A conserved axon type hierarchy governing peripheral nerve assembly

In gnathostome vertebrates, including fish, birds and mammals, peripheral nerves link nervous system, body and immediate environment by integrating efferent pathways controlling movement apparatus or organ function and afferent pathways underlying somatosensation. Several lines of evidence suggest that peripheral nerve assembly involves instructive interactions between efferent and afferent axon types, but conflicting findings challenge this view. Using genetic modeling in zebrafish, chick and mouse we uncover here a conserved hierarchy of axon type-dependent extension and selective fasciculation events that govern peripheral nerve assembly, which recapitulates the successive phylogenetic emergence of peripheral axon types and circuits in the vertebrate lineage.


INTRODUCTION
Nervous system evolution seems to have frequently proceeded via the use of pre-existing axon pathways, rather than through de novo formation of nerve tracts, to accommodate novel features within extant circuits (Katz, 1983;Katz et al., 1983). The segmental organization of the vertebrate body axis, for example, places constraints on peripheral axon growth that force primary somatosensory axons (SAs) to extend through peripheral nerve tracts that are also occupied by more ancestral motor efferent axons (MEs) (Bonanomi and Pfaff, 2010;Butler and Hodos, 2005). Such co-confinement to narrow substrate corridors may effectively foster interactions between axons that extend from phylogenetically newer or older neuron types, and accelerate their incorporation into common functional assemblies.
Indeed, peripheral nerve assembly has long been thought to involve prerequisite association of SAs with earlier extending MEs en route to peripheral targets (Hamburger, 1929;Taylor, 1944;Honig et al., 1986;Swanson and Lewis, 1986;Honig et al., 1998). For example, surgical or laser-mediated removal of ventral neural tube segments, including motor neurons, in amphibian and avian embryos was observed to frequently result in the development of aneural limb muscle (Hamburger, 1929;Honig et al., 1986;Swanson and Lewis, 1986;Taylor, 1944). Our own recent data provided a mechanistic basis for some of these ideas by showing that, in mouse, SAs are guided to targets in the dorsal trunk by molecular labels on earlier extending epaxial MEs (Wang et al., 2011). Genetic manipulations that completely blocked ME extension resulted in randomized extension of SAs along dorsal or ventral trajectories, whereas ME-restricted elimination of the EphA receptor tyrosine kinases EphA3 and EphA4 triggered selective dorsal-to-ventral misrouting of SAs (Wang et al., 2011;Wang and Marquardt, 2013).
Other recent data on mouse limb innervation, however, support a minimal cooperative model according to which mutual interactions between MEs and SAs have only a limited influence on the establishment of peripheral nerve trajectories (Huettl et al., 2011). For example, upon partial genetically induced elimination of MEs in mouse embryos, only mild SA extension defects were observed, whereas elimination of DRG neurons or DRG neuron-derived neuropilin receptor expression led to pronounced defasciculation, but not to mistargeting, of MEs (Huettl et al., 2011). Studies based on surgical manipulations in frog and chick embryos arrived at yet different conclusions, proposing that establishment of limb SA and ME trajectories can be entirely dissociated from each other (Wang and Scott, 1999;Wenner and Frank, 1995). Consolidating these conflicting lines of evidence remains difficult because differences in animal models, methodologies and positional identities of the peripheral nerve segments studied currently preclude their direct comparison.

Conserved reliance of sensory axon extension on pioneer motor axons
To explore the interactions between primary somatosensory afferent axons (SAs) and motor efferent axons (MEs) we first systematically investigated the relationships between molecularly identified peripheral axon types in three different vertebrate species: zebrafish (D. rerio: Fig. 1A), chick (G. gallus domesticus: Fig. 1J) and mouse (M. musculus: Fig. 1S). In anamniotes, including zebrafish, the first emerging sensory-motor circuits are dedicated to simple larval escape reflexes that are mediated by an early central nervous system (CNS) neuron population, Rohon Beard cells (RBs), which feed primary sensory inputs from dermis to motor neurons controlling trunk musculature (Spitzer, 1984). RBs are eventually replaced by neural crest-derived dorsal root ganglion (DRG) neurons that become incorporated into circuits facilitating a wider spectrum of motor outputs (Butler and Hodos, 2005;Dasen, 2009;Holland, 2009). In zebrafish, these circuit rearrangements are reflected by a ∼20 h delay between initiation of primary ME extension, visualized by a motor neuron-specific regulatory module of the NBT (Xenopus β-tubulin) gene to drive red fluorescent protein expression (dsRed) (Peri and Nusslein-Volhard, 2008), and the emergence of SAs from DRGs, visualized by exploiting the DRG neuron-restricted activity of the neurogelin 1 promoter to drive green fluorescent protein (GFP) expression (Blader et al., 2003) (Fig. 1B-E;supplementary material Fig. S1A-E). SAs thus invariably extended along preformed ME trajectories.
To visualize SAs and MEs in chick, we used previously identified enhancer modules of the Isl1 and Hb9 genes (Lee et al., 2004) to confine green and red fluorescent protein expression to DRG and motor neurons, respectively (supplementary material Fig. S2A-F). In mouse, DRG and motor neurons axons were respectively visualized by the previously established DRG and motor neuron-specific transgenes Brn3a tau:lacZ and Hb9::eGFP (Lee et al., 2004;Trieu et al., 2003). In amniotes, including chick and mouse, formation of larval Rohon Beard circuits is skipped, and primary sensory-motor circuits directly assemble from motor neurons and DRG neurons (Dasen, 2009;Holland, 2009). Despite this altered configuration, the principal chronological sequence of peripheral axon extension observed in zebrafish appears to be preserved in chick and mouse: SAs invariably extended along trajectories pioneered by MEs, as reported previously (  (Honig et al., 1986;Wang et al., 2011).
We next tested whether this rigid axon type-dependent extension order would reflect a conserved primacy of the first-extending MEs during peripheral nerve assembly. To address this, we studied SA extension upon preventing ME extension. In zebrafish, this was achieved by injecting morpholino oligonucleotides targeting the Isl1 gene that were previously found to selectively interfere with motor neurogenesis (Hutchinson and Eisen, 2006). The resulting absence of MEs led to severely reduced initial extension rates of SAs ( Fig. 1F-I) and frequently triggered highly erratic patterns of SA extension (supplementary material Fig. S1L-Q). In chick, the prevention of ME extension, by introducing Hb9-driven cellautonomous diphtheria toxin (DTA) into the neural tube prior to motor neurogenesis, resulted in a randomized loss of dorsal or ventral trunk SA pathways within the confines of normal peripheral nerve trajectories (supplementary material Fig. S2G-J), resembling those previously reported by us upon DTA-mediated ablation of motor neuron progenitors in mouse (Wang et al., 2011).
We next asked whether the apparently conserved role of MEs in establishing trunk SA trajectories would similarly govern assembly of peripheral nerve pathways that co-evolved with the tetrapod limb (Butler and Hodos, 2005;Luria et al., 2008;Ma et al., 2010). Similar to zebrafish, genetic ablation of motor neurons or their progenitors prior to ME extension in chick and mouse (supplementary material Figs S2G,H and S3A-J) resulted in markedly reduced SA extension rates ( Fig. 1O-R,X-AA). In contrast to the effects of ME removal on chick and mouse trunk innervation, where SAs projected in a randomized manner within largely normal peripheral pathways (see supplementary material Fig. S2I,J) (Wang et al., 2011), lumbar SAs exhibited highly aberrant projections that frequently deviated from the trajectories normally chosen by peripheral axons (Fig. 1Q,Y,Z). These defects culminated in the failure or severe delay of most SAs to extend beyond the limb plexus (  extension, however, impacts on the establishment of peripheral SA trajectories were considerably milder in both chick ( Fig. 2A-C) and mouse ( Fig. 2D-G), resembling results obtained by previous studies relying on late-stage surgical or incomplete genetic removal of motor neurons (Huettl et al., 2011;Wang and Scott, 1999;Wenner and Frank, 1995).

Sensory axons are invariably guided by motor axons
Do the trajectories chosen by MEs therefore serve as a template for the assembly of peripheral nerves by guiding later-extending axon types? We further tested this by genetically forcing MEs to choose aberrant trajectories, followed by visualizing SAs. In zebrafish, this was achieved by injecting morpholino oligonucleotides targeting a splice variant of the MUSK (muscle-specific kinase receptor) gene that were previously established to selectively alter the pattern of ME extension (Zhang et al., 2004). In these larvae, SAs faithfully recapitulated aberrant trajectory choices made by MEs ( Fig. 3A-I), with SAs invariably skipping trajectories not occupied by MEs (Fig. 3J,K). In parallel, we took advantage of two previously established gene disruptions in mouse that cause varying degrees of ME misrouting at a binary axon choice point at the base of the limb (supplementary material Fig. S4A-H) (Helmbacher et al., 2000;Kramer et al., 2006;Luria et al., 2008). First, the motor neuronrestricted expression of the receptor tyrosine kinase (RTK) EphA4 was selectively eliminated in the motor neuron lineage through Cre/loxP recombination in Epha4 fx/fx ;Olig2 Cre embryos. Second, the expression of the RTK Ret, which is largely motor neuron restricted at the relevant embryonic stages (E9.5-E12), was eliminated in Ret −/− embryos. Both models selectively affected the dorsal choice of MEs originating from the lateral division of the lateral motor column by altering their responsiveness towards mesenchymal guidance cues (  (Helmbacher et al., 2000;Kramer et al., 2006;Luria et al., 2008). As in zebrafish, aberrant ME projections were closely mirrored by SA projections in the mouse hindlimb ( Fig. 3L-V; supplementary material Fig. S4I-Q) and revealed a linear relationship between the extent of ME projections and that of the later-extending SAs (Fig. 3W). In both models, SAs thus continued to tightly adhere to MEs, indicating a restriction of EphA4 and Ret function to ME-mesenchyme signaling, but not in SA-ME interactions, during limb innervation. Taken together, SAs thus appear to invariably favor and tightly adhere to trajectories occupied by earlier extending MEs.

Sensory axons are dispensable for motor axon guidance
We next asked whether these data reflected a hierarchical axon typedependent relationship by testing whether SAs would conversely influence ME extension by DTA-mediated genetic ablation of DRG neurons in mouse embryos (supplementary material Fig. S5A-F). Consistent with previous data (Huettl et al., 2011), genetic removal of DRG neurons in mouse led to defasciculation of MEs (Fig. 4A,B), which could reflect the loss of repulsive activities exerted by SAs on MEs (Gallarda et al., 2008). As the absence DRG sensory neurons prevents the assembly of DRGs proper, which in turn provide a niche for the expansion of Schwann cell precursors, the defasciculated appearance of MEs could have been alternatively (or additionally) caused by reduced numbers of Schwann cell precursors. At the same time, absence of SAs did not influence the accuracy of trajectory or target selection by MEs (Fig. 4C-F), consistent with MEs normally extending ahead of SAs. SAs thus appear to invariably depend on pre-extending MEs for establishing normally patterned peripheral trajectories, but not vice versa, whereas SAs exert repulsive activities that prevent aberrant intermingling with (and possibly defasciculation of ) MEs. Sympathetic axon trajectories are configured by pre-extending sensory and motor axons Consistent with previous observations Coughlin et al., 1977;Yip, 1990), sympathetic efferent axons (SEs) emerging from sympathetic chain ganglia (SCGs), visualized by tyrosine hydroxylase (TH) immunodetection, were the last axons to extend peripherally in zebrafish ( Fig. 5A-E), chick ( Fig. 5F-L) and mouse ( Fig. 5M-Q), thus recapitulating the phylogenetically late innervation of neural crestderived autonomic circuits (Butler and Hodos, 2005;Holland, 2009). At trunk levels, SEs emerging from SCGs follow three principal routes to access effector organs (Elfvin, 1983): (1) (Nakao and Ishizawa, 1994). In amniotes, the lateral route SEs eventually project along pre-extending cutaneous SAs to innervate dermal glands and smooth muscle as part of the circuits underlying skin thermoregulation (Figs 6D-J and 7C,D) (Elfvin, 1983;Nakao and Ishizawa, 1994).
We next tested whether this sequential extension pattern reflected a dependence of SEs on preformed SAs and/or MEs in mouse  models that allowed us to separately address the contribution of each of the three principal axon types to peripheral nerve assembly (supplementary material Fig. S6A-D). This was achieved by separately interbreeding the Cre recombination-controlled DTA expression lines R26 fxDTA (Ivanova et al., 2005) or Isl2 lxstopDTA (Yang et al., 2001) with previously established mouse lines that drive Cre expression in motor neuron progenitors (Ivanova et al., 2005), DRG neurons (Hasegawa et al., 2007) or SCG neurons (Parlato et al., 2007) (supplementary material Fig. S6A-D). We found that the aberrant cutaneous SA projection patterns resulting from the absence of MEs (Wang et al., 2011) were precisely mirrored by SEs (Fig. 7F,G), including frequent failure of dermal target innervation by both SEs and SAs (Fig. 7H,I). This suggested that MEs are dispensable for initiating the peripheral extension of SEs, but indirectly influence their trajectory by determining the pattern of SA projections (Fig. 7J). We further tested this by studying SE projections upon selective genetic removal of SAs. In the absence of SAs, SEs extended peripherally along motor projections (Fig. 7K,L), but consistently failed to enter cutaneous trajectories, leaving the trunk dermis entirely devoid of innervation ( Fig. 7M-O). At the same time, intersegmental blood vessels developed normally in the absence of MEs and SAs (supplementary material Fig. S6E-J), indicating that failure of cutaneous SE projections upon SA or ME removal was not indirectly caused by impacts on vascular patterning. Last, selective genetic removal of SEs did not result in detectable alterations of peripheral ME or SA projections (Fig. 7P-T), consistent with the initiation of SE extension after most ME and SA trajectories have been established. Thus, although a subset of SEs uses the developing vasculature to access peripheral end organs (Makita et al., 2008), trunk cutaneous SE projections are absolutely reliant on their association with pre-formed SA trajectories that, in turn, are initially guided by MEs (Fig. 8A).

DISCUSSION
Our data uncover a conserved hierarchy of axon type-dependent extension and selective fasciculation events governing vertebrate peripheral nerve assembly (Fig. 8A), the temporal order of which recapitulates the successive phylogenetic emergence of peripheral axon types and circuits in the vertebrate lineage ( Fig. 8B-F). First, MEs actively navigate to skeletal muscle targets guided by mesenchymal cues (Bonanomi and Pfaff, 2010;Dasen, 2009), thus establishing an initial grid of peripheral trajectories that provides a template for subsequent peripheral nerve assembly (Fig. 8B). Second, SAs use preformed ME trajectories as gateways to their peripheral target organs (Fig. 8C-E). Third, subsets of SEs eventually follow these pre-established trajectories, presumably by responding to cues on SAs (Fig. 8D-F).
MEs extending from cholinergic motor neurons to muscles involved in locomotion represent the sole common feature of peripheral nerves in extant chordates, predating both vertebrates and neural crest-derived circuits (Fig. 8B) (Candiani et al., 2012;Denes et al., 2007;Fetcho, 1992). SAs and SEs, which emerged in agnathostomes and gnathostomes, respectively (Butler and Hodos, 2005;Holland, 2009), were thus from the outset able to rely on preevolved axon pathways for accessing peripheral end organs (Fig. 8C-F), This pattern could have been enforced by the segmental remodeling in gnathostomes that brought ventral and dorsal peripheral nerve roots in closer apposition (Butler and Hodos, 2005;Jorgensen, 1998).
The timing of axon extension can profoundly influence the outcome of heterotypic axon-axon encounters (Wang and Marquardt, 2013), which may have been a key factor in preserving peripheral axon type hierarchies from fish to mouse. For example, at least in mouse, the early extension of MEs seems to promote selective fasciculation of SAs with pre-extending MEs by encouraging reverse activation of ephrin A proteins on SA growth cones by their cognate EphA RTKs located on MEs (Wang et al., 2011). In addition, this configuration effectively discourages illicit forward activation of repulsive signaling by the same EphAs on ME growth cones by their cognate ephrin A interaction partners on SAs (Gallarda et al., 2008;Wang et al., 2011), which could otherwise jeopardize the role of MEs in pioneering peripheral nerve tracts. The retention of this hierarchy of axon type-dependent interactions could have been further promoted by the prioritized assembly of ME and RB-based escape reflex arcs over more advanced SA and SE-based somatosensory and autonomic circuits in anamniotes developing through pelagic larval stages (Fig. 8B,C).
Resolving axon type-dependent relationships in the context of the embryo remains challenging, because the underlying signaling mechanisms may operate independently from those determining the responsiveness of axons towards non-axonal guidance cues (Raper and Mason, 2010;Wang and Marquardt, 2013). Thus, altering axon-axon signaling tends to produce changes of axonal trajectories within the confines of relatively fixed pathways that are determined by tissue tracts that are permissive or non-permissive to axon growth. Advancing our understanding of the contribution of axonaxon interactions to nervous system development will therefore depend on expanding the toolkit for unambiguously distinguishing or manipulating the different axon types involved.
Although some of the ontophyletic considerations put forward here presently remain speculative, our findings pave the way for systematically exploring the cellular and molecular basis of the axon-axon interactions contributing to peripheral nerve assembly, and may ultimately serve as a generalized model for how, during nervous system development, phylogenetically newer and older neuron types assemble into common circuitries.

Chick
Fertilized eggs were staged according to Hamburger and Hamilton (HH) (Hamburger and Hamilton, 1992) and transfected using plasmid injection into the neural tube and in ovo electroporation at HH 12/13 as previously described (Marquardt et al., 2005). To label MEs, a previously characterized 4.5 kb promoter fragment of the Hb9 gene, Hb9 MN , containing motor neuron-specific regulatory elements (Lee et al., 2004) was fused to mCherry (Cherry red fluorescent protein with a myristoylation signal 'm' for membrane tethering). To co-label SAs, a previously characterized DRGspecific regulatory module of the Isl1 gene, Isl1 DRG , was fused to a minimal TATA box and mGFP (m-green fluorescent protein) (Uemura et al., 2005). In addition to DRG neurons, Isl1 DRG labels a subset of Schwann cell precursors (SCPs), presumably originating from their proliferation niche in the DRG (e.g. Fig. 1N, asterisks). For imaging, transverse sections were thus selected containing fewer SCPs to allow unimpeded visualization of SAs. To prevent ME extension, motor neurons were ablated by co-injecting PGKneolox2DTA (Addgene: Plasmid 13449) with Hb9 MN ::Cre for Cre-mediated activation of DTA expression in motor neurons. After electroporation, eggs were incubated at 38°C until the desired stages.

Mouse
All mouse work conformed to regulations by the University Medical Center Göttingen animal welfare committee and German animal welfare laws. MEs and SAs were co-labeled by interbreeding the previously established motor neuron-specific marker line Hb9 MN ::GFP (Lee et al., 2004) and DRG-neuron marker line Brn3a tau:lacZ (Trieu et al., 2003). For preventing ME extension, the Cre-controlled ubiquitously expressed diphtheria toxin (DTA) expression line Rosa26 lxstopDTA (Jax stock #006331) was interbred with the motor neuron progenitor ( pMN)specific Cre-driver line Olig2 Cre to selectively ablate pMNs (Ivanova et al., 2005). For preventing SA extension, the neural crest-restricted Cre-deleter line Wnt1 Cre (Jax stock #009107) (Hayashi and McMahon, 2002) was interbred with the Cre-controlled diphtheria toxin (DTA) expression line Isl2 lxstopDTA (Jax stock #007942) (Yang et al., 2001), thus leading to ablation of Isl2 + DRG neurons derived from Wnt1 + neural crest cells, but not Isl2 + motor neurons. To study impacts on SEs, and to prevent potential ablation of Isl2 + SCG neurons, the DRG neuron-restricted Cre line Advillin Cre (Hasegawa et al., 2007) was interbred with Isl2 lxstopDTA . SE extension was prevented by interbreeding DBH Cre (Parlato et al., 2007) with Rosa26 lxstopDTA . To selectively misroute MEs in the hindlimb, homozygous Epha4 flox (Herrmann et al., 2010) mice were interbred with  Olig2 Cre to specifically inactivate Epha4 in pMNs or embryos homozygous for a Ret-null allele (Ret ko ) (Schuchardt et al., 1994) were studied. The mouse lines were genotyped as described previously.

Retrograde tracing
Hb9::eGFP transgenic mouse embryos were used to identify motor trajectories. DiI (Sigma)-labeled embryos were incubated in 4% paraformaldehyde overnight at 37°C to permit diffusion, prior to vibratome sectioning at 120 μm for analysis.

Live imaging of zebrafish
Injected embryos were dechorionated, anesthetized in 0.004% tricaine and mounted in 0.8% low melting agarose on 35 mm glass-bottom dishes. Live imaging was performed at 28°C using a LSM5 Live confocal microscope (Carl Zeiss Microimaging).

Quantifying the relationship between motor axon and SA extension
For zebrafish larvae, two segments from each were randomly selected for quantification. For each segment, the main branches of motor axon and SA axons were classified in two categories: 'yes', indicating motor axon or SA axons project in close association with their counterparts; and 'no', indicating motor axon or SA axons extend independently. Next, the percentages of 'yes' and 'no' among total incidences were calculated for motor axon and SA, respectively. In mouse embryos, the diameters of all motor axons and SAs in peroneal nerves were separately measured at midsection for each embryo.