RPM-1 regulates axon termination by affecting growth cone collapse and microtubule stability

The axonal growth cone interprets a milieu of attractive and repulsive guidance cues in order to reach target cells (Dent et al., 2011; Gomez and Letourneau, 2014). Correct nervous system development requires that the axon guidance program eventually be terminated, via a process we refer to as axon termination.

Accurate and timely axon termination is a conserved part of nervous system development. Several examples include: retinal axons, which terminate in the superior colliculus (Frisén et al., 1998; Suetterlin and Drescher, 2014); thalamic axons that terminate in cortical layers (Agmon et al., 1993; Jones, 2001); and pseudo-unipolar sensory axons that terminate accurately in the spinal cord and periphery (Chen et al., 2006; Li et al., 2011). Accurate axon termination is also essential for fly retinal patterning (Garrity et al., 1996; Pecot et al., 2013), and occurs throughout the worm nervous system (Schaefer et al., 2000; White et al., 1986). Despite the importance of axon termination for nervous system construction, this process remains relatively poorly understood.

Time-lapse imaging studies in cultured and intact fly retina (Akin and Zipursky, 2016; Feoktistov and Herman, 2016; Ozel et al., 2015) and mammalian brain slice (Halloran and Kalil, 1994; Hand et al., 2015) have begun to explore how growth cones transition at axonal branch points and during terminal bouton synapse formation in vivo and ex vivo. Although careful temporal and spatial analysis can aid in distinguishing axon termination from synapse formation in fly retinal axons (Ting et al., 2005), axon termination has yet to be explored under normal and perturbed conditions using time-lapse imaging in this system. It is possible that formation of terminal bouton synapses by other types of axons, such as thalamic axons innervating the cortex, could present challenges for anatomically or molecularly separating axon termination from synapse formation. This highlights the importance of systems in which axon termination can be clearly distinguished from synapse formation on a molecular and anatomical level.

The posterior lateral microtubule (PLM) mechanosensory neurons of C. elegans are excellent for studying axon termination for several reasons: (1) PLM neurons share anatomical similarities with vertebrate mechanosensory neurons, which also have a single bifurcated axon (Fig. 1A); (2) axon termination can be rapidly and accurately assessed in PLM neurons (Fig. 1B); and (3) glutamatergic chemical synapses are formed on a collateral branch at a distinct location from where the primary axon terminates (Fig. 1C). Thus, one can distinguish effects on axon termination, effects on chemical synapse formation and manipulations that impact both. Recently, it has been shown the terminated PLM axon tip contains electrical synapses (Fig. 1B) (Zhang et al., 2013). It remains uncertain whether the axon termination program impacts electrical synapse formation.

C. elegans is an ideal model system for analyzing growth cones in vivo. Previous studies on motor axons explored growth cone dynamics and collapse during axon guidance (Knobel et al., 1999; Norris et al., 2014). In PLM neurons, axonal growth cones have also been observed, but only over very short timeframes (Mohamed et al., 2012). Despite this progress, important questions about growth cone changes prior to axon termination remain unanswered. For example, does growth cone collapse prior to termination differ from collapse during axon guidance? What are the signaling proteins that regulate growth cone collapse as part of the axon termination program in vivo?

C. elegans regulator of presynaptic morphology 1 (RPM-1) is an intracellular signaling hub that regulates axon termination, and is orthologous to Drosophila Highwire, mouse Phr1 and human Pam/MYCBP2 (Borgen et al., 2017; Feoktistov and Herman, 2016; Grill et al., 2016; Opperman and Grill, 2014; Schaefer et al., 2000). RPM-1 acts as a ubiquitin ligase to inhibit MAP3K proteins, such as DLK-1 and MLK-1 (Baker et al., 2015; Collins et al., 2006; Nakata et al., 2005). Notably, RPM-1 also uses signaling mechanisms that are independent of its ubiquitin ligase activity (Baker et al., 2014; Grill et al., 2007; Grill et al., 2012; Tulgren et al., 2014). At present, the cellular and developmental abnormalities in rpm-1 mutants that lead to axon termination defects are unknown. Furthermore, upstream regulators of RPM-1, and its orthologs, have remained stubbornly elusive.

Previous studies using cultured neurons from Phr1 mutant mice and fish indicated Phr1 affects microtubules in growth cones (Hendricks and Jesuthasan, 2009; Lewcock et al., 2007). However, these studies came to opposing conclusions about impacts on microtubule stability. It remains untested how RPM-1 or Phr1 affect growth cones and microtubules during axon development in vivo. Furthermore, the relationship between RPM-1 and microtubule stabilizers, such as tubulin acetyltransferases and Tau, is unknown.

We explored growth cone collapse prior to axon termination in vivo, and identified two cellular hallmarks of this process. First, growth cone collapse requires the growth cone to transition from a dynamic to a static state. Second, growth cone collapse prior to termination is extremely long compared with collapse during axon guidance. We also found that RPM-1 is required for growth cone collapse prior to axon termination. This provides a long absent cellular explanation for why axon termination defects occur in rpm-1 mutants, and provides insight into how collapse is regulated in vivo. Interestingly, RPM-1 is not required for electrical synapse formation at the axon termination site, which suggests axon termination can be molecularly distinguished from electrical synapse formation.

Because cytoskeletal changes are required for growth cone collapse, we tested how rpm-1 loss of function influences genetic and pharmacological perturbations that impact the cytoskeleton. Our results yielded several important findings. First, RPM-1 shows genetic interactions with CRMP, RHO-1 and the Rac isoform MIG-2. This provides a functional link between RPM-1 and known regulators of growth cone collapse previously identified in vitro. Second, RPM-1 signaling opposes microtubule stabilizers, such as tubulin acetyltransferases, during axon termination. Finally, our genetic analysis suggests the Tau ortholog protein with tau-like repeats 1 (PTL-1) potentially functions as an upstream inhibitor of RPM-1. This final observation is particularly interesting given the importance of Tau in neurodegenerative disease, and how little we know about Tau function in vivo under non-disease conditions (Hannan et al., 2016; Wang and Mandelkow, 2016).

In C. elegans, posterior gentle touch is sensed by two PLM mechanosensory neurons. Each PLM neuron has a single axon that terminates growth prior to the ALM cell body (Fig. 1A,B). Electrical synapses form at the termination point and can be visualized with the innexin UNC-9, which accumulates at these gap junctions (Fig. 1A,B) (Zhang et al., 2013). The PLM axon extends a collateral branch that forms glutamatergic chemical synapses (Fig. 1C).

Previous work has shown RPM-1 functions cell autonomously in the PLM neurons to regulate axon termination and chemical synapse formation (Schaefer et al., 2000). It remains unknown whether axon termination defects in rpm-1 mutants arises from failed electrical synapse formation. To test this, we generated rpm-1 mutants that express mRFP and GFP::UNC-9 in the PLM neurons. UNC-9 puncta were present at the axon tip in wild-type animals (Fig. 1B). Despite severe axon termination defects in rpm-1 null mutants, UNC-9 accumulated normally (Fig. 1D,E). This is consistent with prior results showing electrical synapses also form normally on the PLM cell bodies of rpm-1 mutants (Meng et al., 2016). It is possible electrical synapses in rpm-1 mutants could be functionally defective, despite accumulation of UNC-9. However, this is unlikely because rpm-1 mutants respond normally to gentle touch (Giles et al., 2015; Schaefer et al., 2000), which is dependent upon electrical synapses (Chalfie et al., 1985; Wicks and Rankin, 1995). Furthermore, we reproduced a previous finding that genetic ablation of the BDU neuron, the postsynaptic partner of electrical synapses at the PLM axon tip, does not affect PLM axon termination (Fig. S1) (Zhang, et al., 2013). These results indicate RPM-1 impacts on axon termination are unlikely to influence electrical synapse formation, and impairing electrical synapse formation does not influence axon termination.

PLM neurons complete long-range, growth cone-mediated axon guidance in the embryo. Upon hatching, PLM axons have reached the site of eventual axon termination in L1 larvae. Although PLM growth cones have been observed in L1 animals (Fig. 2A) (Gallegos and Bargmann, 2004), the developmental and cellular changes growth cones undergo prior to axon termination are unknown. Furthermore, the nature of cellular and developmental abnormalities in rpm-1 mutants that lead to axon termination defects remain unclear.

To address these gaps in our knowledge, we performed developmental analysis on growth cones leading up to axon termination in vivo. Axon and growth cone morphologies were observed using muIs32, a transgene that expresses GFP in mechanosensory neurons. We observed several growth cone morphologies at 1 h post-hatch (PH) (Fig. 2A), which were labeled with an RFP F-actin marker, the Utrophin calponin homology domain (UtrCH) (Fig. 2B). By 7 h PH, growth cones were absent and had been replaced by terminated axon tips (Fig. 2A). Growth cone frequency declined from 1 to 7 h PH (Fig. 2C).

Unlike wild type, growth cones in rpm-1 mutants persisted at 7 h PH (Fig. 2A,C). This was rescued by transgenic extrachromosomal arrays that express RPM-1 using its native promoter (Fig. 2C). These results suggest axons lacking RPM-1 fail to undergo growth cone collapse, leading to defective axon termination. Consistent with this, axon termination defects in rpm-1 mutants emerge after 7 h PH (Fig. 2D,E).

Impaired growth cone collapse in rpm-1 mutants might also be reflected in changes in growth cone size. Indeed, growth cones of similar morphology were larger in rpm-1 mutants than wild type (Fig. 3A), and quantitation showed a significant increase in growth cone size (Fig. 3B). Growth cone size decreased gradually over time in wild type (Fig. 3B). In contrast, rpm-1 mutants partially decreased growth cone size from 1 to 2 h PH, but no further decrease occurred (Fig. 3B). The enlarged growth cone phenotype in rpm-1 mutants was rescued by transgenic extrachromosomal arrays that expressed RPM-1 using its native promoter (Fig. 3C). Taken together, these results are consistent with RPM-1 regulating growth cone collapse prior to axon termination. Our in vivo findings also support the previous observation that cultured neurons from Phr1 mutant fish have enlarged growth cones (Hendricks and Jesuthasan, 2009).

Because the earliest point we examined was 1 h PH, it was possible that the growth cone morphology and size observed in wild type and rpm-1 mutants might reflect effects on growth cones that have reached the target zone for electrical synapse formation. Indeed, work in vertebrate and fly retina has suggested that growth cones enlarge and become more dynamic when they enter guidance choice points or regions where synapses will be formed (Mason and Erskine, 2000; Ozel et al., 2015; Raper et al., 1983). To address this, we examined growth cone shape and size in embryonic animals when PLM growth cones are mediating axon extension, and have not reached the site of axon termination and electrical synapse formation. Although imaging is challenging in live embryonic animals (because they cannot be anesthetized), careful rapid collection of images allowed us to analyze growth cone morphology and size in embryos. Embryonic axonal growth cones showed similar morphologies to growth cones observed 1 h PH in both wild type and rpm-1 mutants (Fig. 3D). Quantitation showed rpm-1 mutants had modest increases in growth cone size at this early developmental point, suggesting that loss of RPM-1 results in changes to growth cone morphology before the collapse process even starts (Fig. 3E). Thus, changes in growth cone size and frequency in rpm-1 mutants that are observed between 1 and 7 h PH are likely to reflect abnormalities in growth cone collapse and axon termination, rather than alterations in growth cones as they near the target zone for formation of electrical synapses.

Time-lapse imaging of growth cones in vivo has only been performed using flies and worms. Previous studies in fly retina have begun to explore how growth cones transition to synapse formation (Akin and Zipursky, 2016; Langen et al., 2015). Although temporal and spatial changes can be used to differentiate axon termination from synapse formation in fly retinal axons, nothing is known about growth cone dynamics during axon termination in vivo. Furthermore, it cannot be ruled out that, in these types of terminal bouton synapses, it could be difficult to molecularly distinguish axon termination from synapse formation. As we have shown, axon termination in PLM neurons is molecularly distinct from electrical synapse formation, which occurs in the same location (Fig. 1). Thus, PLM neurons are ideal for exploring how axon termination is regulated while reducing potential complications from synapse formation.

At present, we know nothing about growth cone collapse prior to axon termination in C. elegans. Thus, several simple questions remain unanswered. Does a series of collapse events and growth cone reemergence occur prior to termination? Does a single protracted collapse event precede axon termination? What changes in growth cone dynamics happen prior to axon termination?

To address these questions, we turned to time-lapse imaging of PLM growth cones in vivo. Although the growth cones of PLM neurons were imaged previously, imaging sessions lasted only a few minutes and did not explore growth cone changes during execution of the axon termination program (Mohamed et al., 2012). Consistent with this study, immediately after hatching we observed a population of dynamic growth cones with active extension and retraction of membrane protrusions (Fig. 4A; Movie 1). Unexpectedly, we also observed a population of growth cones that were static, with no changes in membrane protrusions (Fig. 4B; Movie 2). Quantitation showed 40% of growth cones were dynamic and 60% of growth cones were static in newly hatched L1 animals (Fig. S2). The presence of dynamic and static growth cones in a population of newly hatched animals most likely reflects stochastic variation in PLM axon development. Importantly, the static growth cone state we observed has not been described previously.

Time-lapse analysis showed that dynamic growth cones shifted over time to a static state (Fig. 4A; Movie 1). In contrast, static growth cones did not become dynamic again, and gradually decreased in size over time (Fig. 4B; Movie 2). Extremely long imaging sessions beyond 120 min were technically challenging owing to animal movement or bleaching. Nonetheless, we did obtain a small number of imaging sessions of ∼5 h. These long movies showed that once growth cones shift from dynamic to static, they remain static (Movie 3).

Our developmental time-course analysis for growth cone frequency and size suggested that rpm-1 mutants fail to complete growth cone collapse (Figs 2 and 3). However, it remained unclear whether growth cones collapse and reemerge in rpm-1 mutants, or if a particular stage in the collapse process is impaired in rpm-1 mutants. Time-lapse imaging of rpm-1 mutants was used to address these issues. We observed examples of both dynamic (Fig. 5A; Movie 4) and static (Fig. 5B; Movie 5) growth cones in newly hatched rpm-1 mutants. Consistent with this, the frequency of dynamic and static growth cones did not differ between rpm-1 mutants and wild type (Fig. S2). These results suggest that growth cones in rpm-1 mutants can successfully transition from dynamic to static states during collapse.

As an alternative to long imaging sessions (4-5 h), which are extremely difficult to carry out, we generated time-lapse imaging for rpm-1 mutants during the critical window of 4-6 h PH, in which growth cone collapse and axon termination is completed in wild-type animals. Our results show that during this window rpm-1 growth cones that are static decrease in size but fail to completely collapse (Fig. 5C; Movie 6). These time-lapse imaging results also confirm that growth cones do not undergo rapid collapse and reemergence, nor do they become dynamic again after shifting to a static state in rpm-1 mutants.

Overall, these results support several conclusions: (1) a single slow collapse event, in which growth cones transition from a dynamic to a static state, precedes axon termination; (2) once a growth cone shifts from dynamic to static, it does not shift back again during the termination process; this differs notably from axon guidance where more rapid growth cone collapse and reemergence occurs; (3) RPM-1 is not required for the first step of collapse, the shift from dynamic to static, but is necessary for a static growth cone to complete collapse and form a terminated axon tip. Presumably, static growth cones in rpm-1 mutants can still facilitate axon growth, as axon termination defects will eventually emerge in rpm-1 mutants (Figs 1 and 2). Thus, we define ‘static’ growth cones as lacking visible membrane protrusion and retraction, but our terminology does not imply that static growth cones cannot migrate.

To better understand how RPM-1 affects growth cone collapse, we examined RPM-1 localization in developing PLM neurons. Previous immunohistochemistry with cultured neurons arrived at different conclusions regarding the localization of murine Phr1 in growth cones. In cultured cortical neurons, Phr1 localized in the axon and throughout the growth cone (Murthy et al., 2004). More recently, Phr1 was observed in the axon shaft, but excluded from the growth cone in cultured motor and sensory neurons (Lewcock et al., 2007). In C. elegans, RPM-1 is concentrated at the terminated axon tip and the presynaptic terminals of mechanosensory neurons and motor neurons in adults (Opperman and Grill, 2014). Where and when RPM-1 or Phr1 are localized during development in vivo remains unknown.

We generated rpm-1 mutants carrying transgenic extrachromosomal arrays that expressed RPM-1::GFP in the mechanosensory neurons. Transgenic lines co-expressed tdTOMATO to visualize growth cone and axon morphology. These lines rescued axon termination defects in rpm-1 mutants (Fig. S3). Consistent with prior work, RPM-1::GFP concentrated at the terminated axon tip in adult animals (Fig. 6D). In PLM growth cones 1 h PH, RPM-1 was often expressed diffusely at low levels in the PLM soma, and was absent from the axon and growth cone (Fig. 6A,B, upper panels). In a subset of animals, very low levels of RPM-1 could be detected within the growth cone (Fig. 6B, lower panels). In these animals, RPM-1 was also concentrated in puncta near the nucleus, which are likely Golgi. Shown at 5-6 h PH is RPM-1 accumulated at the terminated axon tip, and localized to puncta in the soma (Fig. 6C). Similar RPM-1 localization patterns were observed in adults (Fig. 6D). Shown in Fig. 6E is quantitation of RPM-1 accumulation in the growth cone or the terminated axon tip across development.

Our observations suggest that a small amount of RPM-1 localizes to the growth cone to facilitate growth cone collapse. RPM-1 in the cell body could also contribute to collapse, as RPM-1 signaling is known to regulate the transcription factors CEBP-1 and TCF/POP (Tulgren et al., 2014; Yan et al., 2009). Accumulation of RPM-1 at the axon tip is consistent with RPM-1 maintaining the termination site after collapse has occurred.

In vitro growth cone collapse assays have revealed several intracellular regulators of collapse, including: CRMP2 (Goshima et al., 1995; Schmidt and Strittmatter, 2007), and the small GTPases Rho (Conrad et al., 2007; Wahl et al., 2000) and Rac (Deinhardt et al., 2011; Jin and Strittmatter, 1997). To provide further evidence that RPM-1 regulates growth cone collapse in vivo, we explored how axon termination is impacted by genetic interactions between rpm-1 and regulators of collapse previously identified in vitro.

Our previous work evaluated axon termination in rpm-1 mutants using the muIs32 transgene to visualize PLM morphology. Using muIs32, rpm-1 null mutants show severe axon termination defects, referred to as ‘hook’ defects, which occur at extremely high frequency (Fig. 7A,B). Less severe termination defects in which the PLM axon only overextends, referred to as ‘overextension’ defects, occur at very low frequency (Fig. 7A,B). Extremely high-frequency hook defects are useful for assessing rpm-1 suppressors, but less so for enhancers because of potential saturation. Therefore, we tested axon termination in rpm-1 mutants using zdIs5, another well-characterized transgene (Chen et al., 2014). rpm-1; zdIs5 animals showed a moderate frequency of both more severe hook defects and less severe overextension defects (Fig. 7A,B).

We further validated zdIs5 using null mutations for different proteins that mediate RPM-1 function, including FSN-1, GLO-4 and PPM-2 (Baker et al., 2014; Grill et al., 2007). No difference in the frequency of axon termination defects occurred in double mutants of rpm-1 with fsn-1, glo-4 or ppm-2 (Fig. 7B). This indicates these RPM-1 binding proteins function in the same genetic pathway as rpm-1, and that zdIs5 is suitable for analyzing rpm-1 genetic interactions. Unless otherwise indicated, all experiments discussed below used zdIs5.

The first regulator of growth cone collapse identified previously in vitro we tested for genetic interactions with RPM-1 was UNC-33, the ortholog of mammalian CRMP2. In unc-33 (mn407) null mutants, overextension defects occurred at moderate frequency (Fig. 7C). Interestingly, the frequency of hook defects was strongly enhanced in unc-33; rpm-1 double mutants compared to rpm-1 single mutants (Fig. 7C). Enhancement also occurred with another unc-33 allele, e204, which generates mutant UNC-33 that is not transported out of the cell body (Tsuboi et al., 2005). These results support several conclusions. First, the zdIs5 transgene is suitable for detecting enhancers of rpm-1 loss of function (lf). Second, UNC-33 regulates axon termination by functioning in the axon. Third, unc-33 and rpm-1 function in parallel genetic pathways to regulate a shared process required for axon termination, presumably growth cone collapse. Finally, the greater frequency of axon termination defects in rpm-1 mutants compared to unc-33 mutants suggests RPM-1 is a more prominent regulator of growth cone collapse in vivo than UNC-33.

Rho and Rac GTPases are also required for growth cone collapse in vitro. Therefore, we tested C. elegans Rho-family GTPases for genetic interactions with rpm-1. C. elegans contains a single Rho isoform, RHO-1, which is lethal when deleted. Therefore, we transgenically overexpressed a RHO-1 T19N dominant-negative (DN) construct using a pan-neuronal promoter. RHO-1 DN caused mild termination defects, and strongly enhanced hook frequency in rpm-1 mutants (Fig. 7D). Hook defects were also enhanced in rpm-1; cdc-42 double mutants (Fig. 7D). These results suggest that rpm-1 functions in parallel to rho-1 and cdc-42 to regulate axon termination. However, this conclusion is drawn cautiously because cdc-42 is maternally rescued, and we used a RHO-1 DN construct.

We tested two C. elegans Rac isoforms, MIG-2 and CED-10/Rac1. Gain or loss of mig-2 function results in uncoordinated locomotion and neuronal migration defects. Therefore, we only scored mig-2 mutants with normal mechanosensory neuron migration (∼50% of animals, data not shown). Whereas axon termination was normal in mig-2(lf) mutants, termination defects were strongly suppressed in mig-2; rpm-1 double mutants compared with rpm-1 single mutants (Fig. 8A,B). Suppression occurred when zdIs5 or muIs32 transgenes were used, and suppression was rescued by transgenic expression of MIG-2 in PLM neurons (Fig. 8B). These results suggest RPM-1 negatively regulates MIG-2 signaling in PLM neurons. This was further supported by our finding that a mig-2 gain-of-function allele (gf) impairs axon termination (Fig. 8C,D).

There are two reasons why mig-2 might suppress rpm-1. (1) RPM-1 inhibits the MAP3K DLK-1, and dlk-1 is a strong rpm-1 suppressor (Nakata et al., 2005). Therefore, MIG-2 could function as part of the DLK-1 pathway. (2) Alternatively, RPM-1 might inhibit MIG-2 signaling independent of DLK-1. To test these possibilities, we generated double mutants between the mig-2(gf) allele and a dlk-1 null allele. Axon termination defects were completely suppressed in mig-2(gf); dlk-1 double mutants compared with mig-2(gf) single mutants (Fig. 8C,D). This indicates mig-2(lf) suppresses rpm-1 because the MIG-2 Rac isoform functions as an upstream activator of DLK-1 signaling.

To provide further evidence MIG-2 affects axon termination by impacting growth cone collapse, we tested how MIG-2 affects growth cone size defects in rpm-1 mutants. Indeed, increased growth cone size in rpm-1 mutants was suppressed in both rpm-1; mig-2 and rpm-1; dlk-1 double mutants (Fig. 8E).

A loss-of-function mutation in ced-10/rac1 did not affect axon termination (Fig. 8F). Although mig-2(lf) was a strong rpm-1 suppressor, only less severe overextension defects were mildly suppressed in rpm-1; ced-10 double mutants (Fig. 8F). These results suggest CED-10 has a minor role in axon termination, but maternal rescue in sterile ced-10(lf) animals might mask stronger suppression.

These results indicate that several regulators of growth cone collapse identified previously in vitro affect axon termination in vivo, and interact genetically with rpm-1. This is further evidence supporting the conclusion that RPM-1 regulates growth cone collapse. Our results also show that MIG-2 functions cell autonomously in mechanosensory neurons to activate DLK-1 signaling.

We next tested the relationship between rpm-1(lf) and genetic or pharmacological perturbations that affect microtubule stability. We pursued these experiments for several reasons. First, how changes in microtubule stability affect growth cone collapse and axon termination in vivo is unclear. Second, two prior in vitro studies arrived at opposing conclusions about how Phr1 influences microtubule stability (Hendricks and Jesuthasan, 2009; Lewcock et al., 2007). Finally, how RPM-1 influences microtubule stability during axon development remains untested.

We started by examining loss-of-function mutants for various known microtubule stabilizers, including the tubulin acetyltransferases MEC-17 and ATAT-2, the Tau ortholog PTL-1, and the microtubule minus-end binding protein PTRN-1. Genetic loss of microtubule stabilizer function resulted in premature termination, in which the PLM axon terminated growth at or posterior to the vulva (Fig. 9A,B). Although microtubule stabilizer mutants were examined previously, premature termination was not described (Cueva et al., 2012; Marcette et al., 2014; Topalidou et al., 2012). This is likely because the zdIs5 background sensitizes for premature termination. Premature termination was rescued by transgenic extrachromosomal arrays expressing PTL-1 or MEC-17 using native promoters (Fig. 9B). In contrast to loss of function in microtubule stabilizers, rpm-1(lf) results in the opposing phenotype: axon termination defects in which axons overgrow (Fig. 9A and Fig. 7). These genetic results suggest RPM-1 opposes microtubule stabilizers. This is consistent with transgenic overexpression of RPM-1 driving premature termination (Fig. 9C).

Next, we tested how a null mutation in kin-18, the ortholog of Tao kinase 1/MARKK, affects axon termination. Previous studies have shown Tao-1 is a microtubule destabilizer (Liu et al., 2010). Therefore, microtubule stability is expected to increase in kin-18 mutants. Consistent with this, hook frequency was enhanced in kin-18; rpm-1 double mutants (Fig. 9D). This indicates rpm-1 acts in a parallel pathway with kin-18 to regulate axon termination, presumably by destabilizing microtubules.

To reinforce genetic results, we tested the drugs taxol and colchicine, which increase and decrease microtubule stability, respectively. Taxol enhanced the frequency of hook defects in rpm-1 mutants compared with DMSO (sham control) (Fig. 9D). In contrast, colchicine strongly suppressed the rpm-1 hook defect (Fig. 9D). Because the rpm-1 hook phenotype is more penetrant in the muIs32 transgenic background (Fig. 7), we used this transgene to evaluate axon termination in colchicine experiments. Suppression by colchicine suggests that increased microtubule stability caused by rpm-1(lf) can be overcome by microtubule destabilizing effects of colchicine. These pharmacological results further support a role for RPM-1 in microtubule destabilization.

Because rpm-1 mutants had the opposite phenotype to microtubule stabilizer mutants, such as mec-17 or ptl-1/tau, axon termination was ideal for analyzing genetic epistasis. Therefore, we constructed double mutants of rpm-1 with mec-17, atat-2, ptrn-1 or ptl-1/tau, and scored both axon termination defects and premature termination (Fig. 9E). For rpm-1 double mutants with mec-17, atat-2 and ptrn-1, both premature termination caused by loss of function in microtubule stabilizers and axon termination defects caused by rpm-1(lf) were suppressed (Fig. 9E). These results indicate RPM-1 and microtubule stabilizers have opposing effects on axon termination, which most likely results from opposing effects on microtubule stability.

As a cellular assay for microtubule stability, we examined axonal swelling. The Chalfie lab previously showed reduced microtubule stability causes axonal swellings in ALM mechanosensory neurons (Topalidou et al., 2012). Similarly, we observed axonal swellings in PLM neurons of mec-17 animals (Fig. 9F,G). In general, rpm-1 mutants lacked axonal swellings (Fig. 9F). However, quantitative analysis revealed very low frequency swellings in rpm-1 mutants (Fig. 9G). Swellings were suppressed in rpm-1; mec-17 double mutants compared with mec-17 single mutants (Fig. 9F,G). These cellular results further suggest RPM-1 and MEC-17 have opposing effects on microtubule stability. However, work in worm motor neurons has shown axonal swellings can arise from impaired axonal transport, suggesting an alternative interpretation (Ikenaka et al., 2013). Nonetheless, a combination of cellular, genetic and pharmacological approaches collectively point to the conclusion that RPM-1 is a microtubule destabilizer, and increased microtubule stability in rpm-1 mutants leads to axon termination defects (Fig. 9H).

Although rpm-1 mutants and ptl-1 mutants also had opposing axon phenotypes (Fig. 9A,E), epistasis indicated a different interesting result. In rpm-1; ptl-1 double mutants, premature termination caused by ptl-1(lf) was completely suppressed (Fig. 9E). However, axon termination defects caused by rpm-1(lf) were not significantly changed in rpm-1; ptl-1 double mutants compared with rpm-1 single mutants (Fig. 9E). This was not simply because ptl-1 mutants have low frequency premature termination defects. ptrn-1 and atat-2 mutants had similar low-frequency premature termination defects, but axon termination defects caused by rpm-1(lf) were still significantly suppressed in these double mutants (Fig. 9E). These results show that rpm-1 is epistatic to ptl-1. A simple and intriguing explanation for this result is that PTL-1 regulates axon development by inhibiting RPM-1.

Axon termination is essential for accurate and efficient nervous system construction. However, we know very little about the growth cone collapse process that occurs prior to axon termination in vivo. Signaling proteins that regulate growth cone collapse as part of the axon termination program also remain unknown.

Using worm mechanosensory neurons, we found that dynamic growth cones transition to a static state, and undergo a single protracted collapse event that culminates in axon termination. Our results indicate that the intracellular signaling hub RPM-1 regulates growth cone collapse to facilitate axon termination in vivo. RPM-1 functions in parallel to molecules that can regulate the actin cytoskeleton, such as RHO-1, CDC-42 and CRMP/UNC-33, while opposing the function of microtubule stabilizers, such as tubulin acetyltransferases. Furthermore, genetic results suggest RPM-1 is potentially inhibited by the microtubule stabilizer Tau/PTL-1. This finding not only provides insight into how RPM-1 is regulated, but has important implications for how Tau functions during axon development.

Using C. elegans mechanosensory neurons, we analyzed axonal growth cones prior to axon termination in vivo. We found growth cone frequency and size decrease over a 5-7 h developmental window in wild-type animals (Figs 2 and 3). In vivo time-lapse imaging showed these developmental changes reflect a single protracted growth cone collapse event, and ruled out the possibility that multiple, rapid collapse events occur prior to axon termination (Fig. 4). Time-lapse imaging also revealed that growth cones transition from a dynamic to a static state, with static growth cones undergoing a slow reduction in size that leads to axon termination. To our knowledge, this static growth cone state has not been described previously in any system.

Protracted growth cone collapse prior to axon termination differs notably from collapse in other contexts. Prior studies with worm motor neurons in vivo and in brain slice showed that growth cones are dynamic, collapse over 10-40 min and reemerge during axon guidance (Halloran and Kalil, 1994; Knobel et al., 1999). This suggests protracted growth cone collapse, in which dynamic growth cones transition to a static state, is a cellular hallmark of axon termination.

Much progress has been made on the mechanisms of growth cone collapse in vitro. In contrast, we know little about growth cone collapse in vivo. Our developmental time-course results (Figs 2 and 3) and time-lapse imaging of PLM growth cones (Figs 4 and 5) demonstrate that RPM-1 is required for growth cone collapse during axon termination in vivo.

Our observation that growth cone size initially decreases in rpm-1 mutants and then halts highlights an interesting possibility: growth cone collapse prior to axon termination could occur through two stages, the second requiring RPM-1. Time-lapse imaging confirmed this model by showing RPM-1 is required for growth cones to collapse from a static state to a terminated axon tip. However, it is important to note that the persistence of static growth cones in rpm-1 mutants eventually results in axon termination defects. We envision this could occur in two ways. (1) It is possible static growth cones in rpm-1 mutants can still mediate axon growth. Consistent with this possibility, retinal axon growth cones maintain a streamlined structure during axon extension, suggesting that highly dynamic growth cones might not be necessary to facilitate axon growth (Mason and Erskine, 2000; Ozel et al., 2015; Raper et al., 1983). (2) Static growth cones in rpm-1 mutants might transiently become dynamic at time points later than those we examined in this study.

Importantly, our results show that the effects of RPM-1 on growth cone collapse and axon termination do not influence electrical synapse formation between the PLM neuron and the postsynaptic BDU neuron. This differs notably from chemical synapse formation in PLM neurons, which is impaired in rpm-1 mutants (Grill et al., 2016; Schaefer et al., 2000).

Further evidence indicating RPM-1 regulates growth cone collapse came from genetic interactions between rpm-1 and molecules previously shown to regulate collapse in vitro, including Crmp, Rho and Rac. Our results show CRMP/UNC-33 and RHO-1 function in parallel to RPM-1 to regulate axon termination (Figs 7 and 10). Interestingly, the Rac isoform mig-2 suppresses axon termination defects in rpm-1 mutants, indicating MIG-2 opposes RPM-1 function (Fig. 10).

Suppression occurs because MIG-2 functions in PLM neurons to activate DLK-1 signaling (Fig. 8), while RPM-1 is an important inhibitor of DLK-1 (Nakata et al., 2005). Our results expand on prior evidence that chemical synapse formation in PLM neurons is regulated by MIG-2 activation of DLK-1 signaling (Chen et al., 2014). In flies, protein kinase A (PKA) and mTORC1 phosphorylate and activate DLK-1 after axon injury and during synapse formation (Hao et al., 2016; Wong et al., 2015). Thus, MIG-2 Rac signaling, PKA and mTORC1 might converge on DLK-1 as an activation code. Alternatively, different signals might activate DLK-1 based on the cellular or functional context.

Beyond the RPM-1 signaling network, very few molecules are known to affect axon termination in C. elegans. Our findings now suggest some of these players, such as the SAD-1 and SAX-1 kinases (Crump et al., 2001; Gallegos and Bargmann, 2004) or MEC-7 tubulin (Chen et al., 2014), might influence growth cone collapse. Similar logic suggests that regulators of axon termination identified using fly retina, such as Dreadlocks or p21 activated kinase (PAK), could influence collapse (Garrity et al., 1996; Hing et al., 1999). Finally, protracted collapse prior to axon termination is consistent with transcription factors like Brakeless (Rao et al., 2000), Runt (Kaminker et al., 2002) and Runx3 (Chen et al., 2006) regulating axon termination in flies and mammals.

Prior in vitro studies with mouse and fish neurons showed the RPM-1 ortholog Phr1 regulates microtubule dynamics (Hendricks and Jesuthasan, 2009; Lewcock et al., 2007). However, these studies yielded opposing conclusions about whether Phr1 stabilizes or destabilizes microtubules. How Phr1 or RPM-1 influence microtubule stability in vivo remains uncertain.

We explored how genetic and pharmacological manipulation of microtubule stability affects axon termination and interacts with rpm-1(lf) in vivo. Our results indicate RPM-1 opposes the function of microtubule stabilizers, such as the tubulin acetyltransferases MEC-17 and ATAT-2 (Figs 9 and 10). In contrast, RPM-1 functions in parallel to the microtubule destabilizer KIN-18/Tao1. Consistent with genetic outcomes, the microtubule-stabilizing drug taxol enhanced and the microtubule-destabilizing drug colchicine suppressed axon termination defects caused by rpm-1(lf) (Fig. 9). These results indicate that RPM-1 signaling is likely to destabilize microtubules during growth cone collapse and axon termination. There are several ways this might occur. The first is by RPM-1 inhibiting DLK-1, as DLK-1 signaling can stabilize microtubules during regeneration and synapse remodeling (Ghosh-Roy et al., 2012; Kurup et al., 2015). Another possibility is that RPM-1 binds to the microtubule-binding protein RAE-1 (Grill et al., 2012). A recent study has shown that RPM-1 forms separate signaling complexes with RAE-1 and FSN-1, the F-box protein that facilitates RPM-1-mediated degradation of DLK-1 (Baker and Grill, 2017). As a result, RPM-1 might regulate RAE-1 and DLK-1 independently to impact microtubules. Finally, RPM-1 could destabilize microtubules more directly, as Phr1 associates with microtubules (Lewcock et al., 2007). Regardless of how RPM-1 signaling facilitates microtubule destabilization, our results show RPM-1 effects on microtubules have important repercussions for growth cone collapse and axon termination (Fig. 10).

Our observation that rpm-1 is enhanced by crmp/unc-33 might appear puzzling, as CRMP can stimulate microtubule assembly and unc-33 mutants display premature termination in other neurons (Fukata et al., 2002; Tsuboi et al., 2005). A likely explanation for our result is CRMP-mediated regulation of actin (Hall et al., 2001; Kawano et al., 2005; Schmidt and Strittmatter, 2007). This is also consistent with unc-33 enhancing rpm-1 to similar levels as the actin regulators rho-1 and cdc-42. Alternatively, microtubule networks in PLM neurons might respond differently to loss of unc-33 compared with other neurons, or microtubule assembly regulators such as unc-33 could have different genetic interactions with rpm-1 than microtubule stabilizers.

An interesting genetic outcome occurred between rpm-1 and ptl-1/tau. We found premature termination defects, but not axon termination defects, were suppressed in rpm-1; ptl-1 double mutants (Fig. 9). This indicates rpm-1 is epistatic to ptl-1. A likely explanation for this result is that PTL-1 inhibits RPM-1 (Fig. 10). Given that DLK-1 is activated by cytoskeletal disruption (Valakh et al., 2015) and RPM-1 orthologs regulate axon degeneration (Babetto et al., 2013; Xiong et al., 2012), it is not unreasonable a microtubule stabilizer like PTL-1/Tau could inhibit RPM-1. Our findings also dovetail with the recent observation that PARK9, an ATPase associated with Parkinson's, negatively regulates the human ortholog of RPM-1 in cultured cells (Bento et al., 2016).

Our results suggesting PTL-1/Tau inhibits RPM-1 is potentially the first evidence that RPM-1, or its orthologs, can be regulated in vivo in neurons. Experiments aimed at determining how PTL-1 inhibits RPM-1 and whether PTL-1 is part of an RPM-1 regulatory network now becomes an intriguing direction for future investigation.

C. elegans strains were maintained using standard procedures. Alleles used included: rpm-1 (ju44), dlk-1 (ju476), glo-4 (ok623), ppm-2 (ok2186), unc-33 (mn407), unc-33 (e204), ced-10 (tm597), cdc-42 (gk388), mig-2 (ok2273)(lf), mig-2 (gm103)(gf), mec-17 (ok2109), ptrn-1 (tm5597), ptl-1 (ok621), atat-2 (ok2415) and kin-18 (ok395).

All double mutants were constructed following standard procedures, and were confirmed by associated phenotypes, PCR genotyping or sequencing as needed. rpm-1 glo-4 double mutants were constructed by recombination and confirmed by PCR. Primers and PCR conditions are available upon request. rpm-1; ced-10/nT1 animals were constructed using the pharyngeal GFP marker expressed by nT1, and confirmed by PCR. GFP-negative, non-nT1 offspring, which are maternally rescued, were collected and axon termination was evaluated. rpm-1; cdc-42/mIn1 animals were constructed using the pharyngeal GFP marker expressed by mIn1, and confirmed by PCR. Axon termination was evaluated in non-mIn1 offspring. Integrated transgenic strains used in this study to label PLM mechanosensory neurons are: muIs32 (P_mec-7GFP), zdIs5 (P_mec-4GFP) and jsIs973 (P_mec-7mRFP). jsIs973 was a kind gift from Dr Michael Nonet (Washington University, St. Louis, MO, USA).

The transgenic extrachromosomal array used to visualize electrical synapses, xdEx1999 (P_mec-7GFP::UNC-9), was a gift from Dr Mei Ding (Chinese Academy of Sciences, Beijing, China). Dr Ding also provided the transgenic extrachromosomal array, xdEx963 (P_unc-53CED-3; kyIs262), which genetically ablates BDU while simultaneously labeling ALM and PLM neurons. mRFP::UtrCH, which was previously used to visualize F-actin in mammalian growth cones (Hand et al., 2015), was a gift from Dr Alex Kolodkin (Johns Hopkins University, Baltimore, MD, USA). All transgenic extrachromosomal arrays were constructed by injecting DNA of interest with a co-injection marker, P_ttx-3GFP or P_ttx-3RFP (50 ng/µl), and pBluescript to reach a total DNA concentration of 100 ng/µl. All Transgenic extrachromosomal arrays and injection conditions are listed in Table S1.

The transgenic strain muIs32 (P_mec-7GFP) was used to label PLM growth cones for analysis of growth cone frequency, morphology and size. Freshly hatched L1 larvae were collected and allowed to age over 1 h intervals at room temperature (22°C) prior to analysis.

At select time points between 1 and 16 h PH, animals were mounted in 5 mM levamisole and M9 buffer on agarose pads. Fluorescent images of L1 PLM growth cones were collected under 100× magnification on a Leica CTR6500 epifluorescent microscope, a DFC360 FX camera and Leica Application Suite software. Image analysis of growth cone width and area were carried out using ImageJ software from NIH image (http://rsb.info.nih.gov/ij/). To define growth cones compared with terminated axon tips, we used both morphological assessment and quantitative measurements of the ratio of growth cone width to axon width. A growth cone was defined as having a ratio of growth cone width to axon width of 1.5 or greater.

For embryonic growth cone analysis, adults were placed on a seeded plate, laid for 2 h and were removed. Embryos were aged at room temperature (22°C) until ∼0.5 to 2 h prior to hatching, when PLM axon extension is occurring. Because the eggshell prevented embryos from being anesthetized, they were mounted on agarose pads in M9 buffer alone. Images of PLM growth cones in embryos were captured under 63× magnification using the microscope described above. Because embryos were moving freely, we rapidly acquired multiple images until a single well-focused image of a growth cone was obtained.

The transgenic strain muIs32 (P_mec-7GFP) labeled PLM growth cones for time-lapse imaging. To obtain animals for imaging, embryos were collected on a plate, and freshly hatched L1 larvae were collected over intervals of 10 min or less. L1 larvae were mounted on agarose pads and immobilized immediately using 5 mM levamisole in a 50% mixture of glycerol and M9. rpm-1 mutants required a higher levamisole concentration (10 mM) to achieve sufficient paralysis.

For extended time-lapse imaging, preparations were sealed using CoverGrip Coverslip Sealant (Biotium) to prevent dehydration. Confocal images of L1 PLM growth cones were collected using a Leica SP8 confocal microscope under 63× magnification and 2× zoom. Leica Application Suite software was used to define z-stacks collected at 0.5 μm intervals. For all movies, ∼2-5 min intervals were used depending upon the z-stack depth required to image the growth cone. Analysis of time-lapse imaging was carried out using ImageJ. Stacks for each time point were aligned using the Stack Reg plug-in, and compressed using Maximum Projection or Extended Depth of Field, as necessary. Growth cones were considered dynamic if at least one distinct membrane protrusion extended and retracted during an imaging session. Labels in movies were added using ImageJ and iMovie.

Transgenic ju44 mutants expressing RPM-1::GFP using the mec-3 promoter (P_mec-3RPM-1::GFP) and tdTOMATO using the mec-7 promoter (P_mec-7tdTOMATO) were anesthetized with levamisole. L1 larvae were imaged at 0-1 h PH or were allowed to age to 5-6 h PH at 22°C before imaging. L1 animals and adults were imaged on a Leica SP8 confocal microscope under 63× or 40×, respectively, with 2× zoom. Enhanced sensitivity of tdTOMATO was essential for visualizing PLM morphology in L1 larvae because of their small size and relatively weak expression of the mec-7 promoter in L1 (data not shown).

Our previous studies used the muIs32 transgene to visualize axon termination. To avoid potential issues with saturation of axon termination phenotypes in rpm-1; muIs32 animals, we relied upon another widely used transgenic background, zdIs5. Frequency of severe (hook) axon termination defects are not saturated in zdIs5; rpm-1 animals. This allowed us to assess enhancement and suppression.

Images of PLM axons in young adult animals were collected under 40× magnification using a Leica CFR5000 B epifluorescent microscope, CCD camera (Leica DFC345 FX) and Leica Application Suite software. Animals were raised at 23°C and anesthetized with 5 mM levamisole for manual scoring. For image collection, animals were immobilized in 1% (v/v) 1-phenoxy-2-propanol in M9 buffer. Defects in the axons of PLM mechanosensory neurons were classified as premature termination (when termination occurs at or before the vulva), overextended (an axon termination defect in which the PLM axon terminates after the ALM soma) or hook (an axon termination defect in which the PLM axon grows past the ALM soma and hooks ventrally towards the nerve cord). Swellings in the PLM axon for each genotype were analyzed using zdIs5 and scored manually ina double-blind analysis.

Five to 10 L4 animals (P0) were transferred to fresh 3.5 cm plates that contained either 2 µm taxol or DMSO and E. coli. P0 animals laid eggs overnight and were removed. F1 progeny developed normally on taxol. F2 offspring were scored for axon termination as young adults.

For colchicine experiments, colchicine or DMSO were spread on nematode growth medium (NGM) plates at a final concentration of 0.25 mM and left overnight. Plates were seeded with E. coli and 16 h later 3-5 P0 adults were placed on plates. F1 progeny developed normally on colchicine and axon termination was scored in late L4 larvae or young adult F1 animals.

We thank Drs Erik Dent and Mike Nonet for helpful discussions. We thank the C. elegans knockout consortium for generating several alleles and the C. elegans Genetics Center (NIH Office of Research Infrastructure Programs, P40 OD010440) for providing strains.