Completion of neuronal remodeling prompts myelination along developing motor axon branches

Myelin enables saltatory conduction and provides trophic support to the sheathed axons (Huxley and Stämpeli, 1949; Vabnick and Shrager, 1998; Yin et al., 2006; Simons and Trotter, 2007; Nave, 2010). In addition, recent observations in the central nervous system (CNS) indicate that myelin contributes to fine-tuning of neural circuits (Fields, 2015; Chang et al., 2016; Kaller et al., 2017). For instance, myelin sheaths and nodes of Ranvier, ion channel–enriched axon segments interspersed between myelin sheaths, show activity-dependent plasticity (Huff et al., 2011; Gibson et al., 2014; Mensch et al., 2015; Etxeberria et al., 2016; Korrell et al., 2019; Bacmeister et al., 2020) that appears to shape “patchy” myelination patterns in neocortex (Tomassy et al., 2014). While activity-regulated myelination is less studied in the peripheral nervous system (PNS; Stevens and Fields, 2000; Fields, 2015), in the PNS, the axon–glial unit is more accessible than in the CNS, and the signaling pathways governing peripheral myelination are better understood (Taveggia et al., 2010; Pereira et al., 2012; Grigoryan and Birchmeier, 2015). Thus, PNS development offers a privileged window into the intersection of axonal remodeling and myelin plasticity.

To capitalize on these advantages, we turned to a major site of PNS remodeling, the neuromuscular junction (NMJ). At mouse NMJs, axonal remodeling follows a predictable course during the first two postnatal weeks and can be followed at the single-axon-branch level (Lichtman and Sanes, 2003; Walsh and Lichtman, 2003). At birth, multiple motor axon branches innervate the same postsynaptic site (Tapia et al., 2012). Subsequently, all but one of these presynaptic inputs are eliminated by a two-step process that first involves activity-driven competition and then axon branch removal by cytoskeletal degradation (Buffelli et al., 2003; Brill et al., 2016) until lifelong innervation by a single axon is established (Tapia and Lichtman, 2012). Already during embryonic development, Schwann cells (SCs), the glia of the PNS, surround growing motor axons and accompany them to the target muscle (Jessen and Mirsky, 2005). SCs initiate myelination perinatally, after SCs have been sorted to sheath individual axon branches (Jessen and Mirsky, 2005; Monk et al., 2015; Rasband and Peles, 2015). Overall, myelination follows a proximal-to-distal gradient along motor axons, with myelination of terminal branches occurring last and asynchronously (Hildebrand et al., 1994; Yamamoto et al., 1996). This temporal correlation between axon–glial differentiation and cessation of developmental axon plasticity is a general feature across the nervous system and in the CNS can, e.g., be observed in visual cortex (Luo and O’Leary, 2005; McGee et al., 2005; Simons and Trotter, 2007). In the PNS, myelination onset is determined by the level of Neuregulin-1 (Nrg1) type III on the axonal surface. Nrg1 binds to glial ErbB2/3 receptors on SCs, leading to phosphorylation of downstream effectors, such as ERK1/2 and AKT (Garratt et al., 2000; Michailov et al., 2004; Taveggia et al., 2005; Iwakura and Nawa, 2013; Basak et al., 2015). While Nrg1 signaling is known to affect synapse development, the underlying signaling takes place at the NMJ itself, involving “terminal” nonmyelinating SCs rather than myelinating SCs along the axon (Loeb, 2003; Hayworth et al., 2006; Lee et al., 2016). Thus, whether the timing of branch-specific myelination also depends on local availability of Nrg1 and whether Nrg1 signaling is locally regulated to coordinate axon remodeling and myelination remain to be elucidated. Hence, using the NMJ as a model, we asked how axonal competition and axon–glial differentiation are coordinated at the single-branch level and what signaling mechanism is involved.

To study the intersection of axon remodeling and myelination, we took advantage of a thoracic nerve-muscle explant, including the triangularis sterni muscle, which is uniquely suited to study the cell biological dynamics of single axon branches (Fig. 1 A; Kerschensteiner et al., 2008; Brill et al., 2013). During the second postnatal week, most NMJs transition from double to single innervation (abbreviated in the figures as din and sin, respectively), while myelin and nodes of Ranvier appear on terminal branches (Fig. 1 B). On three postnatal days (postnatal day 7 [P7], P9, and P11), we quantified the number of NMJs still engaged in synaptic competition using triangularis sterni muscles derived from Thy1-XFP mice, where motor axons are fluorescently labeled (Fig. 1, C and D; Feng et al., 2000). In parallel, we determined the state of axon–glial differentiation on singly innervating terminal branches based on the presence of immunostained Caspr1 (contactin-associated protein 1; Rasband and Peles, 2015) accumulations along a given terminal branch. We did not distinguish further between fully formed nodes (paranodes on both sides) and partially formed heminodes, resulting in a binary score (Caspr⁺ vs. Caspr⁻ terminal branches; Fig. 1, C and D). We observed a concomitant resolution of synaptic competition (i.e., decreasing percentage of doubly innervated synapses) and increasing paranodal formation on “winner” branches (Fig. 1 D). Finally, at P13, all NMJs established single innervation (100 ± 0%, n = 3 mice, 100 NMJs per animal), and all terminal branches started to form paranodes (100 ± 0%, n = 3 mice, 30 NMJs per animal). Next, we immunostained for myelin and other nodal components in Thy1-XFP mice at P9. In parallel to Caspr, nodal (voltage-gated sodium channel [Nav]), juxtaparanodal (contactin-2 [CNTN2]), and internodal (myelin protein zero [MPZ]) markers emerged on terminal branches (Fig. 1, E and F; Doyle and Colman, 1993). Thus, as myelin and nodal compartments coassemble rapidly (Girault and Peles, 2002; Schafer et al., 2006), we used Caspr immunostaining as a surrogate for overall axon–glial differentiation. Notably, when we focused on the branches still engaged in competition, we found significantly fewer branches immunopositive for emerging nodal structures, resulting in an overall delay of axon–glial differentiation of roughly 2 d (∼33% of the full 6-d myelination period) compared with their winner siblings. This finding was consistent across all markers tested (Fig. 1 F). Thus, ongoing competition, and hence sustained plasticity, of terminal axon branches coincide with a transient stall of myelination and node formation. We considered two explanations for this delay: (1) slower assembly of structural components of the axon–glial unit or (2) reduced promyelinating signals. To disambiguate these scenarios, we analyzed the dynamics of node formation during axonal remodeling.

To measure the progress of axon remodeling, as well as the onset of node formation on individual motor axon branches, we characterized transgenic mice expressing GFP-tagged Caspr (Caspr-GFP; Fig. S1; Brivio et al., 2017) and generated mice expressing the β1 subunit of the Nav tagged with GFP (β1-Nav-GFP; Fig. S1), both under control of the Thy1 promotor. In both lines, progress of synapse elimination and onset of node formation were unchanged compared with WT littermates at P9 (Fig. S1 legend). We assessed the paranodal/nodal protein dynamics by FRAP (see Materials and methods) and related the recovery rate to axonal competition status in β1-Nav- or Caspr-GFP crossed to Thy1-OFP3 mice (Brill et al., 2011). GFP-positive clusters forming heminodes were photobleached to approximately one third of their original fluorescence intensity (33 ± 3%, n = 135 nodes in 46 mice) and visualized again 3 h later. The recovery was normalized to nonbleached control nodes in the same field of view to account for imaging-related fluorescence loss (Fig. 2, A and C). Surprisingly, at P9–P11, we found significantly higher recovery rates of β1-Nav- or Caspr-GFP on competing doubly innervating branches compared with singly innervating ones (Fig. 2, B and D), suggesting that once initiated, node formation progressed swiftly. We also found an age-dependent decline (Rios et al., 2000). Recently established nodal structures recover much faster than mature ones (sin P9–P11 vs. sin 6 wk; Caspr-GFP ∼4.3-fold; β1-Nav-GFP ∼4.4-fold; Zhang et al., 2012). At the same time, P9–P11 nodes in proximal positions (stem; Fig. 2, B and D) resembled distal mature (i.e., 6 wk) nodes, consistent with the known myelination gradient (Hildebrand et al., 1994). Hence, our data favor a mechanism where axonal competition delays the initiation, but not progress, of axon–glial maturation. However, myelination and node formation are still initiated on a subset of competing, doubly innervating axon branches (cf. Fig. 1 F). Thus, we wondered whether disparate progress of axon–glial maturation influences the competition outcome.

To address the effect of a branch’s axon–glial maturation status on competition, we related initiation of node formation to synaptic territory (i.e., the fraction of an NMJ that a terminal branch innervates). Synaptic territory is a well-established indicator of probable competition outcome (Gan and Lichtman, 1998; Walsh and Lichtman, 2003; Brill et al., 2016). We determined synaptic territory using the Brainbow approach to individually color motor units (ChAT-Cre x Thy1-Brainbow-1.1; Fig. 3 A; Livet et al., 2007; Rossi et al., 2011) and immunostained for Caspr to reveal node formation along terminal branches at P9. Throughout all stages of axonal competition (1–99% territory), less than ∼20% of the branches were Caspr positive, and there was no correlation of myelination onset to synaptic territory (Fig. 3 B). However, once competition was resolved, Caspr was present on ∼50% of the singly innervating terminal branches (100% territory; χ² test, 1–99% vs. 100%, P < 0.0001, n = 749 axon branches in 45 mice; Fig. 3 B), suggesting a swift lift of the brake on axon–glial differentiation once competition was resolved. This lack of correlation contrasts with other cell biological features of terminal branches, including cytoskeletal stability, organelle transport, or caliber, which are highly correlated to synaptic territory (Keller-Peck et al., 2001; Brill et al., 2016). Moreover, the measured distribution of node formation patterns on competing axons, such as the fraction of NMJs where node formation had started on the winning branch (51–99% territory), the losing branch (1–49% territory), or neither of the competing branches, matched a random binomial distribution (14% din myelinated across 1–99% territory, n = 520 din NMJs in 35 mice; Fig. 3 C). This, however, does not rule out caliber as a central driver of node formation onset, as known for other PNS settings, where 1 µm represents a critical threshold for myelination (Voyvodic, 1989; Peters et al., 1991). Therefore, we analyzed the diameter of competing branches with or without emerging nodes (Fig. 3 D) and found no difference between the groups during all stages of competition. We even found (albeit rarely) partially myelinated axon branches that lost against a nonmyelinated competitor (Fig. 3, F and G). Only in retreating branches were the pruning axons with emerging nodes significantly thicker than their unmyelinated counterparts (Fig. 3 D), possibly due to the protective effect of myelin on axonal structures, as axon stretches covered by MPZ are significantly thicker than MPZ-negative stretches along the same retreating branches (Fig. 3 E; Nave, 2010). Hence, at the NMJ, axon–glial differentiation neither decides competition nor prohibits axon pruning (McGee et al., 2005). Together, the data suggest a unidirectional relationship, with ongoing axon remodeling delaying axon–glial maturation, but not the converse. Next, we wanted to test which phase of synapse remodeling impacted axon–glial maturation to narrow down possible underlying signaling mechanisms.

Synapse remodeling can be divided into several phases (Kano and Hashimoto, 2009; Turney and Lichtman, 2012), with an early activity-dependent competition phase driven by neurotransmission (Buffelli et al., 2003) followed by a late execution phase involving cytoskeletal breakdown and glial engulfment (Bishop et al., 2004; Brill et al., 2016). First, we intervened during competition by irreversibly blocking postsynaptic acetylcholine receptors (AChRs) using unilateral thoracic injection of α-bungarotoxin (BTX; Akaaboune et al., 1999; Kummer et al., 2004) of P7 Thy1-XFP mice (Fig. 4 A). Paired analysis on contra- versus ipsilateral triangularis sterni muscles 2 d later (P9; Fig. 4, B and C) revealed that more multiple innervation was maintained after BTX injection (Fig. 4 D; Loeb et al., 2002; Buffelli et al., 2003). Notably, the number of Caspr-positive winner branches (sin) was significantly reduced (Fig. 4 E), suggesting that blocking neurotransmission delays the initiation of node formation. At the same time, we did not measure a change of axonal SC number or internode or terminal branch length after BTX treatment (Fig. S2, A–D). Under physiological conditions, the number of SCs slightly increases as competition resolves (indicated by reduced SC length; Fig. S2, E and F). We therefore hypothesized that BTX injection maintained terminal branches in a more juvenile state. Thus, we turned to the microtubular cytoskeleton as an important indicator for axonal maturation, since microtubular mass increases as the branch gains synaptic territory (Brill et al., 2016). The initiation of myelination correlates with an increase in tubulin content (Fig. S2 G and H). Following BTX injection, microtubular mass on winner axons (sin) decreased to ∼60%, while competing axons (din) were not affected (Fig. 4, F and G). This hints at the possibility that the delayed node formation following transmission block is due to reduced microtubular mass.

To manipulate axonal microtubules, we genetically deleted spastin, a microtubule-severing enzyme (spastin KO), where we confirmed a delay in axon branch removal (Fig. 4, H–J; Brill et al., 2016). Indeed, loss of spastin led to accelerated node formation in competing axons (din) compared with WT (Fig. 4 K). That this represented a cell-autonomous effect in motor neurons was corroborated by inducing subset deletion in conditional spastin^fl/fl × TdTomato reporter mice (Brill et al., 2016) using a Cre-encoding adeno-associated virus (AAV; AAV9-CMV-iCre; Fig. S3). While we again found delayed axon remodeling (Brill et al., 2016), node formation was now accelerated on competing branches (Fig. S3 C), where TdTomato expression indicated spastin deletion. Overall, microtubular mass was increased in spastin-deleted terminal axon branches (Fig. 4, L and M; Brill et al., 2016), while axonal caliber was unaffected (Fig. S3 D), contrasting the increase in nodal formation specifically on competing branches. This suggests that the microtubular cytoskeleton is the limiting factor to initiate node formation in competing branches, but not on winner axons, perhaps arguing for a two-component system, where each can be limiting in different stages. Moreover, the divergent axon–glial differentiation outcomes of postsynaptic block versus spastin deletion, which both delay remodeling, points to a mechanism that is blocked by ongoing activity-dependent competition but can be overcome by increasing microtubular mass. As the microtubular cytoskeleton sustains axonal transport (which requires tracks and cargoes), we next tested if reducing anterograde transport would affect local initiation of axon–glial differentiation.

To reduce transport in motor neurons, we overexpressed the cargo-binding domain (CBD) of kinesin-1 heavy chain (KHC), a key molecular motor driving anterograde transport (Hirokawa et al., 2009). This results in a dominant-negative mutant (KHC-CBD), which still binds cargoes but lacks the motor domain and competes with endogenous kinesin-1, thus impairing transport of organelles and nodal components in vitro (Cai et al., 2005; Barry et al., 2014). To test the efficacy of this approach in vivo, we turned to zebrafish as an easily accessible model for assaying effects of myelination-regulating signals (Czopka and Lyons, 2011). We used the Gal4/UAS system to transiently coexpress UAS-GFP-KHC-CDB and UAS-mitoTagRFP-T in Rohon–Beard sensory neurons, in which mitochondrial transport can be easily monitored (Plucińska et al., 2012). KHC-CBD overexpression in this system substantially reduced mitochondrial transport per minute (reduction from 0.61 ± 0.11 in control to 0.10 ± 0.02 in anterograde and 0.28 ± 0.04 to 0.10 ± 0.04 in retrograde direction at 2 d postfertilization [dpf]; P < 0.01, Mann–Whitney test, n ≥ 4 zebrafish per group, ≥4 axons). We then analyzed spinal motor neurons, which start to be myelinated at 3 dpf in zebrafish (D’Rozario et al., 2017). To down-regulate axonal transport while monitoring myelination progress, we expressed GFP-KHC-CBD or GFP alone under the neuronal cntn1b promoter in Tg(mbp:RFP) zebrafish, where all compacted internodes are fluorescently labeled by a membrane-targeted RFP (Fig. S4, A–E; Auer et al., 2018). On 6 dpf, axon length in Tg(mbp:RFP) zebrafish injected with cntn1b-GFP-KHC-CBD was similar to controls (Fig. S4 F), but strikingly, the myelinated axon length was only half compared with controls (Fig. S4 G). This supports the notion that PNS myelination depends on axonal transport.

We next probed whether this was true in murine motor axons during remodeling. In Thy1-β1-Nav-GFP animals, emerging β1-Nav-GFP clusters correlate with higher anterograde particle transport in terminal branches (Fig. 5). Among all β1-Nav-GFP-positive branches, winner axons (sin) had the highest anterograde transport rate (Fig. 5 B). Together, this is in line with our previous observation that microtubular mass correlates with node formation (Fig. S2 H). We then injected an AAV9 encoding KHC-CBD and iCre under control of the neuronal human synapsin promoter (AAV9-hSyn-iCre-p2a-KHC-CBD) into neonatal mice (Fig. 5 A). In AAV-injected Thy1-β1-Nav-GFP × TdTomato reporter mice, we found a significant reduction in β1-Nav-GFP anterograde transport, while retrograde rates were unaffected (Fig. 5 B). Notably, the onset of node formation was significantly delayed in reporter-positive branches compared with negative ones, which served as internal controls (Fig. 5, C and D). This points to a transport-delivered signal, which locally times the onset of axon–glial differentiation of terminal axon branches.

Nrg1 type III is a candidate for a transported promyelination signal, as this signaling factor locally needs to reach a critical threshold to initiate axon–glial differentiation (Taveggia et al., 2005; Nave and Salzer, 2006; Birchmeier and Nave, 2008; Velanac et al., 2012) by activating downstream effectors in SCs such as ERK1/2 and AKT (Ogata et al., 2004; Taveggia et al., 2005; Basak et al., 2015; Duregotti et al., 2015). To investigate Nrg1 type III function during axon remodeling, we crossed floxed Nrg1 type III (Velanac et al., 2012) to TdTomato reporter mice and injected neonates with AAV9-CMV-iCre (Fig. S3 E). As expected, myelination was severely impaired in TdTomato-positive branches compared with internal control axons (Fig. S3, F–H). We further tested if increased Nrg1 type III levels are sufficient to remove the competition-dependent block on myelination using Thy1-Nrg1 type III-HA mice, where Nrg1 type III is tagged with hemagglutinin (Fig. 6, A and B). Here, axon remodeling was transiently accelerated (Lee et al., 2016), and overall nodes form significantly earlier (Fig. 6, C–F; Velanac et al., 2012). Notably the myelination delay on competing din branches was preserved, most likely reflecting the distribution of Nrg1 type III (Fig. 6 F).

To analyze local distribution of Nrg1 type III with single-axon precision, we immunostained for the HA-tag and visualized SCs and axons (Fig. 7, A and B). We detected a higher HA signal along winner sin branches compared with competing din axons, in line with differential trafficking regulated by competition-regulated cytoskeletal maturation (Fig. 7 C). Corroborating differential Nrg1 signaling, we measured significantly higher levels of activated forms of both ERK1/2 (pERK) and AKT (pAKT) surrounding winner sin axons (Fig. 7, D–G). As myelination initiation on winner branches was reduced following neurotransmission blockade, HA-tagged Nrg1 type III and pERK signals also significantly decreased in sin branches (Fig. 7, H and I). Transgenic expression of Nrg1 type III did not change the density or area of AChRs (Fig. 7, J and K). Therefore, Nrg1 type III likely impacts myelination via its promyelinating effects rather than by modulating postsynaptic feedback (Velanac et al., 2012; Kamezaki et al., 2016). Indeed, the only phenotype we observed at the endplate was a premature shape change of the AChR distribution in Thy1-Nrg1 type III-HA mice compared with WT littermates (Fig. 7 L; Lee et al., 2016).

Taken together, our data suggest that during motor axon remodeling, a promyelination signal (such as Nrg1) paces branch myelination, which is limited by axonal transport and depends on local regulation of cytoskeletal integrity. Thus, axon–glial differentiation is delayed until competition resolves and the axonal cytoskeleton matures. Notably, according to this model, axon dismantling and myelination initiation both depend on mechanisms that regulate the microtubular cytoskeleton (Brill et al., 2016). This model also links anterograde transport to a signaling function, which previously was mostly considered for retrograde transport, such as delivering neurotrophic factors (Je et al., 2012; Yamashita, 2019).

The intersection of axon remodeling and myelination is widespread (Feinberg, 1982; Bernstein and Lichtman, 1999; Woo and Crowell, 2005; Barres, 2008) and might play a general role in the activity-dependent sculpting of efficient neuronal networks (Luo and O’Leary, 2005; Tapia and Lichtman, 2012; de Hoz and Simons, 2015; Chang et al., 2016). The prevailing notion has been that myelination might terminate axonal plasticity by “cementing” axons in place, thus contributing to closing the critical remodeling period (Caroni and Schwab, 1988; McGee et al., 2005; Geoffroy and Zheng, 2014; Kalish et al., 2020). Indeed, we observed myelination initiation predominantly on winner axon branches (sin), i.e., after competition was resolved (Fig. 1). However, while determining the exact start and progression speed of myelination is technically challenging (thus, we resorted to a “binary” readout of Caspr-positive vs. Caspr-negative branches), our observations clearly reveal that competition delayed myelination, but not vice versa (Fig. 3). A subset of competing axons still initiated myelination, but this did not convey a measurable advantage during competition, as no relationship between myelination and synaptic territory was apparent in our data. Even some retreating axons were myelinated, including in cases where the likely competing branch was not (Fig. 3 G). Still, in general, myelination was prevented on axon branches that were fated for removal. As myelin is an extremely stable structure (Simons and Trotter, 2007; Hughes et al., 2018), which might be metabolically “expensive” to build (Nave and Trapp, 2008; Harris and Attwell, 2012) and requires a dedicated mechanism for dismantling, it seems economical to delay myelination until pruning is resolved (McGee et al., 2005; Cheng and Carr, 2007). Thus, our results support a view where myelin might act as participant in, but not as the terminator of circuit plasticity (Mount and Monje, 2017).

How do competing axon branches delay myelination? A number of cell biological features of such branches scale with its synaptic territory during competition, including cytoskeletal stability, organelle transport, or axon caliber (Keller-Peck et al., 2001; Brill et al., 2016). We can rule out the hypothesis that axon caliber dictates myelination onset, since axon caliber did not differ between myelinated and unmyelinated competing branches (Fig. 3 D; Goebbels et al., 2010), even though we found evidence that along a given branch, myelination has an impact on local diameter (Fig. 3 E). To further probe the mechanism that times myelination onset, we manipulated activity-driven competition (Buffelli et al., 2003) using local BTX injections (Fig. 4). This intervention is muscle specific and therefore less likely to affect axon–SC communication compared with blocking axonal action potential conduction or acetylcholine release (Misgeld et al., 2002; Lorenzetto et al., 2009). Notably, while there exists some cholinergic axon–SC communication, this typically involves BTX-insensitive receptors (Rousse and Robitaille, 2006). In the past, chronic blockade of neuromuscular transmission, such as in chicken embryos treated with curare, has been shown to result in AChR cluster fragmentation and axonal sprouting (cf. Loeb et al., 2002; Loeb, 2003). Also, constitutive genetic ablation of choline acetyl transferase in motor axons leads to premature myelination in the phrenic nerve at birth (Misgeld et al., 2002). Still, these outcomes likely reflect the combination of presynaptic and sustained blockade, prone to elicit homeostatic compensation (Davis, 2013). Here, by using brief and local postsynaptic blockade, we found the expected delay in axon remodeling but also a commensurate hiatus in myelination (Fig. 4, D and E). Thus, myelination onset appears to be part of the BTX-sensitive competition program, revealing an indirect role of neurotransmission in regulating the progress of PNS myelination (for the CNS, cf. Stevens et al., 2002; Gibson et al., 2014; Krasnow et al., 2018). Moreover, postsynaptic block induced a reduction in presynaptic microtubular mass on winner branches (Fig. 4, F and G), suggesting a silencing-induced delay in maturation, which chimes with a cytoskeletal mechanism of inducing myelination.

Indeed, in terminal axon branches, the microtubular cytoskeleton matures in parallel to an increase in synaptic territory (Brill et al., 2016). Accordingly, we observed that spastin deletion, which increased microtubular mass (Fig. 4, L and M), resulted in accelerated myelination specifically on competing branches, breaking the link between remodeling and myelination delay (Fig. 4 K). However, spastin deletion in winner branches did not affect the initiation of myelination. This suggests that the limiting factor in this setting could be cargo instead of track availability, as we reported previously for mitochondria (Marahori, 2020). Microtubular content can locally regulate axonal transport (Kapitein and Hoogenraad, 2015; Roll-Mecak, 2019), hinting that myelination might depend on branch-specific transport. To test this, we expressed a dominant-negative kinesin mutant in vivo, which affects anterograde organelle delivery (Cai et al., 2005; Barry et al., 2014). In both zebrafish and mouse motor neurons, this delayed myelination, despite the transport blockade being partial and short (Fig. 5 and Fig. S4). Taken together, our data suggest that in competing branches, transport of promyelinating cargos is restricted by an immature and increasingly severed microtubular cytoskeleton (Brill et al., 2016).

Since our FRAP experiments suggest that nodal components are not limited in competing branches (Fig. 2; but cf. Roche et al., 2014), we focused on Nrg1 type III as the putative factor determining myelination onset for the following reasons: Nrg1 signaling (1) is the master regulator of PNS myelination (Birchmeier and Nave, 2008; Grigoryan and Birchmeier, 2015); (2) acts in a threshold-based manner (Garratt et al., 2000; Michailov et al., 2004; Taveggia et al., 2005; Nave and Salzer, 2006); and (3) is steeply up-regulated during the synapse remodeling period (Lee et al., 2016), while its axonal presence is limited (Velanac et al., 2012). Indeed, we demonstrated that HA-tagged Nrg1 accumulates faster on winner than on competing branches (Fig. 7 C), and that downstream pathways of Nrg1 signaling are preferentially activated in SCs around winner branches (Fig. 7, F and G). It would have further strengthened our argument if increased Nrg1 could be observed in spastin-deleted axons. Likely due to chromosomal incompatibility of the Nrg1 transgene insertion site and the spastin locus, we failed in showing this, but instead tested the converse setting and showed that HA-tagged Nrg1 distribution and downstream signaling decrease upon BTX blockade (Fig. 7, H and I). The fact that Nrg1 overexpression accelerates both myelination and synapse elimination (Fig. 6; Lee et al., 2016) further strengthens our conclusion that myelination per se does not terminate remodeling. However, it is technically challenging to disambiguate whether Nrg1 signals directly from the axonal surface to myelinating SCs or through a more complicated feedback via muscle and/or terminal SCs. Yet we did not measure a difference in postsynaptic AChR density or area between Thy1-Nrg1 type III-HA and WT littermates (Fig. 7), suggesting Nrg1 likely functions via its promyelinating effects rather than by modulating muscular depolarization.

In summary, our experiments reveal an intercellular signaling mechanism that regulates myelination on a branch-to-branch level in the developing PNS. The extent of branch-specific Nrg1 accumulation, and hence the strength of the promyelination signal, are regulated by the axonal cytoskeleton as a spatially resolved signaling hub (Janke, 2014). A similar local regulation between neuronal remodeling and myelination can be relevant in many developing neural circuits, as certain cortical axon types are myelinated in a highly local fashion (Tomassy et al., 2014; Micheva et al., 2016). Moreover, when disturbed, such signaling could contribute to the disrupted timing of developmental events characteristic of some neuropsychiatric disorders, where axonal transport, neuronal remodeling, and myelination all show subtle defects (Coleman and Perry, 2002; Luo and O’Leary, 2005; Mei and Nave, 2014).

In all experiments, mice from both sexes were included. Animals were housed in individually ventilated cages with food and water ad libitum. All animal experiments were performed in accordance with the regulations by the local authorities (e.g., Government of Upper Bavaria). Experimental animals were kept together with littermates.

For labeling axons, we used Thy1-XFP mice, which express OFP (Thy1-OFP3; Brill et al., 2011), YFP (Thy1-YFP16; Feng et al., 2000), or membrane-RFP (Thy1-Brainbow-1.1 line M; Livet et al., 2007) under control of the Thy1-promotor (Feng et al., 2000). For FRAP experiments and node visualization in living explants, we used Thy1-Caspr-GFP (Brivio et al., 2017) and Thy1-β1-Nav-GFP transgenic mice (generated for this study; see below) crossed to Thy1-XFP mice. Synaptic territory of competing axonal branches was defined in Thy1-Brainbow-1.1 line M mice crossed to Cre-expressing lines: CAG-CreERT (gift from Dr. J. Livet, Institut de la Vision, Paris, France) or ChAT-IRES-Cre (Jackson; #6410; Rossi et al., 2011), which leads to individual fluorescent color combinations of membrane-targeted RFP, YFP, and CFP. For the crossing involving CAG-CreERT, 20 µl of 1.5 mg/ml tamoxifen was subcutaneously injected on P3 to induce expression. Delayed synapse elimination was analyzed in spastin knockout (KO) mice (Brill et al., 2016) or spastin floxed (^fl/fl) mice bred to ROSA-CAG-TdTomato or YFP reporter mice (Jackson; #7914, #7903; Madisen et al., 2010) injected with AAV9-CMV-iCre (Brill et al., 2016). Effects of transport modulation on myelin and nodal development was observed in TdTomato reporter mice crossed to Thy1-β1-Nav-GFP or ROSA-CAG-YFP mice injected with AAV9-hSyn-iCre-p2a-KHC-CBD. Precocious myelination was investigated in Thy1-Nrg1 type III-HA mice (“HANI”; Velanac et al., 2012). To investigate delayed nodal development, we injected AAV9-CMV-iCre into conditional Neuregulin KO mice (Nrg1 type III^fl/fl; Velanac et al., 2012) crossbred with ROSA-CAG-TdTomato reporter. To visualize SCs, we used Plp-GFP transgenic mice (Mallon et al., 2002). All experiments on ROSA-CAG-TdTomato or -YFP reporter mice were performed in homozygous animals.

Genomic DNA was extracted from tail biopsies using a one-step lysis (lysis buffer in mM: 67 Tris, pH 8.8, 16.6 (NH₄)₂SO₄, 6.5 MgCl₂, 5 β-mercaptoethanol, 10% Triton, and 50 µg/ml Proteinase K; incubation at 55°C for 5 h, followed by inactivation step 5 min at 95°C). PCR was performed with GoTaq Green Master Mix (Promega; #M7121) following a standard protocol, and then DNA was separated on a 1.5–2% agarose gel. Genotyping primers and expected products are listed in Table S1.

Transgenic mice expressing the β1 subunit of the sodium channel (β1-Nav) fused to GFP at the C-terminus under control of the Thy1.2 promoter (Caroni, 1997) were generated by pronuclear injection. β1-Nav-GFP cDNA (McEwen et al., 2009) was cloned into the blunted XhoI site of the pTSC21k vector (Lüthi et al., 1997), released using Not I (Zonta et al., 2011), and used for pronuclear injection (Sherman and Brophy, 2000).

Nerve-muscle explants from the thorax, including the triangularis sterni muscle, were prepared from young (P7–P14) or adult mice (6 wk; Kerschensteiner et al., 2008; Brill et al., 2011, 2016). The rib cage was isolated from euthanized animals, and the skin and pectoral muscles over the rib cage were removed. The diaphragm was cut and the thorax was released by cutting the ribs close to the vertebral column. The dissection was continued in oxygenated precooled Ringer’s solution (in mM: 125 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 26 NaHCO₃, 2 CaCl₂, 1 MgCl₂, and 20 glucose, oxygenated with 95% O₂/5% CO₂) in a 10-cm dish, and remnants of muscles, thymus, pleura, and lung were removed. The clean nerve-muscle explant was pinned onto a Sylgard-coated 3.5-cm dish, superfused with oxygenated Ringer’s solution, using shortened insect pins (Fine Science Tools; 26001–25, 0.25 mm), exposing the triangularis sterni muscle, the intercostal nerve, and terminal motor neuron branches. During imaging, the explant was kept at 31–33°C with a heating ring connected to an automatic temperature controller (Warner Instruments; TC-344C) and steadily perfused with Ringer's solution.

Trafficking measurements of β1-Nav-GFP particles in the terminal branches were performed with an Olympus BX51WI epifluorescence microscope equipped with ×20/0.5 NA and ×100/1.0 NA water-immersion objectives, an automated filter wheel (Sutter Instruments; Lambda 10–3), a charge-coupled device camera (Visitron Systems; CoolSnap HQ2), and a GFP ET filter set (AHF Analysentechnik). All devices were controlled by μManager 1.4 (Edelstein et al., 2014). Per movie, we acquired 200 images at 1 Hz using an exposure time of 500 ms. Total imaging time on explants was restricted to maximum 2 h, except for FRAP analysis (below).

For FRAP analysis, we used Thy1-Caspr-GFP and Thy1-β1-Nav-GFP mice and the same setup as described above for transport measurements. The laser (Rapp OptoElectronic; 473 nm, DL-473) for photobleaching was manually focused on a labeled node of Ranvier (∼5 µm²), and the sample was bleached for 1–3 s. We performed FRAP on branches with heminodes during development, since fully developed nodes are rare at the investigated age. In adults, all measurements were performed on fully developed nodes. The GFP signal was imaged with 800-ms exposure time before and immediately after photobleaching with a GFP/mCherry dualband ET filter set (AHF Analysentechnik) and then in 1-h intervals for 3 h again with 800-ms exposure time.

The thorax was fixed in 4% PFA for 1 h in 0.1 M phosphate buffer (PB) on ice and the triangularis muscle was dissected and extracted (Brill et al., 2011). For HA staining, the sample was additionally treated for 1 h at 37°C in 5% CHAPS. The fixed thorax was incubated overnight (or 72 h for HA, pERK, and pAKT) at 4°C in the respective primary antibodies diluted in blocking solution (5% BSA and 0.5% Triton X-100 in 0.1 M PB). To label postsynaptic nicotinic AChRs, Alexa Fluor 488, Alexa Fluor 594, Alexa Fluor 647, or biotin conjugated to BTX (Invitrogen, B13422, B13423, B35450, B1196; 50 µg/ml, 1:50) was added to the primary antibody mixture. The following primary antibodies were used in this study: anti-βIII-tubulin conjugated to Alexa Fluor 488 (BioLegend; AB_2562669; mouse IgG2a, 1:200), Alexa Fluor 555 (BD PharMingen; #560339; mouse monoclonal, 1:200), and Alexa Fluor 647 (BioLegend; AB_2563609; mouse IgG2a, 1:200). For labeling of nodal components antibodies against Caspr (Abcam; AB_869934, polyclonal rabbit; 1 mg/ml 1:400), MPZ (Aves Labs; combined chicken IgY, 200 µg/ml, 1:200), CNTN2 (R&D Systems; AB_2044647; polyclonal goat IgG, 1:200), and pan Nav subunit α (Abgent; AG1392; polyclonal rabbit, 1.0 mg/ml, 1:400). HA was stained with anti–HA-tag antibody (Cell Signaling; AB_1549585; rabbit, 1:50), and phosphorylated ERK1/2 was stained with anti-phospho-p44/42 MAPK (Cell Signaling; AB_331646; rabbit, 1:200), while pAKT was stained with anti–phospho-Akt (Cell Signaling; AB_329824; rabbit, 1:200). Here, we used tyramide signal amplification (PerkinElmer; TSA Cyanine 3 System, AB_2572409). Muscles were washed in 0.1 M PB, incubated for 1 h at room temperature with corresponding secondary antibodies coupled to Alexa Fluor 488, Alexa Fluor 594, or Alexa Fluor 647 (Invitrogen; rabbit: #A-11070, #A-11072, #A-21246, #A-32790; mouse: #A-11005; chicken: #A-11042; #A-21449; goat: #A-11058) and washed again in 0.1 M PB. Muscles were mounted in Vectashield (Vector Laboratories) or Fluoromont-G (Southern Biotech), and image stacks were recorded using a confocal microscope (Olympus; FV1000) equipped with ×20/0.8 NA and ×60/1.42 NA oil-immersion objectives (Olympus).

To generate the pTREK1-hSyn-iCre-p2a-KHC-CBD construct for the AAV9-hSyn-iCre-p2a-KHC-CBD production, we used the Gibson Assembly Master Mix (NEB). We recombined fragments from pEGFP-C1-KHC-CBD (Cai et al., 2005), the dsCMV-iCre and pAAV-hSyn-DIO-HA-hM3D(Gq)-IRES-mCitrine plasmid (Addgene; #50454), and the p2a sequence. The produced amplicon was ligated into the single-stranded AAV backbone plasmid pTREK1. The following oligonucleotides were used: 5′-AGT​ACT​TAA​TAC​GAC​TCA​CTA​TAG​GAT​GGT​GCC​CAA​GAA​G-3′, 5′-TCC​ACG​TCG​CCG​GCC​TGC​TTC​AGC​AGG​GAG​AAG​TTG​GTG​GCG​TCC​CCA​TCC​TCG​A-3′, 5′-TGC​TGA​AGC​AGG​CCG​GCG​ACG​TGG​AGG​AGA​ACC​CCG​GCC​CCA​GTG​CTG​AGA​TTG​ATT​CT-3′, and 5′-ATC​ATG​TCT​GGA​TCC​TCG​ATA​GTT​TAA​ACT​TAC​ACT​TGT​TTG​CCT​C-3′.

For zebrafish injections, we generated pDestTol2CG2_UAS:GFP-KHC-CBD-polyA and pTol2_cntn1b:KHC-CBD-GFP vectors using the Gateway system (Thermo Fisher Scientific). To produce the middle entry clone pME_GFP-KHC-CBD, the GFP-KHC-CBD sequence was amplified from the template plasmid (Cai et al., 2005). The PCR product was then recombined into the vector pDONR221 using BP clonase (Thermo Fisher Scientific). The final expression constructs pTol2_UAS:GFP-KHC-CBD and pTol2_cntn1b:GFP-KHC-CBD were then generated in multisite LR recombination reactions (Thermo Fisher Scientific) with the entry clones, p5E_UAS, p5E_cntn1b, pME_GFP-KHC-CBD, p3E_pA, and pDestTol2_pA of the Tol2Kit (Kwan et al., 2007).

HEK293-T cells were grown in 10-tray Cell Factories (Thermo Fisher Scientific) using Dulbecco's modified essential medium (Gibco) with 10% fetal bovine serum (Gibco) and 1% penicillin/streptomycin (Gibco). The cells were split into the Cell Factories 24 h before transfection to reach 80% confluence at the time of transfection. Then, 420 µg of the pTREK1-hSyn-iCre-p2a-KHC-CBD plasmid and 1.5 mg of the helper plasmid (pDP9rs, kindly provided by Roger Hajjar, Phospholamban Foundation, Amsterdam, Netherlands) were introduced into the HEK293-T cells using polyethyleneimine (Polysciences). 3 d later, the cells were harvested, lysed, and benzonase treated, and the virus was isolated by ultracentrifugation through an iodixanol density gradient (Progen; OptiPrep). Ringer lactate buffer (Braun) was used to replace iodixanol with the help of Vivaspin 20 columns and molecuar weight cut-off 100,000 polyethersulfone membrane (Sartorius). Two 10-tray Cell Factories were pooled and concentrated to a total volume of 500 µl. AAV9 titers were determined by real-time PCR using SYBR Green Master Mix (Roche). Titers in the range of 10¹⁴ viral genome copies per milliliter were acquired.

AAV9 was injected into neonatal pups according to previously published protocol (Passini and Wolfe, 2001). In short, P3 pups were anaesthetized with isoflurane (Abbott) and injected with 3 µl AAV9-CMV-iCre or AAV9-hSyn-iCre-p2A-KHC particles into the right lateral ventricle at a rate of 30 nl/s using a fine glass pipette (Drummond; 3.5”, #3-000-203-G/X) attached to a nanoliter injector (World Precision Instruments; Micro4 MicroSyringe Pump Controller connected with Nanoliter 2000). All surgeries were conducted under ultrasound guidance (Visualsonics; Vevo1100 Imaging System, with a Microscan MS550D transducer). 0.05% (wt/vol) trypan blue was added to the viral solution for visualizing the filling of the injected ventricles. Only whole litters were injected, and pups were allowed to recover on a heating mat before the litter was returned to their mother into the home cage and sacrificed on P9 for immunohistochemistry. To monitor Cre-mediated recombination, mice carried in addition to the respective genes (spastin or Nrg1 type III conditional KO) two TdTomato or YFP reporter alleles (homozygous), which resulted in robust expression of the reporter in a subset of motor neurons.

Injection of BTX on P7 was administered in a similar manner, but the needle was inserted laterally under the skin of the right thorax, and 1 µl of a 50 mg/µl Alexa Fluor 488– or Alexa Fluor 594–conjugated BTX solution (Invitrogen; B13422, B13423) was injected. The contralateral (left) triangularis sterni muscle was unaffected and used as control. The injected pups were viable and active after the treatment and not distinguishable from untreated controls. Ipsi- and contralateral triangularis sterni muscles were then post hoc stained with Alexa Fluor 594– or Alexa Fluor 488–conjugated BTX, respectively, resulting in complementing stainings for blocked and unblocked AChRs. We confirmed a substantial degree of persisting blockade at P9 (11.9 ± 4.5 fold change of BTX staining on injected vs. noninjected side, n ≥ 50 NMJs in five mice) and the absence of denervation (>100 NMJs per mouse, n = 3 mice).

Fertilized Tg(mbp:RFP)^tum102Tg eggs (Auer et al., 2018) at the one-cell stage were pressure microinjected with 1 nl solution containing 20–40 ng/µl plasmid DNA (cntn1b:GFP-KHC-CBD or control cntn1b:GFP; Auer et al., 2018) and 25–50 ng/µl transposase mRNA. For immunohistochemistry, larval zebrafish at 6 dpf were euthanized with 4 mg/ml MS-222 (PHARMAQ) and then fixed overnight in 4% PFA in 0.1 M PB. After fixation, the samples were washed three times in PBS and 0.1% Tween20, immersed for 2 h at room temperature in blocking solution (5% BSA and 0.5% Triton X-100 in 0.1 M PB), and then incubated in primary antibody against α-tubulin (Sigma-Aldrich; #00020911, mouse, 1:200) at 4°C for 48 h in blocking solution. Samples were washed and incubated in secondary antibody conjugated to Alexa Fluor 647 (Invitrogen; goat-anti-mouse #A-28181) overnight at 4°C (Hunter et al., 2011). Samples were washed again and embedded in Vectashield (Vector Laboratories). Image stacks were recorded using a confocal microscope (Olympus; FV1000) equipped with a ×20/0.8 NA oil-immersion objective.

To label Rohon–Beard neurons, fertilized embryos from WT fish were coinjected with a sensory neuron–specific Gal4 driver construct (containing enhancer elements from isl1; Sagasti et al., 2005) together with UAS:KHC-CBD-GFP and UAS:mitoTagRFP-T plasmids (each at 5 ng/ul). Alternatively, UAS:KHC-CBD-GFP and UAS:mitoTagRFP-T plasmids were coinjected into fertilized eggs from the isl2b:Gal4 line (Ben Fredj et al., 2010). At 2 dpf, embryos were anesthetized using 0.2 mg/ml MS-222 (PHARMAQ) and embedded in UltraPure Low Melting Point Agarose (Thermo Fisher Scientific) on a glass coverslip. After selecting double-labeled Rohon–Beard neurons, mitochondrial transport was imaged for at least 50 min in the stem axon using the wide-field microscope configured as in the FRAP experiments. We acquired movies with an imaging frequency of 2 Hz and an exposure time between 200 and 500 ms for each fish (Plucińska et al., 2012).

Innervation patterns were determined by counting the number of innervating terminal branches ending on each BTX-stained NMJ in ImageJ/Fiji (Schindelin et al., 2012). The myelination status of a terminal branch (axon from last bifurcation until NMJ) was determined by any presence of clustered markers of nodal or internodal differentiation (Caspr, CNTN2, MPZ, and Nav). Immunostaining on branchpoints was excluded due to difficulties to discern from the more prominent nodal structures on the stem axon. Axon diameter was determined by measuring the area of the entire terminal branch divided by the length of the branch, resulting in an averaged axon diameter. We verified the precision of our axonal diameter measurement by comparison with other methods to determine axonal diameter (e.g., averaged multisite measurements and determination of smallest diameter) and found no significant difference between Caspr-positive and Caspr-negative axons in measured caliber with the different approaches we tested.

Tubulin content of axons was determined by manually placing regions of interest in a single optical section within an axon, and the mean gray values were averaged for each channel. We have previously established that immunohistochemically determined tubulin content correlates linearly with microtubule density as measured by electron microscopy, once corrected for an offset likely representing nonpolymerized tubulin (Brill et al., 2016).

For FRAP analysis, in-focus images were manually aligned, and the intensity of the bleached area was measured with the polygon tool. The background intensity was measured in a dark and even area, and another GFP-positive paranode in the same field of view was used as control to correct for the recovery rate.

For zebrafish myelination, the motor axon length was determined using the segmented line tool based on α-tubulin staining, and the length of the myelinated stretch was likewise determined based on mbp:RFP fluorescence.

Colocalization of GFP-positive Nav puncta and antibody staining was analyzed in single optical sections of unprocessed images.

To determine transport rates of mitochondria or β1-Nav-GFP particles traveling along the axons, we counted the number of fluorescent particles passing through a region in focus of the axon quantified.

For image representation, maximum intensity projections were generated from confocal image stacks with ImageJ/Fiji and then further processed in Adobe Photoshop, where channels were adjusted individually. For better visibility of dim structures, gamma was adjusted in images that only represent morphological detail; no gamma adjustment was performed in quantitative images (all panels in Fig. 2, A and C; Fig. 4, F and L; Fig. 6, A and B; Fig. 7, B and E; and Fig. S2 G).

All analyses were performed with the experimenter blinded to the treatment or genotypes during imaging and scoring.

Statistical tests were performed using GraphPad Prism software. Statistical significance was determined using the Mann–Whitney test (nonparametric test for two groups), following the Kruskal–Wallis test with post hoc Dunn’s multiple comparisons test (nonparametric test for three or more groups), respectively. An unpaired t test was used when the dataset passed the D’Agostino and Pearson normality test. The χ² test was used for comparing expected frequencies between groups, and the P value calculated from the test was shown. Group sizes were determined using experience values from prior studies (e.g., Brill et al., 2016; Plucińska et al., 2012). P < 0.05 was considered to be significant (*, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001). Bar graphs show mean + SEM. Violin plots depict median and quartiles excluding the outliers, which were identified with Tukey’s test (Fig. 2, B and D; Fig. 5, B; Fig. S2, H; and Fig. S3 D).

Fig. S1 shows characterization of Thy1-Caspr-GFP and Thy1-β1-Nav-GFP mice. Fig. S2 shows that innervation and myelination status correlate with axonal tubulin content and SC length. Fig. S3 shows that AAV9-mediated spastin deletion promotes myelination on competing branches. Fig. S4 shows that microtubule-dependent axonal transport affects myelination onset. Table S1 lists the mouse lines and genotyping primers used.

We would like to thank M. and N. Budak, K. Wullimann, and Y. Hufnagel for excellent technical assistance. We thank Dr. L. Godinho for critically reading the manuscript. We are grateful to Dr. J. Livet (Institut de la Vision, Paris, France) for the generous gift of CAG-CreERT mice and Dr. J. Trimmer (University of California, Davis, Davis, CA) for the Caspr antibody.

Work in T. Misgeld’s laboratory is supported by the Deutsche Forschungsgemeinschaft (CIPSM EXC114, CRC870, FOR 2879, Mi 694/7-1, 8-1, and 9-1-428663564, TRR274 - TRR 274/1 2020; ID 408885537, projects B03, C02), the European Research Council (FP/2007-2013; ERC grant agreement 616791), the Gemeinnützige Hertie Stiftung, the German-Israeli Foundation for Scientific Research and Development (I-1200-237.1/2012), the Elite Network Bavaria (S-LW-2016-351), the German Center for Neurodegenerative Diseases, and Technische Universität München’s Institute for Advanced Studies. T. Misgeld and M.S. Brill were further funded by the Deutsche Forschungsgemeinschaft–funded Excellence Cluster SyNergy (EXC 2145 – ID 390857198). M.S. Brill is recipient of a Deutsche Forschungsgemeinschaft research grant (LE 4610/1-1; 450131873). M. Wang received support from the Technische Universität München PhD program Medical Life Sciences and Technology. D.L. Sherman and P.J. Brophy were supported by the Wellcome Trust (grant 107008).

The authors declare no competing financial interests.

Author contributions: Conceptualization, M. Wang, T. Misgeld, and M.S. Brill; investigation, M. Wang, T. Kleele, G. Plucinska, and Y. Xiao; methodology, D.L. Sherman and P.J. Brophy; resources, P. Avramopoulos, S. Engelhardt, M.H. Schwab, P.J. Brophy, D.L. Sherman, M. Kneussel, and T. Czopka; writing – original draft, M. Wang, T. Misgeld, and M.S. Brill; writing – review and editing, all authors; supervision, T. Misgeld and M.S. Brill; funding acquisition, T. Misgeld.