drops to 19% (Busch et al., 2012). During domestication the standard N2 (Bristol) laboratory strain lost these behaviors, due to a gain-offunction mutation in a neuropeptide receptor, npr-1 F215V, and a loss-of-function mutation in glb-5, glb-5 (McGrath et al., 2009;Weber et al., 2010).
Here, we explore the functions of GLB-5 neuroglobin further. Since globins are proposed to protect neurons against hypoxia, we ask whether a functional glb-5 allele from a Hawaiian wild strain, glb-5(Haw), alters C. elegans' responses to hypoxia/reoxygenation stress. We show that without glb-5 hypoxia experience strongly attenuates both O 2 -evoked Ca 2ϩ responses in URX neurons and bordering behavior. GLB-5(Haw)'s ability to influence O 2 responses depends on a prenyl binding protein, PDL-1/ PrBP, which promotes traffic of the soluble guanylate cyclases to dendritic endings. Our data suggest GLB-5 and the soluble guanylate cyclases operate in close proximity. However, dendritic localization is not essential for GCY-35/GCY36 to respond to O 2 variation.

Behavioral assays
Bordering experiments were performed by transferring 40 L4 hermaphrodites (grown in 21% O 2 at room temperature, RT) to a 5 cm NGM plate that had been seeded 2 d before with OP50 bacteria. Assay plates were put in 1% O 2 (Coy hypoxia chamber; Coy Lab Products) RT for 3-18 h (as indicated), brought back to 21% O 2 , and the percentage of animals bordering calculated after 30 min or longer, as indicated. The bordering index is the fraction of worms found on the bacterial lawn border divided by the total number of worms on the plate, multiplied by 100. To quantify speed 10 -15 animals were picked onto a 3.5 cm Petri dish containing NGM agar that was seeded 2 d before with Escherichia coli OP50. We placed the Petri dish into a custom-designed Perspex chamber (48 ϫ 48 ϫ 17 mm; w ϫ d ϫ h) containing an inner circular cavity designed to fit the 3.5 cm dish. Two metal pipes provided an inlet and an outlet into the chamber for gas supply. Gas was delivered via silicone tubing using a custom-built computer-controlled manifold. The manifold delivered a constant gas flow of 7 ml/min via gas mass flow controllers (FMA5510; Omega) from gas tanks containing defined O 2 :N 2 mixtures (British Oxygen Company), and enabled gas switches to be made at defined times. Behavior was recorded using a Grasshopper 20S4M-C CCD camera (Point Gray Research) mounted onto a Leica dissecting microscope using an appropriate C-mount adapter (Leica) at 2 frames/s. Videos were analyzed using Zentracker, a custom software that tracks the center of mass of each animal and calculates instantaneous speed. The software is available on request.

Ethyl methanesulfonate mutagenesis screen
We mutagenized glb-5(Haw); npr-1 worms as described previously (Brenner, 1974). Two-hundred and fifty mutagenized P o hermaphrodites were divided into 50, 9 cm seeded plates. After 4 d, F1 animals (ϳ6000 gravid hermaphrodites) were divided into 60, 9 cm seeded plates and allowed to lay eggs for 2 h. To kill most of the F1 worms without harming the eggs, 3 ϫ 25 l drops of chloroform were put on the lid of each plate. After 3 d, we transferred ϳ250 aggregating L4 worms from each plate to a new seeded plate and incubated them in 1% O 2 for 18 -20 h. Then, after a 30 min incubation in 21% O 2 , we transferred worms that were not on the bacterial lawn border to new plates, and repeated the assay. Nonbordering worms were singled into 3.5 cm plates. Putative mutants were re-assayed and lines in which we verified phenotypes kept.

Mapping
We mapped db805 to a 4 Mb genomic interval using single nucleotide polymorphisms (SNPs; Davis et al., 2005). To identify the gene disrupted by db805, we prepared amplification-free genomic libraries from db805 mutant DNA (Kozarewa et al., 2009;Weber et al., 2010) and sequenced them with the Illumina Genome Analyzer II. Sequence analysis was performed as described previously (Davis et al., 2005).
Genotyping. The pdl-1(db805) allele was followed by amplifying and sequencing the interval flanking the splice-donor region of exon 3. The gk157 deletion was followed by PCR using primers that flank the deletion. The glb-5(Bri) mutation was tracked by examining the size of PCR fragments amplified with primers that flank the duplication in this allele (Persson et al., 2009). The npr-1(ad609) allele was followed by amplifying the interval containing its associated DNA lesions, and using restriction enzyme digestion to distinguish wild-type and mutant alleles.
pdl-1, gcy-35, and gcy-33 cDNA. We extracted total RNA from N2 worms by vortexing with TRIzol (Life Technologies) and acid-washed glass beads, and purifying using an RNeasy Midi Kit (Qiagen). cDNAs were generated by RT-PCR unless otherwise mentioned (OneStep RT-PCR kit; Qiagen). cDNA of gcy-33 was amplified from a cDNA clone kindly provided by Y. Kohara. Oligonucleotides were designed using the sequences in Wormbase (http://www.wormbase.org/).
Transgenes. Microinjections were performed as described previously (Mello et al., 1991). Plasmids (2.5-50 ng/l) were coinjected with fluorescent reporters, and multiple independent lines isolated. In brief, N2 genomic DNA was used to amplify the pdl-1 gene, including 3 kb of upstream sequences. The genomic fragment was cloned into pPD95.75 (A. Fire, personal communication), and then modified by adding an outron-mCherry fragment to make a polycistronic expression vector, as described previously (Coates and de Bono, 2002;Persson et al., 2009). This expression construct was modified to create other expression constructs. To create a pdl-1 cDNA expression construct, the genomic part of the pdl-1 gene was replaced with pdl-1 cDNA. To create constructs that expressed pdl-1 in the BAG or AQR, PQR, and URX neurons, the promoter region of pdl-1 was replaced with the 3.3 kb or the 1.3 kb promoter of flp-17 and gcy-37, respectively. To express pdl-1 in the BAG, AQR, PQR, and URX neurons, the promoter region of pdl-1 was replaced with the 3.1 kb promoter of glb-5. To generate a functional pdl-1::GFP expression construct, the pdl-1 stop codon was deleted by mutagenesis and sequences encoding GFP inserted instead of the outron-mCherry fragment to create an in-frame translational reporter gene. To create a functional gcy-35::gfp expression construct, we inserted the GFP reporter sequences after the codon encoding Ser671, thereby retaining intact N and C termini. Similarly, to create a functional gcy-33::gfp expression construct, we inserted the GFP reporter sequences after Ala900. To examine how hif-1 affected glb-5 responses to hypoxia, we injected a genomic copy of glb-5(Haw) into hif-1(ia4); npr-1(ad609) animals.

Y2H interaction assays
We used the Matchmaker Gold Yeast two-hybrid system (Clontech Laboratories) to investigate protein interactions with PDL-1. In brief, pdl-1 cDNA was cloned into pGBKT7, transformed into Y2HGold yeast cells, and grown on selective Trp plates (according to the manual from Clontech Laboratories). Positive and negative control vectors, pGBKT7-53 and pGBKT7-Lam, were transformed into Y2HGold yeast cells and grown under the same conditions. The cDNAs of gcy- 35, were cloned into pGADT7, transformed into yeast strain Y187, and grown on selective -Leu plates. The negative control vector, pGADT7-T, was also transformed into Y2HGold yeast cells and grown under the same conditions. Positive colonies from the -Trp and -Leu plates were mixed with a toothpick on a YPDA plate, and grown at 30°C overnight. Y2HGold/Y187 diploids were streaked onto -Trp/Leu double-selection plates and grown for 3-4 d at 30°C. Colonies were then grown overnight in liquid culture, after which 5 l of diluted and nondiluted diploid suspensions were put on Trp/Leu/Aureobasidin A/X-␣-Gal or -Trp/-Leu/-Ade/-His/Aureobasidin A/X-␣-Gal plates. Colony growth was monitored after 3-4 d to detect protein interaction.
For imaging experiments, we glued the worms to agarose pads (2% agarose in M9) using Dermabond tissue adhesive (Closure Medical). Immobilized worms were trapped inside a 500 m deep rectangular PDMS chamber. Alternatively, we anesthetized the worms using 10 mM levamisole and put them in a custom-made, airtight microfluidic chamber. Humidified gases were delivered to the microfluidic chambers using a PHD 2000 syringe pump (Harvard Apparatus) at a flow rate of 1 ml/min. Teflon valves, regulated by a ValveBank Controller (Automate Scientific), were used to rapidly switch between gas mixtures. Image analysis was performed using custom-written MATLAB software.

Confocal microscopy
We mounted worms on 2% agarose pads (in M9 buffer supplemented with 10 mM sodium azide), and imaged them on a Zeiss LSM 510 laser scanning confocal microscope using a Plan-Apochromat 63ϫ 1.4 NA objective (Zen operating software; Carl Zeiss). Image analysis was performed using ImageJ (Abramoff et al., 2004).

Results
The neuroglobin GLB-5 regulates C. elegans responses to hypoxic exposure Transcription of many globin genes, including some C. elegans globins, is induced by hypoxia (Sun et al., 2001;Hoogewijs et al., 2007). This regulated expression is proposed to protect against hypoxia/re-oxygenation damage (Burmester and Hankeln, 2009). To investigate if GLB-5 altered C. elegans responses to hypoxia/re-oxygenation stress, we exposed animals bearing the functional glb-5(Haw) allele found in natural isolates or the lossof-function glb-5(Bri) allele found in N2 Bristol lab strain, to 1% O 2 , returned them to 21% O 2 , and looked for differences in behavior. In our assays we used animals that also carried the npr-1 215F allele or the npr-1(ad609) functional null allele, since the npr-1 215V gain-of-function allele found in N2 suppresses the activity of some O 2 -sensing circuits (Gray et al., 2004;Cheung et al., 2005;Rockman and Kruglyak, 2009;Weber et al., 2010). We incubated the glb-5(Bri); npr-1(ad609) and glb-5(Haw); npr-1(ad609) animals (hereafter referred to as npr-1 and glb-5(Haw); npr-1 animals, respectively), and animals from the Hawaiian wild strain CB4856 (bearing the natural glb-5(Haw) and npr-1(215F ) alleles) in 1% O 2 for 3, 6, 9, and 18 h, transferred them to 21% O 2 , and examined their behavior after 30 min. When grown in 21% O 2 animals from all three strains gather where the bacteria grow thickest, at the border of the lawn ( Fig. 1A; de Bono and . Upon exposure to 1% O 2 , animals from each strain left the food and accumulated outside the bacteria. When returned to normoxia, glb-5(Haw); npr-1 and CB4856 animals immediately accumulated at the lawn border, regardless of time in hypoxia. In contrast, npr-1 animals incubated in hypoxia for 18 h showed little preference for the border when returned to 21% O 2 (Fig. 1B). The hypoxia-induced inhibition of bordering was reversed within 4 h of return to normoxia (Fig. 1C). Bordering reflects an O 2 preference, with [O 2 ] ϳ13% in the thicker bacteria at the border, and ϳ17% O 2 at the lawn center (Gray et al., 2004). These results suggest that GLB-5(Haw) signaling enables C. elegans exposed to hypoxia and returned to normoxia to remain sensitive to these O 2 differences and to border.

C. elegans PDE6␦/PrBP prenyl binding protein is required for GLB-5 function
How does GLB-5(Haw) alter O 2 sensing? To investigate this question we turned to forward genetics, and sought genes required for glb-5 signaling. Originally, we identified the glb-5(Haw) allele because it enables natural isolates of C. elegans to switch from slow to rapid movement on a bacterial lawn when O 2 rises from 17 to 21% (Persson et al., 2009). Although robust, this assay is too labor intensive for forward genetic screens. In contrast, identifying mutations that prevent glb-5(Haw); npr-1 animals from bordering after hypoxia exposure is amenable to this approach. We mutagenized glb-5(Haw); npr-1 animals and sought mutants that, after experiencing 1% O 2 , failed to border immediately after return to 21% O 2 . To exclude mutations that indiscriminately disrupted bordering, we selected lines that resumed bordering after recovering at 21% O 2 for 4 h. We isolated four mutants that satisfied these criteria. By combining Snip SNP mapping with Illumina whole genome sequencing (Wicks et al., 2001;Sarin et al., 2008) we mapped one allele, db805, to a 3.8 Mb genomic interval on chromosome II, and identified 10 mutations in this interval. Only one of these mutations altered a predicted open reading frame, disrupting a splice acceptor site in pdl-1 (Fig. 1D). pdl-1 encodes a 17 kDa protein orthologous to mammalian PDE6␦ (phosphodiesterase 6, ␦-subunit). PDL-1 and human PDE6␦ share 69% amino acid identity at the protein level.
PDE6␦ was originally identified as a subunit of PDE6 from bovine retina (Gillespie et al., 1989). Biochemical studies showed that it is a prenyl binding protein (PrBP) that can extract prenylated proteins such as small lipidated GTPases (e.g., Ras) from membranes, sequestering them in the cytoplasm, and facilitating their traffic to different membrane compartments . Mice lacking PDE6␦ are small but viable; they exhibit progressive cone-rod dystrophy and defective localization of prenylated rhodopsin kinase (GRK1) and PDE6 catalytic subunits to rod outer segments (Zhang et al., 2007).
To confirm the pdl-1 phenotype, we obtained a deletion allele, gk157, which removed most of the pdl-1 coding region and ϳ0.5 kb of the upstream promoter region. pdl-1( gk157 ) delayed the recovery of bordering in glb-5(Haw); npr-1 animals exposed to hypoxia more strongly than the pdl-1(db805) mutation ( Fig. 2A). Since pdl-1(gk157 ) is probably a null allele, we used it in subsequent experiments, unless otherwise indicated.

PDL-1 functions in the same neurons as GLB-5
Where does PDL-1 function to modify GLB-5-dependent behaviors? glb-5 is expressed in the AQR, PQR, URX, and BAG O 2 -sensing neurons. Previous work showed that selectively expressing a glb-5(Haw) transgene in AQR, PQR, and URX neurons confers on npr-1 animals the sharply tuned O 2 responses of glb-5(Haw); npr-1 animals (McGrath et al., 2009;Persson et al., 2009). To identify where PDL-1 is expressed, we generated a transgene in which genomic DNA for pdl-1 was fused in-frame with a sequence-encoding GFP (Fig. 2B). Using behavioral assays, we showed that this transgene was fully functional (Fig. 2C,F ). We then crossed it with a strain expressing a functional pglb-5::glb-5(Haw)::mCherry fusion construct, and looked for colocalization of green and red fluorescence in animals bearing both transgenes. pdl-1 and glb-5(Haw) were coexpressed in the AQR, PQR, URX, and BAG neurons (Fig. 2B). pdl-1 was also expressed in other neurons in the head and tail.
PDL-1 also acts in the AQR, PQR, URX, and BAG neurons to enable normoxia-cultivated glb-5(Haw); npr-1 animals to switch from rapid movement at 21% O 2 to slow movement at 17% O 2 . This phenotype could be rescued by expressing pdl-1 from its own promoter, or from the glb-5 promoter (Fig. 2F ). As for bordering behavior, expressing pdl-1 specifically in the AQR, PQR, and URX neurons, using the gcy-37 promoter, or only in BAG neurons, using the gcy-33 promoter, failed to rescue glb-5(Haw)dependent responses to a switch from 21 to 17% O 2 . In contrast, coexpressing the two transgenes rescued the behaviors (Fig. 2F ). To confirm our results, we made RNAi constructs under the control of the gcy-37 promoter that targeted pdl-1 or GFP sequences. We injected these into pdl-1; glb-5(Haw); npr-1 mutant animals that expressed a pdl-1(cDNA)-GFP fusion construct from the pdl-1 promoter. Whereas animals not expressing the RNAi constructs strongly reduced movement when switched from 21 to 17% O 2 , animals expressing the RNAi transgenes did not (Fig. 2G). These data suggest that PDL-1 acts in AQR, PQR, URX, and BAG neurons to promote glb-5(Haw)-dependent O 2 responses in behavioral paradigms that operate at very different timescales.  . These heme proteins are thought to directly bind O 2 , andO 2 binding is thought to modulate their cGMP production and to change neural activity by gating cGMP ion channels (Couto et al., 2013). The functional importance of the CAAX motif has been directly tested for the GCY-36 soluble guanylate cyclase. Mutating sequences encoding the CAAX motif to encode SAAX disrupted GCY-36 localization to dendritic endings and resulted in a nonfunctional gcy-36 gene (Cheung et al., 2004).

Zimmer
To examine if PDL-1 regulates trafficking of GCY-35 and GCY-33, we made transgenes that expressed functional fluorescently tagged versions of these proteins. Tagging GCY-35 N terminally or C terminally interfered with its biological activity. We therefore inserted gfp cDNA into sequences encoding a predicted flexible loop in the GCY-35 C-terminal tail. Like GLB-5(Haw), GCY-35-GFP was enriched at the dendritic tips of URX neurons (Fig. 3A). Moreover, this transgene largely (although not completely) restored O 2 control of locomotor activity to gcy-35; npr-1 mutants (Fig. 3B). We compared the accumulation of GCY-35::GFP in the dendrite and cell body of URX neurons in gcy-35; npr-1 and gcy-35; pdl-1; npr-1 animals (Fig. 3C,D). While fluorescence in the cell body was very similar in the two genotypes, loss of pdl-1 essentially abolished GCY-35-GFP accumulation at the tip of URX dendrites. Thus, pdl-1 is required for correct localization of GCY-35 at dendritic endings. However, localization of GCY-35 to dendrites is not essential for O 2 -evoked responses.
To test if PDL-1 can directly interact with GCY-35, we performed yeast two-hybrid experiments (Y2H). We used PDL-1 as bait, mated it with a prey strain expressing GCY-35, and examined diploid growth on selective plates (Fig. 3F ). The rapid growth of the pdl-1/gcy-35 diploids suggested that GCY-35 and PDL-1 physically interact. To determine whether this interaction required prenylation of GCY-35, we disrupted the CAAX prenylation motif by a cysteine-to-serine codon change. PDL-1 failed to interact with this mutant form of GCY-35. These results suggest that PDL-1, like its mammalian counterpart, is a prenyl binding protein involved in protein transport. More specifically, it is required to traffic GCY-35 to the nonciliated dendritic endings of URX by a prenylation-dependent physical interaction.
Mammalian PrBP/PDE6␦ can also interact with the nonprenylated Arf-like (Arl) small GTPases, Arl2-GTP and Arl3-GTP. We therefore used the Y2H assay to examine if PDL-1 can also interact with arl-3 and arl-13. No significant growth was observed (Fig. 3F ). Interestingly, we did observe a weak but reproducible interaction with the rac-2 GTPase (Fig. 3F ).
Our expression studies and cell-specific rescue experiments suggested that pdl-1 functions not only in AQR, PQR, and URX but also in BAG sensory neurons (Fig. 2C,F ). BAG neurons express GCY-33, a soluble guanylate cyclase that also has a C-terminal CAAX prenylation motif. GCY-33::GFP accumulated strongly at the dendritic endings of BAG (Fig. 3I ). This dendritic fluorescence was reduced in pdl-1; gcy-33; npr-1 mutants although not abolished (Fig. 3 I, J ), suggesting that PDL-1 contributes to but is not solely responsible for localizing GCY-33 to BAG dendritic endings. Levels of GCY-33::GFP in the cell body were not altered in pdl-1 mutants (Fig. 3J ).

5(Haw)
; npr-1 animals (Fig. 2 D, F ). Our results suggest that localizing the GCY-35 O 2 sensor to the dendritic endings of URX is not essential for transducing O 2 stimuli, but that PDL-1 activity modifies O 2 -evoked Ca 2ϩ signaling in URX.

GLB-5 and PDL-1 modify O 2 -evoked Ca 2؉ responses in BAG neurons
We next investigated how the glb-5(Haw) allele and pdl-1 contributed to O 2 -evoked Ca 2ϩ responses in BAG neurons in normoxia-grown animals. Previous work showed that a fall in O 2 concentration evokes a rise in Ca 2ϩ in BAG neurons . Our behavioral experiments suggested BAG neurons were part of the circuit that enables feeding glb-5(Haw); npr-1 animals to reduce their locomotor activity sharply when O 2 decreases from 21 to 17% (Fig. 2F ). To monitor BAG Ca 2ϩ we expressed YC2.60 using the flp-17 promoter. We measured Ca 2ϩ changes evoked by a series of O 2 downsteps in well fed animals in the absence of food. We could detect small but robust rises in Ca 2ϩ when O 2 fell from 21 to 17%; as expected Ca 2ϩ responses became bigger as we reduced O 2 levels further (Fig. 5A). Thus, in glb-5(Haw); npr-1 animals even small decreases in O 2 evoked an increase In BAG Ca 2ϩ .
The small size of the Ca 2ϩ response evoked in BAG by a switch from 21 to 17% O 2 (Fig. 5A) precluded us from genetically dissecting this response. Instead, we studied responses evoked by a switch from 21 to 7% O 2 . As expected from previous work , O 2 -evoked Ca 2ϩ responses in BAG were eliminated in mutants defective in gcy-33 (Fig. 5B).

Hypoxia modifies O 2 -sensor physiology
Having explored how GLB-5(Haw) and PDL-1 activity shaped the O 2 responses of BAG and URX neurons in animals grown in normoxia, we explored how these genes influenced the physiology of O 2 -sensing neurons after hypoxia experience. We kept animals in 1% O 2 for 18 h and then imaged O 2 -evoked Ca 2ϩ responses. For these experiments, individual worms were transferred from the hypoxia chamber and rapidly glued to agarose pads in a 21% O 2 environment. All experiments were done in the presence of food as described previously, and begun within 5 min from the time of transfer. For URX, we chose a 21%-17%-21% O 2 stimulus train to evoke neuronal activity, since it mimics the O 2 concentration difference between the bacterial lawn (ϳ17% O 2 ) and the surrounding agar (ϳ21% O 2 ) in the bordering assay plates. Ca 2ϩ Figure 6. O 2 -evoked Ca 2ϩ responses in URX neurons after prolonged hypoxia. Ca 2ϩ responses evoked by 21%-17%-21% O 2 stimulus trains in URX after 18 h incubation at 1% O 2 measured using YC2.60. A, B, Prolonged hypoxia abolishes URX responsiveness to changes in O 2 in npr-1 animals (A), but has much weaker effects on glb-5(Haw); npr-1 animals (B). C, Deleting pdl-1 reduced URX responsiveness of glb-5(Haw); npr-1 animals in both normoxia and hypoxia cultivated animals, and led to higher Ca 2ϩ levels at 17% O 2 . D, Bar graph comparing tonic URX Ca 2ϩ levels at 21% O 2 , just after the switch from 17% O 2 to 21% O 2 , in various genotypes; data taken from A-C (21%-17%-21% O 2 responses after 18 h in hypoxia); baseline normalized ratios. Asterisks indicate significant differences compared with glb-5(Haw); npr-1. Mann-Whitney test, *p Ͻ 0.05. Error bars represent SEM. E, 21%-7%-21% O 2 stimulus trains evoked strong Ca 2ϩ responses in BAG in npr-1, ; npr-1 animals. F, The 21%-7%-21% O 2 -evoked Ca 2ϩ responses in BAG were smaller after hypoxia (for comparison see E). Asterisks indicate significant differences between Ca 2ϩ level just after the shift from 21 to 7% O 2 at times indicated by the bars, Mann-Whitney test. Gray shading represents SEM. G, Disrupting hif-1 did not prevent the fast recovery of bordering behavior in glb-5(Haw); npr-1 animals following exposure to hypoxia. Asterisks indicate significance for comparisons with glb-5(Haw); npr-1 animals. Kruskal-Wallis test with Dunn's post-test, n Ն 4, *p Ͻ 0.05, **p Ͻ 0.01, and ***p Ͻ 0.001. Error bars represent SEM. responses were undetectable in npr-1 animals following hypoxic cultivation (Fig. 6A). In contrast, glb-5(Haw); npr-1 animals retained strong O 2 -evoked Ca 2ϩ responses in URX after exposure to hypoxia (Fig. 6B). These data suggest that hypoxia/reoxygenation reconfigures the activity of URX in npr-1 animals, and that the glb-5(Haw) allele either suppresses this reconfiguration or enables rapid recovery. In hypoxia-conditioned pdl-1; glb-5(Haw); npr-1 animals, URX neurons responded less robustly than glb-5(Haw); npr-1 controls to O 2 stimuli (Fig. 6C,D), suggesting that the protective effect of GLB-5(Haw) is partly dependent on PDL-1. These imaging results are consistent with our behavioral observations that hypoxia experience inhibits bordering in npr-1 and pdl-1; glb-5(Haw); npr-1 but not glb-5(Haw); npr-1 animals (Fig. 2 A, C).
To explore if the physiology of BAG neurons was also reconfigured by hypoxia, we compared O 2 -evoked Ca 2ϩ responses in animals kept for 18 h at 21% or 1% O 2 (Fig. 6 E, F ). The 21%-7%-21% O 2 stimulus trains elicited smaller Ca 2ϩ responses in npr-1 and glb-5(Haw); npr-1 animals after exposure to hypoxia ( Fig. 6 E, F ). These results suggest that like in URX, hypoxia experience reconfigures BAG O 2 responses; however, unlike in URX, the presence of GLB-5(Haw) or absence of PDL-1 did not appear to have strong effects. One caveat is that expression of the YC2.60 sensor in BAG neurons was decreased in hypoxia-treated animals, complicating comparisons across O 2 conditions.

GLB-5(Haw) suppresses behavioral reconfiguration by hypoxia in an HIF-1-independent mechanism
The reconfiguration of bordering behavior by hypoxia occurred over hours (Fig. 1B), suggesting that transcriptional regulation is involved. Since HIF-1 is a key factor in adaptation to hypoxia (Semenza and Wang, 1992;Jiang et al., 2001), we wondered if HIF-1 was involved. To test this, we examined the bordering behavior of hif-1(ia4); npr-1; pglb-5:glb-5(Haw) animals after hypoxia (Fig. 6G). hif-1(ia4 ) animals survived well in 24 h hypoxia, and upon exposure to 1% O 2 , left food and accumulated outside the bacteria. When returned to 21% O 2 , hif-1(ia4);  animals rapidly resumed bordering behavior, like glb-5(Haw); npr-1 animals. These results suggest that HIF-1 is not required for GLB-5(Haw) to promote rapid recovery from hypoxia exposure, although we have not examined if it required for npr-1 animals to inhibit bordering following hypoxia experience.  counteracts the effects of prolonged hypoxia on behavior and URX physiology after animals are returned to normoxia. We speculated that GLB-5 could do this in one of two ways: by preventing hypoxia from reconfiguring the physiology of URX or by enabling rapid recovery of URX response properties when animals exposed to prolonged hypoxia are returned to 21% O 2 . In the second model URX recovery would need to be fast, since glb-5(Haw); npr-1 animals showed substantial O 2 -evoked Ca 2ϩ responses within 5 min of transfer from hypoxia to 21% O 2 (Fig. 6B).

GLB-5(Haw) enables URX neurons adapted to hypoxia to rapidly retune their O 2 response set point upon return to normoxia
To differentiate between these possibilities we built an airtight microfluidic-imaging chamber that enabled us to measure URX response properties in hypoxia-grown animals without prior exposure to 21% O 2 . Since gluing animals in the hypoxia chamber was difficult, we immobilized animals using the cholinergic agonist levamisole (see Materials and Methods). We grew npr-1 and glb-5(Haw); npr-1 animals for 18 h in either 1% or 21% O 2 , transferred them to the imaging scope, and shifted them from 1 to 7% O 2 . To compare the effects of acute and prolonged hypoxia, we also exposed npr-1 and glb-5(Haw); npr-1 animals cultivated at 21% O 2 to a 21%-1%-7-21% O 2 stimulus train. Animals adapted to 1% O 2 for 18 h exhibited large persistent increases in URX Ca 2ϩ when shifted to 7% O 2 (Fig. 7 A, B), whereas animals exposed to 1% O 2 for only 5 min exhibited a smaller, transient increase in Ca 2ϩ to the same 7% O 2 stimulus (Fig. 7C-E). The effects of prolonged hypoxia on URX activity after O 2 levels were raised for the first time from 1 to 7% were independent of GLB-5(Haw) activity (Fig. 7 A, B). This contrasts with the marked difference in O 2 -evoked Ca 2ϩ responses in URX when hypoxiatreated npr-1 and glb-5(Haw); npr-1 animals were exposed to 21% O 2 (Fig. 6 A, B). These data suggest that hypoxia can reprogram URX activity in both npr-1 and glb-5(Haw); npr-1 animals, but that recovery after further exposure to 21% O 2 is much faster in glb-5(Haw); npr-1 than npr-1 animals.

Discussion
The neuroglobin GLB-5(Haw) enables fast behavioral recovery after exposure to hypoxia. At 21% O 2 wild C. elegans rapidly accumulate where bacterial food is thickest. This foraging behavior, called bordering, is inhibited by prior hypoxia experience if glb-5 is defective. The effects of hypoxia exposure on subsequent behavior at 21% O 2 build up gradually and plateau after 18 h in 1% O 2 . C. elegans relies on the URX O 2 -sensing neurons to accumulate where bacteria are thickest; these neurons are activated by a rise in O 2 . If GLB-5 is defective, prolonged hypoxia and reoxygenation disrupts O 2 -evoked Ca 2ϩ responses of URX neurons. A simple hypothesis is that GLB-5 mitigates the effects of hypoxia/re-oxygenation by controlling O 2 -evoked Ca 2ϩ dynamics during this transition. The exact molecular mechanism by which this is achieved is unclear, but GLB-5, like many other hexacoordinated globins, oxidizes rapidly in 21% O 2 (Persson et al., 2009;Yoon et al., 2010), making it a potential electron donor that signals by generating reactive oxygen species.
GLB-5(Haw)'s ability to promote rapid recovery of bordering after hypoxia experience, and to tune URX neurons to changes in O 2 between 21 and 17%, depend on PDL-1, the C. elegans ortholog of the mammalian prenyl binding protein PDE␦. PDL-1 is required in both the URX and BAG neurons to sustain these GLB-5(Haw)-dependent behaviors. BAG, like URX, is an O 2 sensor, but is activated by a decrease instead of an increase in O 2 due to the activity of the GCY-33/GCY-31soluble guanylate cyclase . PDL-1 mediates GCY-35 and GCY-33 O 2 sensor localization to the dendritic endings of URX and BAG neurons, respectively (Fig. 3C,I ). Since GLB-5 is also localized to the dendritic endings of URX, one interpretation of our data is that physical proximity helps GLB-5 modify O 2 sensory transduction by GCY-35/GCY-36. In BAG neurons GCY-33 and GLB-5 likely function in a complex, since GLB-5(Haw) can target GCY-33 to cilia in the absence of PDL-1. Although a simple model in which the effects of PDL-1 are mediated via localization of soluble guanylate cyclase is appealing, PDL-1 may play a more complex role involving trafficking of other prenylated proteins. For example, in mammals PDE␦ traffics heterotrimeric guaninenucleotide binding protein gamma subunits; the C. elegans orthologs of these genes, gpc-1 and gpc-2, are also predicted to be prenylated.
PDL-1 activity is required in BAG for GLB-5(Haw) to confer a strong behavioral response to switches between 21% and 17% O 2 (Fig. 2F ). Consistent with this, a 21%-17%-21% O 2 stimulus train evoked a significant rise and fall in Ca 2ϩ levels in BAG in well fed glb-5(Haw); npr-1 animals (Fig. 5). Thus, the same small changes in ambient O 2 concentration induce reciprocal O 2 responses in URX and BAG, and act antagonistically to control locomotor activity: at 21% O 2 high URX/low BAG activity promotes rapid movement, whereas at 17% O 2 low URX/high BAG activity promotes slow movement (Figs. 4,5). Previous work has already demonstrated reciprocal O 2 responses in URX and BAG, but BAG responses in those studies were evoked at much lower O 2 concentrations . Like URX, BAG neurons appear to have tonic signaling activity, since BAG Ca 2ϩ levels remained elevated while O 2 levels were reduced. Consistent with our data, previous work has shown that activating BAG induces animals to reduce locomotor activity .
PDE6␦ is considered to be a promiscuous prenyl binding protein (Zhang et al., 2004). However, in our yeast two-hybrid assays PDL-1 did not interact with several well known prenylated proteins such as RAS-1 and RAB-18, or ARL2-like proteins such as ARL-3 and EVL-20 (Fig. 3F ). Moreover, the interaction of PDL-1 with GCY-33 was significantly weaker than its interaction with GCY-35 in the same assay. One explanation is that expression of these prenylated proteins in yeast is variable. An alternative explanation is that the substrate specificity of PDL-1 for prenylated proteins is complex.
The neuronal functions of mammalian PDE6␦ have been previously studied in rods and cones, where it regulates transport of phosphodiesterase PDE6, rhodopsin kinase, and G t␣ -all prenylated proteins-to the outer segment of these ciliated neurons (Li et al., 1998;Norton et al., 2005;Zhang et al., 2007;Luo et al., 2008). Like cone and rod photoreceptors, BAG has a ciliated dendritic ending, but URX is a nonciliated neuron (Ward et al., 1975). Thus, prenyl binding proteins can facilitate protein traffic to both ciliated and nonciliated dendritic endings. The localization of GCY-35 and GCY-33 may regulate O 2 sensory transduction level in two ways. First, localization of these molecular O 2 sensors probably alters their O 2 environment: the dendritic end- The Ca 2ϩ sensor is YC2.60. Animals that were incubated in 1% O 2 did not experience 21% O 2 before the 7% O 2 stimulus. The transition from hypoxia to 7% O 2 elicited an increase in Ca 2ϩ in npr-1 and glb-5(Haw); npr-1 animals. Asterisks indicate significant differences at times indicated by bars between the Ca 2ϩ levels before the transition to 7% O 2 , Mann-Whitney test. C, Brief exposure to 1% O 2 followed by a 7% O 2 stimulus did not elicit a strong and enduring increase in Ca 2ϩ concentration in npr-1 and glb-5(Haw); npr-1 animals. Asterisks indicate significant differences compared with the Ca 2ϩ level in 1% O 2 , at times indicated by bars just before the transition to 7% O 2 , Mann-Whitney test. D, E, Prolonged exposure to hypoxia reconfigures URX O 2 responses. A comparison of responses evoked by a switch from 1 to 7% O 2 in npr-1 and glb-5(Haw); npr-1 animals (from A and C; B and C, baseline normalized ratios). Asterisks indicate significant differences compared with the Ca 2ϩ level in 1% O 2 after 3 min in 7% O 2 . Mann-Whitney test, *p Ͻ 0.05, **p Ͻ 0.01, and ***p Ͻ 0.001. Gray shading represents SEM. F, Model for PDL-1 activity in the URX and BAG neurons. Prenylated GCY-33 and GCY-35 interact with PDL-1, which facilitates their traffic to dendritic endings. There, they act together with GLB-5(Haw) to tune the dynamic range of O 2 responses, and to rapidly retune O 2 -sensing properties after return to normoxia following prolonged hypoxia.
ings of URX and BAG are located close to the surface at the C. elegans nose tip, where O 2 concentrations are likely to be higher than those found in the cell bodies, which lie buried inside the animal. Second, as highlighted above, components of the O 2 sensory transduction machinery, including GLB-5 (Persson et al., 2009) and cGMP-gated channels (Arellano-Carbajal et al., 2011), are enriched at the dendritic endings of URX, facilitating compartmentalized signaling. Interestingly, although pdl-1 mutants have little detectable GCY-35 at URX dendritic endings, they retain strong O 2 -evoked Ca 2ϩ responses in these neurons, indicating that cGMP signal transduction can still function. This suggests that soluble guanylate cyclases and cGMP channels remain in close proximity in pdl-1 mutants.
Prolonged hypoxia reconfigures responses to O 2 in other C. elegans behavioral paradigms, although the role of globins was not studied (Cheung et al., 2005;Chang and Bargmann, 2008;. Our Ca 2ϩ imaging experiments demonstrate that previous O 2 experience can retune URX O 2 responses ( Fig.  7 A, B). Interestingly, previous studies of HIF-1 activation showed that the HT22 mouse neuronal cell line and cardiofibroblasts can adjust their O 2 sensitivity in response to prolonged incubation in 5% or 30% O 2 (Roy et al., 2003;Khanna et al., 2006). Prolonged exposure of HT22 cells to 5% O 2 attenuates subsequent HIF-1 activation by 0.5% O 2 . Moreover, HIF-1 activity increases when HT22 cells grown in 30% O 2 are transferred to 20% O 2 . These studies suggest that in these cells the O 2 balance point for HIF-1 induction is regulated by experience.
In summary, our results suggest that the GLB-5 neuroglobin confers on C. elegans O 2 sensors the ability to retain their tuning properties following prolonged hypoxia and re-oxygenation. This mechanism allows fast behavioral and cellular activity adjustment following re-oxygenation. It would be interesting to test if mammalian neuroglobin regulates the threshold for O 2 responses in glomus cells of the carotid body (Di Giulio et al., 2006. Our results also suggest that PDL-1, the C. elegans ortholog of PDE␦, facilitates traffic of prenylated GCY-35 and GCY-33 from the cell bodies of BAG and URX neurons to their ciliated and nonciliated dendritic endings, respectively. PDL-1 is highly conserved between C. elegans and humans, plays a major role in vision physiology, and spatially regulates K-RAS activity in human cells (Zhang et al., 2012;Zimmermann et al., 2013). Our studies establish a tractable in vivo system to study PDL-1 function in vivo, enabling its mechanisms of action to be explored further and potential small molecule inhibitors to be tested.