ROS regulation of RAS and vulva development in Caenorhabditis elegans

Reactive oxygen species (ROS) are signalling molecules whose study in intact organisms has been hampered by their potential toxicity. This has prevented a full understanding of their role in organismal processes such as development, aging and disease. In Caenorhabditis elegans, the development of the vulva is regulated by a signalling cascade that includes LET-60ras (homologue of mammalian Ras), MPK-1 (ERK1/2) and LIN-1 (an ETS transcription factor). We show that both mitochondrial and cytoplasmic ROS act on a gain-of-function (gf) mutant of the LET-60ras protein through a redox-sensitive cysteine (C118) previously identified in mammals. We show that the prooxidant paraquat as well as isp-1, nuo-6 and sod-2 mutants, which increase mitochondrial ROS, inhibit the activity of LET-60rasgf on vulval development. In contrast, the antioxidant NAC and loss of sod-1, both of which decrease cytoplasmic H202, enhance the activity of LET-60rasgf. CRISPR replacement of C118 with a non-oxidizable serine (C118S) stimulates LET-60rasgf activity, whereas replacement of C118 with aspartate (C118D), which mimics a strongly oxidised cysteine, inhibits LET-60rasgf. These data strongly suggest that C118 is oxidized by cytoplasmic H202 generated from dismutation of mitochondrial and/or cytoplasmic superoxide, and that this oxidation inhibits LET-60ras. This contrasts with results in cultured mammalian cells where it is mostly nitric oxide, which is not found in worms, that oxidizes C118 and activates Ras. Interestingly, PQ, NAC and the C118S mutation do not act on the phosphorylation of MPK-1, suggesting that oxidation of LET-60ras acts on an as yet uncharacterized MPK-1-independent pathway. We also show that elevated cytoplasmic superoxide promotes vulva formation independently of C118 of LET-60ras and downstream of LIN-1. Finally, we uncover a role for the NADPH oxidases (BLI-3 and DUOX-2) and their redox-sensitive activator CED-10rac in stimulating vulva development. Thus, there are at least three genetically separable pathways by which ROS regulates vulval development.

One of the ways ROS can participate in signalling, is by modulating the activities of small GTPases. For examples, the Ras signalling pathway is known to be modulated by ROS in mammals [26,27] and C. elegans [4,28]. One of the ways the RAS pathway is affected by ROS in mammals is by direct oxidation of a sensitive cysteine (C118) of the RAS protein itself, by nitric oxide (NO) and superoxide [29,30] as well as by hydrogen peroxide in the presence of transition metals [31]. This has been mostly studied in vitro [29][30][31][32][33], and in cultured cells [27,[34][35][36] but to a much lesser extent in vivo [37]. Redox-sensitive cysteines have also been studied in other small GTPases, such as Rac, where the C18 cysteine has been identified as redoxsensitive [38] as well as Rho, where it's the C20 cysteine that is sensitive [39]. Studies with Ras and Rac have generally concluded that oxidation of these cysteines (C118 and C18, respectively) leads to the activation of the protein by stimulating guanine nucleotide release. In contrast, oxidation of C20 generally inhibits Rho proteins because of the presence of an additional cysteine, C16. Oxidation of C20 promotes guanine nucleotide release but subsequent disulfide bridge formation between C16 and C20 prevents guanine nucleotide binding, thus inactivating the protein [39]. A powerful tool to study the roles of these sensitive cysteines in vitro and in vivo is by replacement of the cysteine by a serine that cannot be oxidized [6,33,38] or by an aspartic acid to mimic oxidation [38,40].
In C. elegans the Ras pathway has been particularly well characterized for its role in the development of the C. elegans vulva, the egg-laying organ [41,42] (Fig 1A). let-60ras encodes the C. elegans orthologue of mammalian Ras and is most similar to K-Ras [43]. Severe loss-offunction mutants of let-60ras cannot survive, but a gain-of-function mutation (n1046gf), resulting in a G13E substitution, is viable [44] (below we denote this mutation as let-60rasgf). Such oncogenic mutations at G13 (Fig 1B and S1 Fig) favor the active GTP-bound state [45,46]. In C. elegans, this mutation induces the formation of multiple vulvas instead of only a The multivulva phenotype of let-60rasgf is sensitive to ROS. let-60ras(n1046gf) is denoted in the figure as let-60gf. A The genetic pathway by which LET-60ras promotes vulval development. C. elegans gene names are in lowercase italics, and the corresponding mammalian protein homologues are in uppercase. The pathway depicts the genetic epistatic relationships between the genes rather than the biochemical interactions of the proteins. Gain-of-function mutations that activate LET-60ras or loss-offunction mutations in lin-1 lead to the development of multiple vulvas. B Space-filling model of human K-Ras bound to GTP, showing the residue (G13) that is mutated in let-60rasgf, and the redox-sensitive cysteine (C118). The structure (PDB ID: 3GFT) was derived using Pymol. C Sample images from mutants scored in D (asterisks denote the vulvas and arrowheads the pseudovulvas). The scale bar represents 50 μm. D Number of vulvas of the wild type and the let-60rasgf mutant after treatment with PQ and NAC. ��� P = 0.0001 compared to let-60rasgf.
Here we use C. elegans vulva formation in the sensitized let-60rasgf gain-of-function background to dissect aspects of ROS signalling in a fully in vivo situation. We do this with genetic and pharmacological manipulations that are in the physiological range, with no impact on the organism's health. We monitor real developmental outcomes (vulva formation) and, by controlling sources, sinks, and especially targets, we could identify genetically distinct ROS signalling pathways, despite that fact that they act in parallel and use the same enzymes and the same active molecules (superoxide and peroxide). We also provide strong evidence that it is hydrogen peroxide, rather than any other ROS species, that is responsible for cysteine oxidation and that in C. elegans, in vivo, oxidation inhibits, rather than activates LET-60ras and CED-10rac.

RAS signalling is sensitive to ROS
The let-60rasgf mutant provides a sensitized background to score changes in RAS signalling, and allowed for extensive characterization of the pathway by identifying suppressors and enhancers [42]. We used the same logic to identify and characterize mechanisms of ROS signalling acting on the RAS pathway and on vulva formation in general. We treated let-60rasgf mutants with the prooxidant paraquat (PQ). PQ can potentially generate superoxide at many cellular sites [49], but its main site of superoxide production is believed to be in the mitochondrial matrix [50,51]. We used a very low concentration of PQ (0.1-0.2 mM) that has been shown to increased mitochondrial superoxide [6,52,53] but without toxicity [52]. Such a low concentration might provide for alteration of ROS signals within the physiological range. In the presence of mitochondrial and cytoplasmic SODs, increased superoxide generation by PQ is expected to lead to increased hydrogen peroxide generation. N-acetyl cysteine (NAC) is a precursor of glutathione and thus functions as an antioxidant by facilitating the removal of peroxides, including hydrogen peroxide [54]. We used NAC at 9 mM, a concentration that has been shown to lower ROS [55] but has no effect on wild type viability [52]. Like other studies using prooxidants on let-60rasgf mutants [28], we found that although PQ had no effect on vulva formation in the wild type, it partially suppressed the Muv phenotype of let-60rasgf ( Fig  1D). Conversely, we found that NAC treatment had no effect on the wild type but enhanced the Muv phenotype of let-60rasgf (Fig 1D). We tested two additional alleles of let-60 in addition to the canonical n1046gf allele (S2 Fig). The n1700 allele leads to the same amino acid change as n1046 (G13E), but was independently isolated and is thus in a different background. It leads to a slightly more severe Muv phenotype. The effects of PQ and NAC on this allele were qualitatively the same, suppression and enhancement, respectively. However, the suppression was relatively greater and the enhancement relatively less, likely due to the more severe baseline Muv phenotype. We also tested ga89, a weaker temperature-sensitive allele [56]. However, the allele is so weak, even at the restrictive temperature (26˚C), that only the suppressing effect of PQ could be reasonably inferred (S2 Fig). At this stage we could infer that oxidation inhibits a target that could be either LET-60ras itself or another activator of the RAS pathway. All numerical values for all Muv data shown in bar graphs in all figures are given in S1 Table. We report the total number of vulvas (which includes both the main vulva and the ectopic pseudovulvas) and the controls shown are always scored in parallel for every experiment.

Cysteine C118 of LET-60ras is the target of oxidation by PQ
As described in the introduction, cysteine C118 of mammalian RAS is a known potential target of oxidation by NO and superoxide, with oxidation resulting in activation of the protein.
Nitrosylation is not relevant in C. elegans, which lacks nitric oxide synthases [57]. We tested whether C118 was involved in the PQ and NAC sensitivity that we observed (although we observed inhibition rather than activation by PQ, and activation by an antioxidant, NAC). We used CRISPR to replace C118 with a serine (C118S). Serine has a similar structure to cysteine but cannot be oxidized by cellular ROS [58] (Fig 2A and 2B). We produced C118S alleles, qm226 and qm225, in the wild type and let-60rasgf backgrounds, respectively. Below we also describe alleles in which C118 is replaced by aspartic acid (C118D). For clarity, we use the following formalism when new alleles that affect C118 are involved: we denotate the wild type let-60 allele as let-60(+) and we use let-60ras(+)-C118S (or C118D) and let-60rasgf-C118S (or C118D) to denotate single and double mutants in which C118 has been replaced by another amino acid.
The C118S replacement had no effect on vulva formation in the wild type, but strongly enhanced vulval induction in let-60rasgf mutants (Fig 2C and S3 Fig). We scored both the Muv phenotype by counting visible pseudovulvas in adults (Fig 2C and 2D) as well as invaginations in L4-stage larvae (S3 Fig). The C118S replacement fully suppressed all effects by PQ or NAC (Fig 2D). The degree of enhancement produced by C118S was very similar to that produced by NAC (Fig 2D). We conclude that in the let-60rasgf background, PQ acts on the Muv phenotype by increasing hydrogen peroxide levels, leading to increased oxidation of C118, and NAC acts by preventing oxidation of C118 by lowering hydrogen peroxide levels. When C118 is replaced by a serine that cannot be oxidized, neither compound has any effect. The fact that the C118S replacement leads to increased vulva formation indicates that C118 is normally partially oxidized. The picture of RAS oxidation by ROS is thus very different in living intact C. elegans from that in vertebrate cells: both the mechanism of oxidation (by hydrogen peroxide rather than by superoxide or nitric oxide) and the consequence of oxidation (inhibition rather than activation) are different.

Mimicking constitutive oxidation of C118 fully suppresses the oncogenic let-60rasgf allele
Oxidation of cysteine produces cysteine sulfenic acid, which can be further oxidised to form sulfinic acid and then sulfonic acid [59] (Fig 2B). The molecular shape and charges of cysteine sulfinic acid is mimicked by aspartic acid (D) [40] (Fig 2A and 2B). We used CRISPR to replace C118 with aspartic acid (C118D), creating two alleles, qm227 and qm228, in the let-60rasgf and in the wild-type backgrounds, respectively. This should mimic an intermediate but permanent degree of oxidization of C118 (Fig 2B). Strikingly, both let-60ras(+)-C118D and let-60rasgf-C118D mutants have only a single vulva and, like the wild type, are completely insensitive to PQ and NAC (Fig 2C and 2E). This is consistent with our interpretation of the effects of PQ and NAC treatment on let-60rasgf and let-60rasgf-C118S: oxidation at C118 inhibits Ras signalling. Note that the down-regulation produced by the C118D substitution doesn't prevent the formation of a vulva, whereas loss-of-function mutations in let-60ras are vulvaless [44]. Thus, one possibility for the complete suppression of the multi-vulva phenotype of the let-60rasgf oncogenic gain-of-function mutation (G13E) by C118D is that the mode of action of oncogenic mutations at G13 might be by interference with the normal regulation by oxidation at C118 (Fig 1B). In other words, the C118D mutation could be specifically counteracting the effects of the gain-of-function mutation rather than simply down-regulating LET-60rasgf activity. However, our findings of the level of expression of the LET-60rasgf-C118D protein suggest that there may be other possible reasons for the complete suppression of the Muv phenotype in these double mutations (see below).

The degree of oxidation of C118 or the inhibition of oxidation by the C118S mutation does not act on vulva formation through MPK-1 phosphorylation
Activation of the Ras pathway leads to increased phosphorylation of the downstream effector MPK-1 in C. elegans as in other systems [47,60] (Fig 1A). To investigate whether changes of oxidation of LET-60rasgf by compound treatment or by mutations (C118S or C118D) counteracts or enhance the effects of let-60rasgf on MPK-1 phosphorylation, we used commercial antibodies to quantify MPK-1 phosphorylation in vivo, by Western Blot (Fig 3A and 3B and S4  Fig). There are two isoforms of MPK-1 (a and b), with MPK-1a believed to be the relevant isoform to signalling events in the animal's soma, including vulval development [61]. Using a variety of controls to ensure accurate quantification (S4 Fig and S5 Fig), we observed increased phosphorylation of MPK-1a in let-60rasgf mutants (Fig 3A and 3B) as expected. However, phosphorylation of MPK-1a was not affected in any way in let-60ras(+)-C118S or let-60rasgf-C118S mutants and neither was it significantly suppressed by PQ treatment or enhanced by NAC treatment of let-60rasgf, despite the dramatic effects of the treatments and the mutation on the Muv phenotype (Fig 3A-3D). These findings suggest that activated Ras might impinge on vulva formation through more than one effector and ROS levels regulate vulval development independently of MPK-1 phosphorylation (see Discussion). However, in contrast to the lack of effects of the treatments and of the C118S mutation, the C118D mutation was capable of suppressing the increases level of MPK-1 phosphorylation that is observed in let-60rasgf mutants, such that the level of MPK-1 phosphorylation in let-60rasgf-C118D mutants is the same as that observed in the wild-type. See below for additional discussion of this observation.

The effects of altered oxidation and the C118S mutation are not mediated by altered protein stability
We sought to investigate whether changes of protein stability could be the mechanism by which oxidation affects LET-60rasgf and RAS signaling. The possibility existed because overexpression of wild type LET-60ras can lead to the Muv phenotype [62]. We examined how the C118S mutation, and PQ and NAC treatments affect the LET-60ras protein levels for both LET-60ras(+) and LET-60rasgf using Ras-specific commercial antibodies (Fig 3E-3H

The LET-60rasgf-C118D double mutation lowers protein levels
In contrast to what we observed with drug treatments and for LET-60rasgf-C118S, the protein level of LET-60rasgf-C118D but not of LET-60ras(+)-C118D is significantly lower than that of LET-60ras(+) (Fig 3E and 3F). Possibly therefore, the full suppression of the Muv phenotype observed in let-60rasgf -C118D mutants could be due to the lower level of protein acting on the MPK-1 phosphorylation pathway. Alternatively, the depth of the effect (complete suppression of Muv) could be the result of a double effect: an effect on the MPK-1 pathway via lower LET-60rasgf protein expression and an effect on the MPK-1-independent pathway that appears to mediate the other effects of changes to the oxidation status of C118 of LET-60ras.

C118 oxidation depends on the cytoplasmic pool of hydrogen peroxide regulated by SOD-1
Any excess O 2 •produced by PQ treatment is expected to be converted to hydrogen peroxide by superoxide dismutases. Although there are 5 SODs in C. elegans, cytoplasmic SOD-1 and mitochondrial SOD-2 account for virtually all SOD activity [20,63]. SOD-1 may also be present in the mitochondrial inter-membrane space as is the case in other organisms [64]. To determine the enzymatic source of the relevant hydrogen peroxide, we constructed a let-60rasgf;sod-1 double mutant. These double mutant animals displayed an enhanced Muv phenotype (Fig 4A), similar to that obtained by treating let-60rasgf with NAC or replacing cysteine C118 with serine ( Fig 2D). Furthermore, the mutants' Muv phenotype is no longer enhanced by NAC or suppressed by PQ (Fig 4A). These observations suggest two, not mutually exclusive, mechanisms concerning the origin of the hydrogen peroxide that can oxidize C118: 1) Wherever it is produced, the superoxide produced by PQ reaches the compartment in which SOD-1 is present, and/or 2) in the absence of SOD-1, any hydrogen peroxide produce by other SODs from PQ-dependent superoxide cannot produce a sufficient elevation of cytoplasmic hydrogen peroxide to affect C118 of LET-60ras. Other elements of the data presented in Fig  4A, such as the enhanced vulva formation produced by PQ and the effect of the C118S mutation in the absence of SOD-1, are discussed further below.

Ras signalling is sensitive to increased mitochondrial superoxide levels in mitochondrial mutants
Mitochondria are a site of ROS formation that has been studied extensively. Mitochondriallyderived ROS are believed to be involved in signals that affect mitochondrial dynamics [65], mitophagy and autophagy [66], apoptosis, responses to changes in oxygen levels particularly hypoxia, inflammatory responses [67,68], wound healing [6,69] and aging [70]. We used three mutations that have elevated mitochondrial superoxide levels. sod-2 mutants completely lack the main mitochondrial matrix superoxide dismutase SOD-2 [71]. isp-1 and nuo-6 are point mutants in subunits of the mitochondrial respiratory chain that lead to low electron transport, low ATP levels but high level of superoxide generation [72,73]. All three mutations suppress the Muv phenotypes in double mutant combinations with let-60rasgf (Fig 4B). sod-2 suppresses by about 40% but isp-1 and nuo-6 suppresses almost completely (Fig 4B).
As described above, the loss of SOD-1 enhances the Muv phenotype (Fig 4A) while the loss of SOD-2 suppresses it (Fig 4B). We found that suppression of the Muv phenotype by loss of SOD-2 is partially abrogated by the loss of SOD-1 (Fig 4C) or by treatment with NAC ( Fig  4D). These observations suggest that loss of mitochondrial SOD-2 suppresses let-60rasgf via a SOD-1-dependent increase in cytoplasmic or mitochondrial inter-membrane hydrogen peroxide (Fig 4E). This conclusion was supported by the finding that replacement of C118 with serine in the let-60rasgf background suppresses the effect of loss of SOD-2 on the Muv phenotype (Fig 4B). Fig 4E suggests a model of how the increased superoxide produced in the mitochondrial matrix in the absence of SOD-2 reaches the inter-membrane space and/or cytoplasm, where SOD-1 is located. The model includes that in the absence of a mitochondrial matrix SOD, superoxide can be transported out of both mitochondrial compartments into the cytoplasm through specialised channels [74,75]. Thus, our results suggest that in the absence of (pMPK-1a), total MPK-1a and Histone H3 as a loading control. Unprocessed original scans of blots and additional analyses are shown in S4  Fig. B The mean ratio of phosphorylated MPK-1a (pMPK-1a) to total MPK-1a ± SEM of three different Western blots. C A representative Western blot for phosphorylated MPK-1a (pMPK-1a), total MPK-1a and Histone H3 as a loading control. Unprocessed original scans of blots and additional analyses are shown in S5 Fig. D The mean ratio of phosphorylated MPK-1a (pMPK-1a) to total MPK-1a ± SEM of three different Western blots. E A representative Western blot for LET-60 with Tubulin as a loading control. Unprocessed original scans of blots are shown in S6A Fig. F The mean band intensity ± SEM of three different Western blots normalised to band intensities in the wild type. G A representative Western blot for LET-60 with Histone H3 as a loading control. Unprocessed original scans of blots are shown in S6B Fig. H The mean band intensity ± SEM of three different Western blots normalised to band intensities in the wild type. Molecular weight markers are shown for E and G as different loading controls were used for these blots. All bars are compared to the wild-type control bar or as indicated. �� P<0.01, ��� P<0.001.
https://doi.org/10.1371/journal.pgen.1008838.g003 Similarly, we tested whether the Muv suppression by isp-1 and nuo-6 depended on C118 by scoring the Muv phenotype in isp-1 let-60rasgf-C118S and nuo-6; let-60rasgf-C118S double mutants (Fig 4B). The C118S replacement also suppressed the Muv suppression by these mitochondrial mutants. Since functional wild-type SOD-2 is present in isp-1 and nuo-6 mutants the effect of their known increased superoxide generation might act through a SOD-2-dependent production of hydrogen peroxide in the mitochondria. This excess mitochondrial hydrogen peroxide could participate to the cytoplasmic pool by exiting the mitochondria through passive membrane diffusion or through specialized channels [76] (Fig 4E). Consistent with what we observed for let-60ras(+)-C118S and let-60rasgf-C118S (Fig 3C and 3D), the isp-1 and nuo-6 mutations do not affect the levels of MPK-1 phosphorylation (Fig 4F and 4G and S8  Fig). In addition, the fact that the suppression of the effect of the ETC mutants by C118S is not complete suggests that other consequences of the mutations, such as low electron transport and low ATP levels might also participate in suppressing the Muv phenotype.
For our observations on the effect of sod-2, isp-1, and nuo-6 on the Muv phenotype we cannot exclude an alternative interpretation: that all the effects we observe are additive, rather than revealing interactions. However, given the known relationships between the cytoplasmic and mitochondrial ROS pools discussed and reviewed in the above paragraphs, this alternative model appears less likely.

The intracellular superoxide pool affects vulva formation by a separate pathway that is independent from hydrogen peroxide generation
As we have seen above, the C118S mutation produces an elevated Muv phenotype that is very similar to that produced by NAC, and NAC has no additional effects on C118S mutants. In addition, the suppression of the Muv phenotype by PQ is fully abolished by the C118S mutation (Fig 2D). Thus, let-60rasgf-C118S mutants, with or without NAC or PQ treatment, have almost the same Muv phenotype as let-60rasgf treated with NAC (Fig 2D). However, loss of SOD-1 in the let-60rasgf background produces an increase in the Muv phenotype that is significantly greater than that produced by NAC (Fig 4A). Furthermore, in the absence of SOD-1, PQ not only fails to suppress the Muv phenotype, it actually increases it (Fig 4A). In addition, the Muv increasing effect of PQ in the sod-1 background is even stronger in the absence of C118 in sod-1;let-60rasgf-C118S mutants (Fig 4A). Thus, increased superoxide formation by PQ in the absence of SOD-1 (which would convert it to hydrogen peroxide), and in the absence of a target for hydrogen peroxide (in the let-60rasgf-C118S background), leads to a stimulation of the Muv phenotype. Note that the fact that PQ simulates more in the let-60rasgf-C118S background means that there is likely residual hydrogen peroxide formation from PQ-dependent superoxide even in the absence of SOD-1. This residual hydrogen peroxide can act in an inhibitory fashion on C118. This effect is eliminated in the C118S mutant combination with PQ or NAC treatments. ###P = 0.0001 for PQ treatment compared to PQ treatment of let-60rasgf. +++P = 0.0001 for PQ treatment of sod-1; let-60rasgf-C118S compared to PQ treatment of sod-1; let-60rasgf. B Number of vulvas of long-lived mitochondrial mutants in an otherwise wild-type background and in combination with the let-60rasgf mutant, with and without the C118S mutation. ��� P = 0.0001 compared to let-60rasgf for the first set of bars, and compared to let-60rasgf-C118S for the second set of bars. ###P = 0.0001 compared to the mutation in the let-60rasgf background. C Number of vulvas of sod-2 mutants in combination with sod-1 and the let-60rasgf mutation. D Number of vulvas of sod-2 mutants, in combination with the let-60rasgf mutation and NAC treatment. E A model that suggests a path by which PQ, SOD-1, and the loss of SOD-2, all increase cytoplasmic H 2 O 2 and thereby suppress let-60rasgf. For simplicity SOD-1 is shown in the cytoplasm although it may also be in the mitochondrial inter-membrane space. F A representative Western blot for phosphorylated MPK-1a (pMPK-1a), total MPK-1a and Histone H3 as a loading control. Unprocessed original scans of blots and additional analyses are shown in S8 Fig. G The mean ratio of phosphorylated MPK-1a (pMPK-1a) to total MPK-1a ± SEM of three different Western blots. All bars are compared to the wild-type control bar. � P<0.01.
https://doi.org/10.1371/journal.pgen.1008838.g004 leading to even greater vulva formation under the action of PQ. Together, the observations presented in Fig 4A indicate the existence of a distinct pathway in which superoxide stimulates the Muv phenotype, independently of both SOD-1-dependent hydrogen peroxide formation and oxidation of C118.
The superoxide-stimulated pathway affects vulval development by acting downstream of LIN-1 let-60ras-dependent signalling is implemented by the ETS-transcription factor LIN-1 [48] (Fig  1A). For vulval development loss of LIN-1 is epistatic to changes in the LET-60ras-dependent pathway. Thus, processes that act downstream of LIN-1 to affect the Muv phenotype are likely to be at least partially epistatic to LET-60ras signalling. We observed that loss of SOD-1 has the same effect on lin-1 mutants as on let-60rasgf mutants, that is, it enhances the Muv phenotype (Fig 5A). However, PQ treatment has opposite effects on let-60rasgf and lin-1 mutants: while it suppresses the Muv phenotype of let-60rasgf it enhances lin-1 (Fig 5B). The effect of PQ is also additive to the effect of sod-1 (Fig 5B), which is the same pattern we observed for let-60rasgf in Thus, an increase in intracellular superoxide resulting from loss of SOD-1 appears to stimulate a second pathway, independent of C118 of let-60ras and acting downstream of lin-1 (Fig 5C).

The two C. elegans NADPH oxidases, BLI-3 and DUOX-2 affect vulva formation
NADPH oxidases (NOXs) are the main enzymatic sources of ROS in the cell. As described in the introduction, NOXs are membrane-bound enzymes that produce superoxide. The Duox sub-class possesses an additional peroxidase domain and can thus convert superoxide into hydrogen peroxide. While DUOXs have well-established roles at the plasma membrane, there is evidence that they may also have intracellular roles, such as at the ER-membrane (reviewed in [25]). The C. elegans genome encodes for two DUOX proteins: BLI-3 and DUOX-2. BLI-3 is expressed in the hypodermis, where it is required for cross-linking of collagen to form the cuticle. Complete loss of bli-3 is lethal while reduction-of-function mutations in bli-3 lead to a blistered cuticle phenotype (Bli), molting defects and altered pathogen susceptibility [25,77]. Significant expression of duox-2 has not been observed, and loss of duox-2 has not previously been associated with any phenotype [25]. In order to test whether NOX-derived ROS participates in vulva formation, we examined how mutants of the C. elegans BLI-3 and DUOX-2 affect the let-60rasgf Muv phenotype. The bli-3 mutation we used (e767) is in the peroxidase domain, and thus likely prevents hydrogen peroxide formation but not superoxide formation by BLI-3 [23]. The duox-2(ok1775) mutation is likely a strong loss-of-function mutation. We found that each of the NOX mutations leads to strong suppression of the Muv phenotype of let-60rasgf in double mutants (Fig 6A), suggesting that the wild-type function of both NOXs participate in stimulating the Muv phenotype. This is the first function that can be ascribed to DUOX-2. In addition to suppressing the Muv phenotype of let-60rasgf, loss of either of the NOXs renders the let-60rasgf mutants almost insensitive to NAC and PQ (Fig 6B and 6C). This indicates that the effect of the NOX mutations is epistatic to the effect of hydrogen peroxide on LET-60rasgf. However, we have not been able to link NOX activation to the elevation of the SOD-1-dependent intracellular superoxide pool that acts downstream on lin-1 (Fig 5) (see below).

Evidence for the activation of the NOX pathway by ROS
The activity of NOX enzymes is stimulated by the small GTPase Rac [78]. There are three Raclike GTPases in C. elegans (CED-10, MIG-2, and RAC-2/3), although it is not certain that RAC-2/3 is a functional protein [79]. CED-10 and MIG-2 regulate cytoskeletal dynamics and function in multiple processes in the C. elegans, including phagocytosis of cell corpses, cell migration, axon pathfinding and growth cone protrusion [79]. ced-10 and rac-2 mutants have defects in vulval development, which can mainly be attributed to their roles in vulval cell migrations, but they also have a weak, synthetic defect in vulval cell specification [80]. Rac GTPases can be regulated by ROS. In particular, oxidation of cysteine C18 (S9 Fig) has been found to lead to activation of Rac1 in vitro and in cultured mammalian cells [38]. In light of this, we focused on the Rac1 homologue ced-10 and used CRISPR to create a C18S allele, ced-10rac(qm229), which should be insensitive to ROS. This allele significantly enhanced the Muv phenotype in the let-60rasgf background (Fig 6A). This was the reverse of the effect expected from mammalian studies, as we would expect that inhibition of CED-10 would phenocopy duox-2 and bli-3 loss-of-function mutations, but it had the opposite effect. We interpret this to mean that oxidation of C18 of CED-10 inhibits the protein. This would be similar to what we observed for C118 of LET-60. In both cases, we observed that the loss of the cysteine leads to activation rather than to inhibition of the activity. Since the C18S mutation had the opposite effect to that of the duox-2 and bli-3 mutations, we tested whether Rac was acting through each of the C. elegans Duoxs by creating bli-3(e767); ced-10rac(qm229) and duox-2(ok1775); ced-10rac(qm229) double mutants. The enhancement by C18S was abolished by either of the NOX mutations (Fig 6A), suggesting that the C18S replacement stimulates the Muv phenotype through both of the NOXs (Fig 6D).
Next we wondered whether the increased superoxide due to loss of SOD-1, which stimulates the Muv phenotype (Fig 4A), acted through CED-10rac. To test for this we constructed the let-60gf; sod-1;bli-3 and let-60gf;sod-1;duox-2 triple mutants. However, the effects of loss of SOD-1 and that of the NOX mutations were additive (Fig 6A). In any case, oxidation of C18 of CED-10rac would be inhibitory (as revealed by the activating effect of C18S), but the loss of SOD-1 is activating (Fig 4A). Thus, if there was a link between the effect of loss of SOD-1 and activation of the NOXs it would need to include additional relays, which would remain to be identified. Thus, we consider the NOXs to promote vulval formation in a ROS-regulated manner, by acting in a pathway that is parallel to both LET-60ras and to intracellular superoxide (Fig 7).

Discussion
Our observations demonstrate that, by using molecular genetics, ROS signalling can be studied in a fully in vivo situation with the same detail as other forms of signal transduction. Findings The source of hydrogen peroxide is SOD-1 and is either converted from cytoplasmic superoxide or superoxide that exits the mitochondria in the absence of the mitochondrial superoxide dismutase SOD-2. SOD-2-dependent hydrogen peroxide produced in the mitochondria can also participate in the SOD-1-dependent cytoplasmic pool. Cytoplasmic superoxide inhibits vulval development by acting downstream of LIN-1, via an unknown target. The NADPH oxidases BLI-3 and DUOX-2 promote vulva development by an independent pathway. This pathway is ROS-dependent as the NADPH oxidases are activated by CED-10Rac, which is inhibited by ROS via C18, however, the source and nature of this ROS are not known. The targets of NADPH oxidase-produced ROS have not been identified. It is also not known whether for vulva development NADPH oxidases act at the plasma membrane and produce ROS extracellularly, or whether they act at the ER membrane and produce ROS in the ER. All pathways are depicted as acting in the same cell although it is possible that the different signalling pathways act in different cell types and could thereby participate in intercellular communication during vulva development. in other systems such as yeast [81] and with other pathways such as the UPR ER in C. elegans [5], reinforce this conclusion. ROS signalling resembles signalling through cAMP in that diffusible messages (ROS species) have sources and sinks. It resembles Ca ++ signalling in its clear-cut compartmentalization, and it resembles phosphorylation in reversibly targeting specific residues and thus altering protein function. This last feature is particularly powerful for molecular genetic dissection of ROS signalling pathways and functions. There have been many efforts to obtain fluorescent probes for visualizing ROS in living cells similar to tools that allow to visualize calcium [82]. However, for many applications these techniques suffer from lack of resolution and specificity. Although better tools for visualization would be highly desirable, our findings demonstrate that detailed information about ROS signalling can be obtained despite the lack of sufficiently powerful visualization tools.
Our findings suggest that ROS produced in mitochondria can participate in cytoplasmic ROS signalling in multiple ways (Figs 4E and 7). In the presence of mitochondrial superoxide dismutase (SOD-2), superoxide produced in the mitochondrial matrix is converted to hydrogen peroxide and can contribute to the cytoplasmic pool of this species. In the absence of SOD-2, superoxide exits the mitochondrial matrix, where it is converted to hydrogen peroxide by the cytoplasmic superoxide dismutase (SOD-1) and participates in ROS signalling in the cytoplasm. At this stage we don't know whether hydrogen peroxide directly modifies C118 as observed with other redox sensitive cysteines [83], or whether it is required for oxidation via additional steps.
In addition to a role for hydrogen peroxide, we also identified a specific signalling role for superoxide (Figs 5C and 7). By elevating the cytoplasmic superoxide pool by removal of SOD-1, with or without concomitant PQ treatment to enhance the effect, we showed that it can affect the Muv phenotype downstream of LIN-1, by acting on an as yet unidentified target. Interestingly, Xu and Chisholm have also demonstrated a specific role for superoxide in C. elegans wound healing. They showed that mitochondrial superoxide targets the C. elegans Rho GTPase RHO-1 [6]. Epidermal wounding causes a local increase in mitochondrial superoxide, which inhibits RHO-1 via the redox sensitive cysteines C16 and C20, and this promotes actindependent wound closure.
The NOX system is a transmembrane signalling system that can release ROS extracellularly or intracellularly [12] including in C. elegans [5,25] We found that the activity of the NOXs could affect vulva formation independently of ROS regulation of LET-60rasgf via C118 (Figs  6B, 6C and 7). At this stage we don't know if for this function the NOXs act intracellularly or extracellularly. The NOXs appear themselves to be regulated by intracellular ROS via oxidation of C18 of CED-10rac (Fig 6A and 6D). However, the species (e.g. superoxide or hydrogen peroxide) and the origin of the ROS that regulate CED-10rac, and thus the NOXs, is still unknown. We have also yet to identify the direct targets of cytoplasmic superoxide, and the targets of the NOX-derived ROS.
We identified two key targets of cytoplasmic ROS regulation of vulva formation as known redox sensitive cysteines in the small GTPases LET-60ras (C118) and CED-10rac (C18). Strikingly, we observed that in C. elegans, oxidation of C118 in LET-60ras and C18 in CED-10rac is inhibitory, while in vitro and cell culture experiments in mammalian cells have suggested they would be stimulatory by promoting guanine nucleotide dissociation. How to explain these differences? One important difference is that C. elegans does not produce, and likely doesn't use, nitric oxide (NO) [57]. Thus, the cysteines that are susceptible to oxidation by NO are free to be oxidized by different reactive species, which might result in different effects on the target protein. It is also possible that the cellular redox conditions are different in C. elegans and mammalian cells, including differences in the concentration of various ROS species or the amplitude of variations in these concentrations. More generally redox conditions in vivo, in C. elegans and in mammals, might be very different from those in cultured cells. For example, most mammalian cell types experience low levels of oxygen, low levels of hormonal stimulation, and possibly very low or no amounts of NO. Thus, some of the findings in mammalian cells might be only pertain to the conditions in cultured cells. In contrast, our tools to manipulate ROS levels are all compatible with normal or extended survival of the animals.
In addition, the cellular redox conditions (as well as the composition of the cellular milieu in the direct vicinity of a ROS target) will contribute to determining whether a modification is inhibitory or stimulatory. That is to say, reactive nitrogen, superoxide and hydrogen peroxide promote Ras guanine nucleotide dissociation. But this by itself is not activating, it's only activating under conditions that allow Ras to be competent to re-associate with GTP. Of note, in vitro, prolonged exposure of Ras in the presence of oxidants and GTP do not result in significant fraction of GTP-bound Ras, unless a radical scavenger is also included in the reaction [31].
We observed that PQ and NAC treatment and the C118S mutation, as well as isp-1 and nuo-6 mutations, have no effect on the enhanced MPK-1 phosphorylation resulting from the n1046gf allele, suggesting that the state of oxidation of LET-60ras affects a pathway that does not affect MPK-1 phosphorylation but that modulates vulva formation in the presence of elevated MPK-1 phosphorylation. We have no information yet as to the nature of this pathway. However, there is other evidence that such pathways may exist. For example, let-60ras mutants have many phenotypes in addition to vulval development, that indicate that LET-60ras acts in multiple cell types and tissues. mpk-1 mutants share most of these phenotypes, suggesting that LET-60ras mainly signals through MPK-1 [47]. However, there are also some let-60ras phenotypes which are either not shared by mpk-1 mutants, or for which an involvement of mpk-1 has not been examined [42,47], suggesting that there may be other effectors of LET-60ras. Even for vulval development, where it has been clearly established that LET-60ras acts via MPK-1 [84], it has also recently been shown that there is also MPK-1-independent signaling. In some VPCs LET-60ras acts through an alternative pathway, via RGL-1(RalGEF) and RAL-1 (Ral), to specify the secondary vulval cell fate [85].
The possibility that mimicking constitutive oxidation of C118 suppresses a classical gainof-function allele similar to oncogenic alleles in mammals is potentially of great interest. It suggests that the reason why mutations in the N-terminal G12 and G13 of Ras produce a gain-offunction could be intimately linked to events that involve the oxidation of C118. When C118 looks like it is permanently oxidized (with the C118D mutation) the G13E oncogenic mutation is fully suppressed. On the other hand, we found that the LET-60rasgf-C118D double mutant protein is expressed at a significantly lower level than LET-60rasgf. This lower level might be the basis for the complete suppression of the effect of n1046gf on MPK-1. However, given that PQ has no effect on protein levels or MPK-1 phosphorylation, we favor the model that both effects (the mimicking of constitutive strong oxidation and low level of expression) might be necessary for the complete suppression conferred by the C118D mutation. Identification of the hypothetical pathway downstream of LET-60ras oxidation could help resolve this issue in the future.
For testing the effects of Methyl viologen dichloride hydrate/Paraquat (PQ; Sigma-Aldrich 856177) and N-Acetyl-Cysteine (NAC; Sigma-Aldrich A7250), compounds were dissolved in water and stored at 4˚C. The compounds were added to the NGM just before pouring the plates. The final concentrations of PQ and NAC used for vulva studies were 0.1 mM and 9 mM, respectively.

Multivulva phenotype scoring
Adult worms were transferred to control NGM plates or plates containing PQ or NAC for a 3hour limited egg-laying. Once hatched worms reached adulthood with completed vulval formation, animals were scored for the presence of a normal vulva as well as the ectopic pseudovulvas. We report the total number of vulvas, which includes both. The controls shown are always scored in parallel for every experiment. All numerical data are presented in S1 Table.

Statistical analyses
For scoring of the multivulva phenotype, experimental strains were compared to the control using one-way ANOVA followed by Dunnett's multiple comparison test, which corrects for multiple comparisons. P-values and samples sizes are reported in S1 Table. For comparing % Muv, the Chi-square test was used. For Western Blot quantifications, ratios of phosphorylated pMPK-1a to total MPK-1 or LET-60 levels normalised to the wild-type control were compared using one-way ANOVA followed by Dunnett's multiple comparison test. All statistical analyses were carried out using Graphpad Prism 7.03. Both the n1046 and n1700 mutation harbor the G13E mutation but were isolated independently. The ga89 mutation is an L19F substitution that is a temperature-sensitive gf mutation. A Data is graphed as % Muv to allow for better visualisation of the effects of NAC and PQ on the weak Muv phenotype of ga89, even at the restrictive temperature. Data for let-60(n1046) comes from Fig 1D. B Same data as in A but graphed as Average # of Vulvas. ��� P = 0.0001 and �� P = 0.005 compared to control as indicated.  -1 and nuo-6 mutants. A Relative expression levels of total MPK-1a relative to the loading control histone H3. Values are shown as a fraction of the ratio of the indicated proteins compared to wild-type worms. No significant differences were detected, illustrating that the significant differences shown in Fig 3G and 3F arise from differences in the levels of pMPK-1a not total MPK-1a. B Relative expression levels of total MPK-1b and pMPK-1b relative to the loading control histone H3. MPK-1b level was significantly decreased in nuo-6 and nuo-6; let-60gf compared to wild-type. It also appeared to be decreased in isp-1 and let-60gf, although these differences were not statistically significant. The changes of dpMPK-1b levels mirrors the changes of total MPK-1b but the decrease of dpMPK-1b levels in isp-1 and let-60gf reached statistical significance. Mean and standard error of the mean (SEM) of 3 independent experiments are indicated in the graphs. C Original scans of western blots. 1: wild-type N2; 2:nuo-6; 3:isp-1; 4: let-60gf; 5:nuo-6; let-60gf; 6: isp-1 et-60gf.

Supporting information
The scanned images were cropped to improve clarity and focus upon the specific proteins.