Genetic specification of left–right asymmetry in the diaphragm muscles and their motor innervation

The diaphragm muscle is essential for breathing in mammals. Its asymmetric elevation during contraction correlates with morphological features suggestive of inherent left–right (L/R) asymmetry. Whether this asymmetry is due to L versus R differences in the muscle or in the phrenic nerve activity is unknown. Here, we have combined the analysis of genetically modified mouse models with transcriptomic analysis to show that both the diaphragm muscle and phrenic nerves have asymmetries, which can be established independently of each other during early embryogenesis in pathway instructed by Nodal, a morphogen that also conveys asymmetry in other organs. We further found that phrenic motoneurons receive an early L/R genetic imprint, with L versus R differences both in Slit/Robo signaling and MMP2 activity and in the contribution of both pathways to establish phrenic nerve asymmetry. Our study therefore demonstrates L–R imprinting of spinal motoneurons and describes how L/R modulation of axon guidance signaling helps to match neural circuit formation to organ asymmetry. DOI: http://dx.doi.org/10.7554/eLife.18481.001


Introduction
The diaphragm is the main respiratory muscle of mammalian organisms, separating the thoracic and abdominal cavities. Many diseases, including congenital hernia, degenerative pathologies and spinal cord injury, affect diaphragm function and thereby cause morbidity and mortality (Greer, 2013;McCool and Tzelepis, 2012). Despite the large interest given to diaphragm function in various physiological and pathological contexts (Lin et al., 2000;Misgeld et al., 2002;Strochlic et al., 2012), little attention has been paid to the embryological origin of left-right (L/R) asymmetries in diaphragm morphology and contraction, in part because they were inferred to be simply an adaptation to the structure of other, surrounding asymmetric organs such as the lungs (Laskowski et al., 1991;Whitelaw, 1987). In the present study, we investigated the origin and the mechanisms responsible for the establishment of the diaphragm asymmetries, including motor innervation by the left and right phrenic motoneurons that arise in the spinal cord at cervical levels C3 to C5 (Greer et al., 1999;Laskowski and Owens, 1994). Our findings show that both the diaphragm muscle and phrenic nerves have asymmetries, which are established independently of each other during early embryogenesis.

Results
As many L/R asymmetries are determined prenatally (Sun et al., 2005), we analyzed the diaphragm innervation of mouse embryos on embryonic day (E) 15.5, when synaptic contacts begin to be established in this organ (Lin et al., 2001). We observed that the phrenic nerves split into primary dorsal and ventral branches when reaching the lateral muscles, whereby the distance from the end-plate to the nerve entry point differs between the left and right side and results in a characteristic 'T' -like pattern on the left and 'V' -like pattern on the right ( Figure 1A; Figure 1-figure supplement 1A, B). Similar differences in the L/R branching patterns are present in the human diaphragm (Hidayet et al., 1974) (Figure 1-figure supplement 1C). Additionally, we observed an asymmetric number of branches defasciculating from the left and right primary nerves to innervate the motor end-plates ( Figure 1A; Figure 1-figure supplement 1A,B). We further found that the L/R distribution of acetylcholine receptor (AchR) clusters at the nascent neuromuscular junctions also differed, with a 2.1 ± 0.2-fold increase in the medio-lateral scattering of AchR clusters on the right side of the diaphragm compared to the left side (N = 11, p<0.001 Wilcoxon) ( Figure 1B Thus, phrenic branch patterns exhibit clear asymmetries before synapse formation and fetal respiratory movements (Lin et al., 2001(Lin et al., , 2008, and are therefore unlikely to be induced by nerve activity or muscle contraction. We therefore asked whether diaphragm nerve asymmetry was genetically hard-wired downstream of Nodal signaling, which initiates a left-restricted transcriptional cascade to establish visceral asymmetry (Komatsu and Mishina, 2013;Nakamura and Hamada, 2012). To answer this question, we examined two complementary types of mouse mutants that have defective Nodal signaling and ensuing lung isomerism. First, we examined Pitx2 DC/DC embryos lacking PITX2C, a transcription eLife digest The diaphragm is a dome-shaped muscle that forms the floor of the rib cage, separating the lungs from the abdomen. As we breathe in, the diaphragm contracts. This causes the chest cavity to expand, drawing air into the lungs. A pair of nerves called the phrenic nerves carry signals from the spinal cord to the diaphragm to tell it when to contract. These nerves project from the left and right sides of the spinal cord to the left and right sides of the diaphragm respectively.
The left and right sides of the diaphragm are not entirely level, but it was not known why. To investigate, Charoy et al. studied how the diaphragm develops in mouse embryos. This revealed that the left and right phrenic nerves are not symmetrical. Neither are the muscles on each side of the diaphragm. Further investigation revealed that a genetic program that establishes other differences between the left and right sides of the embryo also gives rise to the differences between the left and right sides of the diaphragm. This program switches on different genes in the left and right phrenic nerves, which activate different molecular pathways in the left and right sides of the diaphragm muscle.
The differences between the nerves and muscles on the left and right sides of the diaphragm could explain why some muscle disorders affect only one side of the diaphragm. Similarly, they could explain why congenital hernias caused by abdominal organs pushing through the diaphragm into the chest cavity mostly affect the left side of the diaphragm. Further studies are now needed to investigate these possibilities. The techniques used by Charoy et al. to map the molecular diversity of spinal cord neurons could also lead to new strategies for repairing damage to the spinal cord following injury or disease.  factor downstream of Nodal (Essner et al., 2000;Liu et al., 2001;Schweickert et al., 2000). In the absence of PITX2C, Nodal signaling is interrupted, which causes a right pulmonary isomerism (i.e. the left lung has three main lobes like the right lung, instead of only one) (Liu et al., 2001(Liu et al., , 2002. Second, we examined Rfx3 -/embryos lacking RFX3, which is essential for cilia function that helps to distribute Nodal to the left side of the body. As a result, some Rfx3 -/embryos exhibit bilateral Nodal expression and left pulmonary isomerism (i.e. the right lung has one lobe like the left lung) (Bonnafe et al., 2004). We found that diaphragm L/R nerve asymmetries were lost in both Pitx2 DC/DC and Rfx3 -/embryos with impaired visceral asymmetries at E14.5 (Figure 2A-E) (number of secondary branches, Wt versus mutant with lung isomerism: PITX2C, p=4.493E-5; RFX3, p=0.002884; defasciculation distance, Wt versus mutant with lung isomerism: PITX2C, p=0.001268; RFX3, p=2.719E-6, Mann-Whitney). Thus, the Nodal pathway is essential for the establishment of diaphragm nerve asymmetry. We next asked whether phrenic nerve asymmetry has an environmental origin, because it is conceivable that the lung buds confer L/R asymmetry-inducing signals to nerves that are navigating   close by ( Figure 3A,B). However, the analysis of Pitx2 DC/DC and Rfx3 -/mutants showed that the pattern of nerve asymmetry did not always correlate with the pattern of lung asymmetry; for example, in 2/10 Pitx2 DC/DC embryos, nerve patterns were normal even though the lungs were isomerized (Pitx2 ∆c/∆c ) (Rfx3 -/-) Pitx2 ∆c/∆c Figure 3. The asymmetry of phrenic circuits results from an intrinsic neuronal program. (A) Schematic representation of the organisation of the phrenic nerves as they pass through the lungs and reach the diaphragm. (B) Photomicrographs of the expected L/R asymmetry of lungs and diaphragm muscles at E14.5 in wild-type embryos and the altered L/R asymmetry observed in the Rfx3 -/and Pitx2 DC/DC mutant embryos. Quantification of diaphragm muscle asymmetry: Pitx2 +/+ and Pitx2 DC/+ 6.25 ± 0.68, N = 20, versus Pitx2 DC/DC iso 0.26 ± 0.6, N = 6; Rfx3 +/+ and Rfx3 -/+ 7.02 ± 0.74, N = 17 versus Rfx3 -/-iso 0.72 ± 1.6, N = 7 (see methods). (C) Schematic representation of L/R asymmetries in the lungs, diaphragm muscles and phrenic nerves. A colour code is used to show the uncoupling occurring between phrenic nerve and lung asymmetries or phrenic nerve and diaphragm muscle asymmetries. Any structure represented in green is indicative of its left characteristics, whether it is observed on the left or the right side of the embryo, whereas red structures represent right characteristics. (20%; Figure 3C; Figure 3-figure supplement 1A,B). Moreover, 1/13 Rfx3 -/embryos exhibited nerve isomerism together with pulmonary situs inversus, and nerve patterns were reversed in 1/13 embryo with lung isomerism (7.7% and 7.7%; Figure 3C; Figure 3-figure supplement 1A,B). Alternatively, it is conceivable that muscle asymmetry controls nerve asymmetry. In agreement with this possibility, L/R asymmetry of the lateral diaphragm muscles was lost in both Pitx2 DC/DC and Rfx3 -/mutants ( Figure 3B). However, muscle width did not correlate with changes in nerve patterns in 2/ 10 Pitx2 DC/DC embryos or in 6/13 Rfx3 -/embryos (20% and 46.2%, respectively). For example, muscle isomerism could be observed in 1/13 Rfx3 -/embryos that have normal nerve patterns (7.7%) or in 1/13 Rfx3 -/embryos with reversed nerve patterns (7.7%). Finally, nerves were isomerized in 2/13 Rfx3 -/embryos that exhibit normal L/R muscle asymmetry (15.4%) ( Figure 3C; Figure 3-figure supplement 1C). Together, these findings raise the possibility that phrenic motoneurons possess intrinsic L/R differences that are established independently of visceral and muscle asymmetries. 3D reconstructions of cervical spinal cord tissue immunolabeled with Pou3f1/Oct6, whose expression has been reported in motoneurons (Philippidou et al., 2012), did not reveal any obvious differences in the L/R organization of the cervical motoneuron pools in the spinal cord ( Figure 3D-E). We therefore explanted phrenic motoneuron-enriched Hb9::GFP spinal cord tissue (Wichterle et al., 2002) to follow the behavior of motor axons as they extended from the explants independently of the surrounding organs ( To identify molecular determinants of L/R differences in phrenic axon growth, we laser-captured left versus right GFP-positive cervical motoneurons from Hb9::GFP transgenic E11 embryos for microarray analysis ( Figure 4A). The presence of several markers for phrenic motoneurons (e.g. Pou3f1/Oct6, Islet1 and ALCAM) in the microarray data demonstrated the accuracy of the dissection procedure ( Figure 4-figure supplement 1A-B). Consistent with the lack of obvious anatomical differences distinguishing left and right Pou3f1/Oct6 + cervical motoneuron populations, none of these markers had asymmetric expression levels. We further observed that amongst 22,600 transcripts expressed above background, 146 were enriched on the left and 194 on the right, with a predominance of transcripts encoding nuclear proteins (differentially enriched transcripts: right 35.56% versus left 26.02%; Figure 4B; Figure 4-source data 1 and 2). Immunoblotting confirmed that Morf4l1, a protein involved in histone acetylation/deacetylation and chromatin remodeling and reported to be essential for neural precursor proliferation and differentiation (Chen et al., 2009;Boije et al., 2013), was enriched in the left cervical motoneuron domain (L/R fold-change 1.81 ± 0.163, p=0.0022, Mann-Whitney; Figure 4C-E). Xrn2, a protein regulating RNA processing and miR stability that regulates miR expression in neurons (Kinjo et al., 2013), was also enriched in the left cervical motoneuron domain (L/R fold-change 1.37 ± 0.13, p=0.028; Mann-Whitney; Figure 4-figure supplement 1C). Thus, cervical motoneurons are intrinsically L/R-specified.
To determine whether molecular differences in L/R specification manifest themselves in differential axon guidance responses, we studied mice lacking Slit/Robo signaling, which is known to regulate the fasciculation of phrenic axons (Jaworski and Tessier-Lavigne, 2012). In agreement with prior reports, we observed defective nerve defasciculation in Robo1 -/-;Robo2 -/double mutants ( Figure 5A). Notably, defasciculation of the left nerve was as high as that of the right nerve and assumed a similar pattern in the left and right diaphragm, rather than adopting the normal asymmetric pattern seen in wild-type littermates ( Figure 5A). Partial symmetrization was observed in double heterozygous mutants, indicating concentration-dependent sensitivity of phrenic nerve axons to Slit signals ( Figure 5A).
To determine whether differential levels of Slit/Robo signaling dictate the L/R pattern of phrenic nerve fasciculation, we examined their transcript levels, but found no evidence for lateralized expression of the transcripts for Robo1, the major regulator of diaphragm innervation (Jaworski and Tessier-Lavigne, 2012), or the ligands of Robo1: Slit1, Slit2 and Slit3 ( Figure 5-figure supplement 1A -B). By contrast, we identified L/R differences in Robo1 protein by immunoblotting of phrenic motor neuron-enriched cervical spinal cord tissue. Robo1 was detected in one long and two short forms ( Figure 5-figure supplement 1C), whereby the long Robo1 form migrating as a 250 kDa protein was enriched in the left samples and the short forms migrating as 120 kDa and 130 kDa proteins were enriched in the right samples (R/L ratio-1.22 ± 0.1, p=0.01587; Mann-Whitney, Figure 5B, Figure 5-figure supplement 1D). Even though 12 alternatively spliced isoforms have been predicted for mouse Robo1, the predicted changes in protein sequence are unlikely to account for the short forms we observed in our immunoblots, because they are predicted to change the molecular Left (146) Right (  weight by just 7.1 kDa. However, both human and drosophila Robo1 have been shown to be processed by metalloproteases (Seki et al., 2010;Coleman et al., 2010), and potential cleavage fragments have been reported in mouse brain tissues (Clark et al., 2002). These findings raise the possibility that differential post-translation processing of Robo proteins may be involved in creating L/R asymmetries in diaphragm innervations. Next, we investigated whether axon guidance effectors that were revealed by our transcriptomic analysis to exhibit asymmetric expression levels could also contribute to the L/R phrenic nerve patterns. Given that metalloproteases have emerged as important regulators of axonal behaviors during development and regeneration (Bai and Pfaff, 2011;Łukaszewicz-Zają c et al., 2014;Small and Crawford, 2016;Verslegers et al., 2013b), we concentrated on these effectors. Consistent with previous expression data (GSE41013) (Philippidou et al., 2012), our transcriptome analysis indicated that cervical motoneurons express several metalloproteases. Interestingly, among the 7 Mmps and 13 ADAMs expressed by cervical motoneurons, Mmp2 and ADAM17 were expressed at higher levels in the right motoneurons. We focused on MMP2 because it was shown to control axon development in mouse and motor axon fasciculation in drosophila (Miller et al., 2011;Gaublomme et al., 2014;Zuo et al., 1998;Miller et al., 2008).
The microarray analysis showed that Mmp2 transcripts were enriched in the Hb9-positive right motoneurons, which was confirmed using qRT-PCR and quantitative in situ hybridization (log2(R/L) Embryo 1 -0.22 ± 0.08; Embryo 2 -0.63 ± 0.11, RNAscope) ( Figure 5C; Figure 5-figure supplement 2A-D). Moreover, in situ zymography with DQ-Gelatin (Hill et al., 2012), which is effectively cleaved by MMP2 (Snoek-van Beurden and Von den Hoff, 2005), showed that gelatinase activity on the axon shaft and growth cones was 1.6 times higher in right than in left motoneuron cultures ( Figure 5D; Figure 5-figure supplement 2E). Remarkably, this difference was absent in motoneuron cultures prepared from Rfx3 -/embryos with phenotypic left isomerism ( Figure 5D). These results provide evidence that the differential L/R MMP activity is controlled by the Nodal pathway and further suggest that MMP2 contributes to the establishment of phrenic nerve asymmetry.
We therefore analyzed the diaphragm nerve patterns in Mmp2 -/mice (Itoh et al., 1997). At E14.5, Mmp2 -/embryos exhibited normal lung asymmetry and well-developed phrenic branches on both sides ( Figure 5E). Interestingly, we observed a partial symmetrization of the phrenic branches, with a right pattern that resembled the one observed on the left in control littermates in E14.5 Mmp2 -/embryos ( Figure 5E). Thus, higher right MMP2 activity could contribute to promote the right pattern of phrenic nerve defasciculation.

Discussion
Taken together, our work shows that the first asymmetry instruction in diaphragm patterning is provided by early Nodal signaling, which sets the L/R axis and visceral asymmetry of the embryo. Beyond this early mechanism, phrenic motoneurons have an intrinsic, genetically encoded L/R asymmetry that manifests itself in the differential activation of molecules that have key roles in axon guidance, including Robo1 and MMP2.
Future work should aim to address how and at which stage phrenic motoneurons are imprinted. For example, an early Nodal signal might be propagated from the lateral plate mesoderm (LPM) to the cervical spinal cord. In agreement with this idea, it has been suggested that Lefty expression in the prospective floor plate of the neural tube prevents Nodal diffusion to the left LPM (Shiratori and Hamada, 2006). Moreover, Lefty expression is confined to the left prospective floor plate and is reversed or expanded bilaterally in 'iv' or 'inv' mutants, which exhibit reverse visceral asymmetry (Meno et al., 1997). Given the key role of the floor plate in the patterning and   (Goulding et al., 1993;Placzek et al., 1991), left and right floor plate cells might also differently imprint left and right spinal cord. Alternatively, or additionally, endothelial cells invading the ventral spinal cord could convey early Nodal signaling from the LPM to the spinal cord. Indeed, these cells exhibit L/R asymmetries that can be preserved during their migration (Chi et al., 2003;Klessinger and Christ, 1996). The resulting L/R imprints could occur early on during neurogenesis or later on during motoneuron differentiation. The two hypothesizes might not be exclusive. Indeed, recent work in zebrafish habenula suggests that differences in both the timing of neurogenesis and exposure to lateralized signal during neuron differentiation act in parallel to set L/R asymmetries (Hüsken et al., 2014). Interestingly, early imprinting of progenitors in Caenorhabditis elegans induces an epigenetic mark for L/R identity that drives differential genetic programs during neuron differentiation (Cochella and Hobert, 2012;O'Meara et al., 2010).
Our work provides evidence to show that a L/R imprint confers specific axon behaviors to the left and right phrenic motoneurons. For example, we found that Slit/Robo signaling is required for the establishment of asymmetric nerve patterns, which suggests that left and right phrenic motoneurons have different Slit/Robo signaling levels. Interestingly, Slit/Robo signaling has previously been shown to control phrenic axon fasciculation (Jaworski and Tessier-Lavigne, 2012). Consistent with an intrinsic control of Slit/Robo signaling as the cause of this axonal asymmetry, we discovered L/R differences in Robo1 protein in motoneurons, which may arise through differential proteolysis and may help to modulate responsiveness to Slit signaling, even though Slit and Robo genes are expressed similarly in the left and right motoneuron pools. In support of the idea that differential proteolysis contributes to the emergence of different Robo1 forms in the left and right phrenic motoneuron pools, Robo processing has previously been reported in other contexts (Seki et al., 2010;Coleman et al., 2010). This Robo1 processing could have different outcomes on Slit/Robo signaling. In drosophila, cleavage of Robo by ADAM10 is required for recruitment of downstream signaling molecules and the axon guidance response (Coleman et al., 2010). Metalloproteases can also decrease the amount of available receptors and/or terminate adhesion and signaling (Bai and Pfaff, 2011;Hinkle et al., 2006;Romi et al., 2014;Hattori et al., 2000;Gatto et al., 2014).
Slit/Robo signaling can control many different aspects of axon development, such as axon growth, branching, guidance or fasciculation. As primary and secondary branches are formed by selective defasciculation and because Slit/Robo signaling controls phrenic axon fasciculation, our interpretation is that the different Slit/Robo signaling abilities of left and right phrenic axons result in different axonaxon fasciculation states, with right axons having greater defasciculation behavior than the left ones. Alternatively, the Slit/Robo pathway may differentially regulate axon and branch growth, or branch trajectories, as it does for other systems of neuronal projections Brose et al., 1999;Blockus and Chédotal, 2016). These ideas have to be taken cautiously. Differential Robo forms were assessed from spinal cord tissue essentially containing neuronal soma, and not peripheral phrenic axons. Furthermore, the tissue samples, although enriched in phrenic motoneurons by the procedure, contained additional neuronal sub-types. Further investigations are thus needed to assess with more specific tools Robo protein dynamics and distribution along phrenic axons and in the growth cones. This work will provide a better characterization of the functional outcome determined by the balance of short and long Robo forms in the establishment of phrenic nerve patterns.
Asymmetries in several genes implicated in axon guidance were observed in our transcriptome analysis. In particular, we found differences in the expression of regulators of guidance receptors activities, such as metalloproteases. Mmp2 expression level and gelatinase activity were found to be higher in right cervical motoneurons. Moreover, differential gelatinase activity between left and right motoneurons was lost in cultures from Rfx3 -/-mutants with symmetrical Nodal signaling, suggesting that early Nodal signaling impacts on gelatinase activity in motoneurons. Mmp2 genetic loss reduced the asymmetry of the diaphragm branch pattern, suggesting that asymmetric expression of Mmp2 in motoneurons contributes to set phrenic nerve patterns.
However, in contrast to embryos lacking Pitx2 and Rfx3, embryos lacking Mmp2 only exhibited a partial symmetrization of the phrenic nerve branches. Rfx3 and Pitx2C transcription factors act at the onset of the left-right imprinting, and their genetic loss is therefore expected to abolish the entire program of L/R nerve asymmetry. By contrast, the subsequent construction of individual neuronal circuits relies on the concerted action of many different signaling pathways, whereby loss of a single pathway is not expected to disrupt the entire asymmetry program. The partial defect may be due to the presence of other effectors of the Nodal pathway that contribute to L/R nerve asymmetries independently of MMP processing, to the co-expression of several MMPs acting with partial redundancies with each other (Prudova et al., 2010;Kukreja et al., 2015) and to the fact that MMPs have many different substrates with potentially opposite effects on the same biological process. For example, proteomic studies have identified more than 40 secreted and transmembrane substrates for MMP2 (Dean and Overall, 2007), of which we found 32 to be expressed in cervical motoneurons including Adam17, which is enriched in right motoneurons ( Figure 5-figure supplement 2F, Figure 4-source data 2).
The MMP substrates that are responsible for asymmetric phrenic nerve patterning remain to be determined, but Slit/Robo signaling appears to be an obvious candidate. First, cleavage of human Robo1 has been suggested to be MMP-dependent, although in drosophila, Robo1 is cleaved by Adam10/Kuzbanian (Coleman et al., 2010;Seki et al., 2010). Second, short forms of Robo1, lprobably generated by proteolysis, are enriched in right motoneurons, in which MMP activity is the highest. In support, incubation of cervical spinal cord tissue with active MMP2 significantly increased the short Robo1 forms (fold change: 1.60 ± 0.23, p=0.00285, Mann-Whitney, four independent western blots, Supplementary file 1). Nevertheless, the L/R ratio of Robo protein forms in cervical motoneuron tissue collected from Mmp2 null embryos, although showing a tendency towards reduction, was not statistically different from the wild-type ratio (WT: 1.22 ± 0.10, N = 5; Mmp2 -/-: 1.14 ± 0.01, N = 3; p=0.78, Mann-Whitney; Supplementary file 2). This might be due to an insufficient number of tested embryos. Alternatively, because short Robo1 forms were still detected, this L/R ratio might rather reflect the activity of other proteases, either compensating for MMP2 loss or also contributing to Robo processing.
An additional MMP candidate is NCAM, which is highly expressed by developing phrenic axons, controls axon-axon fasciculation, and is cleaved by MMPs (Dean and Overall, 2007;Hinkle et al., 2006). In the light of MMP redundancy and the possible involvement of other proteases in the processing of axon guidance receptors and their ligands, the in vivo assessment of these hypotheses will be challenging.
Finally, the genetic program for L/R identity in spinal cord motoneurons that we have described here may provide important insights into motoneuron development and diseases. For example, the L/R imprinting of spinal motoneuron might also explain why right-sided fetal forelimb movements are far more frequent than left-sided movements at developmental stages when motoneurons have not yet received any input from higher brain centers (Hepper et al., 1998). In addition, our description of early events controlling diaphragm formation may have broad implications for our understanding of several human conditions. Examples include congenital hernias, which generally affect the left hemi-diaphragm and can cause perinatal lethality (Pober, 2008), and some types of congenital myopathies that impair diaphragm function only on one side (Grogan et al., 2005). Our data thus provide a novel basis for investigations of molecular diversity in spinal cord neurons and for functional studies of diaphragm physiology and pathology.

Genotyping of mouse lines
This work was conducted in accordance with the ethical rules of the European community and French ethical guidelines. Genotyping of transgenic mouse lines was performed as described in Liu et al. (2001) for Pitx2

Diaphragm immunolabeling
Diaphragms were dissected from embryos fixed overnight in 4% paraformaldehyde. After permeabilization and blocking in PBS with 5% BSA with 0.5% Triton X-100, diaphragms were incubated overnight at room temperature with the primary antibody, Neurofilament 160 kDa (1/100, RMO-270, Invitrogen, France; RRID:AB_2315286). Diaphragms were then incubated with the secondary antibody, a-mouse Alexa-555 (1/400, Invitrogen, France) with or without Alexa488-coupled a-BTX (1/50, Molecular probes, ThermoFischer Scientific, France; RRID:AB_2313931), for 4 hr at room temperature in blocking solution. The procedure was performed entirely on freely floating diaphragms. Diaphragm imaging was then performed under an inverted microscope and a montage was constructed using the metamorph software (Molecular device, Sunnyvale, CA).

Immunofluorescent labeling and in situ hybridization
Cryosections (20 mm) were obtained from embryos fixed overnight in 4% paraformaldehyde, embedded in 7.5% gelatin with 15% sucrose. For immunolabeling, embryonic sections or cultured neurons were incubated overnight at 4˚C with Oct6 antibody (1/50; Santa Cruz, Germany) and then for 2 hr at room temperature with anti-goat secondary antibody, Alexa-488 (1/400; Invitrogen, France). Nuclei were stained with bisbenzimide (Promega, Madison, WI). In situ hybridization was performed as described previously (Moret et al., 2007). The probes were synthetized from the Mmp2 IMAGEclone plasmid (n6813184). Mmp2 in situ hybridization and Pou3f1 (Oct6) immunolabeling were performed on adjacent sections because the antibody could no longer recognize the Oct6 epitope after in situ hybridization.

Images processing and quantifications
Serial Pou3f1/Oct6-labeled sections were imaged using a confocal microscope. Series of images were converted into a single stack using the ImageJ plugin Stack Builder. Images were aligned manually using morphological structures and labeled nuclei were extracted. A three-dimensional reconstruction of the Pou3f1 (Oct6) labeling from the cervical to the brachial part of the embryos was then generated in ImageJ (3D Project command). All quantifications were done using ImageJ. For quantification of defasciculation distance, we first traced the tangential straight line of the endplate (Figure 1-figure supplement 1B). We then traced a perpendicular line to the tangent that goes through the nerve entry point. Finally, we measured the distance from the entry point to the intersection of both lines. For branch number quantification, we traced a parallel to the tangential straight line of the endplate. The line was placed at a distance of one quarter of the defasciculation distance. We then counted the number of secondary branches that crossed the line. The endplate thickness was evaluated from the a-Btx staining. The a-Btx-positive region was outlined and divided into 30 rectangles. The average width of the rectangles was calculated. Width evaluation of endplate from the plot profile of a-Btx staining gave similar values.

Western blot
Cervical ventral spinal cords were dissected from E11.5 HB9::GFP embryos in cold HBSS with 6% glucose (as shown in Figure 4-figure supplement 1D) and directly frozen in dry-ice cooled eppendorf tubes. Typically, left and right dissected tissues from 6-8 embryos were pooled and lysed in RIPA buffer with protease inhibitors for 30 min at 4˚C. Western blots were performed using primary antibody (Anti-Morf4l1 (1:1000, Abcam, France -ab183663), anti-Xrn2 (1:1000, Abcam, France -ab72181, RRID:AB_2241927), anti-Robo1 (1:500 [Seki et al., 2010]) and secondary antibody (Antigoat or -mouse HRP [A5420 and A4416, Sigma-Aldrich, France] at 1/5000). Image quantification was done with Image Lab4.0 software (Bio-Rad, France). Left and right data were normalized to the tubulin level for Morf4l1 and Xrn2 and to Robo1 full-length or tubulin for Robo1 short forms. To allow comparison between replicates left and right values were then normalized to have the same left plus right sum for all western blots.
Motoneuron explant culture E12.5 GFP-positive mouse embryos (4-6 per experiment) from HB9::GFP transgenic mice were selected and dissected using the fluorescence GFP-positive pool. Ventral cervical spinal cords were isolated (left and right parts separated) and cut into explants. Explants were cultured as described in Moret et al. (2007). Immunohistochemistry was performed using Anti-Tuj1 (1:100, Millipore, France -MAB1637, RRID:AB_2210524) and anti-GFP (1:100, Invitrogen, France -A11122, RRID:AB_ 221569). Axon outgrowth was calculated using the ImageJ plugin NeuriteJ (Torres-Espín et al., 2014), which creates regions of interest (ROI) corresponding to radial concentric rings separated by 25 pixels. NeuriteJ extracted the signal from GFP-positive axons and measured the labeled surfaces between two ROIs. To quantify the total area occupied by GFP-positive axon, we summed the surface of all ROIs. To calculate the proximo-distal index, the width of the labeled axons was calculated in the second ROI (proximal ring) and in the ROI at 30% of the maximal distance of growth (distal ring) (see Figure 4-figure supplement 1F). The index was calculated by dividing the width of the proximal fascicles by the width of the distal fascicles.

Microarray analysis and quantitative real-time PCR
The GFP+ motor pool was laser-captured from E11.5 GFP + mouse embryos from HB9::GFP transgenic mice frozen in À45˚C isopentane. Captured tissues were lysed in the lysis buffer provided with the RNA purification kit (RNAeasy microkit, Qiagen, France ). RNA quality was assessed on an Agilent 2100 bioanalyser (Agilent Technologies, USA). L/R matching samples that had a RNA integrity number (RIN) above 9 were amplified (ExpressArt PICO mRNA amplification kit, Amp-tec-Exilone, France) and reverse transcribed (BioArray HighYield RNA Transcript Labeling, ENZO, France). cDNA quality was assessed on an Agilent 2100 bioanalyser before fragmentation and hybridization on an Affymetrix microarray (GeneChip Mouse 430 2.0, Affymetrix, ThermoFischer scientific, France). Expression normalizations and present or absent calls were performed in Affymetrix Expression Console Software. Fold change and filtering were performed in Excel. Transcripts were considered as being expressed if scored as present in at least one sample of each embryo. Transcripts were classified as differentially expressed if the fold change (FC) between left and right samples had the same trend for all embryos (same sign for log2 ratio) and was over 1.5 (FC > 0.58 or FC < À0.58 in log2) on average and for at least two of the embryos. Transcripts with very low expression (maximal normalized expression <200) were removed. Raw data are available on GEO under the accession number GSE84778.
Real-time PCR was performed using MIQE pre-validated Mmp2 (qMmuCID0021124) and Mnx1 (qMmuCED0040199) primers (BioRad, france). Data were normalized to GAPDH expression values (primers Fw: AGAACATCATCCCTGCATCC; Rv: ACACATTGGGGCTAGGAACA). Real time PCRs were performed in duplicate on amplified RNA prepared as described for the microarray. Lasercapture microdissection, RNA preparation, microarray and qPCR were performed at the ProfileXpert core facility (France).

RNAscope in situ hybridization
RNAscope in situ hybridization (Advanced Cell Diagnostic, Ozyme, France) was performed on 14-20-mm cryosections according to the manufacturer's recommendations for fresh frozen samples, using Mmp2 C3 and C1 proprietary probes (references 315937 and 315931-C3, ACD, Ozyme, France). Both probes gave the same pattern, which mirror the distribution observed using the conventional in situ procedure. UBC and DapB probes were used as positive and negative controls, respectively (references 310777 and 312037, ACD, Ozyme, France). All incubations were performed in the HyBez hybridation system (ACD, Ozyme, France). Sections were fixed in 4% paraformaldehyde for 15 min before dehydratation and incubated in pretreat buffer 4 (Advance Cell Diagnostic, Ozyme, France) for 15 min at room temperature. DAPI staining was performed at the end of the procedure. The left and right side of the cervical spinal cord were imaged at 20x on a FV1000 confocal microscope (Olympus, France) using the same acquisition parameters. Labeled surfaces were quantified in ImageJ in ROI drawn from the DAPI staining. The threshold calculated on the sum of the Z-stack image of one side was applied to the other side. Surface ratios were calculated after normalization to the selection area.

Statistical analyses
Control and mutant embryos were from the same litters. All analyzable samples (diaphragm, western blots, cells or explants) were included, no outliers were removed. Left and right samples were from the same embryos. Analyses of the diaphragm innervation and Mmp2 quantitative in situ were performed blind. No blinding was done on other data collections or analyses. Sample sizes, statistical significance and tests are stated in each figure and figure legend. All statistical analyses were done using Biostat-TGV (CNRS). Mann-Whitney (method: Wilcoxon rank sum) or Wilcoxon signed rank were used for small-sized samples or when distributions were not normal. Wilcoxon signed rank was used when paired analysis was needed (left versus right from the same embryo). DevWeCAN of the Université de Lyon, within the program 'Investissements d'Avenir' (ANR-11-IDEX-0007) operated by the French National Research Agency (ANR). CC was funded by a doctoral fellowship from the Fondation pour la Recherche Mé dicale (FRM FDT20130928169) and a postdoctoral fellowship from the International Brain Research Organisation (IBRO). DMM was supported by R01 DC009410. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Additional information
Author contributions CC, performed the experiments, analyzed the results and contributed to writing the manuscript; SD, YC, IS, MB, KK, performed the experiments and analyzed the results; LMor, BD, JMS, provided embryos and provided information for protocols; LDG, LMoo, provided embryos and information for protocols; MS, provided antibodies and information for protocols; CR, provided embryos and information for protocols and wrote the manuscript; JFM, provided embryos; DMM, provided embryos, provided information for protocols and contributed to writing of the manuscript; JF, performed the experiments, analyzed the results, designed the experimental plan and supervised the study and wrote the manuscript; VC, designed the experimental plan, supervised the study and wrote the manuscript