Developmental fidelity is imposed by genetically separable RalGEF activities that mediate opposing signals

The six C. elegans vulval precursor cells (VPCs) are induced to form the 3°-3°-2°-1°-2°-3° pattern of cell fates with high fidelity. In response to EGF signal, the LET-60/Ras-LIN-45/Raf-MEK-2/MEK-MPK-1/ERK canonical MAP kinase cascade is necessary to induce 1° fate and synthesis of DSL ligands for the lateral Notch signal. In turn, LIN-12/Notch receptor is necessary to induce neighboring cells to become 2°. We previously showed that, in response to graded EGF signal, the modulatory LET-60/Ras-RGL-1/RalGEF-RAL-1/Ral signal promotes 2° fate in support of LIN-12. In this study, we identify two key differences between RGL-1 and RAL-1. First, deletion of RGL-1 confers no overt developmental defects, while previous studies showed RAL-1 to be essential for viability and fertility. From this observation, we hypothesize that the essential functions of RAL-1 are independent of upstream activation. Second, RGL-1 plays opposing and genetically separable roles in VPC fate patterning. RGL-1 promotes 2° fate via canonical GEF-dependent activation of RAL-1. Conversely, RGL-1 promotes 1° fate via a non-canonical GEF-independent activity. Our genetic epistasis experiments are consistent with RGL-1 functioning in the modulatory 1°-promoting AGE-1/PI3-Kinase-PDK-1-AKT-1 cascade. Additionally, animals lacking RGL-1 experience 15-fold higher rates of VPC patterning errors compared to the wild type. Yet VPC patterning in RGL-1 deletion mutants is not more sensitive to environmental perturbations. We propose that RGL-1 functions to orchestrate opposing 1°- and 2°-promoting modulatory cascades to decrease developmental stochasticity. We speculate that such switches are broadly conserved but mostly masked by paralog redundancy or essential functions.

Ras is the most mutated mammalian oncoprotein: more than a quarter of all tumors harbor activating mutations in Ras [13]. Three main oncogenic Ras effector cascades have been identified (S1 Fig). Two of these, the Raf-MEK-ERK MAP kinase and the PI3-Kinase-PDK-Akt cascades, have been extensively studied and pharmacologically targeted [14][15][16]. The RalGEF-Ral effector signal is poorly characterized, though it may be as important for oncogenesis as the Raf and PI3K cascades [17][18][19][20][21][22]. Ras binds and activates RalGEF, an exchange factor that stimulates GTP loading on the Ral small GTPase. Ral is a Ras-like small GTPase subfamily, in the Ras family of the Ras superfamily, and is conserved throughout Metazoa [20]. For most Ras family members, including Ral, this GDP/GTP cycle is controlled by activating GEFs and inactivating GAPs. Despite their structural and primary sequence similarities, Ral interacts with a very different set of effectors than does Ras [23].
Here we pursue the unexpected findings that RGL-1/RalGEF is functionally non-equivalent to RAL-1 in two distinct ways. First, while RAL-1 is essential for viability and fertility [10,25], RGL-1 is inessential. Second, RGL-1 and RAL-1 are non-equivalent in VPC fate patterning.
While RAL-1 functions as a simple intermediary to propagate the 2˚-promoting LET-60-RGL-1-RAL-1 signal, RGL-1 additionally performs an opposing, putative 1˚-promoting function that offsets its canonical 2˚-promoting function. As a consequence, deletion of RGL-1 has no net effect on the delicate balance of 1˚-and 2˚-promoting signals in sensitized mutant backgrounds. Using GEF-specific mutations and genetic bypass experiments, we show that the opposing functions of RGL-1 in VPC fate patterning are genetically separable. In the context of mammalian studies that argue that RalGEF physically interacts with PDK and Akt as a scaffold [26,27], our genetic epistasis results are consistent with RGL-1 functioning as a scaffold for PDK-1 and AKT-1 in the modulatory 1˚-promoting AGE-1/PI3K cascade. Our analysis raises the question of how activity in two apparently opposing cascades contribute to VPC fate patterning. Comparing VPC fate patterning in different environments, we found that the error rate in patterning was 15-fold higher in rgl-1 deletion mutants than in the wild type. We hypothesize that the two opposing activities of RGL-1, which tie together the two opposing 1˚and 2˚-promoting modulatory cascades, are orchestrated to reduce the level of noise in the signaling network, and hence reduce the rate of ambiguous fates or mis-patterning events.

Results
The impetus for this analysis was two enigmatic genetic observations. First, deletion of ral-1 but not rgl-1 confers severe deficits in growth, cell polarity, and fertility. Second, deletion of RAL-1 but not RGL-1 confers a net effect on 1˚vs. 2˚VPC induction. Thus, despite the linear Ras-RalGEF-Ral signaling module defined in mammals and validated in C. elegans VPC fate patterning, in two different ways RGL-1 is non-equivalent to RAL-1. By pursuing these two genetic observations, we arrived at important models regarding the multiple roles of RalGEF and Ral in signal transduction and development.
We previously described that ral-1(tm2760) animals are sterile but otherwise wild type. Efforts to feed or inject dsRNA in the RNAi hypersensitive rrf-3 background failed to phenocopy this sterility. Yet consistent with our depletion by bacterially mediated RNAi and injected dsRNA, tm2760 abrogated the 2˚-promoting activity of RAL-1 [10]. Subsequent analysis of ral-1(tm5205) and a ral-1(M -Z -) construct led to the conclusion that RAL-1 function is maternally rescued and necessary for various facets of embryonic, postembryonic, and germline development that require function of the PAR complex in cell polarity [25]. This function is ascribed to a central role of RAL-1 in the exocyst complex, as described for mammals [28][29][30][31][32][33][34]. No phenotypes have been ascribed to LET-60/Ras that could be associated with exocyst and PAR complex function [35]. We observed that ral-1(tm5205) larvae from a heterozygous mother are sickly in the fourth larval stage and become sterile adults, but VPCs in an otherwise wild-type background are patterned normally (N = 52).
Mammalian studies suggest that Ral associates with the exocyst in an activity-dependent manner to promote complex assembly [28][29][30][31][32][33][34]. Thus, abrogation of GTP-loading, either through disruption of RalGEF activity or Ral GTP-loading activity, should phenocopy loss of Ral and result in defective exocyst function. Since C. elegans encodes only a single RalGEF ortholog [10], we tested whether the RGL-1/RalGEF is required for all described RAL-1 activities.
Strikingly, four different deletion mutants of RGL-1 are superficially wild type, develop normally, are fertile, and can be grown indefinitely in culture ( Table 1). One of these mutations, fax-1(gm27), deletes several neighboring genes on the X chromosome, including rgl-1 [36]. We developed robust primer sets to detect ok1921 and tm2255 (see Methods; S6 Fig, S7 Fig). These primers failed to amplify rgl-1 sequences from the gm27 mutant animal. Mutants for the heterodimeric RalGAP we also healthy and fertile [37].The observation that deletion of RGL-1/ RalGEF does not result in the same sickly/sterile and maternal affect lethality phenotypes as deletion of RAL-1 raises the interesting possibility that Ral and RalGEF are functionally nonequivalent, contrary to the model derived from biochemical experiments in mammalian cell culture.

Using sensitized genetic backgrounds to evaluate parallel signals in VPC induction
Four signaling cascades-two central and two modulatory-control 1˚and 2˚fate induction.
(Wnt signaling also contributes to 1˚induction, but interpretation of Wnt signaling is confounded by addition roles in vulval development: competency of VPCs, polarization of 2˚lineages, and fusion of 3˚cells to the surrounding hypodermis, failure of which can itself potentiate inductive signals [37][38][39][40][41][42][43][44][45][46]. Consequently, Wnt signaling was not included in this analysis.) To aid understanding of genetic experiments presented in this study, we foreshadow our conclusions by presenting a schematic of the signaling cascades discussed in this study, along with mammalian orthologs of proteins discussed here (Fig 1B and 1C; S1 Fig). Core 1˚-and 2˚-promoting signals are detectable by direct mutation: since they are necessary and sufficient to induce their respective fates, mutational perturbation of them causes loss or gain of vulval cell types. In contrast, the role of the two modulatory cascades is not revealed through single mutant analysis, but rather requires sensitized genetic backgrounds ( [24]; this study). Parallelism has been used extensively by us and others to analyze modulatory signals in VPC fate induction, primarily through use of modifier genetics, e.g. [5-7, 10, 12, 47, 48].
To analyze such signals, we used let-60(n1046gf), a moderately activating G13E mutation analogous to mutations found in a subset of mammalian cancers [49]. In this background, gain and loss of the RAL-1 2˚-promoting signal resulted in decrease and increase of ectopic 1c ells, respectively [10,12]. We have similarly used the let-23(sa62gf) activating mutation in the LET-23/EGFR [10]. For an under-induced background, we used lin-45(n2506) [10]. Through combined use of genetic principles of parallelism and epistasis, we are able to dissect the modulatory signals. We present these principles as a network circuitry diagram ( Fig 1C). We have exploited such techniques to delineate a 2˚-promoting signaling cascade downstream of RAL-1 [12]. Here we use these techniques to similarly dissect the two roles of RGL-1/RalGEF in 2˚and 1˚-promoting activities.

RGL-1 performs a function in VPC cell fate patterning that opposes its canonical 2˚-promoting function
We previously showed that depletion of rgl-1 by RNAi revealed a role of RGL-1 in promoting 2˚fate consistent with the established Ras-RalGEF-Ral signal in mammals [10]. We reproduce these RNAi-based experiments here (Fig 2A, 2C and 2E). We also found that ral-1(tm5205) confers enhancement of 1˚induction in the let-60(n1046gf) background (Fig 2F), validating our previous results with RAL-1.
To summarize, in all of our analyses, rgl-1(RNAi) and ral-1(RNAi) confer decreased 2˚-promoting signaling. In contrast, deletion of ral-1 but not rgl-1 decreases induction of 2˚fate. Thus, our results are consistent with the working model that RGL-1 performs an additional, Ral-independent function that antagonizes its canonical function, perhaps by promoting 1f ate. The discrepancy between RNAi-and mutational-based analyses is consistent with RGL-1 having different functional thresholds in level of gene product for the two opposing activities, which we have been unable to further investigate.

Tagged endogenous RGL-1 is expressed ubiquitously, including uniform expression throughout VPC development
We used the self-excising cassette (SEC) approach [50] to tag the endogenous 5' end of the endogenous rgl-1 locus with sequences encoding mNeonGreen fluorescent protein (mNG; FP) and a 3xFlag epitope tag (S3A and S3B Fig). We observed uniform signal throughout vulval lineages (Fig 3A-3E). We also observed ubiquitous somatic and germline expression (Fig 3G-3J).

The rgl-1 transcriptional fusion is expressed in both 1˚and 2˚lineages
We previously described a transgenic ral-1 transcriptional fusion that expressed GFP dynamically over the time course of VPC fate induction [10]. To summarize published results, early in the 3 rd larval stage (L3), GFP was expressed consistently in all six VPCs. Later in L3, after induction, GFP expression was excluded from presumptive 1˚cells while persisting in presumptive 2˚cells. These observations provided a potentially important mechanistic insight: by reducing inferred RAL-1 expression in presumptive 1˚cells while retaining expression in presumptive 2˚cells, the signaling network attenuates inappropriate RAL-1 activation in presumptive 1˚cells, thereby preventing potentially contradictory LET-60 signaling through RGL-1-RAL-1 in presumptive 1˚cells.
We similarly tested the expression pattern of a transgene harboring the rgl-1 promoter transcriptional fusion to GFP, sEx14985 [51]. As with the ral-1 promoter transcriptional fusion, the transgenic rgl-1 reporter was expressed in all VPCs early in L3 (S3C Fig). But, unlike the ral-1 transcriptional fusion, GFP from the rgl-1 transcriptional fusion persisted in presumptive 1˚cells throughout vulval patterning and proliferation. After the first VPC cell division, GFP expression was consistently higher in 1˚relative to 2˚cells.
The significance of this change is unclear, since many transcriptional fusions of vulval signaling genes change expression patterns around this time [5-7, 10, 12, 47, 48]; reviewed in [52]. One interpretation is that increased expression of RGL-1 in presumptive 1˚cells accounts for resistance to rgl-1(RNAi) of this putative 1˚-promoting activity of RGL-1.
Our observation with a rgl-1 transcriptional reporter transgene also contradicts our observation of endogenous tagged RGL-1, where we observed ubiquitous RGL-1 expression (Fig 3   Fig 2. RGL-1 and RAL-1 are functionally non-equivalent in VPC patterning. 1000x photomicrographs of late L4 A) wild-type, B) let-60 (n1046gf); gfp(RNAi) compared to 600x photomicrographs of C) let-60(n1046gf); rgl-1(RNAi), and D) let-60(n1046gf); rgl-1(tm2255) animals. White lines = normal 2˚-1˚-2˚L4 vulvae, white arrows = L4 ectopic 1˚pseudovulvae. Black scale bars = 5 μm. Anterior is left and ventral down. E) RNAi depletion of rgl-1 and ral-1 enhance 1˚induction relative to gfp(RNAi) control. Data are the mean ± standard error of the mean (SEM). For statistical reasons single, non-pooled assays are shown, and white numbers represent animals scored therein. Significance was calculated by Kruskal-Wallis, Dunn test. Data shown were scored concurrently and are representative of 4 independent assays (this study) and 6 prior independent assays [12]. (The let-60(n1046gf) 1˚induction baseline is consistently higher when grown on HT115 bacterially-mediated RNAi food compared to the standard OP50; [10,12,48]). F) Deletion of ral-1 but not rgl-1 enhances ectopic 1˚induction by let-60(n1046gf). Data shown are representative of 4 assays, each scored concurrently. G) Re-constructed strains show the same result: ok1921, tm2255 and gm27 deletions fail to significantly enhance ectopic 1˚induction by let-60(n1046gf). Left: Three n1046-containing isolates, with and without rgl-1 mutations and scored concurrently. The concurrently scored MT2124 1˚induction baseline was not significantly different from outcrossed lines DV2214 and DV2215 (see Table 1), N = 30 for each, and from assays that showed the most deviation of double mutants from the single mutant, but are still not significantly different. Right: DV2251 let-60(n1046gf); lon-2(e678) vs. DV2252 let-60(n1046gf); lon-2(e678) gm27 animals scored concurrently but separate from the left group, representative of two assays. H) A general model for opposing RGL-1 GEF/RAL-1-dependent and -independent functions (green = bifunctional, blue = 1˚-promoting, rose = 2˚-promoting; see We observed similar discrepancy between dynamic expression from the ral-1 transcriptional reporter transgene [10] and CRISPR-tagged endogenous RAL-1 [12]. We speculate i) that our promoter fusions lack regulatory sequences that drive ubiquitous expression, but ii) may reflect subtle alterations in expression in VPCs after induction that may shift responsiveness to different signals. We also note that the transcriptional reporter transgene is highly over-expressed, which may alter expression patterns and intensity. Our observations also suggest that CRISPR tags and traditional transcriptional fusion transgene provide complementary views of gene expression.

RGL-1 performs opposing GEF-dependent and GEF-independent functions in VPC fate patterning
Here we describe a putative reduced function (rf) mutation in ral-1. A list of ral-1 and rgl-1 mutations is shown (Fig 4A and 4B; Table 1). The Million Mutation Project (MMP) described random sequence identification of a large collection of mutagenized C. elegans lines, demanding only that mutant animals be viable and fertile over many generations [53]. We analyzed the sole non-synonymous mutation in ral-1, gk628801, which caused an R139H mutation. Arg-139 is conserved in all Ras family members in metazoans. Outcrossed ral-1(gk628801rf) single mutant vulvae were superficially wild type (N = 83). In the let-60(n1046gf) background, ral-1(gk628801rf) resulted in elevated 1˚induction (Fig 4C), consistent with our previously published analysis using ral-1(RNAi) and deletion mutations (Fig 2; [10]). Thus, ral-1 (gk628801rf) reduces RAL-1 2˚-promoting signaling. Notably, ral-1(gk628801rf) animals are viable and fertile, in contrast with PAR-and exocyst-associated phenotypes described for  Light green exon 2: by RNAseq this is a rare mRNA species, and the 20 residues coded for by Exon 2 are not conserved among Drosophila and mammalian RalGEFs. Purple, pink and bright green lines: REM, GEF and RA (Ras Association) domains (The REM domain is a structural component of some but not all Ras family GEFs). Red gk alleles: W163 � and R361Q mutations from the million mutation project. Orange: canonical R324E GEF-deficient mutation in transgenes reEx94 and reEx95. Light blue: coverage of bacterially mediated RNAi clone (library location; [54]). Red lines: sequences deleted by ok1921, tm2255 and gm27 deletions (dotted line indicates that deletion ends out of frame). B) An alignment of a portion of the RalGEF domain containing the canonical GEF-deficient R324E and gk275304 R361Q mutations (top to bottom: Human RGL2, in which the GEF-deficient mutation was validated, fly RalGEF, C. elegans RGL-1). C) The ral-1(gk628801rf) R139H missense mutation enhances 1˚induction in the let-60(n1046gf) background. Data shown were scored concurrently and are representative of two assays. D) rgl-1 mutation gk275304 (R361Q) but not gk275305 (W163 � ) enhances 1˚induction in the let-60(n1046gf) background. Data shown were scored concurrently and are representative of two assays. E) Transgenic rescue of rgl-1(tm2255) in the let-60(n1046gf) background. Transgenic array-bearing animals harboring P lin-31 ::rgl-1 cDNA with P myo-2 ::gfp co-injection marker and their non-array-bearing siblings were scored. The reEx94 (shown) and reEx95 (P<0.006, N = 96 and N = 88) R324E mutant transgenes enhanced 1˚induction relative to their non-transgenic siblings, as scored in separate concurrent assays for each array, suggesting that a GEF-independent 1˚promoting activity of RGL-1 functions cell autonomously in the VPCs. (reEx94 rgl-1(R324E) mutant transgenic animals failed to respond to ral-1(RNAi), S4C Fig) The reEx109 (shown) and reEx110 (also n.s., N = 63 and N = 57) wild-type transgenes failed to alter 1˚induction relative to their non-array-bearing siblings, as scored in separate concurrent assays for each array, suggesting that the GEF-dependent 2˚-promoting activity of RGL-1 also functions cell autonomously in the VPCs. (reEx109 rgl-1(+) transgenic animals responded to ral-1(RNAi), S4D Fig) F) reIs10[P lin-31 ::ral-1A(Q75L), P myo-2 ::gfp] ("ral-1(act)") suppressed ectopic 1˚induction in the let-60 (n1046gf) background; in a single concurrent assay, 1˚induction was further suppressed by rgl-1(tm2255), revealing an opposing RGL-1 signal that may be 1˚promoting. Further validation of reIs10 "ral-1(act)" shown in S4B Fig. Data are the mean ± standard error of the mean (SEM). For statistical reasons single, non-pooled assays are shown, and white numbers represent animals scored therein. Significance was calculated by Kruskal-Wallis, Dunn test. G) A schematic of the bypass experiment in F. Left: canonical and non-canonical RGL-1 activities are opposed and roughly equivalent (in the backgrounds assayed). Right: constitutive, VPC-specific activation of RAL-1 bypasses the GEF activity while also revealing a GEF-independent function of RGL-1, resulting in increased 2˚-promoting signal and concomitant loss of 1˚-promoting signal, with a net decrease of ectopic 1˚cells.
Using the VPC-specific lin-31 promoter [55], in the let-60(n1046gf); rgl-1(tm2255) double mutant background we generated transgenic extrachromosomal arrays expressing VPC-specific RGL-1(+) or RGL-1(R324E), a mutation deficient in GEF catalytic activity in mammalian RalGDS (RalGEF; [56]). Animals bearing the R324E transgenes showed significant increase in ectopic 1˚induction compared to non-array-bearing siblings. This result is consistent with rescue of a GEF-independent function of RGL-1. Animals bearing the wild-type transgenes were not different than their non-array-bearing siblings (Fig 4E). Consequently, we propose that RGL-1 performs GEF-dependent and GEF-independent functions that are genetically separable by mutating the GEF domain. We also hypothesize that the GEF-independent RGL-1 activity can function cell autonomously in the VPCs.

RGL-1 interacts with the 1˚-promoting PI3K-PDK-Akt cascade
Mammalian RalGDS binds to PDK and Akt1 in cultured cells, possibly functioning as a scaffold for PDK and Akt [26,27]. The AGE-1/PI3K-PDK-1 signal has been described as promoting 1˚fate in VPC fate patterning [24], but AKT-1 was not implicated downstream of this process, potentially because of redundancy of AKT-1 and AKT-2 in C. elegans [57,58]. Previously, a gain-of-function mutation in AKT-1 was tested in a genetic background with reduced 1˚induction, not let-60(n1046gf). No effect was found, leading to the model that AKT-1 did not contribute to vulval induction [24]. We re-evaluated the akt-1(mg144gf) mutation in the let-60(n1046gf) background and observed significant increase in ectopic 1˚induction (Fig 5A).

Fig 5. RGL-1 interacts genetically with the 1˚-promoting AGE-1/PI3K-PDK-1-AKT-1 cascade. A)
The constitutively activating akt-1(mg144gf) mutation increased promotion of 1˚fate in the let-60(n1046gf) background, and this effect was partially blocked by rgl-1(tm2255), though the triple mutant was not suppressed to the double mutant baseline level. Animals were scored concurrently. These results were replicated with a re-built strain (S5A Fig) and the strain with rgl-1(ok1921) (S5B Fig). Data are the mean ± standard error of the mean (SEM). For statistical reasons single, non-pooled assays are shown, and white numbers represent animals scored therein. Significance was calculated by Kruskal-Wallis, Dunn test. B) The constitutively activating pdk-1(mg142gf) mutation increased promotion of 1˚fate in the let-60 (n1046gf) background, and was completely suppressed to baseline level by rgl-1(tm2255). Animals were scored concurrently and are representative of two assays. C, D) Mutation of the negative regulatory PTEN ortholog by daf-18(ok480) similarly increased 1˚induction, and was completely suppressed to baseline level by rgl-1(tm2255) (C) and rgl-1(ok1921) (D). Animals for each were scored concurrently and scoring was repeated once, with the same general results. rgl-1(gk275305) (nonsense; S5C  -18(ok480) n1046gf. (S5I Fig). We speculate that JIP-1 functions in the PI3K cascade, but may also function elsewhere in VPC fate patterning, perhaps in its canonical role as a scaffold for JKK and JNK MAP kinases. Further investigation of the role of JIP-1 is beyond the scope of this study.
A common target of the C. elegans PI3K-Akt cascade is inhibition of the DAF-16/FoxO transcription factor [61,62]. We tested the role of daf-16 alleles mu26, mu86 and mgDf47 in different backgrounds, with inconclusive results. Consequently, we were unable to determine the role, if any, of DAF-16/FoxO in VPC fate patterning. Perhaps AKT-1 has a target other than DAF-16 in this system, or the diversity of functionally distinct DAF-16 isoforms results in inconclusive results with daf-16 mutations that perturb all isoforms [63,64].
Taken together, these genetic epistasis experiments, using an assay of parallelism with let-60 (n1046gf), are consistent with RGL-1 contributes to the AGE-1/PI3K-PDK-1/PDK-AKT-1/ Akt cascade in VPC fate patterning. Alone among the genetic tools used, the gain-of-function mutation in the downstream AKT-1 was only partially suppressed by rgl-1(tm2255), while the effect of other Akt cascade activators was completely suppressed by rgl-1(tm2255). Perhaps RGL-1 is essential for the PDK-1 1˚-promoting signal but only partially required for the AKT-1 1˚-promoting signal. We note that mammalian Akt is activated via parallel mechanisms: phosphorylation by upstream PDK and binding of PIP 3 , resulting in recruitment to the plasma membrane and activation [65]. If RGL-1 functions as a scaffold for PDK-1 and AKT-1, its deletion would be expected to result in reduction of the PDK-1 phosphorylation of AKT-1 but not PIP 3 -dependent recruitment of AKT-1 to the plasma membrane. Thus, the pattern of suppression by mutated RGL-1 is consistent with RGL-1 functioning as a scaffold for PDK-1-AKT-1 signaling in 1˚fate induction. However, we cannot rule out the possibility of parallelism between a GEF-independent RGL-1 activity and the AKT-1 cascade, or RGL-1 functioning in a bifurcated cascade downstream of AKT-1 that is not revealed by the mechanism of constitutively activated PDK-1 or DAF-18/PTEN deletion upstream.

Deletion of RGL-1 decreases fidelity of VPC patterning
Based on a relatively small sample size, RGL-1 deletions cause no gross VPC patterning defects, consistent with both AKT-1 and RAL-1 cascades being modulatory, not central. We hypothesized that RGL-1 orchestrates activation of these two potentially opposing cascades-Akt output to presumptive 1˚cells and Ral output to presumptive 2˚cells-to improve robustness of the VPC developmental system in response to environmental stressors. A previous study investigated the impact on VPC fate patterning of environmental stressors: the N2 baseline error rate under laboratory conditions was 0.2% [66]. We similarly investigated the impact of environmental stressors of starvation, heat, and osmotic stress, compared to non-stressful conditions, on wild-type, rgl-1(ok1921), and rgl-1(tm2255) animals.
We evaluated 300 animals per genotype under each condition, totaling 1,200 animals per genotype and 3,600 animals overall (Fig 6A). Similar to previous results [66], wild-type animals exhibited very low error rates (0.0-0.3%), not only in control but also in stressful conditions. In contrast, the error rate in VPC patterning of both rgl-1 mutants was substantially increased across all experimental environments (1.0-3.7%), including control environments, in which patterning error rates were increased more than 15-fold relative to the wild type. These results indicate an environmentally-insensitive increase in error rate in the absence of functional RGL-1. Consequently, we propose that RGL-1 does not impact robustness in response to environmental stressors. Rather, we hypothesize that RGL-1 mitigates stochasticity within the complex signaling network that regulates VPC fate patterning. This observation represents intriguing and unprecedented insight into the role of signaling networks in developmental fidelity.

Discussion
We previously showed that LET-60/Ras-RGL-1/RalGEF-RAL-1/Ral signaling promotes 2V PC fate in support of the necessary and sufficient LIN-12/Notch. Here we analyze two facets of RGL-1 function in C. elegans. First, in marked contrast to RAL-1, we find that deletion of RGL-1 does not overtly perturb development or fertility. This observation contradicts the expectation from mammalian cell culture: since C. elegans RAL-1 is essential and required for exocyst-associated functions in multiple organisms, the expectation was that RGL-1 would be similarly required. We hypothesize either that GTP-loading of RAL-1 is not required as part of its function in the exocyst complex, or that a thus far unknown RalGEF functions redundantly with RGL-1 in exocyst-related by not 2˚-promoting signals.
Second, we find evidence that RGL-1 performs opposing GEF-dependent and GEF-independent functions that promote fidelity of VPC patterning. Our genetic epistasis results are consistent with RGL-1 functioning in the PDK-1-AKT-1 1˚-promoting signal. Our results raise the interesting possibility of opposing and inessential modulatory cascades functioning to combat stochasticity in a signaling network.

RAL-1 but not RGL-1 is essential for development
RAL-1 was previously implicated in essential developmental events in C. elegans [10,25]. Yet despite the ostensible linearity of the Ras-RalGEF-Ral signaling model in mammals [14,23], we unexpectedly found that RGL-1 is not essential. Given that deletion of neither the GEF nor the GAP disrupts exocyst-dependent developmental events but deletion of RAL-1 does so, we hypothesize that the RAL-1 role as a membrane tether for the exocyst is independent of nucleotide-bound state. This hypothesis does not imply that there is no activation-dependent alteration of exocyst function by RAL-1, merely that such is not essential for the described developmental events.
An alternative hypothesis is that other GEFs function redundantly with RGL-1 to regulate the GDP/GTP cycle of RAL-1. We find that RGL-1 and RAL-1 are required for the 2˚-promoting signal. But this other GEF would need to be specific for RAL-1 association with the exocyst and not signaling to promote 2˚fate. While we cannot exclude this possibility, no GEFs fitting that role have been described. RalGPS, a mammalian Ral-selective GEF that contains SH3 and PH domains, is not encoded in the C. elegans genome; [10], nor is TD-60/RCC2, similar to the Ran GEF RCC1 that controls nuclear import/export, which has been proposed to be an atypical GEF for RalA [67].

RGL-1 performs genetically separable and opposing functions in VPC fate patterning
To our surprise, we found that deletion alleles of rgl-1 caused no net alteration of the balance of 1˚and 2˚VPCs in the let-60(gf) background. By genetic analysis, RGL-1 performs its canonical role in promoting 2˚fate as a signaling intermediary between LET-60/Ras and RAL-1 ( [10]; this study), a role that mirrors extensive biochemical and cell biological evidence from mammalian cell culture (reviewed in [14,23]). Here, further genetic analysis reveals an additional, unexpected role for RGL-1, that of a non-canonical signaling participant opposing the canonical role as an activator of RAL-1. This non-canonical role apparently counteracts or "cancels out" the canonical role. In sensitized backgrounds, putative null mutations have no net effect on the sensitive balance between 1˚and 2˚VPC fates. We do not intend to imply that opposing functions are "equal", only that in our assays they appear to counter-balance each other with approximate equivalency. The observation that deletion of RGL-1 increased the basal error rate of patterning 15-fold supports the idea that RGL-1 serves an important function, that of fidelity.
In mammalian cell culture RalGEF-Ral and PDK-Akt cascades worked in concert [26]. Yet mammalian RalGEF-Ral signaling has also been found to oppose canonical Ras effectors in cell culture [68]. We speculate that the relationship between these cascades may depend on cell context, and in the case of vertebrates, which paralog of RalGEF is expressed.
RGL-1 is not the first protein in the VPC fate patterning network found to be bifunctional and promote both 1˚and 2˚fates. Depending on signaling dose and via mechanisms we do not understand, LIN-3/EGF-LET-23/EGFR signaling was found to promote both 1˚and 2f ates, resulting in the graded signal model [8,9,69]. We found that LET-60/Ras can promote 1˚or 2˚fate depending on its use of effector, LIN-45/Raf or RGL-1/RalGEF [10,11]. Yet the situations are not equivalent. LIN-3, LET-23, and LET-60 are essential for vulval induction: RGL-1/RalGEF in VPC fate patterning strong loss results in complete absence of 1˚fate induction and hence a vulvaless phenotype [70][71][72][73], and thus their roles in 2˚fate induction were teased out only in sensitized backgrounds or special assays [8][9][10]. In contrast, RGL-1 functions are solely modulatory and dispensable for 1˚and 2˚fate induction, permitting dissection of its balanced and opposing functions. Furthermore, deletion of RGL-1 does not alter any other known developmental events.

Linking opposed signaling cascades and mitigating development noise
To occur with high fidelity, the VPC fate patterning system must strictly define developmental fields in response to an initial point source of LIN-3/EGF ligand. In other words, it must generate the precise 3˚-3˚-2˚-1˚-2˚-3˚pattern with 99.8% accuracy [66] without mis-specified or ambiguous fates that might block mating and egg laying, which would negatively impact reproductive fitness. A critical question, then, is how the programming of signal transduction networks decreases the potential for errors (developmental stochasticity or "noise"). Previously, the fidelity of VPC fate patterning was thought to be a property that emerges from the combination of three mechanisms: 1) sequential induction sets up the basic pattern, 2) graded signal collaborates with Sequential Induction to more precisely sculpt the initial pattern, and 3) mutual antagonism serves to exclude potentially conflicting signals from cells that are initially specified, thus preventing assumption of ambiguous fates (reviewed in [52]). We speculate that the roles we describe here for RGL-1 define a fourth method that is woven into the other three: orchestration of two modulatory cascades to more sharply demarcate fates as a function of the VPC's spatial relationship to the AC and other VPCs.

C. elegans handling and genetics
Nomenclature is as described [74,75]. All strains were derived from the N2 wild type. Except where noted, animals were cultured on NGM agar plates with OP50 bacteria on at 20˚C [76]. Strains used are shown in S1 Table. Data were analyzed with GraphPad Prism software (GraphPad Software Inc., La Jolla, CA).

VPC induction assays
VPC induction was analyzed by DIC/Nomarski optics (Nikon eclipse Ni with images captured using NIS-Elements AR 4.20.00 software) in late L4 animals on an agar pad (molten 3% NG agar with 5 mM sodium azide) in a 5 μl drop of M9 buffer. Ectopic pseudovulvae were scored as invaginations at 600x or 1000x. Ectopic 1˚vulval induction index, from 0 to 3 ectopic 1˚s, is described elsewhere [10,12]. Under-induced backgrounds were scored as total VPCs induced, typically 0-3 (in under-induced backgrounds the entire vulva is frequently uninduced). To summarize, we scored the wild-type vulval induction based on the stereotypical 2˚-1˚-2˚lineages centered on the AC at the A-P midpoint of the gonadal primordium (forming the "Christmas Tree" or "Stanley Cup" shape). In the let-60(n1046gf) and let-23(sa62gf) backgrounds, the morphology of ectopic 1˚cells generally conformed with the described "cap" structure where the entire 1˚lineage has pulled away from the cuticle [9]. We did not observe ectopic pseudovulvae with the characteristic asymmetrical "beret" lineage of 2˚lineages, where one side remains attached to the cuticle.
As we have described previously [10,12,48], we occasionally observed drift of the severity of the let-60(n1046gf) but not let-23(sa62gf) Muv phenotype. Consequently, we employed a stringent protocol for all VPC induction scoring experiments. We analyzed parental MT2124 and outcrossed strains harboring the n1046gf single mutant, and established that the baseline of undrifted strains averaged 1.2-1.5 ectopic 1˚s. (When grown on bacterially mediated RNAi food source HT115, induction was consistently higher: 1.5-1.8; [10,12,48]). We additionally always worked with freshly thawed or chunked strains; our animals were always freshly derived from a cross or thaw. For all strains harboring the n1046gf mutation we always employed stringent scoring criteria: the n1046gf single mutant was scored first, and experiments deviating from the aforementioned expected range of induction were discarded. Additionally, we only compared genotypes that were scored concurrently, thus minimizing variation from assay to assay.

DIC, epifluorescence and confocal microscopy
For epifluorescent imaging, animals were mounted in 2 mg/ml tetramisole/M9 buffer and visualized using a Nikon Eclipse TE2000U microscope equipped with a DVC-1412 CCD camera (Digital Video Camera Company), with Hamamatsu SimplePCI acquisition software. Confocal images were captured by A1si Confocal Laser Microscope (Nikon) using NIS Elements Advanced Research, Version 4.40 software (Nikon).

Assessment of patterning error rate with environmental insults
The vulval cell lineages and Pn.p fates (and errors) were assessed in early to mid L4 individuals as previously described [66,78]. Nematode populations of N2 wild type and the two rgl-1 mutations were kept in identical environmental conditions for at least three generations prior to experiments. Animals were age-synchronized by hypochlorite treatment and liquid arrest for~24 hours, then randomly allocated to the four experimental environments: control (20˚C), heat stress (29˚C), osmotic stress (250 mM NaCl NGM plates, 20˚C; [79]), and starvation (20˚C; [80]). In starvation conditions, L1 larvae were cultivated on standard NGM plates until they reached the mid L2 stage, at which point they were transferred to unseeded NGM plates containing 1 mg/ml of ampicillin to prevent bacterial growth. After 48 hours, starved animals were transferred to regular NGM plates seeded with E. coli OP50 and the vulval phenotype was scored when animals had reached the early or mid L4 stage [80].