Polarized epidermal growth factor secretion ensures robust vulval cell fate specification in Caenorhabditis elegans

Intercellular communication relies on the spatially and temporally controlled release of signaling molecules by signal-emitting cells. Members of the epidermal growth factor (EGF) family are involved in a variety of cell fate decisions in all metazoans (Massagué and Pandiella, 1993). Secreted EGF ligands bind to receptor tyrosine kinases of the ErbB/EGFR family, which activate different intracellular signaling pathways, such as the RAS/MAPK, PI3K/AKT, JAK/STAT and the PLC pathways, depending on the cellular context (Hynes and MacDonald, 2009). Although many studies have focused on the mechanisms controlling the polarized secretion, internalization and recycling of EGF receptors (Sorkin and Goh, 2008), less is known about the factors controlling the intracellular trafficking of the EGF family ligands. EGF ligands are typically produced as transmembrane precursor proteins. They can either act as a membrane-bound form in a juxtacrine manner or be cleaved by intracellular rhomboid family proteases (Urban et al., 2001) and extracellular metalloproteases, allowing them to be released and signal at a distance (Massagué and Pandiella, 1993). The concentration and temporal duration of an EGF signal can lead to differential responses in the signal-receiving cells, for example during cell fate specification (Katz et al., 1995), apoptosis (Iwamoto et al., 1999) or cell migration (Wyckoff et al., 2004). Moreover, by controlling the subcellular localization of EGF ligands, cells can regulate ligand availability and add directionality to the signal (Dempsey et al., 1997). The basolateral versus apical sorting of transforming growth factor alpha (TGFα) in cultured epithelial cells is important for restricting EGFR activation to the basolateral compartment (Singh and Coffey, 2014). One of the few examples demonstrating polarized EGF secretion during animal development in vivo is the release of the Drosophila EGF ligand Spitz by photoreceptor neurons (Yogev et al., 2010).

The Caenorhabditis elegans genome encodes a single EGF-like growth factor, termed LIN-3, and an EGFR receptor homolog, termed LET-23 (Sundaram, 2013). Thus, the lack of genetic redundancy has greatly simplified analysis of the EGF signaling network in C. elegans. The LIN-3/LET-23 pathway controls many cell fate decisions during larval development as well as various adult functions. In particular, development of the C. elegans hermaphrodite vulva is an excellent model for studying cell fate decisions controlled by EGF signaling at single cell resolution (Sternberg, 2005). During the first larval stage (L1), 12 epidermal Pn.p cells (P1.p to P12.p) are generated in the ventral epidermis (Fig. 1). Six Pn.p cells in the mid-body region (P3.p to P8.p) are specified by a Wingless signal to become vulval precursor cells (VPCs) that express LET-23 and are competent to differentiate into vulval cells (Eisenmann et al., 1998). VPC differentiation is induced by the uterine anchor cell (AC) located dorsally to the VPCs in the somatic gonad (Kimble, 1981). From the L2 stage on, the AC expresses LIN-3 (Hill and Sternberg, 1992). Even though LIN-3 is secreted to the plasma membrane as a type I transmembrane protein similar to TGFα, the AC can induce distant VPCs without making direct contact (Hill and Sternberg, 1992). Thus, at least a fraction of LIN-3 must be released from the AC, which probably occurs by shedding of the extracellular domain rather than by intracellular proteolysis, as the intracellular rhomboid protease ROM-1 does not act in the AC (Dutt et al., 2004). The LIN-3 signal activates, via its receptor LET-23, the RAS/MAPK pathway, which specifies the primary (1°) cell fate in the nearest VPC P6.p (Fig. 1) (Sternberg, 2005; Sundaram, 2013). Strong RAS/MAPK signaling in P6.p leads to the upregulation of LET-23 and of DSL family NOTCH ligands (Chen and Greenwald, 2004; Greenwald and Kovall, 2013). The latter activate the LIN-12 NOTCH receptor in the adjacent VPCs (P5.p and P7.p) by lateral signaling to repress the 1° and induce the secondary (2°) cell fate (Berset et al., 2001; Yoo et al., 2004). The distal VPCs (P3.p, P4.p and P8.p) that receive neither the inductive LIN-3 nor the lateral DSL signal adopt the uninduced, tertiary (3°) cell fate. The 3° VPCs divide once and then fuse with the surrounding epidermis (hypodermal syncytium; hyp7). This interplay between the EGFR/RAS/MAPK and NOTCH pathways results in an invariant 2°-1°-2° vulval cell fate pattern.

Vulval induction in wild-type larvae is extremely robust (Félix and Barkoulas, 2012). P6.p, the VPC located closest to the AC adopts the 1° cell fate when animals are raised under standard conditions. Rare errors in vulval fate specification, such as shifts in the 1° fate from P6.p to another VPC or the hyperinduction of extra VPCs, only occur when animals are grown under suboptimal conditions, for example at low temperature or under food starvation (Braendle and Félix, 2008). The AC produces a limiting amount of the inductive LIN-3 signal, whereas LET-23 expressed by the VPCs provides excess LIN-3 binding sites (Barkoulas et al., 2013; Simske and Kim, 1995). Hence, the nearest VPC P6.p sequesters most of the inductive signal and thereby prevents induction of additional, more distal VPCs (Hajnal et al., 1997). Efficient LIN-3 sequestering and vulval induction both require the localization of LET-23 to the basolateral membrane compartment of the VPCs facing the AC (Whitfield et al., 1999). Basolateral LET-23 localization is maintained by the tripartite LIN-2/LIN-7/LIN-10 receptor localization complex (Kaech et al., 1998). The apical mislocalization of LET-23, for example in lin-2(0) mutants, leads to a strong reduction in vulval induction and a penetrant vulvaless (Vul) phenotype (Hoskins et al., 1996). However, in combination with a loss-of-function (lf) mutation in the RAS GTPase-activating protein GAP-1, which causes mild hyperactivation of the RAS/MAPK pathway, apical LET-23 mislocalization causes ectopic induction of VPCs distant from the AC and a multivulva (Muv) phenotype (Hajnal et al., 1997). This finding has been attributed to an expanded range of the LIN-3 signal caused by loss of LIN-3 sequestering by P6.p.

Here, we show that not only basolateral LET-23 localization in the VPCs, but also the polarized secretion of LIN-3 by the AC towards the 1° VPC, is necessary for robust vulval induction. Furthermore, an unbiased screen for genes controlling polarized LIN-3 secretion identified a putative neuropeptide-like ligand and a GPCR required for LIN-3 polarity and robust vulval induction.

In wild-type C. elegans larvae, the three proximal VPCs (P5.p, P6.p and P7.p) are induced to adopt vulval cell fates. Thus, the vulval induction index (VI) in the wild-type is exactly 3. An lf mutation in the RAS-GAP gene gap-1(ga133lf) enhances the activity of the inductive RAS/MAPK signaling pathway, but not enough to cause induction of additional VPCs (Hajnal et al., 1997). Only one in 300 gap-1(lf) animals exhibited a hyperinduced phenotype (Fig. 2A,E,F). gap-1(lf) can thus be utilized as a sensitized genetic background to detect changes in LIN-3 distribution, such as the reduced LIN-3 sequestering caused by apical LET-23 mislocalization (Hajnal et al., 1997).

The unc-6 gene encodes a Netrin homologue that is secreted by ventral cord motor neurons and polarizes the AC along the dorso-ventral axis towards the VPCs that are aligned on the ventral midline (Ziel et al., 2009). Surprisingly, unc-6(ev400lf); gap-1(ga133lf) double mutants exhibited a 20% penetrant hyperinduced phenotype as a result of ectopic induction of distal VPCs (Fig. 2C,F). Similar hyperinduced phenotypes were observed after combining gap-1(lf) with mutations in other known AC polarity regulators, such as the Netrin receptor unc-40 (Ziel et al., 2009) or the unc-40 downstream effector madd-2 (Fig. 2D,F) (Morf et al., 2013). Furthermore, depolarization of the AC by unc-6(lf) enhanced the Vul phenotype and reduced the VI of the lin-3(e1417) reduction-of-function allele or the let-23(sy1) receptor mislocalization mutant (Fig. 2E,G).

To test whether the hyperinduced vulval phenotype is caused by depolarization of the AC, we introduced an unc-40::gfp rescue minigene (zhEx668) driven by the AC-specific mk62-63 cdh-3 promoter fragment (Ziel et al., 2009) into the unc-40(lf); gap-1(lf) background. The unc-40::gfp transgene was expressed in the AC and in ventral nerve cord (VNC) neurons of mid-L2 to early L3 larvae, but not in the VPCs (Fig. S1A,B). Three independent transgenic lines exhibited partial suppression of the hyperinduced phenotype and a reduced VI for unc-40(lf); gap-1(lf) double mutants (Fig. S1C). Because VNCs neurons are unlikely to affect VPC differentiation, unc-40 probably acts in the AC to inhibit vulval induction.

In wild-type larvae, the AC invariably induces the most proximal VPC (P6.p) to adopt the 1° cell fate (Braendle and Félix, 2008). Besides the changes in vulval induction, we also observed shifts of the 1° fate from P6.p to P5.p, or rarely to P7.p, in unc-6(lf) single mutants as well as in unc-6(lf); gap-1(lf) double mutants (Fig. 2E,H).

These findings suggested that dorso-ventral AC polarity is needed for the robust induction of the 1° fate in P6.p. Similar to the case of apical LET-23 mislocalization, loss of AC polarity may affect the range and distribution of the inductive LIN-3 signal among the VPCs.

One possible explanation for the vulval phenotypes observed in AC polarity mutants is that the AC normally secretes LIN-3 in a polarized fashion in the direction of the closest VPC. To directly observe LIN-3 localization in the AC, we inserted an mNeongreen (mNGr) fluorophore cassette into the lin-3 locus directly after the predicted signal sequence (after Leu24 in LIN-3A) and N-terminally to the EGF domain, using CRISPR/Cas9-mediated genome engineering [mNGr::lin-3(zh112), Fig. S2] (Dickinson et al., 2015). mNGr::lin-3(zh112) animals exhibited an overall wild-type phenotype without any obvious defects in vulval development, indicating that the mNGr::LIN-3 fusion protein retains its activity at levels comparable with the wild-type LIN-3 protein. In L2 and L3 larvae, mNGr::LIN-3 expression was detected in intracellular punctae and on the basal cortex of the AC (Fig. 3A,C,E). No extracellular mNGr::LIN-3 signal could be detected, suggesting that the majority of mNGr::LIN-3 remains attached to the AC surface after its secretion to the plasma membrane, or that a cleavage product of the mNGr::LIN-3 protein is released from the AC and rapidly taken up by the adjacent VPCs or degraded. The same localization pattern was observed with a multicopy GFP::LIN-3 transgene (zhIs67), in which the GFP tag had been inserted at the same position as in the endogenous mNGr::lin-3(zh112) reporter strain (Fig. S3).

In wild-type larvae, mNGr::LIN-3 was enriched towards the ventral cortex of the AC from the mid-L2 stage until the mid-L3 stage (Fig. 3A,C,E). To quantify LIN-3 polarity, we recorded optical sections through the AC of animals expressing mNGr::LIN-3 together with the mCherry::PLCδ^PH reporter that labels phosphatidylinositol-(4,5)-bisphosphate (PIP₂)-rich membranes and serves as a global AC polarity marker (Ziel et al., 2009). After measuring the average intensities in summed z-projections in the ventral and dorsal half of the AC, we calculated for each animal a dorso-ventral polarity index (I_DV) for both mNGr::LIN-3 and PIP₂ by dividing the signal intensity in the ventral by the intensity in the dorsal half of the AC (see Materials and Methods). We first measured the mNGr::LIN-3 and PIP₂ I_DV in wild-type larvae between the early L2 and mid-L3 stages before the first round of VPC divisions. The developmental stage was assigned according to the animal's gonad length (Kimble and Hirsh, 1979). The average mNGr::LIN-3 I_DV was around 1.3 at all three stages, with only a few cases showing an I_DV below 1.0 (Fig. 3H,J,L). Around 90% of the animals across the three stages analyzed had an I_DV higher than 1.1 (Fig. 3H,J,L). The PIP₂ global polarity marker showed a slightly stronger polarization of the AC, with an average I_DV above 1.5 and an I_DV≥1.2 in around 90% of the animals at the three stages analyzed (Fig. 3G,I,K).

Taken together, the endogenous mNGr::lin-3 and transgenic gfp::lin-3 reporters revealed a polarized distribution of LIN-3 in the AC before and during vulval induction, but no extracellular signal was detectable.

We next tested whether global depolarization of the AC affects LIN-3 polarity. The unc-6(ev400) loss-of-function allele causes a strong reduction in AC polarity that manifests in a more uniform PIP₂ localization. Accordingly, the average PIP₂ I_DV was decreased in unc-6(lf) mutants from the mid-L2 stage on, such that fewer than 50% of the mid- to late L2 larvae had a PIP₂ I_DV greater than 1.2 (Fig. 3B,D,G,I). PIP₂ polarity remained reduced in early to mid-L3 unc-6(lf) larvae, although the effect was less pronounced (Fig. 3F,K).

LIN-3 polarity was also decreased in unc-6(lf) mutants (Fig. 3B,D,F). In mid- to late L2 larvae, the time period when the AC signal selects the 1° VPC, the LIN-3 I_DV was below 1.1 in around half of the unc-6(lf) mutants analyzed (Fig. 3H,J). By the mid-L3 stage, the effect of unc-6(lf) on LIN-3 polarity was less pronounced (Fig. 3L). Moreover, we observed a strong correlation between LIN-3 and PIP₂ polarity on a per animal basis. There was a significant increase in the fraction of animals showing both PIP₂ and LIN-3 depolarization in unc-6(lf) animals compared with the wild-type (Fig. S4).

We conclude that the polarized distribution of LIN-3 in the AC depends on the global polarity established by the UNC-6 Netrin signal.

The observation that depolarization of the AC in the sensitized gap-1(lf) background causes ectopic vulval induction and a Muv phenotype opened the possibility to conduct a systematic screen for global AC polarity regulators and possibly also for specific regulators of LIN-3 polarity. To this aim, we performed a genome-wide RNA interference (RNAi) screen in an rrf-3(pk1426); gap-1(ga133) hypersensitive background and searched for animals exhibiting a Muv phenotype (Simmer et al., 2002). Through this approach, we identified 51 primary candidate genes that showed a reproducible Muv phenotype upon RNAi knockdown in the rrf-3(pk1426);gap-1(ga133) background (Table S1). Among the candidates identified were three genes previously known to act in the UNC-6 Netrin pathway that polarizes the AC: unc-40, unc-73 and madd-2 (highlighted in green in Table S1) (Alexander et al., 2010; Morf et al., 2013; Ziel et al., 2009). To search for genes specifically regulating polarized LIN-3 localization in the AC, we screened again in a strain carrying the PIP₂ global AC polarity marker together with the multicopy GFP::LIN-3 transgene zhIs67 to increase signal intensity. This approach allowed us to identify candidates that specifically change LIN-3 distribution without affecting global AC polarity. Three of the 51 primary candidates exhibited altered GFP::LIN-3 but unchanged PIP₂ localization after RNAi knockdown. Those genes were sra-9, srh-247 and nlp-26 (Table S1, highlighted in orange). Genes sra-9 and srh-247 encode putative seven-pass transmembrane G-protein coupled receptors (GPCRs) (Troemel et al., 1995); nlp-26 encodes a secreted neuropeptide-like protein (Nathoo et al., 2001). However, a role during vulval development has not been reported for any of these genes so far. Because RNAi of srh-247 showed a weaker and more variable phenotype, we focused further analyses on the sra-9 and nlp-26 genes.

To confirm the RNAi phenotypes, we generated sra-9(lf) and nlp-26(lf) mutations using the CRISPR/Cas9 system (Arribere et al., 2014; Dickinson et al., 2015). The sra-9(zh108) allele contains a 2.5 kb deletion spanning the entire coding region. The gfp::nlp-26(zh113) allele carries a gfp insertion in the predicted signal peptide, generating a premature stop codon before the neuropeptide sequence. The endogenous gfp::nlp-26(zh113) reporter did not show any GFP expression, indicating that NLP-26 is expressed at very low levels.

We detected no significant differences in PIP₂ polarity in sra-9(lf) or nlp-26(lf) mutants at the different stages analyzed (Fig. 4A-F,H,J). Thus, sra-9 and nlp-26 are not required for global AC polarity. However, LIN-3 polarity in sra-9(lf) and nlp-26(lf) mutants was significantly reduced at the mid- and late L2 stages, and to a lesser extent in early to mid-L3 larvae (Fig. 4A-F,G,I). In particular, the fraction of animals showing a LIN-3 I_DV>1.1 was reduced in sra-9(lf) and nlp-26(lf) mutants (Fig. 4G,I). After the mid-L3 stage, when the VPC fates had been determined and the first round of VPC division had been completed, there was no significant difference in LIN-3 polarity between sra-9(lf) and nlp-26(lf) mutants and the wild-type (data not shown). In contrast to unc-6(lf) mutants, LIN-3 and PIP₂ polarity did not correlate in sra-9(lf) and nlp-26(lf) mutants at the level of individual animals. There was a significant increase in the fraction of animals with depolarized LIN-3 that maintained polarized PIP₂ expression (Fig. S4). Thus, SRA-9 and NLP-26 are required for LIN-3 polarity independently of global AC polarity. Moreover, sra-9(lf) and nlp-26(lf) mutants did not show defects in AC invasion or vulval morphogenesis, two processes that depend on global AC polarity and are perturbed in unc-6(lf) mutants (Estes and Hanna-Rose, 2009; Ziel et al., 2009).

These results indicate that the asymmetric distribution of LIN-3 in the overall polarized AC is not the default state, but rather an actively regulated process.

Mutants sra-9(lf) and nlp-26(lf) exhibited hyper- or hypo-induced vulval phenotypes at a low frequency when combined with the gap-1(lf) mutation (Fig. 5A). Rarely, even nlp-26(lf) single mutants exhibited abnormal vulval induction. In addition, we observed 1° cell fate shifts from P6.p to P5.p in nlp-26(lf) single mutants as well as in sra-9(lf); gap-1(lf) double mutants (Fig. 5B). Given the increased error rate in vulval induction in sra-9(lf) and nlp-26(lf) mutants, we examined whether the loss of LIN-3 polarity affects activation of the MAP kinase MPK-1 in VPCs, using an ERK-nKTR biosensor (de la Cova et al., 2017). Briefly, upon phosphorylation by MPK-1 the ERK-nKTR::mClover protein translocates from the nucleus to the cytoplasm of the VPCs. By co-expressing a nuclear mCherry::H2B marker together with the ERK-nKTR::mClover biosensor on a single bi-cistronic mRNA, MPK-1 activity can be quantified as the ratio of the nuclear mCherry::H2B (red) signal intensity divided by the nuclear ERK-nKTR::mClover (green) signal intensity (Fig. 5C-E). For each VPC and animal, the red/green ratio was normalized to the mean of the ratios in P4.p through P8.p in that animal (Fig. 5F-I) (P3.p was not analyzed because it only assumes a VPC fate in around 50% of the animals). In wild-type larvae, MPK-1 activity was highest in P6.p in 91% of the animals, consistent with published data of de la Cova et al. (2017) (Fig. 5F,K). In unc-6(lf) mutants, the difference between MPK-1 activity in P6.p and the other VPCs was diminished overall, and the frequency of animals exhibiting highest MPK-1 activity in P6.p was reduced to 71% (Fig. 5G,K). In sra-9(lf) and nlp-26(lf) mutants, we also measured a greater variability in MPK-1 activity with a significant fraction of animals showing average or below average MPK-1 activity in P6.p (Fig. 5H-K).

Thus, the robust MPK-1 activation observed in P6.p of wild-type larvae is perturbed by global AC depolarization in unc-6(lf) mutants or by specific loss of LIN-3 polarity in sra-9(lf) or nlp-26(lf) mutants. This increased variability in MPK-1 activity after the loss of LIN-3 polarity in the AC may underlie the elevated error rate in vulval induction.

One mechanism for ensuring robust vulval induction involves the progressive alignment of P6.p with the AC during the L2 stage (Grimbert et al., 2016). Despite initial variability in the location of the AC relative to P6.p in wild-type early L2 larvae, the AC and P6.p are precisely aligned to each other by the early L3 stage. LIN-3 signaling is required for the migration of VPCs towards the AC (Grimbert et al., 2016). Together with lateral LIN-12 NOTCH signaling between the VPCs, which compete for the inductive signal, this results in an almost invariant alignment of the AC with the nearest VPC P6.p (Huelsz-Prince and van Zon, 2017).

We thus investigated whether changes in LIN-3 polarity or in global AC polarity affect the AC to P6.p alignment. For this purpose, we measured the distance between the AC and P6.p, as well as the distance between P6.p and P5.p (or P7.p if the AC was located posterior to P6.p) and calculated the relative P6.p to AC alignment index R (Fig. 6A,B,H). In wild-type mid- to late L2 and early L3 larvae, R never exceeded a value of 0.4, signifying that the AC was always situated closest to P6.p. In 18% (n=35) of unc-6(lf) L2 larvae, R exceeded a value of 0.4, and the AC was occasionally situated equidistant or even closer to P5.p or P7.p (Fig. 6A). By the early L3 stage, the AC in unc-6(lf) mutants had aligned with P6.p in 90% of the cases (Fig. 6B). In nlp-26(lf) and sra-9(lf) mutants, the defects in AC to P6.p alignment were more subtle. The mean alignment index R was slightly higher and the variability (i.e. variance) of R was increased in sra-9(lf) and nlp-26(lf) larvae at the mid- to late L2 stage compared with wild-type larvae (Fig. 6A). By the early L3 stage, the AC and P6.p were properly aligned in all sra-9(lf) and most nlp-26(lf) larvae (Fig. 6B).

We conclude that polarized localization of LIN-3 in the AC is not absolutely required for AC to P6.p alignment, but it increases the fidelity of the process. The increased error rate in vulval induction and the occasional 1° cell fate shifts in unc-6(lf), sra-9(lf) and nlp-26(lf) mutants may be caused by the less precise AC to P6.p alignment.

In the course of our analysis, we noticed that the mNGr::LIN-3 distribution in the AC was not only polarized along the dorso-ventral axis, but also biased along the anterior-posterior axis towards P6.p before the AC and P6.p had completely aligned (Fig. 6C). We thus measured the LIN-3 polarity along the anterior-posterior axis, applying the same method as described above for the dorso-ventral polarity index to calculate a P6.p-directed anterior-posterior LIN-3 polarity index I_AP (Fig. 6H). An index I_AP>1 indicates that LIN-3 distribution is biased towards P6.p. We measured I_AP in wild-type, unc-6(lf), sra-9(lf) and nlp-26(lf) mutants between the mid-L2 and early L3 stages, before P6.p and the AC had fully aligned (i.e. in animals with R>0.09). In wild-type larvae, the mean I_AP was significantly higher than 1 (P<0.0001), indicating that LIN-3 distribution in the AC is directed towards P6.p (Fig. 6C,G). In sra-9(lf) and nlp-26(lf) mutants, the mean I_AP was reduced and did not significantly differ from 1 [P=0.23 for sra-9(lf) and P=0.16 for nlp-26(lf)]. Thus, P6.p-directed LIN-3 polarity is lost in sra-9(lf) and nlp-26(lf) mutants (Fig. 6E-G). Interestingly, the mean I_AP in unc-6(lf) mutants was significantly higher than 1 (P<0.01) and comparable with the wild-type I_AP (Fig. 6D,G). Thus, the P6.p-directed LIN-3 polarity in the AC does not depend on unc-6. We also tested whether the relative distance R between the AC and P6.p correlated with the P6.p-directed LIN-3 polarity I_AP. (In those cases where the AC aligned with P5.p or P7.p, we quantified I_AP and R relative to these VPCs.) In wild-type larvae, we observed a tendency to increased P6.p-directed LIN-3 polarity I_AP with decreased AC to P6.p distance R (Fig. S5A). Thus, as P6.p approached the AC LIN-3, localization became more polarized towards P6.p. This trend was weaker in unc-6(lf) and absent in sra-9(lf) and nlp-26(lf) mutants (Fig. S5B-D).

Our data thus far suggest that during the process of AC alignment, the AC senses the proximity of the nearest VPC and progressively channels LIN-3 secretion into this direction. The perception of the nearest VPC appears to be preserved in unc-6(lf) mutants, although the overall efficiency is slightly reduced as a result of loss of dorso-ventral AC polarity. On the other hand, NLP-26 and SRA-9 are required, directly or indirectly, for the communication between the AC and the most proximal VPC in order to induce the asymmetric distribution of LIN-3.

Because nlp-26 encodes a predicted secreted neuropeptide-like protein, we hypothesized that NLP-26 may constitute a signal secreted by the VPCs or the VNC. The gfp::nlp-26(zh113) strain, which carries a gfp insertion in place of the signal sequence, did not show any GFP expression, probably because endogenous nlp-26 expression levels are very low. We thus generated a multicopy transcriptional nlp-26 reporter by fusing 1.4 kb of the nlp-26 5′ regulatory region to an nls::lacZ::gfp reporter cassette in order to confine the GFP signal to the nuclei (zhEx632[P_nlp-26-nls::lacZ::gfp]). This transcriptional nlp-26 reporter was dynamically expressed in the VPCs and their sister Pn.a neurons in the VNC. In early to mid-L2 larvae, P_nlp-26-nls::lacZ::gfp was expressed in all VPCs and their sister Pn.a neurons (Fig. 7A). In addition, nlp-26 was strongly expressed in the hyp7 cell at all stages (Fig. 7A, inset). By the late L2/early L3 stage, nlp-26 transcription was upregulated in P6.p and the P6.a neurons, while expression faded in the other VPCs (Fig. 7B,C). nlp-26 continued to be expressed in the P6.px daughter cells of mid-L3 larvae (Fig. 7D). Note that descendants of the 3° VPCs P3.p, P4.p and P8.p began to express nlp-26 after they had fused with hyp7 (Fig. 7D).

To observe sra-9 expression, we inserted by CRISPR/Cas9-mediated genome editing a gfp::3×Flag cassette in frame at the 3′-end of the sra-9 coding sequences to generate the endogenous sra-9 reporter (zh151[sra-9::gfp::loxP::3×Flag]). In mid-L2 larvae, SRA-9::GFP was expressed in the AC and other uterine cells but not in the VPCs or the VNC (Fig. 7E). From the early L3 stage on, SRA-9::GFP expression was also observed in the VPCs, whereas AC expression faded until it was absent in mid-L3 (Pn.px stage) larvae (Fig. 7F,G).

The nlp-26 and sra-9 expression patterns suggest that NLP-26 might act as a signal secreted by the VPC that polarizes LIN-3 trafficking in the AC, whereas SRA-9 could function in the AC to transduce the VPC signal. We therefore performed Pn.p cell- and uterine-specific RNAi of nlp-26 and sra-9 using the zhEx418[lin-31::rde-1] (Haag et al., 2014) and qyIs102[fos-1ap::rde-1] (Matus et al., 2010) transgenes, respectively, in the rrf-3(lf); rde-1(lf) background, as described in Yang et al. (2017) (Fig. S6A,B). We then quantified the polarity of the GFP::LIN-3 reporter in the AC under the different RNAi conditions. Pn.p cell-specific sra-9 RNAi had no significant effect on GFP::LIN-3 polarity compared with empty vector controls (Fig. 8A,B,G), but nlp-26 RNAi reduced GFP::LIN-3 polarity (Fig. 8C,G). Conversely, uterine-specific sra-9 RNAi reduced LIN-3 polarity, whereas nlp-26 RNAi had no significant effect compared with empty vector controls (Fig. 8D-F,H). To further define the cellular focus of sra-9 in the uterus, we expressed wild-type sra-9 under control of the lin-3 AC-specific element (Hwang and Sternberg, 2004) together with an mCherry reporter in the sra-9(lf) background, using the bi-cistronic transgene zhIs143[P_ACEL>sra-9::SL2::mCherry] (Fig. S6C). AC-specific expression of sra-9 rescued the GFP::LIN-3 polarity defects of sra-9(lf) mutants, suggesting that SRA-9 functions in the AC (Fig. 8I-L). Thus, NLP-26 may act as a feedback signal secreted by the VPCs to polarize, via SRA-9, the secretion of LIN-3 by the AC.

C. elegans vulval cell fate specification serves as an excellent in vivo model to analyze the subcellular localization of the EGF ligand and receptor at single cell resolution. It has been proposed that the limiting amounts of LIN-3 secreted by the AC form a gradient that can act in a dose-dependent manner to specify the different fates of the proximal VPCs (Katz et al., 1995). Moreover, LET-23 expressed on the basolateral surface of the nearest VPC (P6.p) sequesters most of the LIN-3 signal (Hajnal et al., 1997) (Fig. 9A). If LET-23 is mislocalized to the apical VPC compartment, then less LIN-3 is sequestered by P6.p and the inductive signal can reach the distal VPCs (Fig. 9B). Thus, LET-23 localized on the basolateral membrane of P6.p limits the range of the graded LIN-3 signal.

Here, we show that both LET-23 localization in the VPCs and polarized secretion of LIN-3 by the AC regulate the range of the inductive signal (Fig. 9C). Our results indicate that the AC polarizes LIN-3 trafficking from the endoplasmic reticulum towards its ventral side that faces the VPCs. This polarized secretion depends on the global AC polarity established by the ventral to dorsal UNC-6 Netrin gradient (Ziel et al., 2009). LIN-3 secretion is then further enriched in the direction of the nearest VPC, usually P6.p, along the anterior-posterior axis. Although dorso-ventral LIN-3 polarity is strongly reduced in unc-6 netrin(lf) mutants, the secretion of LIN-3 in the direction of the nearest VPC along the anterior-posterior axis occurs independently of the UNC-6-dependent AC polarity.

Using an unbiased, genome-wide approach, we identified two genes, sra-9 and nlp-26, that are specifically required for polarized LIN-3 localization and do not affect global AC polarity: sra-9 encodes an orphan GPCR protein and nlp-26 a secreted neuropeptide-like protein. nlp-26 is expressed in the VNC neurons and the VPCs, where it shows a dynamic expression pattern until the late L2/early L3 stage, when expression becomes upregulated in the 1° VPC that sequesters the LIN-3 signal. Thus, NLP-26 could act as a secreted cue from the VPCs that polarizes LIN-3 trafficking in the AC. This hypothesis is supported by the finding that Pn.p cell-specific depletion of nlp-26, but not of sra-9, reduced LIN-3 polarity in the AC. An endogenous sra-9 reporter was expressed in the AC of mid-L2 larvae, the period when LIN-3 polarity is first observed, and AC-specific expression of sra-9 rescued the LIN-3 polarity defects of sra-9(lf) mutants. Thus, SRA-9 could be required for the transduction of the NLP-26 signal in the AC. Whether SRA-9 functions as a receptor for NLP-26 in the AC remains to be determined. Taken together, SRA-9 and NLP-26 appear to be part of a sensing mechanism used by the AC to detect the location of the nearest VPC and focus LIN-3 secretion into the direction of the future 1° VPC. Because VPCs migrate towards the AC along a LIN-3 gradient while competing for the inductive AC signal via DELTA/NOTCH-mediated lateral inhibition, AC to P6.p alignment and 1° vulval cell fate specification are tightly coupled (Huelsz-Prince and van Zon, 2017). Our data indicate that the AC plays an active part in this feedback mechanism. We propose that the AC directs and channels LIN-3 secretion towards the ‘leader’ in the race of the VPCs for the 1° cell fate (Fig. 9A). As a result, the LIN-3 gradient becomes progressively restricted and the closest VPC (the ‘winner’) receives most of the inductive signal.

Vulval fate specification is extremely robust. In the wild-type larvae grown under standard conditions, P6.p adopts the 1° fate in 99.9% of the cases and always induces the 2° fate in its neighbors P5.p and P7.p (Braendle and Félix, 2008). Previous studies have shown that this robustness is achieved through cross-talk between the inductive EGFR/RAS/MAPK and the lateral DELTA/ NOTCH signaling pathways, which results in the almost invariable alignment of the 1° vulval cell P6.p with the AC (Barkoulas et al., 2013; Berset et al., 2001; Grimbert et al., 2016; Yoo et al., 2004). Here, we show that the polarized and directed secretion of LIN-3 by the AC is an additional mechanism that ensures robust vulval induction. Even though nlp-26(lf) and sra-9(lf) mutants do not exhibit penetrant vulval phenotypes, they do display (at a low penetrance) errors in fate patterning, shifts in 1° fate selection and a less precise AC to P6.p alignment. All these defects are probably due to increased variability in MAPK activation in P6.p, as observed using a MPK-1 biosensor. Thus, LIN-3 secretion by the AC is an actively regulated process that depends on the global AC polarity determined by the ventral to dorsal UNC-6 Netrin gradient and on a sensing mechanism to detect the nearest VPC, which is mediated by the neuropeptide-like protein NLP-26 with the GPCR SRA-9 as a candidate receptor. Whether SRA-9 and NLP-26 regulate the secretion of other proteins by the AC in addition to LIN-3 remains to be determined. Moreover, nlp-26(lf) and sra-9(lf) may act in other tissues besides the VPCs and AC, respectively, and could also affect AC alignment and vulval induction indirectly.

In summary, the directed secretion of a growth factor towards the target tissue may be a common strategy to achieve the robust and efficient patterning of cell fates. In mammalian cells also, the confined subcellular localization of EGF family ligands regulates the range and directionality of their signaling activity. For example, TGFα is sorted to the basolateral membrane compartment of epithelial MDCK cells to restrict signal diffusion and promote juxtacrine signaling (Singh and Coffey, 2014). Similarly, Pro-Epiregulin is sorted to the basolateral compartment of MDCK cells and its mislocalization to the apical compartment causes hyperproliferative, locally invasive tumors (Singh et al., 2013). It will be interesting to investigate whether similar feedback mechanisms between signal sending and signal receiving cells are used in the EGF signaling pathway of other species.

Unless noted otherwise, C. elegans strains were maintained at 20°C on NGM (Nematode Growth Medium) plates applying standard methods (Brenner, 1974). The C. elegans Bristol variety N2 was used as wild-type reference, and strains used for the experiments were derivates of N2. The following alleles and transgenes were used: LGI: unc-40(e271) (Hedgecock et al., 1990), zhIs127[P_{ACEL Δpes10}>mCherry, unc-119(+)]; LGII: qyIs23[P_cdh-3::PLC∂PH::mCherry, unc-119(+)] (Ziel et al., 2009), arTi85[P_lin-31::ERK-KTR(NLS3)-mClover-T2A-mCherry-H2B::unc-54 3′UTR, rps-27p::NeoR::unc-54 3′UTR] (de la Cova et al., 2017), rrf-3(pk1426) (Simmer et al., 2002), let-23(sy1) (Katz et al., 1996), ttTi5605 (Frøkjær-Jensen et al., 2008), sra-9(zh108) (this study) and sra-9(zh151[sra-9::gfp::loxP::3×Flag]) (this study); LGIII: unc-119(ed3), madd-2(tr103) (Alexander et al., 2010), oxTi444, zhIs143[P_{ACEL Δpes10}>sra-9_genomic::SL2::mCherry::unc-54 3′UTR] and unc-119(+) (this study); LGIV: lin-3(e1417) (Hwang and Sternberg, 2004) and lin-3(zh112[mNGr::loxP::3×Flag]); LGV: nlp-26(zh113[gfp::loxP::3×Flag]) (this study) and rde-1(ne219); LGX: unc-6(ev400) (Hedgecock et al., 1990) and gap-1(ga133) (Hajnal et al., 1997); unknown LG: zhIs67[gfp::lin-3, unc-119(+)] (this study) and qyIs102[fos-1ap::rde-1;myo2::yfp; unc-119] (Matus et al., 2010). Extrachromosomal arrays used were zhEx418[lin-31::rde-1, myo2-mcherry] (Haag et al., 2014), zhEx632[P_nlp-26-nls::lacZ::gfp, myo2-mcherry] (this study) and zhEx668.1 to zhEx668.3[P_cdh-3::unc-40_minigene::gfp::unc-54 3′UTR] (this study).

DNA fragments were isolated by PCR using templates and primers described in the supplementary Materials and Methods and assembled using Gibson recombinase (Gibson et al., 2009) or by ligation of restriction fragments.

To insert the mNGr::3×Flag sequence in the 5′ region of the lin-3 locus, the CRISPR/Cas9 system according to Dickinson et al. (2015) was applied. The repair template plasmid pLM5 was injected at a concentration of 8 ng/μl, the two single guides with integrated CAS9 plasmids pLM12 and pLM13 at a concentration of 40 ng/μl together with the recommended co-injection markers pGH8 (Addgene 19359) at 10 ng/μl, pCFJ104 (Addgene 19328) at 5 ng/μl and pCFJ90 (Addgene 19327) at 2.5 ng/μl.

To insert the gfp::3×Flag sequence in the 5′ region of the nlp-26 locus, the repair template plasmid pLM7 was injected at a concentration of 10 ng/μl, the three single guides with integrated CAS9 plasmids pLM8, pLM9 and pLM10 at a concentration of 50 ng/μl together with the same co-injection markers used for zh112.

This deletion allele was generated according to the Arribere et al. (2014) protocol, with the following single guide sequences: TTG GCA AAG TTC TAG TTA T, AC C AAT TGA ATT GCT GGA T.

To insert a gfp::3×Flag cassette into the 3′ region of the sra-9 locus, the repair template plasmid pSS22 was injected at a concentration of 10 ng/μl, the two single guides with integrated CAS9 plasmids pSS20 and pSS21 at a concentration of 50 ng/μl together with the same co-injection markers used for zh112.

Selection of homologous integrands was carried out according to the protocol of Dickinson et al. (2015).

RNAi was performed using the feeding method as described by Kamath and Ahringer (2003). P0 worms were synchronized at the L1 stage, transferred to NGM plates containing 3 mM IPTG and 50 ng/ml ampicillin seeded with the indicated RNAi bacteria and allowed to grow for 5-7 days at 20°C, after which the F1 progeny were analyzed. For Pn.p cell-specific RNAi, a strain of the genotype rrf-3(pk1426)II; unc-119(ed4)III; rde-1(ne219)V; zhEx418[lin-31::rde-1; myo2-mcherry] including the zhIs67[gfp::lin-3, unc-119(+)] reporter was used and LIN-3 polarity was measured in the F1 generation in three independent experiments. To control the tissue specificity of this RNAi strain, we performed global and Pn.p cell-specific lin-3 RNAi and observed expression of the zhIs67 reporter (Fig. S6). For uterine-specific RNAi, a strain of the genotype rrf-3(pk1426)II; unc-119(ed4)III; rde-1(ne219)V; qyIs102[fos-1ap::rde-1; myo-2::yfp] including the zhIs67[gfp::lin-3, unc-119(+)] reporter was used and LIN-3 polarity was measured in the F1 generation in three independent experiments.

Images were acquired using an Olympus BX61 wide-field microscope equipped with a Cr.E.S.T. X-light spinning disc system, a Lumencor SPECTRA X light engine and a Hamamatsu Orca CMOS camera or an iXon Ultra 888 EMCCD camera controlled by the Visitron VisiView 2.1.1 software. Fluorescent image z-stacks of the mNGr::LIN-3 and GFP::LIN-3 reporters were processed using the Huygens Deconvolution software (SVI, Center for Microscopy and Image Analysis, University of Zürich). Images were analyzed with Fiji/ImageJ software (Schindelin et al., 2012).

The developmental stages were determined by measuring the gonad length in Nomarski images, as described (Kimble and Hirsh, 1979). Animals were divided into three groups: early to mid-L2 stage (gonad 30-70 µm), late L2 stage (gonad 70-110 µm) and early to mid-L3 stage (gonad 110-150 µm).

The dorso-ventral and anterior-posterior AC polarity for LIN-3 and PIP₂ was measured using a semi-automated script in Fiji, creating summed z-projections of image stacks across the AC with 0.2-0.13 μm z-spacing with prior deconvolution (for LIN-3::mNGr and GFP::LIN-3) or in which the background had been subtracted (for P_cdh3::mCherry::PLCδ^PH). A threshold was applied to set pixel intensities outside of the AC to zero. The dorso-ventral polarity index I_DV was calculated by dividing the AC into equal areas of the ventral and dorsal halves defined by the center of the AC nucleus in the DIC image and dividing the average ventral signal intensity by the average dorsal intensity. Thus, an I_DV>1 indicates a higher signal intensity in the ventral half of the AC. An analogous method was used to calculate the P6.p-directed LIN-3 polarity index I_AP by dividing the AC into equal halves proximal and distal to P6.p or to the nearest VPC, as shown in Fig. 6H.

Vulval induction was scored by examining worms at the L4 stage under Nomarski optics as described (Sternberg and Horvitz, 1986). The number of VPCs that had adopted a 1° or 2° vulval cell fate was counted for each animal, and the vulval induction index (VI) was calculated by dividing the total number of induced cells by the number of animals scored. Animals with VI>3 were scored as hyperinduced, and animals with VI<3 as hypo-induced. The different strain combinations compared in the vulval induction assays were generated from progeny obtained in the same crosses.

For the alignment quantification, the distance between the AC and either P6.p or the closest VPC was measured as illustrated in Fig. 6H. To calculate the relative alignment index R, this value was divided by the P6.p-P5.p or to P7.p-distance depending on the AC location.

MPK-1 activity in the VPCs was measured using the recently established ERK-nKTR biosensor arTi85, which is based on the MPK-1 activity-dependent nuclear export of the biosensor (de la Cova et al., 2017). Custom-made ImageJ (Schindelin et al., 2012) and Cell Profiler (Carpenter et al., 2006) scripts were used to process and quantify the images taken in the indicated mutant backgrounds under standardized illumination conditions, as described by Maxeiner et al. (2019). Flat field illumination and background corrections were carried out using blank and dark field images, respectively, taken for each experiment. The nuclear red/green (mCherry::H2B/nKTR::mClover) average intensity ratios were measured in each VPC (except for P3.p) in summed z-projections of the five central slices taken with a z-spacing of 0.13 μm, relative to the focus of the nuclear mCherry::H2B signal. Because we could not assume that P4.p was not affected, the red/green nuclear ratio for each VPC was normalized to the average of the red/green ratios in P4.p to P8.p in the same animal, rather than normalizing to the ratio in P4.p, as described by de la Cova et al. (2017). The normalized values are plotted in Fig. 5F-I.

Statistical analyses for vulval induction were performed by bootstrapping the data with a resample size of 10,000 samples. The standard deviation within the bootstrapped samples was used to estimate the standard error of the mean (s.e.m.) and the 95% confidence interval. Statistical analysis of continuous measures (I_DV, I_AP, R) was performed using Student's t-tests or Fisher's exact tests as indicated in the figure legends.

We thank members of the Hajnal laboratory, Esther Stöckli, Markus Affolter and Stefan Luschnig for critical discussion and comments on the manuscript. We are also grateful to Gino Poulin and the Hengartner laboratory for technical support, the C. elegans Genetic Center for providing strains, Andrew Fire for GFP vectors and J. Ahringer for RNAi clones.