Genetic Dissection of the Function of Hindbrain Axonal Commissures

The Robo3 receptor controls midline crossing by axons. Deleting Robo3 in specific commissural neurons with a conditional knockout reveals their contribution to sensory and motor integration, and models human neurological conditions.


Introduction
At all levels of vertebrate nervous systems, axons cross the midline to form commissural projections [1,2]. In the visual system and the corpus callosum in the neocortex, the physiological importance of brain commissures is well established in large part due to the ''split-brain'' studies of Roger Sperry [3] and others. However, the function of commissural projections in the hindbrain and spinal cord has been more difficult to assess. Surgically ablating specific commissures in the hindbrain has proved to be difficult [4]. In addition, so far, no mutant has been created in which these connections are disrupted in a region-specific manner. Therefore, the function of specific commissures in a variety of hindbrain systems remains uncertain.
By contrast, considerable progress has been made over the past two decades in understanding how commissures form during embryonic development. Genetic screens in Drosophila, Caenorhabditis elegans, and mice, as well as biochemical approaches in rodents, have shown that the molecular mechanisms regulating commissure development are highly conserved in evolution [5] and involve axon guidance molecules, including Netrins, Slits, Semaphorins, Ephrins, Morphogens, and IgCAMs [6]. Their functions are tightly regulated by ECM molecules, cytosolic second messengers, receptor dimerization, and protein degradation in addition to transcriptional and posttranslational modifications [6,7].
To broaden our understanding of the function of hindbrain commissures, we undertook to genetically rewire the hindbrain by forcing select commissural tracts to remain ipsilateral, and then examined the cellular and behavioral consequence of these manipulations. We focused on Robo3, a vertebrate roundabout receptor involved in the formation of hindbrain and spinal cord commissures [8,9]. Robo3, also called Rig-1 (retinoblastoma inhibiting gene 1), was isolated as the product of a gene upregulated in retinoblastoma-deficient mice [10]. In humans, mutations in ROBO3 cause a rare syndrome called horizontal gaze palsy with progressive scoliosis (HGPPS) [11,12]. HGPPS patients have both an uncrossed pyramidal tract and dorsal column-medial lemniscus as well as severe scoliosis. Despite defects in conjugate horizontal eye movements [13], they are reasonably coordinated and do not present other obvious neurological deficits. The cause of horizontal gaze palsy, one of two signature traits of HGPPS syndrome, is still uncertain [11]. In addition, a subset of HGPPS patients show an unexplained asymmetry of brainstem auditory evoked potentials and activation of their auditory cortex [13,14].
Studies in mice show that in hindbrain and spinal cord, Robo3 is required for commissural axons to cross the midline and that commissural neurons, deficient in Robo3, extend their axons ipsilaterally [8,9]. The genetic manipulation of Robo3 could thus provide effective means to characterize roles for hindbrain commissures and lead to understanding the etiology of the HGPPS syndrome. Because existing Robo3 knockout mice die perinatally, precluding behavioral analysis [9], we have undertaken the construction of a conditional allele of Robo3 and show that viable animals lacking specific commissures can be obtained.

A Reduced Internuclear Commissure in Robo3-Deficient Mice Induces Abnormal Horizontal Compensatory Eye Movements
In vertebrates, lateral eye movements are controlled by two pairs of cranial motor nuclei, the abducens (VI) and the oculomotor (III) nuclei, which project ipsilaterally to the lateral and medial rectus muscles, respectively [15] ( Figure S1). Conjugate lateral eye movement requires bilateral coordination of those muscles, which relies in part on abducens interneurons that project to the contralateral oculomotor nucleus, forming the abducens-oculomotor (III) internuclear commissural projection ( Figure S1). Furthermore, in functional magnetic resonance imaging studies of HGPPS patients, nucleus VI is hypoplastic, whereas cranial nerves III and VI are present bilaterally, suggesting that this palsy may result from defects in commissural systems [11]. We, therefore, analyzed the oculomotor system in the existing Robo3-null mice (Robo3 2/2 ) [9] as well as in a Robo3-conditional mouse that lacked Robo3 in rhombomere 5 (r5), the rhombomere from which abducens neurons are derived. The latter mice (Krox20::cre;Robo3 lox/lox ) were generated by creating Robo3 lox/lox mice ( Figure S2) and crossing them with Krox20::cre knock-in mice that express Cre recombinase in r3 and r5 [16]. Upon Cre-mediated excision of exons 12-14, a truncated Robo3 protein is produced, which lacks transmembrane and cytoplasmic domains, thereby disabling it as a receptor. R3-r5-specific Robo3 inactivation was confirmed by performing whole-mount labeling with a Robo3 exon-specific probe and Robo3 antibody on embryonic day (E)12 embryos ( Figure 1). As expected, severe reduction of commissural projections in r3 and r5 was observed in whole-mount ( Figure 1; n = 2/2) and coronal sections of Krox20::cre;Robo3 lox/lox E11-E12 embryos labeled with anti-neurofilament and anti-Robo3 antibodies (Figure 2 n = 3/3). By contrast, many commissural axons were still observed in adjacent rhombomeres ( Figure 2). A few Robo3-positive axons still crossed the midline at the r3 and r5 level in Krox20::cre;Robo3 lox/lox embryos. These fibers most likely originate in commissural neurons located outside of r3 and r5 [17] (see Discussion). In Krox20::cre;Robo3 lox/lox Figure 1. Rhombomere-specific deletion of Robo3. Whole-mount control (A, C, and E) or Krox20::cre;Robo3 lox/lox (B, D, and F) E12 embryos hybridized with a Robo3 riboprobe covering exons 12-14 (A and B) or immunostained with anti-Robo3 (C and D) or anti-neurofilament (E and F) antibodies. (B) In Krox20::cre;Robo3 lox/lox embryos, Robo3 exon12-14 transcripts are not expressed in rhombomeres 3 (r3) and 5 (r5). (C and D) Likewise, there is a severe reduction of Robo3 immunoreactive commissural axons in r3 and r5. (E and F) Anti-neurofilament immunostaining confirms the strong reduction of commissures in r3 and r5. The arrowheads in (F) indicate axons that abnormally follow the midline. Scale bars represent 100 mm. doi:10.1371/journal.pbio.1000325.g001

Author Summary
Coordination of the left and right sides of the body requires the action of neurons whose axons cross the nervous system midline. The precise contributions of ''commissural'' neurons to sensory and motor functions remain poorly understood. To probe these crossing circuits, we took advantage of the recent finding that the Robo3 axon guidance receptor is required for midline crossing by axons at most axial levels. A Robo3 conditional knockout mouse line was generated, allowing Robo3 to be deleted in selective neuronal populations. This led to disruption of specific commissures in the sensory, motor, and sensorimotor systems, and resulted in severe but specific functional deficits. Surprisingly, although rerouted axons do not cross the midline, they still project to their appropriate neuronal targets, suggesting that midline crossing is not required to complete the axonal guidance program of those neurons. Moreover, some of the mouse lines represent good models for human syndromes, including horizontal gaze palsy with progressive scoliosis (HGPPS), which is characterized by deficits in coordinated eye movements. This study links defects in commissural axon guidance with specific and dramatic behavioral phenotypes.
Commissural axons that crossed the midline at the level of the abducens nucleus express Robo3 in E13 Robo3 +/2 embryos (Figure 4A-4C; n = 3/3). To determine the need for Robo3, we Rhombomere-specific deletion of commissures. Coronal sections at the level of r3, r5, and r6 of E11 control (A-C9) or Krox20::cre;Robo3 lox/lox (D-F9) embryos immunostained with antibodies against Robo3, neurofilament, and Hb9. The midline is indicated by an arrow, and the Hb9-positive abducens motor neurons by an asterisk. In r3 and r5, there is an almost complete absence of Robo3 and neurofilament-positive commissural axons in Krox20::cre;Robo3 lox/lox (D-E9) compared to controls (A-B9). The density of neurofilament-positive axons is also strongly reduced. By contrast, there is no obvious reduction of commissural axons or Robo3 expression in r6 (C, C9, F, and F9). Note that at E11, abducens shape and position are not altered in mutants compared to controls (see Figure 3 for later stages). VII, migrating facial neurons and facial nerve. Scale bars represent 60 mm. doi:10.1371/journal.pbio.1000325.g002 studied the abducens-oculomotor (III) internuclear commissural projection by injecting 1,19 dioctadecyl-3,3,39,39 tetramethylindocarbocyanine (DiI) tracer unilaterally at the level of the III nucleus. In control newborns (n = 6/6), this injection resulted in retrograde labeling of many neurons at the level of the contralateral abducens nucleus and of their axons crossing the floor plate ( Figure 4 and Figure S5). On the other hand, in Robo3 2/2 (n = 3/3; Figure S5) and Krox20::cre;Robo3 lox/lox (n = 3/3; Figure 4E) newborn mice, contralateral labeling was virtually absent, whereas some neurons were retrogradely labeled in nucleus VI on the ipsilateral side. In addition, in Robo3 2/2 embryos (n = 3/3), no GFP-positive axons were found to cross the midline at the level of the abducens nuclei at E14 (unpublished data), and in E15 Krox20::cre;Robo3 lox/lox embryos (n = 4/4), there was a severe reduction of neurofilamentpositive axons crossing the midline at the level of the abducens nucleus ( Figure 4F-4I). Together, these results indicate that deletion of Robo3 either globally or specifically in r5 neurons results in loss of the abducens-oculomotor (III) internuclear commissural projection without markedly affecting the pathfinding of abducens motor axons.
To analyze the functional consequence of the lack of the VI-III internuclear commissure we studied compensatory eye movements in adult Krox20::cre;Robo3 lox/lox mice. The animals were provided with a pedestal and subjected to optokinetic and/or vestibular stimulation in both the horizontal and vertical plane (for Materials and Methods, see [19], and Text S1 and Table S1 for statistics). All three reflexes in the horizontal plane, including the horizontal optokinetic reflex (OKR), the horizontal vestibulo-ocular reflex (VOR) as well as the horizontal visual VOR (VVOR), showed striking changes in the mutants (n = 6) compared with control (n = 4) littermates ( Figure 5A-5D). Conversely, all eye movement reflexes in the vertical plane were unaffected ( Figure 5E-5G). During OKR, the impairments in the Krox20::cre;Robo3 lox/lox mice appeared to occur predominantly in the low-frequency range, whereas during VOR, they were most prominent at higher frequencies. These frequency-specific deficits reflect the potential role of internuclear commissural connections in high-amplitude eye movements, because OKR and VOR evoke high gain values at the lower and higher frequencies, respectively [20]. This view is further supported by the fact that the gain of eye movements in the Krox20::cre;Robo3 lox/lox mice decreased in a virtually linear manner with respect to stimulus amplitude at higher amplitude values (e.g., for OKR, see Figure 5D; p,0.001 versus control mice curve; ANOVA for repeated measurements). As a consequence, the amplitude of the horizontal VVOR, which is high over the entire frequency range because of its combined OKR and VOR nature, was significantly affected in mutants at both the lower and higher frequency ranges. The finding that all horizontal reflexes, but no vertical reflexes, were disrupted suggests that the abducensoculomotor internuclear commissure was selectively affected. Ipsilateral aVCN-MNTB Projections and Impaired Auditory Brainstem Responses in Krox20;Robo3lox Mice Since Krox20 is also expressed in r3, we next tried to determine which commissures could be affected at this level. Most auditory brainstem nuclei send axons across the midline, allowing binaural sound localization [21]. The projection from globular bushy cells in the anteroventral cochlear nucleus (aVCN), the first central auditory relay in the central nervous system, to the principal cells of the median nucleus of the trapezoid body (MNTB) is one of the most studied auditory commissures ( Figure S6). We found that, at E13-E15, when aVCN axons first cross the floor plate [22], aVCN neurons expressed high-level Robo3 mRNA ( Figure 6A and 6B) and their axons expressed Robo3 protein ( Figure 6C). Genetic fate-mapping has shown that aVCN neurons are generated in the rhombic lip at the r2-r3 level [23]. We therefore studied aVCN-MNTB projections in Krox20::cre;Robo3 lox/lox mice. As previously described [24], DiI injections in the aVCN of P12 control mice labeled aVCN axons projecting to the contralateral MNTB (n = 4/4; Figure 6D and 6E), where they form characteristic giant synapses called ''calices of Held.'' By contrast, in Krox20::cre;Robo3 lox/lox mutants, this projection was exclusively ipsilateral, although typical calices of Held still formed on MNTB neurons (n = 4/4; Figure 6F and 6G). The numbers of labeled calices were not significantly different between controls and mutants (468633 and 480635 calices per side per animal, respectively, p = 0.7; n = 4 animals for each genotype; Student t-test with Mann-Whitney hypothesis). The aVCN ipsilateral projection to the superior olive appeared normal, and the olivocochlear bundle still crossed the midline (unpublished data). These results also incidentally show that globular bushy cells originate Figure 5. Impaired horizontal compensatory eye movements in Krox20::cre;Robo3 lox/lox mice. (A) During optokinetic stimulation horizontal gains are reduced most prominently at the lower frequencies in in Krox20::cre;Robo3 lox/lox mice (p = 0.043 ANOVA; n = 6 versus n = 4 for controls; see Table S1). (B) At higher frequencies, the VOR is severely impaired (p = 0.001 versus control mice curve; ANOVA for repeated measurements; Table S1), confirming the importance of the commissural connections in large-amplitude eye movements. (C) When horizontal visual and vestibular inputs are combined in the VVOR (visual vestibulo-ocular reflex), it results in lower gains over the entire range of frequencies tested (p = 0.004 versus control mice curve; ANOVA for repeated measurements). (D) OKR deficits are strongly correlated to the amplitude of stimulation. (E-G) In marked contrast, in the vertical plane, no significant differences were observed in OKR (E), VOR (F), or VVOR (G), supporting the concept that primarily horizontal eye movements require the presence of commissural connections (see Table S1 for all statistics). Error bars indicate standard error of the mean. Results were obtained from four control and six Krox20::cre;Robo3 lox/lox mice. doi:10.1371/journal.pbio.1000325.g005 exclusively from r3 and that the Robo3lox mice can be used for genetic fate-mapping of hindbrain commissural neurons.
We next examined the consequence of the absence of a crossed aVCN/MNTB projection on auditory brainstem responses (ABR; see Materials and Methods) [25]. The detection thresholds of ABR did not significantly differ in Krox20::cre;Robo3 lox/lox versus control mice and remained in the normal range from 5-40 kHz ( Figure  S6). In all animals, all waves were detected once the stimulus level was set above ABR threshold ( Figure 6H). This indicates that neither auditory sensitivity nor gross neural conduction were affected. In control mice, the ABRs exhibited the four classical waves (I-IV) ipsilaterally ( Figure 6H; for wave interpretation, see Materials and Methods and Text S1). Only two waves (IIIc and IVc) were recorded contralaterally, and their detection occurred with enough delay to be ascribed to contralateral excitation. The shape of mutant ABR waves was markedly different from controls. First, only three waves were present ipsilaterally ( Figure 6H). At 60 dB sound pressure level (SPL), the latencies and amplitudes of waves I and II exhibited no significant difference relative to controls, which was not surprising, because the mutation targeted neural pathways beyond the cochlear nucleus. In contrast, in mutants, the next visible wave (III) had a longer latency ( Figure 6H) than in controls. Moreover, in mutants, the IIIrd wave was not followed by any identifiable waves. Secondly, the contralateral wave IVc still occurred, but with a 0.42-ms average delay compared to the normal IVc. Its magnitude tended to be reduced although not significantly (n = 12 wild type versus n = 8 mutants; p = 0.06). In both mutants and controls, a similar dependency of wave latency on stimulus level was observed (unpublished data). These results show that in Krox20::cre;Robo3 lox/lox mutants, the ipsilateral projection of globular bushy cells to MNTB neurons results in abnormal ABRs both ipsilaterally and contralaterally.

An Impaired Interolivary Commissure Affects Locomotion
Inferior olivary (IO) neurons are the source of climbing fiber input on cerebellar Purkinje cells. During development, IO neurons migrate tangentially from the rhombic lip towards the floor plate. Upon reaching the floor plate, IO neuron cell bodies stop and only their axons cross it. We showed previously that IO axons are unable to cross the midline in Robo3 2/2 embryos and, as a result, project to the ipsilateral cerebellum [8]. However, the functional consequence of this ipsilateral rerouting of IO axons is unknown. To attempt to generate mice lacking the interolivary commissure, we crossed Robo3 lox/lox conditional knockout with mice in which Cre recombinase was knocked into the Ptf1a (Ptf1-p48) locus (Ptf1a::cre) [26], which encodes a bHLH transcription factor expressed by a majority of IO neuron progenitors. The resulting Ptf1a::cre;Robo3 lox/lox mice were viable, until at least 7 mo, but from postnatal day (P)10 onwards, they exhibited profound locomotor deficits ( Figure 6A and 6B; see also Video S1) including an ataxic gait (35/35 cases) that persisted into adulthood. When locomotion was tested on the Rotarod, Ptf1a::cre;Robo3 lox/lox mice showed a significant deficit in motor performance ( Figure 7A and 7B); the Ptf1a::cre;Robo3 lox/lox mice fell off the Rotarod at highly significant shorter latencies than controls (p,0.001, Mann-Whitney U-test, n = 7 mutants versus 9 controls). These motor performance deficits in the Ptf1a::cre;Robo3 lox/lox mice were so severe that it appeared as if their ataxia was worse than that of mice that have no cerebellar output at all. We therefore compared their performance to that of Lurcher mice, which are characterized by a degeneration of their Purkinje cells and therefore by a virtual absence of the output from their cerebellar cortex [27,28]. Although not significant (p = 0.2; Mann-Whitney U-test), the impaired locomotion in the Lurcher mice on the Rotarod was indeed milder than in the Ptf1a::cre;Robo3 lox/lox mice ( Figure 7A). On the more discriminative Erasmus Ladder [29], the difference in motor performance between Ptf1a::cre;Robo3 lox/lox mice and Lurcher mice was fully apparent ( Figure 7C and 7D); both the step time and the overall walking pattern (percentage of successful walking trials) were highly significantly more affected in Ptf1a:: cre;Robo3 lox/lox mice than in Lurcher mice (Ptf1a::cre;Robo3 lox/lox versus Lurcher mice for step time in all sessions: p,0.001, multiple comparisons; for walking pattern in all sessions: p,0.001; Mann-Whitney U-test; see also Table S2). These findings indicate that the level of cerebellar ataxia induced by rewiring the climbing fiber projection is more severe than that caused by a functional loss of the cerebellar cortex, suggesting a pathological dominant-negative effect in the Ptf1a::cre;Robo3 lox/lox mice. Despite this clear functional phenotype, the size and foliation of the cerebellum was unaffected in Ptf1a::cre;Robo3 lox/lox mice ( Figure 7E-7J; n = 3 mutants versus 3 controls). In addition, the distribution, morphology and immunostainings of their Purkinje cells, molecular layer interneurons, and granule cells ( Figure S7) as well as the arborization of their climbing fibers ( Figure 7G and 7J) were unaffected.
When DiI was injected unilaterally in the cerebellum of newborn controls, it retrogradely labeled IO neurons exclusively on the side opposite to the injection in control mice (4 out of 4 cases; Figure 7K). In contrast, following injections in Ptf1a::cre; Robo3 lox/lox mice (8/8 cases; Figure 7L) about 67% of the DiIlabeled IO neurons were situated on the side of the injection, whereas 33% of the olivary axons projected contralaterally ( Figure 7L). Immunolabeling of E13 Ptf1a::cre;Robo3 lox/lox embryos (n = 4) with antibodies to Brn3.2, a marker of a large subset of IO neurons, showed that, as previously described in Robo3 2/2 mice, the ipsilateral IO projection did not result from a migration of IO cell bodies across the midline ( Figure S8). The use of multiple IO neuron markers (Brn3.2, CaBP, BEN/SC1, CGRP) revealed that the lamellar organization of the IO was perturbed in Ptf1a::cre; Robo3 lox/lox mice (n = 4; Figure S8), whereas the mossy fiber inputs and their neuronal sources developed normally ( Figure S7).
To confirm that the ataxic behavior of Ptf1a::cre;Robo3 lox/lox mice can be primarily attributed to the ipsilateral rerouting of a majority of olivocerebellar axons, we crossed these mice to a Tau-lox-Stoplox-mGFP-IRES-nls-lacZ knock-in line [30] (here, called Tau mGFP ). Upon Cre recombination in neurons, the Stop cassette is excised, leading to the permanent expression of a myristoylated GFP in axons and of b-galactosidase (bgal) in nuclei. This strategy made it possible to identify the neurons that expressed Cre recombinase in the Robo3lox background. In the inferior olive of Ptf1a::cre; Robo3 lox/lox E16 embryos (n = 3), we found that 8762% of Brn3.2positive neurons also expressed bgal, whereas the remaining Brn3.2-positive neurons were completely negative for bgal ( Figure 8A and 8B). This suggests that Cre is not expressed by all IO neurons in the Ptf1a::cre line, and could explain the maintenance of a contralateral contingent of IO axons in Ptf1a::cre;Robo3 lox/lox animals (see Discussion). To find out whether Cre was expressed by spinal cord commissural neurons during midline crossing, we studied bgal and GFP expression in thoracic cross-sections of the spinal cord of E11-E13 Tau mGFP embryos at the level of the forelimbs (n = 3; Figure 8C-8L). We found that in the spinal cord of Ptf1a::cre;Robo3 lox/+ ;Tau mGFP E11-E13 embryos, GFP-expressing axons were not immunoreactive for Robo3 and that commissures were not GFP-positive, with the exception of a very small ventral subset of axons ( Figure 8). Interestingly, this small subset of GFP axons still crossed the midline in Ptf1a::cre; Robo3 lox/lox ;Tau mGFP embryos (Figure 8; n = 3). Moreover, Robo3immunoreactive commissural neurons did not express bgal (n = 3 animals; unpublished data). The use of additional markers such as TAG-1 and neurofilament confirmed that spinal cord commissures were similar in control and Ptf1a::cre;Robo3 lox/lox E13 embryos ( Figure S9). As shown previously [31], in Ptf1a::cre;Robo3 lox/+ ; Tau mGFP E12-E16 embryos, we could not detect bgal or GFP expression in the main sources of mossy fiber projection, the pontine neurons, the external cuneatus nucleus, or the lateral reticular nucleus (unpublished data).

Discussion
We show that a Robo3-conditional allele can be used to genetically disrupt distinct commissural projections and force axons to project ipsilaterally. In all three systems studied here, commissure rewiring induces severe and permanent dysfunction of specific neuronal networks. Unlike surgical ablation of a given commissure, this strategy ends up rewiring commissural neurons on the ipsilateral side, but does not interrupt the activity of the targeted circuit. We show that for the systems studied here, ipsilaterally rerouted axons still connect to their proper targets. Moreover, the fact that human HGPPS patients do not exhibit major motor or cognitive impairments also suggests that the ipsilateral axons are fully functional. However, it is difficult to link the observed behavioral phenotypes exclusively to the loss of contralateral connectivity as there is also a rerouting in the targeting of these neurons that potentially could cause novel and perturbed neuronal signals leading to a defect, which is not caused by the lack of commissural connections per se but rather by the new wiring.
Our results and others [8,9] suggest that Robo3 expression is required for most, if not all, commissural axons to cross the floor plate throughout spinal cord and hindbrain. The persistence of commissural axons that continue to cross the floor plate in domains expected to express cre recombinase in the Robo3 conditional lines presented here may be explained in a number of ways. In Krox20::cre;Robo3 lox/lox embryos, the few axons crossing at r3 and r5 level may originate from neurons located in adjacent rhombomeres as it is well known that many commissural axons extend rostrally and caudally before crossing [17,32], and that during development, mixing of neurons from adjacent rhombomeres does occur [33,34] (one example is provided by facial branchiomotor neurons that migrate from r4 to r6 [35]). Another reason for incomplete rerouting of commissural axons is that Cre recombinase may not be expressed by all r3/r5 neurons, as is the case for IO neurons in Ptf1a::cre;Robo3 lox/lox knockout. Finally, cre might be expressed too late, after these commissural axons have crossed, an hypothesis we favor to explain the persistence of the ventral subset of commissural axons in the spinal cord of Ptf1a::cre;Robo3 lox/lox embryos.
Although Robo3 is required for commissures to form in the hindbrain and spinal cord, this receptor is not sufficient for crossing as it is also expressed by many noncommissural neurons in the forebrain and hindbrain [36], such as the abducens motor neurons that project ipsilaterally. This opposite behavior of some Robo3-expressing axons could be attributed to a differential expression in commissural/noncommissural axons of Robo3 partners, possibly Robo1 and Robo2, or of a different Robo3 isoform. A recent study has shown that two Robo3 splice variants that differ in their cytoplasmic tail have distinct activities (procrossing and anti-crossing) in axons [37].
We found that deletion of Robo3 selectively in two hindbrain rhombomeres, including that comprising the abducens nucleus, yields mice that display selective horizontal eye movement defects reminiscent of those in the human HGPPS patients [12][13][14][15]. Likewise, zebrafish Robo3 mutants exhibit oculomotor defect [38]. Since there is, to the best of our knowledge, no oculomotor center originating from r3, our results in mice indicate that commissural projections from r5, most likely internuclear oculomotor connections, serve to facilitate large amplitude, compensatory eye movements in the horizontal plane. Our results also suggest that the absence of this commissure is one of the primary causes of the eye movement defects in HGPPS patients, who also suffer from a mutation in the Robo3 gene [11,12]. Accordingly, abducens motor axons project normally in Krox20::cre;Robo3 Iox/lox mice. Moreover, we found that horizontal eye movements are not perturbed in Islet1::cre;Robo3 lox/lox mice, which lack Robo3 in abducens motor neurons (unpublished data). These results do not exclude the possibility that oculomotor defects in HGPPS patients may also involve other supranuclear centers such as the pons and cerebellum, which could henceforth be studied in transgenic mice by deleting Robo3 in appropriate regions using the conditional allele.
We show that in mice, in the absence of Robo3, the aVCN-MNTB projection is entirely uncrossed and that this results in strong defects in ABR. The most striking feature of ABRs in Krox20::cre;Robo3 lox/lox mice is the presence on the ipsilateral side of a much-delayed wave III not followed by any later wave. This abnormal wave III could involve wrongly rewired neurons, and its long delay suggests that the corresponding pathway does not function in a strictly normal manner. Contralaterally, the persistence of waves identifiable as IIIc and IVc may be attributed to crossing pathways originating outside of r3-r5. Although there are species differences in the organization of auditory circuits [39,40], our results suggest that the trapezoid body could be uncrossed in HGPPS patients, at least partially. It will be of interest to determine whether sound localization is perturbed in HGPPS patients, some of whom also seem to have abnormal ABR [13,14].
Interestingly, HGPPS patients have a hypoplastic pons, an uncrossed corticospinal tract [12,13], and most likely uncrossed spinal cord and interolivary commissures. Yet these patients are still coordinated and not ataxic. This contrasts with the severe ataxic gait of Ptf1a::cre;Robo3 lox/lox mice. In this mouse line, spinal cord commissures and the corticospinal tract (unpublished data) do not seem to be affected, and among embryonic precerebellar neurons, cre recombinase is only expressed by IO neurons and not by mossy fiber projection neurons. Possibly, a selective reduction of the interolivary commissure results in a more dramatic locomotion impairment than a nonselective rewiring of multiple parts of the motor system in which, presumably, the different parts of the nervous system are ultimately still connected to the correct sides of the other relevant parts. Moreover, the selective reduction of the interolivary commissure in the Ptf1a::cre;Robo3 lox/lox mice also Figure 7. Uncrossed inferior olivary axons cause ataxic gait. (A) Ptf1a::cre;Robo3 lox/lox mice are severely ataxic, which is demonstrated by a very short latency to fall in the Rotarod test (n = 7 mutants versus n = 9 for controls). (B) Image of a 4-mo-old Ptf1a::cre;Robo3 lox/lox mouse displaying an ataxic gait. (C) The step time of Ptf1a::cre;Robo3 lox/lox mice (light blue; n = 6) on the Erasmus Ladder is longer than in controls (violet and red; n = 4; p,0.001; multiple comparisons) or in Lurcher mice (orange; p,0.001; multiple comparisons). (D) Ptf1a::cre;Robo3 lox/lox mice (light blue) also show significantly less successful trials per session than Lurcher mice (orange) (p,0.001 for all sessions; Mann-Whitney U-test, one-tailed). A trial was defined as successful if the mice were able to walk on the ladder without disruption (twisting, turning, walking backwards, etc.). (E to J) The morphology of the cerebellum in controls (E to G) and Ptf1a::cre;Robo3 lox/lox mice (H to J) is comparable, and in both cases, their climbing fibers (the terminal arbors of inferior olivary axons) innervate the cerebellar cortex. results in a stronger locomotion impairment than mice without any cerebellar output (i.e., the Lurcher mice). This phenotype of the Ptf1a::cre;Robo3 lox/lox mice on the Erasmus Ladder can be explained by the fact that the olivocerebellar system may operate in a pushpull fashion [41], because converting the ''pull'' into a ''push'' by reversing the modulation of climbing fiber activities should indeed result in a phenotype that is more severe than having no output of the cerebellar cortex at all. Thus, taken together, our results suggest that having only one component of a commissural circuit misrouted and uncrossed may be functionally worse than having all of its components affected or having no output at all.
Many studies have shown that upon entering and crossing the floor plate, commissural axons receive signals that induce major and long-term modification of their behavior and prevent them from re-entering the midline. This is associated with a loss of attraction for netrin-1 [42] involving DCC silencing and DCC/ Robo dimerization [43], and an activation of Slit-and Semaphorin-mediated repulsion [44]. Moreover, there is also evidence that the translation of some axon guidance receptors [45] or their splicing is modulated by floor plate crossing [37]. However, it was not known if all these molecular changes were required for crossed axons to recognize and follow their correct pathway on the contralateral side and, more importantly, to reach their final target cells. Our results suggest that even in the absence of midline crossing, both olivocerebellar axons and aVCN axons still contact their appropriate target neurons (Purkinje cells and MNTB, respectively), but on the ipsilateral rather than on the contralateral side. These findings raise the possibility that midline crossing in general may not have a major function in axon guidance after crossing, and that its major importance is only to allow, at a physiological level, the integration of inputs coming from both sides of the nervous system. Future studies will have to confirm this at other axial levels and in other phyla like Protostomia.

Mouse Lines
C57BL/6 mice (Janvier) were used for expression studies. Mice were anesthetized with isofluorane (Abbott). The day of vaginal plug is embryonic day 0 (E0), and the day of birth corresponds to postnatal day 0 (P0).
The Robo3 conditional knockout mouse line was established at the MCI/ICS (Mouse Clinical Institute-Institut Clinique de la Souris, Illkirch, France; http://www-mci.u-strasbg.fr). To generate the Robo3 conditional knockout, the targeting vector was constructed as follows. Three fragments of 5.1, 1.1, and 3.7 kb (respectively, the 59, floxed, and 39 arms) were amplified by PCR using 129S2/SvPas DNA as template and sequentially subcloned in an MCI proprietary vector. This MCI vector has a floxed neomycin resistance cassette. The linearized construct was electroporated in 129S2/SvPas mouse embryonic stem (ES) cells. After selection, targeted clones were identified by PCR using external primers and further confirmed by Southern blot with 59 and 39 external probes. Two positive ES clones were injected into C57BL/6J blastocysts, and derived male chimaeras gave germline transmission. The Robo3-null knockout line was described previously [9]. Briefly, it consists on a knock-in of the GFP gene into the first exon of Robo3, leading to the expression of the GFP by Robo3-expressing cells. However, the GFP signal is too weak to be seen without the use of an anti-GFP antibody. The Hb9::GFP mice have a transgene containing a 9-kb-long region of the Hb9 promoter that drives eGFP expression in all postmitotic somatic motor neurons [18]. The GFP signal is bright enough to be seen directly. The Ptf1a::cre and Krox20::cre knock-in lines were previously described [16,26]. Homozygous Krox20 cre/cre are embryonic lethal and were therefore maintained as heterozygous for Cre. The Tau mGFP knock-in line was obtained by replacing the coding sequence of the Tau gene by a lox-Stop-lox-mGFP-IRES-nls-lacZ cassette [30]. Unless otherwise mentioned, controls were Robo3 +/2 or Robo3 lox/lox animals or double heterozygotes that were always found to be undistinguishable from wild-type mice. All mice were genotyped by PCR (see Table S3 for primer sequences). All animal procedures were carried out in accordance with institutional guidelines (UPMC and INSERM).

Histology and Immunocytochemistry
Embryos and mice were processed as described previously [46]. Tissue sections and whole-mount embryos were hybridized with digoxygenin-labeled riboprobes as described previously [46].

Quantifications
The number of abducens motor neurons was quantified on 20mm coronal sections of adult hindbrain double labeled with anti-ChAT and anti-Hb9. Neurons were counted on alternating sections and counts multiplied by two to better reflect the total number of neurons in each abducens nucleus.
Di-labeled IO neurons were counted as follows: on every section containing the inferior olive, four confocal optic planes were acquired (two on the contralateral side and two on the ipsilateral side), and only the cells showing circle-shaped DiI labeling were counted as positive (see Figure S8). The percentage of ipsilateral and contralateral neurons represents the number of ipsilateral-or contralateral-labeled neurons divided by the total number of traced neurons on both sides.
The number of DiI-labeled calyces of Held were counted from stacked confocal planes (step size: 1.16 mm) encompassing 50 mm of 100-mm vibratome sections, on all the sections containing the MNTB. Only typical ''moon crescent''-shaped objects were counted as calyces (see Figure 6E and 6G). Quantification in the text are given as the number of calyces per nuclei (only one side) per animal. This number is an underestimate of the real number of calyces, because the confocal stack does not encompass the entire section.

DiI Tracing
The 4% PFA-fixed embryos or postnatal animals were injected with small crystals of DiI (Molecular Probes) using glass micropipettes. Injected brains were kept at 37uC for 1 wk (aVCN-MNTB and VI-III commissures) to 3 wk (olivocerebellar projection). Brains were cut in 100-mm sections with a vibratome (Leica) and counterstained with Hoechst.

Eye Movement Recordings
Mice were placed in a restrainer, fixed onto the center of a turntable, facing a cylindrical screen, and their OKR and VOR measured as previously described [19]. See Text S1 for a detailed description of this procedure.

Rotarod Training
Mice were placed on the cylinder of a Rotarod apparatus (model 7650, Ugo Basile Biological Research Apparatus) that rotated at four turns per minute, and the time the mice spent on top of the cylinder was recorded. After 300 s, the recording was ended. The animals were tested twice, and between the sessions, there was a 60-min break during which the mice rested in their cages. Mean values of the two trials were calculated for each animal. The Mann-Whitney test was used for statistical analyses (p,0.05 was considered significant).

Erasmus Ladder
The Erasmus ladder [29] is a fully automated system designed to screen motor performance and motor learning. It consists of a horizontal ladder that is composed of 2637 rungs (pressure sensors) and is situated between two sheltered boxes equipped with pressurized air outlets. In order to test motor performance Ptf1a::cre;Robo3 lox/lox mice (n = 6) and their littermates (n = 4) were placed in one of the shelters. After a randomized delay of 9-12 s, the light was turned on in the box, automatically followed after 3 s by an air puff from the pressurized air outlet, encouraging the mice to leave the shelter and walk across the ladder to the shelter on the other side, where the procedure was repeated. The paw placement and overall step time (the time needed to transfer the paw from one sensor to the other) was recorded for each trial in real time using the pressure sensors. One trial was defined as a crossing from one shelter to another. A trial was regarded successful if the mice walked with a consistent pattern, touching the pressure sensors with all paws, and with no disruptions such as rearing, twisting, or turning around. One session consisted of at least 20 trails.

Recordings of Auditory Brainstem-Evoked Responses
Mice were anesthetized with a mixture of ketamine and levomepromazine. ABRs were collected between subcutaneous needle electrodes inserted at the vertex and mastoids on both sides, with the ground electrode in the back (see Text S1). Sound stimuli were short broadband clicks and frequency-specific tone-bursts (two-period rise-fall, 20-period plateau, frequency 5, 10, 15, 20, 32, and 40 kHz) presented at a repetition rate of 17 per second, the sound level varying in 5-dB steps from 10 to 90 dB SPL (with 0 dB SPL = 2.10 25 N.m 22 ). The responses from the electrodes were amplified (6100,000), filtered (10-3,000 Hz), digitally converted and averaged (6500) by a computerized dataacquisition system. On the ipsilateral side, wave I reflects the activity of spiral ganglion cells. Later waves II, III, and IV are ascribed to the sequential activation of more and more central neural generators. On the contralateral side, a short latency wave was observed similar to ipsilateral wave I ( Figure 6H), which precluded its being due to the activation of neural pathways contralateral to the stimulus and was thus likely due to remote detection of responses from the ipsilateral spiral ganglion through electrically conductive neck tissues. The total absence of any deflection in the contralateral ABRs at the latency corresponding to ipsilateral wave II indicates that the artifact affecting this short latency contralateral wave did not come into play for later waves.

Statistical Analysis
Unless mentioned in the text, results are presented as means6standard deviation (sd). All differences of the means between two sample sets were assessed by two-tailed Student t-test with Welsh hypothesis (unequal variances) or a Mann-Whitney Utest, as appropriate, except for the eye movement results. There curves were first tested with an ANOVA for repeated measurements for significant differences, and only for the different curves a Student t-test was used to identify significantly different points. Statistics were carried out with R software (http://www.r-project. org/). Figure S1 Schematic representation of the oculomotor system involved in lateral eye movement. (A) In controls, lateral eye movements are controlled by two pairs of cranial motor nuclei, the abducens (VI, green) and the oculomotor (III, red), projecting ipsilaterally to the lateral rectus muscle (LR) and medial rectus muscle (MR) respectively. Conjugate eye movement involves a commissural connection (arrow) between VI nucleus interneurons (black) and nucleus III. (B) In Robo3-null embryos and Krox20::cre;Robo3 lox/lox mutants, the III and VI still project to the correct muscle, but the internuclear commissure is severely reduced. It is still unknown whether VI interneurons innervate the ipsilateral nucleus III.  (A and B), the internuclear commissure is strongly labeled (arrow), and DiI-labeled cells are observed at the level of the abducens nucleus (VI). (C and D) In Krox20::cre; Robo3 lox/lox mutants, the DiI-labeled internuclear commissure is severely reduced. Some DiI-labeled fibers are still found on the contralateral side, but no cells are traced in the vicinity of the abducens nucleus. In Robo3 2/2 knockout (E and F), the commissure is almost completely absent. No cells are traced on the contralateral side. Scale bars represent 100 mm. Found at: doi:10.1371/journal.pbio.1000325.s005 (1.87 MB TIF) Figure S6 ABR thresholds and binaural difference in control and Krox20::cre;Robo3 lox/lox mice. (A) The detection thresholds of ABR did not significantly differ in Krox20::cre; Robo3 lox/lox versus control mice and remained in the normal range from 5-40 kHz. (B) The binaural-difference wave complex resulting from the fact that the late ABR waves in response to a diotic stimulus are smaller than the sum of waves in response to stimuli in either the right or the left side (see Materials and Methods), was still present in mutants, with similar amplitudes and latencies. It suggests that at least part of the functional coupling of left and right signals remained present in Krox20::cre;Robo3 lox/lox mice, likely in relation to commissural neurons outside r3 and r5. (C and D) Schematic representation of the auditory pathway in control (C) and Krox20::cre;Robo3 lox/lox mice (D). (G) Auditory inputs (red arrowheads) from the hair cells are transmitted to neurons of the spiral ganglion that project ipsilaterally into the brainstem on globular bushy cells in the anterior part of the ventral cochlear nucleus (aVCN). In controls, these cells send large-diameter axons to the contralateral medial nucleus of the trapezoid body (MNTB), forming calyces of Held synapses (arrow). MNTB neurons then project to the lateral superior olive (SO). In Krox20::cre;Robo3 lox/lox mice, globular bushy cell axons only project to the ipsilateral MNTB but still form calyces. (C) is adapted from [4]     Video S1 Ataxic gait of Ptf1a::cre;Robo3 lox/lox mice. Two P15 littermates: the Ptf1a::cre;Robo3 lox/+ mouse walks normally, whereas the Ptf1a::cre;Robo3 lox/lox mouse has an ataxic gait. Found at: doi:10.1371/journal.pbio.1000325.s014 (1.36 MB MOV)