Transient Neuronal Populations Are Required to Guide Callosal Axons: A Role for Semaphorin 3C

Neurons, glia, and callosal axons operate as a “ménage à trois” in the development of the corpus callosum.


Introduction
The largest commissural tract in the human brain is the corpus callosum (CC), with over 200 million axons that act as a conduit for information between the two cerebral hemispheres. Callosally projecting neurons are Satb2-positive pyramidal projection neurons positioned, in rodents, in upper and lower cortical layers and that extend their axons through the CC [1][2][3][4]. More than 50 human syndromes result in agenesis of the CC (AgCC) and have an associated genetic etiology [5,6]. AgCC can result from defects during different steps of callosal development, including cell proliferation, migration, or a failure in axonal guidance within the CC [6].
Studies to date suggest that a specialized population of glia adjacent to the midline are central for the formation of the CC [7][8][9][10][11][12][13]. The primitive astroglial cells of the ''glial'' sling form a bridgelike structure at the midline between the two lateral ventricles and are required for the development of the CC [12,14,15]. Additional glial structures in the CC were described: radial glial cells in the glial wedge (GW) and astrocytes in the indusium griseum (IG) [6,7,10]. Recent observations in mice and humans showed that many neurons are also present within the ''glial'' sling [16,17]. Similarly, scattered neurons were observed within the cat CC during early postnatal life [18,19]. However, whether these populations have a specific function during development has not been investigated.
In this paper, we characterize the embryonic midline cellular organization at times prior to and during the formation of the CC. Through this effort, we discovered that in mouse embryos, uncharacterized GABAergic neurons intermix with glutamatergic neurons within the entire CC white matter. Here, we explore the possibility that these populations act in conjunction with midline glial cells to mediate the formation of the CC. We first investigated the identity of these populations and their spatial organization relative to ingrowing callosal axons. To determine whether these populations are functionally important, we examined the consequences of genetic ablation of a subpopulation of neurons as well as testing whether neuronal cells of the CC contribute to axonal guidance there. We show that the two neuronal populations that transiently populate the CC form a complex cellular network and that CC GABAergic interneurons are required for the proper organization of this network. Furthermore, ex vivo and in vitro experiments indicate that GABAergic and glutamatergic neurons of the CC are able to attract callosal axons.
Although the Semaphorin/Npn-1 signaling had been shown to be essential for CC development, the specific semaphorin ligand involved in this process, its source within the midline, as well as its precise function, remained to be determined [26,29]. In this study, we show that the transient population of CR-positive glutamatergic neurons expresses Sema3C and that either the ectopic transplantation of glutamatergic neurons or the ectopic expression of this ligand is sufficient to attract callosal axons. The use of Sema3C knockout (KO) mice confirms a novel and essential role of this factor in the pathfinding of callosal axons. Taken together, these results reveal that transient GABAergic and glutamatergic neurons are required for the formation of the CC. The present work, therefore, gives new insights into the mechanisms involved in axon guidance and implicates that transient neuronal populations work in conjunction with their glial partners in the guidance of callosal axons.

Glutamatergic and GABAergic Neurons Populate the Neonatal CC
Previous work has implicated the ''glial'' sling as central for the establishment of the CC [11,12]. Despite its name, the sling has been shown to contain at least one neuronal population whose function to date is undetermined [17]. As a starting point for investigating whether this neuronal population, and perhaps others, contribute to the formation of the CC, we undertook an immunological analysis of this region during embryonic and postnatal development. In addition to glial cells, upon immunostaining with antibodies against bIII-tubulin, MAP2A, and NeuN, we detected a large number of neurons, not only within the ''glial'' sling, but also within the entire white matter of the developing CC from embryonic day 12.5 (E12.5) until postnatal day 14 (P14). In particular, our molecular analysis revealed two distinct neuronal subpopulations. One that includes the population previously described as ''sling neurons'' was comprised of a population of differentiated glutamatergic neurons ( Figure 1). We found that this population expressed the homeobox transcription factor Emx1 and T-box transcription factor Tbr1, which are known to promote glutamatergic fate [30][31][32], the type 1 vesicular glutamate transporter (VGLUT1), and the calcium binding protein calretinin (CR) ( Figure 1A i -1A ii to 1B i -1B ii and unpublished data). Nearly all the CR-positive embryonic neurons of the CC intermediate zone (IZ) coexpressed the glutamatergic marker Tbr1 (91.761.3% at E16.5, n = 1,415) (Figure S1A i ). The other population was composed of GABAergic interneurons and was identified using either: i) a GAD67-GFP mouse line in which the green fluorescent protein (GFP) is reliably expressed within GABAergic neurons [33] ( Figure 1C i -1C ii to 1F i -1F ii ) or ii) a Mash1-GFP transgenic mouse line (Ascl1, Mammalian achaete-scute homolog) (GENSAT) that labels telencephalic GABAergic interneurons derived from Mash1-expressing progenitors of the ventral telencephalon [34,35] (Figure S1C i -S1C ii , S1D i -S1D ii and Figure S4E i ). A careful analysis of the colocalization between CR and GAD67-GFP in the neurons of the CC IZ indicates that these two neuronal population are exclusive at embryonic ages (0.760.2% at E16.5, n = 1,907; and 1.460.4%, n = 1,898 at E18.5) ( Figure 1C i -1C ii to 1E i -1E ii and Figure S1B i ). The two neuronal population identified by the expression of CR or GAD67-GFP included half of the CC IZ cells at embryonic ages (46% at E16.5, n = 5,561; 53% at E18.5, n = 6,495). At E16.5, the CR-positive glutamatergic neurons constitute 73.660.012% of these neurons, whereas the GAD67-GFP-positive GABAergic interneurons constitute 25.960.012% (n = 2,580; Figure S1B ii ). At E18.5, the proportion of both neuronal cell types is even (51.060.026% for CR + neurons, 48.360.025% for GAD67-GFP + neurons, n = 3,442; Figure S1B iii ). As expected, the CR-positive glutamatergic neurons and GABAergic interneurons did not express any of the glial markers nestin, GLAST, and GFAP (Figure S1C i -S1C ii to S1E i -S1E vi and unpublished data) previously found on early astroglial cells of the CC, IG, and GW [10].

Transient CC Neuronal Populations Are Located along the Path of Callosal Axons
We wished to determine whether CR-positive glutamatergic and GAD67/Mash1-GFP-positive GABAergic neurons are present at times and in a spatial distribution consistent with their contribution to the formation of CC axonal paths. As such, we undertook a longitudinal analysis of these populations during development to establish their relationship to callosal axons.
Between E12.5 and E15.5, the glutamatergic CR-positive neurons occupy a strategic midline position at the corticoseptal boundary (CSB) ( Figure 1C i -1C ii , arrowheads). As such, they occupy this area prior to any callosal axons entering this region ( Figure S2A i -S2A iv ). At this stage, the GAD67/Mash1-GFPpositive interneurons are still migrating within the marginal zone,

Author Summary
The largest commissural tract in the human brain is the corpus callosum, with over 200 million callosal axons that channel information between the two cerebral hemispheres. Failure of the corpus callosum to form appropriately is observed in several human pathologies and can result from defects during different steps of development, including cell proliferation, cell migration, or axonal guidance. Studies to date suggest that glial cells are critical for the formation of the corpus callosum. In this study, we show that during embryonic development, the corpus callosum, which was considered a neuron-poor structure, is in fact transiently populated by numerous glutamatergic and GABAergic neurons. With the use of in vitro graft experiments and of various transgenic mice, we demonstrate that neurons of the corpus callosum are essential for the accurate navigation of callosal axons. Moreover, we discovered that the guidance factor Semaphorin 3C, which is expressed by corpus callosum neurons, acts through the neuropilin 1 receptor to orient axons crossing through the corpus callosum. The present work therefore gives new insights into the mechanisms involved in axon guidance and implies that transient neurons work together with their glial partners in guiding callosal axons.
subplate, IZ, and subventricular zone of the frontal cortical area ( Figure 1C i ).
At E16.5, GAD67/Mash1-GFP neurons intermix with the CR neurons at the midline and in the lateral part of the CC ( Figure 1D i -1D ii ). At this age, CC pioneer callosal axons start to cross the midline [36,37], whereas later-growing axons originating from the frontal cortex just approach the lateral border of the CC  Double immunohistochemistry for CR and Emx1 (A i -A ii ) and for CR and Tbr1 (B i -B ii ) in coronal sections from mice at E15.5 (A i -A ii ) and E16.5 (B i -B ii ). (A ii and B ii ) are higher power views of a single confocal plan of the CC Z-stack seen in (A i and B i ), respectively (open arrows). The CR + neurons of the CC are glutamatergic in nature since they express Emx1 and Tbr1. A yellow channel highlights the colocalization of markers. (C-F) Immunohistochemical staining for CR (in red) in coronal sections from GAD67-GFP transgenic mice at E14.5 (C i -C ii ), E16.5 (D i -D ii ), E18.5 (E i -E ii ), and P3 (F i -F ii ). (C ii , D ii , E ii , and F ii ) are higher power views of a single confocal plane of the CC Z-stack seen in (C i , D i , E i , and F i ), respectively (open arrows). (C i -C ii ) At E14.5, CR + neurons are located at the corticoseptal boundary (CSB) (arrowheads in [C ii ]), whereas GAD67-GFP + neurons are present in the subventricular zone (SVZ), IZ, and subplate of the frontal cortex. (D i -D ii ) At E16.5, GAD67-GFP + neurons start to intermix with CR + neurons in one compact strip in the midline and lateral CC. (E i -E ii ) At E18.5, although the CC has undergone its dorsoventral compartmentalization, the CR + neurons segregate into three distinct compact strips (open arrowheads) in the lateral extension of the CC, whereas GAD67-GFP + neurons are dispersed in the entire CC white mater. In addition, GAD67-GFP + neurons form a cluster of cells in the extreme lateral part of the CC (*). (F i -F ii ) At P3, CR + neurons have nearly disappeared from the CC, while GFP + neurons are still present. Bar indicates 435 mm in (C i and E i ), 220 mm in (A i , B i , D i , and F i ), 110 mm in (A ii , B ii , E ii , and F ii ), and 70 mm in (C ii and D ii ). CFr, frontal cortex; LV, lateral ventricle; MZ, marginal zone. doi:10.1371/journal.pbio.1000230.g001 Double immunohistochemical staining for CR and Npn-1 (A i -A ii and B i -B ii ) and for CR and ephrinA5 binding sites (C i -C ii ) in coronal CC sections from E16.5 (A i -A ii ) and E18.5 (B i -B ii and C i -C ii ) mice. (A ii ) is a high-power view of the CC seen in (A i ). (B ii and C ii ) illustrate the lateral extensions of the CC seen respectively in (B i and C i ). (A i -A ii ) At E16.5, the CR + glutamatergic neurons extend throughout the entire CC and surround the pioneer axons that start to cross the midline (arrowhead in [A i ]). (B i -B ii and C i -C ii ) At E18.5, the CR + glutamatergic neurons segregate in three strips (open arrows) that delineate the dorsoventral pathways labeled for Npn-1 or ephrin-A5 binding sites. (D i -D ii ) Immunohistochemistry for Npn-1 in coronal telencephalon regions of the main body of the CC ( Figure 1E i -1E ii , open arrowheads). They were located: i) at the border of the IG and of the cingulate cortex (CCi), ii) in the middle of the white matter of the CC, and iii) in the ''glial'' sling at the border of the septum (SEP) and in the region of the GW. The three stripes of glutamatergic CR-positive neurons delineate both a ventral and a dorsal CC axonal path ( Figure 2B i -2B ii , Figure 2C i -2C ii , open arrows, and Figure 2G ii ). Labeling using carbocyanine dyes showed that these dorsal and ventral axonal paths originate from distinct mediolateral cortical areas ( Figure 2E i -2E ii , Figure 2F i -2F ii , and Figure S2D i -S2D iv ). This dorsoventral organization was further delineated by the restricted expression pattern of receptors for axon guidance molecules such as Npn-1 ( Figure 2B i -2B ii and Figure S2F) and Deleted in Colorectal Cancer (DCC) ( Figure  S2H) dorsally, and ephrinA5 binding sites ventrally ( Figure 2C i -2C ii and Figure S2G). By contrast, at this age, GAD67/Mash1-GFP-positive neurons are more diffusely distributed within the entire white matter of the CC and are seen surrounding the growing commissural axons ( Figure 1E i -1E ii , Figure 2E i -2E ii , and 2F i -2F ii ).
Glutamatergic neurons of the CC expressing CR disappeared abruptly between P1 and P3, whereas GAD67/Mash1-GFP GABAergic neurons disappeared progressively in a spatiotemporal gradient, from P7 at the midline, until P21 in the extreme lateral part of the CC ( Figure 1F i -1F ii and Figure S3A i -S3A ii to S3D i -S3D ii ). Cleaved caspase 3 staining and ultrastructural changes showed that both neuronal populations of the CC died at early postnatal ages ( Figure S3). Our ultrastructural study revealed that dying neurons adopt different morphological types: a nonlysosomal vesiculate type (type IIIB) for GAD67-GFP-positive GABAergic interneurons or an autophagic type (type II) for glutamatergic neurons [38].
Our results, therefore, demonstrate that the CC is more heterogeneous than previously thought. Specifically, the entire CC white matter contains transient CR-positive glutamatergic neurons and GABAergic interneurons that correspond to the organization of the callosal projections within this region. Furthermore, their location and the timing of their appearance raise the possibility that these neurons actively participate in the guidance of callosal axons.

Topographic Positioning between the Glutamatergic and GABAergic Neurons, and the Developing Callosal Axons
To study the spatial relationships between CC neurons and callosal axons, we used electron microscopy and 3-D analysis of high-resolution confocal image stacks.
Electron microscopy and pre-embedding immunocytochemistry showed CC neurons apposed to one another, forming a complex cellular network (stars) around callosal axons (arrowheads) ( Figure  S4A to S4D i -S4D ii and unpublished data). To determine how glutamatergic CR-positive and GABAergic GAD67/Mash1-GFPpositive neurons participate in this cellular network, we generated isosurface maps ( Figure S4E ii ) using immunostaining to label both neuronal populations and cell nuclei (CR, GFP, Hoechst) ( Figure  S4E i ). Isosurface reconstructions allowed us to explore the geometry of this cellular organization using the navigator function of IMARIS 4.3 software ( Figure S4E iii -S4E vii and Video S1). The 3-D visualization showed that both CR-positive glutamatergic neurons and GAD67/Mash1-GFP interneurons contributed in forming the ''walls'' of a complex cellular network surrounding callosal axons inside the CC ( Figure S4E iii -S4E vii and Video S1). Our observations thus indicate that the two neuronal populations that transiently populate the CC form a dense cellular network that interacts intimately with the growing commissural axons.

CC Formation Is Impaired in Mash1 Mutant Mice Lacking GABAergic Interneurons
Given the density and complexity of the neuronal network that we identified, it is relatively difficult to unravel its function in CC formation. As a first step, we analyzed the brains of mutant mice defective for the production of GABAergic interneurons. Mash1 is a transcription factor expressed in GABAergic progenitors of the ventral telencephalon and its inactivation severely impairs the production of cortical interneurons [34,35]. Consistently, we found that the CC of Mash1 mutant embryos was nearly devoid of GABAergic interneurons (GABA-positive neurons: 1.39060.146 neurons/mm 2 in CC of wild-type (WT) mice versus 0.16760.055 neurons/mm 2 in CC of Mash1 2/2 mice, p,0.001) (compare Figure S5C i -S5C ii with S5D i -S5D ii ). To investigate how the lack of GABAergic interneurons affects CC formation, we also examined whether the other CC cell types are impaired by Mash1 inactivation. At E16.5, the CC glial cells' localization, morphology, and expression of guidance factors (ephrins, semaphorins, and Slit2) were not affected in Mash1 mutant embryos ( Figure S5E i -S5E ii and S5F i -S5F ii , and Figure S6). By contrast, some glutamatergic CR-positive neurons were displaced ventrally at the midline (compare Figure 3C i -3C ii and 3D i -3D ii , arrowheads). Thus, in Mash1 mutants, although the glial scaffold appears normal, the CC neuronal network is severely affected, with a lack of GABAergic interneurons and a displacement of CR-positive glutamatergic neurons at the midline.
Our analysis showed that Mash1 inactivation leads to major alterations of axonal paths in the CC (compare Figure 3A with 3B, compare Figure 3C i -3C ii with 3D i -3D ii , and see Figure S5G i -S5G v with S5H i -S5H v ). From E16.5 to E18.5, Mash1 2/2 embryos exhibited partial ( Figure 3D i -3D ii and Figure S6) to complete ( Figure 3B and Figure S5H i -S5H iv ) AgCC, with few axons, if any, crossing the midline. Although axons were impaired in midline crossing, they expressed normal levels of L1, Npn-1, and DCC guidance receptors (unpublished data, Figure 3D i -3D ii , and Figure  S6G i -S6G ii and S6H i -S6H ii ). Instead, callosal axons entered the IG or the SEP and formed two large ectopic fascicles known as Probst bundles that are characteristic of acallosal mammalian forebrains ( Figure 3B and Figure 3D i -3D ii , open arrowheads). In addition, DiI-labeled axons that are normally located in the dorsal and ventral paths of the CC intermingle in Mash1 2/2 embryos slices from E16.5 GAD67-GFP transgenic mice. (D ii ) is a high-power view of the CC seen in (D i ). At E16.5, callosal axons encounter GAD67-GFP + GABAergic interneurons while they are migrating tangentially through the IZ. (E i -E ii and F i -F ii ) P0 CC coronal section of GAD67-GFP transgenic mice showing axonal tracing of callosal axons by insertion of DiI crystals, respectively, in the frontal (CFr, arrowhead in [E i ]) and parietal (CPa, arrowhead in [F i ]) cortices. (E ii and F ii ) are high-power views of the corresponding sections in (E i and F i ), respectively. Callosal axons from the medial cortical area grow in the dorsal path of the CC, whereas axons from the lateral cortex grow in the ventral path. GAD67-GFP + GABAergic interneurons are dispersed in the entire CC and surround callosal axon bundles without delineating the dorsoventral paths. (G i -G ii ) Schematic drawing representing the complementary organization of CR + glutamatergic neurons (in red) and of GAD67/Mash1-GFP GABAergic interneurons (in green) in the CC at E16.5 (G i ) and E18.5 (G ii ). The pioneer callosal axons originating from the cingulate (CCi) cortex, as well as the later-growing callosal axons from the frontal and the parietal cortices are represented in blue and are seen to navigate through the CC neurons. Bar indicates 435 mm in (A i , D i , E i , and F i ), 220 mm in (A ii , B i , C i , and C ii ), 110 mm in (D ii , E ii , and F ii ), and 70 mm in (B ii ). HIC, hippocampal commissure; LV, lateral ventricle; SP, subplate. doi:10.1371/journal.pbio.1000230.g002 before reaching the midline (compare Figure S5G i -S5G v and S5H i -S5H v ). On the other hand, in Mash1 2/2 , the area specification of the dorsal telencephalon [39], and the laminar distribution of the Tbr1-positive cortical layers V-VI that contain pyramidal callosal neurons and of Sabtb2-positive callosally projecting neurons were normal (unpublished data, and compare Figure S5A i -S5A ii and S5B i -S5B ii ).
To investigate whether this severe axon guidance phenotype was due to defects in the CC region rather than to altered development of other regions in the Mash1 mutants, we performed transplantations of the CC into E16.5 telencephalic slices, using different combinations of WT and Mash1 2/2 embryos ( Figure 3 and Figure S7). In our slice assays, as in in vivo [36,37], the callosal axons from dorsolateral neocortex develop later than pioneer axons, and after E16.5, their growth cones enter the CC region in successive streams over a period of several days ( Figure  S2B i -S2B ii to S2C i -S2C ii ). When dorsal cortical explants from GFP-positive Mash1 2/2 mice were transplanted into WT slices (n = 7 out of 7), GFP-labeled callosal axons crossed the midline, whereas transplantations of dorsal cortex from GFP-positive WT mice into Mash1 2/2 slices (n = 6 out of 6) lead to an impairment in axonal midline crossing ( Figure S7). These experiments suggested that callosal axons mistargeting in Mash1 mutant embryos is due to defects in CC midline and surrounding structures. To further investigate this issue, we then performed reversion experiments ( Figure 3). When dorsal cortical explants from GFP-positive WT mice and explants of the CC region from WT donors were transplanted into a WT brain slice, a majority of GFP-labeled callosal axons crossed the midline ( Figure 3E   Experimental paradigm used to test whether the CC neuronal network integrity is necessary and sufficient to direct the growth of callosal axons. To this end, WT CC is transplanted in a Mash1 2/2 slice. (G ii -G iii ) GFP immunocytochemistry showing the complete restoral of Mash1 2/2 callosal axons pathfinding. Dashed lines outline the CC transplant localizations. Brain slices in (E ii -E iii , F ii -F iii , and G ii -G iii ) were counterstained with Hoechst. Bar indicates 435 mm in (E ii , F ii , and G ii ), 220 mm in (A, B, C i , D i , E iii , F iii , and G iii ), and 110 mm in (C ii and D ii ). CFr, frontal cortex. doi:10.1371/journal.pbio.1000230.g003 WT CC restored normal axonal guidance of the majority of Mash1 2/2 callosal axons, but only when the transplant comprised the medial and lateral parts of the CC that contain the GABAergic interneuron population we have identified ( Figure 3G i -3G iii ; n = 4 out of 6). Transplantation experiments of GAD67-GFP-positive WT CC into WT slices confirmed that CC GABAergic interneurons remain through the CC transplant and maintained their initial organization after several days in vitro ( Figure 4D i -4D iii ; n = 4 out of 4). Therefore, callosal axons misrouting observed in Mash1 mutant embryos is largely due to defects in the CC region.
Altogether, our experiments indicate that Mash1 inactivation does not impair callosal pyramidal neurons differentiation but leads to a severe modification of the CC neuronal network. These results suggest that CC GABAergic interneurons, which are lacking in Mash1 2/2 mice, may participate in callosal axons guidance and support the idea that the integrity of this neuronal CC network is important for normal callosal axons navigation.

Chemoattractive Activity of the Two CC Neuronal Populations
To further understand how CC neurons contribute to callosal axon navigation, we tested whether a CC region enriched in GABAergic and/or CR-positive glutamatergic neurons could promote the growth of callosal axons in coexplant and heterotopic graft experiments (Figures 4, 5, and 6).
At first, we examined whether the CC region exerts an attractive influence on cortical axons by placing E16.5 lateral CC explants, comprising the two neuronal populations of interest, adjacent to explants of medial (cingulate or frontal) cortex ( Figure 4A i -4A ii ). At E16.5, after 2 d in vitro, outgrowth in the quadrant closest to the CC aggregate was increased for axons originating from the cingulate and the frontal cortical area compared to that in the quadrant furthest away from the aggregate, indicating chemoattraction ( Figure 4A i -4A ii and 4C). By contrast, the septal and the IG regions were found to exert a repulsive action on cortical axons ( Figure 4B i -4B ii , 4C, and unpublished data). At E16.5, after the cerebral hemispheres have fused, it was possible to ascertain the callosal identity of the axons, by using CC organotypic slices. DiI-labeled axons growing in E16.5 slice preparations from GAD67-GFP slices (unpublished data, n = 17 out of 22) or in E16.5 WT slices grafted with a CC from a GAD67-GFP embryo ( Figure 4D i -4D iii ; n = 4 out of 4) navigated across the midline as they normally do in vivo. In contrast, when small explants of E16.5 GAD67-GFP-positive lateral CC containing both neuronal populations were inserted into E16.5 heterotopic septal region of host slices, some DiIlabeled callosal axons were deflected from their normal trajectory, penetrated the SEP and innervated the transplants ( Figure 4E i -4E iii , arrowheads; n = 7 out of 8). The enrichment of both types of CC neurons within small E16.5 lateral CC explants was confirmed by using GAD67-GFP-positive explants ( Figure 4A ii and Figure 4D i -4D iii to 4F i -4F iii ) and CR immunohistochemistry ( Figure 4F i -4F iii ), whereas the lack of astroglial cells was demonstrated by GFAP immunohistochemistry (unpublished data). Thus, these observations reveal the existence of an attractive activity for callosal axons located in the neuron-rich region of the CC.
We next determined the respective contribution of GAD67-GFP-positive GABAergic interneurons and CR-positive glutamatergic neurons of the CC to this guidance activity. To directly test the involvement of CC GABAergic neurons, we grafted in the SEP, explants of E14.5 or E16.5 GAD67-GFP medial ganglionic eminence (MGE) ( Figure 5A i -5A iii ), which generate the GABAer-gic interneurons of the CC (unpublished data). Interestingly, numerous axons left the callosal track, penetrated the repulsive SEP, and grew through migrating GAD67-GFP-positive interneurons originating from the MGE transplant ( Figure 5A i -5A iii ; n = 12 out of 15 for E14.5 MGE, and n = 17 out of 22 for E16.5 MGE). This attraction was specific for MGE-derived interneurons, since control explants of the lateral ganglionic eminence (LGE) did not attract callosal axons ( Figure 5B i -5B iii ; n = 5 out of 6). These observations strongly support the idea that CC GABAergic neurons directly contribute to the attraction of callosal axons. To estimate whether CC GABAergic neurons are the sole contributors of this guidance activity, we compared in coexplant experiments the quantity of cortical axons that were attracted by WT or Mash1 2/2 E16.5 lateral CC explants ( Figure 5C i -5C ii , 5D i -5D ii , and 5E). Mash1 2/2 explants of the lateral CC, that contained glutamatergic neurons but are devoid of GABAergic interneurons ( Figure S5D i and S5F i ), were found to exert a reduced chemoattraction on cortical axons compared to WT lateral CC explants that contain both neuronal populations ( Figure 5E; 240%, p,0.05). The equal number of CR-positive neurons within small lateral CC explants of Mash1 2/2 compared to WT was confirmed by using CR immunohistochemistry (unpublished data). These results show that CC GABAergic interneurons contribute to part of the attractive activity of the CC on cortical axons.
To further test whether CR-positive glutamatergic neurons can also directly attract callosal axons, we took advantage of the fact that the E14.5 developing CC comprises CR-positive glutamatergic neurons and lacks GAD67-GFP-expressing GABAergic interneurons ( Figure 1C  Altogether, coexplant and transplantation experiments indicate that CC neuronal populations exert an attracting influence on callosal axons, which is mediated by both GABAergic and glutamatergic CR-positive neurons.

Sema3C Is Expressed by CR-Positive Glutamatergic Neurons and Contributes to the Attractive Activity on Callosal Axons
In search for candidate molecular signals mediating the attractive activity on callosal axons, we found that Sema3C is strongly expressed only in the subcortical white matter and especially the CC region ( Figure 7A i and Figure S6I i -S6I ii ), as previously observed [26]. In the CC, colabeling experiments revealed that Sema3C mRNA expression is restricted to CRpositive glutamatergic neurons ( Figure 7A ii -7A iv ). The Sema3C mRNAs were never detected in GAD67-GFP-positive interneurons ( Figure S8A i -S8A ii ) or GFAP-positive astroglial cells ( Figure  S8B i -S8B ii ). Since Sema3C has been described to act as an attractive factor for neocortical and cingulate axons in vitro [26,40,41], CR-positive glutamatergic neurons of the CC might exert their attractive effect on callosal axons through the action of Sema3C.
To test this possibility, aggregates of Sema3C-expressing HEK293T cells were placed in the repulsive septal region of E16.5 WT slices (Figure 7B i ). Callosal axons were misrouted from their normal path and invaded cell aggregates expressing Sema3C ( Figure 7B iii -7B iv ; arrowheads; n = 13 out of 16), whereas control To determine the in vivo function of Sema3C in the developing CC, we examined the brains of mutant mice inactivated for the Sema3C gene ( Figure 7D i -7D v and Figure S8D i -S8D ii ). CR and GFAP immunohistochemistry at E16.5 and E18.5 indicated that the position and organization of the CR-positive glutamatergic neurons and glial cell populations within the CC is indistinguishable in WT and Sema3C 2/2 mice, suggesting that their development is not sensitive to the loss of Sema3C (compare Figure  S8C i with S8D i and Figure S8C ii with Figure S8D ii , respectively). Sema3C 2/2 mice exhibited partial to severe AgCC. When the agenesis was partial, all dorsal Npn-1-positive axons failed to cross the midline, whereas part of ventral callosal axons labeled for NPY were able to cross (compare Figure 7C i -7C v with Figure 7D i -7D v ). Misguided callosal axons formed Probst bundles within the IG ( Figure 7D i -7D iv , arrowheads). In some cases, Sema3C 2/2 mice displayed severe AgCC characterized by midline fusion defects and a complete failure of any callosal axons to cross the midline at the level of the CC main body (compare Figure S8C i -S8C ii with Figure S8D i -S8D ii ).
Taken together, these results reveal that guidance mechanisms of callosal axons rely in part on Sema3C, which contributes to the chemoattractive effect of CR-positive glutamatergic neurons on callosal axons.

Sema3C Directs Cortical Axon Growth via Npn-1
The precise identity of the endogenous neuronal receptor for Sema3C remains unclear. In vitro, Sema3C binds with high affinity to both Npn-1 and its close homolog Npn-2 [42]. Since Npn-1, but not Npn-2, is expressed on callosal axons (see Figure 2A Figure 3C i -3C ii , and 7C i -7C iv ; and unpublished data) and Semaphorin/Npn-1 signaling is critical for CC development [26,29], we examined whether Npn-1 was necessary to allow callosal axons to respond to Sema3C. We placed aggregates of Sema3C-expressing HEK293T cells adjacent to explants of E15.5 medial cortex ( Figure 8A i -8A ii ). After 2 d in vitro, axonal growth in the quadrant closest to the aggregate was increased by 70% at E15.5 compared to that in the quadrant farthest away from the aggregate (p,0.01; Figure 8A i -8A ii ), indicating chemoattraction. Consistently, adding recombinant Sema3C (5 to 10 nM) to dissociated neurons from medial cortex increased axon length by 35% compared to the control condition (p,0.001; Figure 8B i -8B ii ). Npn-1 blocking antibodies abolished both the attractive and growth-promoting responses of cortical neurons to Sema3C ( Figure 8A i -8A ii and 8B i -8B ii ) and disturbed DiI-labeled callosal axons navigation in E16.5 brain slices (unpublished data, n = 14 out of 19). To exclude the possibility of nonspecific antibody binding, we knocked down endogenous Npn-1 in dissociated cortical neurons using two different small interfering RNA (siRNA) sequences that efficiently silenced expression of Npn-1 without affecting Npn-2 levels, as assessed by antibody staining (unpublished data). Remarkably, both siRNAs completely abrogated the positive effect of Sema3C on axon growth ( Figure 8C i -8C ii ). Taken together, these results strongly suggest that Npn-1 is necessary for mediating the attractive response of callosal axons to Sema3C.

Discussion
In this study, we show that two transient neuronal subpopulations, one CR/glutamatergic and the other Mash1/GABAergic, occupy a strategic position for contributing to the guidance of nascent callosal axons. Mash1 2/2 mice that lack GABAergic neurons possess major pathfinding defects in the CC, and grafting WT CC comprising midline neurons in a Mash1 2/2 slice rescued this phenotype. In addition, we show that both neuronal populations possess the ability to chemoattract callosal axons and that this activity is dependent in part on Sema3C function in the CR-positive glutamatergic population. Consistent with this observation, Sema3C is required in vivo for the proper development of the CC pathway. Taken together, our work provides strong evidence for a role in callosal axon guidance by the two transient populations we characterized in this study. Notably, these activities appear distinct from those emanating from glial guidepost cells. As such, our findings show that two previously uncharacterized neuronal populations converge towards the midline and form a cellular network that is essential for controlling callosal axon navigation (model in Figure 2G i -2G ii ).

Roles for CC Neurons in the Guidance of Callosal Axons
We have revealed the existence of two populations of glutamatergic and GABAergic neurons that although arising from distinct sources, converge on the interhemispheric fissure prior to the arrival of CC axons. The precise origins of these two CC neuronal populations have yet to be determined. Our observations suggest that CR-positive glutamatergic neurons invade the CC through a tangential subpial migration and may thus correspond to cortical pioneer neurons that originate from the retrobulbar ventricle [43,44]. In contrast, our fate-mapping and tracing experiments indicate that the GABAergic interneurons of the CC originate in the MGE (unpublished data) as described for a majority of cortical interneurons in mice [45][46][47][48][49][50][51].
A few studies have reported the presence of neurons within or around the CC, such as CR-positive neurons in the mouse and human ''glial'' sling [16,17] and scattered neurons in the cat CC during early postnatal life [18,19]. It was proposed that these neurons were migrating through the CC [19] or below the CC in explants. Data are expressed as a P/D ratio, where P and D are the mean fluorescence 6SEM intensity in the quadrants proximal and distal to the CC or SEP explants. Double asterisks (**) indicate significantly different with p,0.01; triple asterisks (***) indicate significantly different with p,0.001. CC neurons cause strong chemoattraction on cingulate (CCi) and frontal axons, whereas the SEP is repulsive. (D i ) In vitro model of organotypic slices at E16.5 used to study callosal axon navigation through CC after transplantation of E16. 5 [10]. Our study provides an evaluation of the positioning, development, and character of these populations and demonstrates that their presence within the CC is transient.
Moreover, our data strongly support a requirement for these neurons in the guidance of callosal axons. First, we identified a close structural association between the neurons of the CC and the callosal axons during embryonic development. The intimate relationship between these neurons and the incoming callosal afferents is further bolstered by our 3-D analysis showing that CRpositive glutamatergic neurons form a complex multicellular network with the transient GABAergic interneuron population we identified. The integrity of the CC multicellular network formed by GABAergic and glutamatergic neurons is required for normal CC axonal navigation, as shown by both our analysis of Mash1 mutant mice and our grafting experiments. Our present work reveals that the guidance of callosal axons is actively mediated through the chemotropic actions of the two novel neuronal populations of the CC that we examined. Moreover, our study demonstrates that CR-positive glutamatergic neurons exert a direct attractive influence on callosal axons via Sema3C expression.
This function of the CC neurons fits well with the emerging notion that migrating neurons may have a role in axon pathfinding. It has recently been found that thalamocortical axon growth relies on the early tangential migration of a population of GABAergic neurons within the ventral telencephalon [52]. In addition, the lateral olfactory tract (LOT) projections are guided by early generated neurons, named ''LOT neurons,'' that migrate tangentially [53,54]. Similar functions have been reported for CD44-positive neurons in the guidance of retinal axons at the optic chiasm [55] and Cajal-Retzius reelin/calretinin-positive neurons in the establishment of hippocampal projections [56].

CC Neurons Participate in Callosal Axon Guidance via Sema3C/Npn-1 Signaling
Our work demonstrates that a neuronal-rich region of the CC attracts callosal axons, at least in part through the expression by glutamatergic CC neurons of the guidance cue Sema3C. Initially, the Sema3C gene is strongly expressed by guidepost CR-positive glutamatergic neurons adjacent to the midline prior to the entrance of callosal axons. In organotypic slices, Sema3Cexpressing HEK293T cells attract callosal axons into heterotopic regions. These observations, using a specific guidance assay for CC axons, extend previous in vitro studies showing that Sema3C acts as an attractive guidance signal for neocortical and cingulate axons [26,40,41]. It was known that the Sema3C repulsive activity is mediated via Npn-1/Npn-2 heterodimers or Npn-2/Npn-2 homodimers [57], but the nature of the receptor mediating the attractive effect was not yet characterized. Here, we show that inhibiting selectively the Npn-1 receptor abolished completely the attractive and outgrowth-promoting effects of Sema3C. Moreover, callosal axons are found to express Npn-1, but not Npn-2. Therefore, our results reveal that Npn-1 can serve as a Sema3C receptor to mediate chemoattraction. Neuropilins require a signaling coreceptor to mediate semaphorin function. For example, PlexinAs and L1-CAM are responsible for transducing Sema3A repulsive response via Npn-1 in cortical neurons [58,59].
Other transmembrane proteins, including the tyrosine kinase receptors Met, ERBB2, OTK, and VEGFR2, participate in semaphorin responses by regulating diverse intracellular signaling events and functional outcomes [60][61][62]. The transducer that mediates Sema3C attractive response remains so far undefined. It will be important to analyze whether the assembly of specific subunits combinations confers unique ligand-binding properties of semaphorin receptors, and whether different Npn-1 receptor complexes coexist on cortical axons, dictating either Sema3Amediated repulsion or Sema3C-mediated attraction.
Although it has long been recognized that Sema3C regulates the formation of the cardiovascular system [63], its in vivo function in the central nervous system remains relatively unexplored. Here, we observed that the development of the CC path depends on Sema3C expression. Indeed, callosal axons fail to grow or navigate correctly through the CC of mutant mice lacking Sema3C gene function. These results shed new light on previous studies showing that mice in which Npn-1 is unable to bind Semas exhibit CC axonal pathfinding defects [26,29]. The similarity between the Npn-1 Sema2 mice and Sema3C 2/2 mice suggests that Sema3C is the ligand required for directing Npn-1-mediated callosal axon navigation. Taken together, these results reveal that CR-positive glutamatergic neurons within the dorsal midline territory control callosal axon navigation, at least in part, through a Sema3Cdependent mechanism. Therefore, transient CC guidepost neurons play a central role in mediating the guidance cues required for callosal axon pathway formation.

Neurons and Glia Cooperate to Guide Callosal Axons
Previous studies on CC development emphasized the role of glial cells. In Silver and Ogawa's study [11], aberrant callosal axons maintained a potential to regrow upon the surface of a gliacovered scaffold after mouse embryos were made surgically acallosal at E16.5. Other studies indicate that astroglial cells of the GW and IG direct callosal pathfinding at the midline by secreting guidance cues [7,8]. Our work demonstrates that, in addition, CC formation requires the presence of specific neuronal populations. What is the relative contribution of neurons and glia to CC formation? Preliminary results indicate that neurons and astroglial cells of the CC intermingle to form a complex 3-D structure and that glutamatergic CR-positive neurons lie along the radial glial processes. Therefore, in addition to secreting guidance factors, glial cell populations may aid in the establishment of the CC neurons through more complex trophic or signaling interactions. Indeed, the interplay between the neuronal and glial cells within the midline will be intriguing to investigate in the future.

Animals
All animal research has been conducted according to relevant national and international guidelines. WT mice maintained in a C57Bl/6 genetic background were used for developmental analysis of the CC. We used heterozygous GAD67 _ GFP (Dneo) mice [33], which will be referred to as GAD67-GFP mice in this cocultured with small explants of Mash1 2/2 CC missing GABAergic interneurons (D ii ) Immunohistochemical staining for bIII-tubulin on E16.5 Mash1 +/+ frontal cortex cocultured with E16.5 Mash1 2/2 CC. (E) Quantification of the axonal guidance responses to CC explants. Data are expressed as a P/D ratio 6SEM, where P and D are the mean fluorescence intensity in the quadrants proximal and distal to Mash1 +/+ or Mash1 2/2 CC explants. An asterisk (*) indicates significantly different with p,0.05. Mash1 2/2 cortical axons are normally attracted by WT CC. By contrast, Mash1 mutant CC is still attractive for cortical axons, but with a reduced efficiency compare to WT CC. Bar indicates 485 mm in (A ii , B ii , C ii , and D ii ) and 120 mm in (A iii and B iii ). LV, lateral ventricle. doi:10.1371/journal.pbio.1000230.g005 work. Experimental animals were obtained by mating C57Bl/6 mice with heterozygous GAD67 _ GFP mice. GAD67 _ GFP embryos can be recognized by their GFP fluorescence. NINDS GENSAT BAC Transgenic mice for Ascl1 (Ascl1-EGFP)1Gsat/ Mmnc (MMRC) referred to as Mash1-GFP in this work were maintained in a C57Bl/6 background and were recognized by their GFP fluorescence. Mash1 KO heterozygous mice were maintained in a mixed C57Bl/6 and DBA background and crossed to produce homozygous embryos [64]. Mash1 heterozygous mice were also crossed with a transgenic mouse line expressing GFP ubiquitously [65] in order to produce GFPpositive Mash1 2/2 embryos. PCR genotyping of these lines was performed as described previously [52]. Heterozygous embryos did not show any phenotype and were used as controls. Sema3C  heterozygous mice were maintained in a CD1 background and mated to obtain Sema3C 2/2 embryos. The genotype of the offspring was determined by PCR as described [63]. For staging of embryos, midday of the day of vaginal plug formation was considered as embryonic day 0.5 (E0.5).
Embryos were collected by Caesarean section and killed by decapitation. Their brains were dissected and fixed by immersion overnight at 4uC in a solution containing 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (pH 7.4). Postnatal mice were deeply anaesthetized and perfused with the same fixative, and their brains postfixed 4 h. Brains were cryoprotected in 30% sucrose, and cut in coronal 50-mm-thick frozen sections for staining.

Production of AP-Sema3C
AP-Sema3C was obtained by cloning cDNA encoding mouse Sema3C in pAPtag-5 vector (GenHunter Corporation), which contains a sequence coding for secreted alkaline phosphatase. To produce AP-tagged proteins, HEK293T cells were transfected with the AP-Sema3C vector or empty pAPtag-5 vector as control, using lipofectamin plus (Invitrogen) or fugen (Roche). After 3 d of culture in Opti-MEM serum-free medium, the supernatant was collected and concentrated using Centricon filters (Millipore). AP activity was assessed as described [29].

Slice Culture Experiments
We developed an in vitro model of CC organotypic slices adapted from a previously published telencephalic slice culture preparation [45,52,[66][67][68] and CC preparation [7]. Embryos were placed in ice cold dissecting medium (MEM Gibco ref 11012-044 with 15 mM glucose and 10 mM Tris [pH 7-9]). Brains were removed and embedded in 3% low-melting point agarose (Invitrogen); 250-mm-thick coronal sections were then cut using a vibratome filled with cold dissecting medium, and slices at the level of the CC were collected in the same medium. CC slices were cultured on nuclepore Track-Etch membrane (1-mm pore size; Whatman) or PET cell inserts (1-mm pore size; Beckton-Dickinson) in tissue dishes containing 1 ml of BME/HBSS (Invitrogen) supplemented with glutamine, 5% horse serum, and Pen/Strep [52].
For CC transplantation experimentation, slices from E16.5 embryos were selected since at this early stage of development, the CC contained its whole complement of guide post cells and only the pioneer CC axons of ventral cingulate origin [36,37]. It is critical that cultured hemispheres are already joined for the differentiation of the CC in vitro. In our slice assay, as in vivo, the callosal axons from dorsolateral neocortex develop later, and after E16.5, their growth cones enter the CC region in successive streams over a period of several days. Our slice assay performed at E16.5 allowed us to study: (1) the function of both CC guidepost neuronal populations that have reached the CC midline at that stage, (2) the outgrowth properties of the majority of callosal axons that are growing through the CC after E16.5, and (3) the effects of transplantations and pharmacological (guidance factors, lesions) manipulation on callosal axons navigations.
To define the putative function of CC neurons in attracting callosal axons, the transplantation assay was performed at E16.5 to analyze the navigation of WT early callosal axons labeled for DiI after insertion of small DiI crystals into the frontal cortex of slices. Small explants of E14.5 corticoseptal boundary comprising only CR-positive glutamatergic neurons or E16.5 lateral CC IZ comprising both neuronal populations were excised using tungsten needles and transplanted into the SEP of E16.5 host slices. After incubation for 48-64 h, the slices were fixed, and axon trajectories through the various regions were analyzed by confocal analysis. In most of our transplantation experiments of CC (.90%), we observed that axons grew without any difficulty through small or large transplants, and only cases with axons penetrating into the grafted explants were counted as positive results for attraction.
We found that CC GABAergic interneurons are generated by the medial ganglionic eminence (MGE) from E14.5 to E16.5 (unpublished data). To define the putative function of the CC GABAergic interneurons we transplanted small explants of E14.5 or E16.5 MGE into the SEP of an E16.5 slice as described above. As a control, we used small explants of E16.5 lateral ganglionic eminence (LGE) that do not generate CC GABAergic interneurons. In this assay, cases with axons growing along GAD67-GFP + interneurons originating from the grafted explants were counted as positive results for attraction.
For the Sema3C study, HEK293T cells were transfected with an AP-control plasmid or an AP-Sema3C plasmid (see above). To highlight HEK293T transfected cells, a pEGFP plasmid was coexpressed. Aggregates of HEK293T transfected cells prepared by high-density culture within an inverted drop of medium were transplanted into the CCi, CC, or SEP of host slices as described before [68]. For the Npn-1 study, the Npn-1-blocking antibody (R&D systems) was added at the final concentration of 5 mg/ml.
For the Mash1 study, the transplantation assay was performed at E16.5 to analyze the growth of WT (Mash1 +/+ ; Mash1 +/2 ) or Mash1 2/2 GFP-positive callosal axons within CC of WT (Mash1 +/+ ; Mash1 +/2 ) or Mash1 2/2 slices. Since heterozygous embryos did not show any phenotype, they were also used as controls. Portions of the frontal cortex with underlying white matter and CC from donor slices were excised using tungsten needles and transplanted into host slices from which the equivalent region had been removed. After incubation for 48-64 h, the slices were fixed and immunostained for GFP before confocal analysis.
For dissociated cell cultures, neurons were dissociated and plated onto polylysine/laminin-coated four-well plates (Nunc) in Neurobasal medium supplemented with 1 mM glutamine, 1:50 B27 (GIBCO), and AP control or AP-Sema3C supernatants (see above). In some experiments, neurons were cultured in the presence of anti-Npn-1 (R&D Systems). Efficient knock-down of Npn-1 was obtained using the following siRNA sequences: 59-AAUCAGAGUUCCCGACAUAUU-39 (Npn-1 siRNA1) and 59-UGUCAAGACUUACAGAGUAUU-39 (Npn-1 siRNA2). Neurons were coelectroporated with a pCAGGS-GFP vector and with different siRNAs (100 pmol) as described [72]. Quantification of axonal growth and guidance was performed as described before [73], or by using a measuring program built in MatLab software that allows to compare the density of immunolabeled axons in the proximal region facing the source of guidance cues and the distal region.

In Situ Hybridization
Sema3C plasmid was linearized with EcoRI (New England Biolabs) for antisense RNA synthesis by T7 polymerase (Promega) and with XhoI (New England Biolabs) for sense RNA synthesis by T3 polymerase (Promega). EphA4, Npn-1, EphB1 plasmids were linearized with SacI (New England Biolabs) for antisense RNA synthesis by T3 polymerase (Promega). ephrinB2 plasmid was linearized with BamH1 (New England Biolabs) for antisense RNA synthesis by sp6 polymerase (Promega). Slit2 plasmid was linearized with Xba1 (New England Biolabs) for antisense RNA synthesis by T7 polymerase (Promega). For in situ hybridization, brains were dissected and fixed by immersion overnight at 4uC in a solution containing 4% paraformaldehyde (PFA) in PBS. Freefloating vibratome sections (100 mm) were hybridized with digoxigenin-labeled cRNA probe as described before [74]. To combine in situ hybridization and immunofluorescence, Fast Red (Roche) was used as an alkaline phosphatase fluorescent substrate.

Imaging
Fluorescent-stained sections were imaged with confocal microscopes (Zeiss LSM 510 Meta or Leica SP5) equipped with 106, 206, 406 oil Plan-NEOFLUAR, and 636 oil, 1006 oil Plan-Apochromat objectives. Fluorophore excitation and scanning were done with an Argon laser 458, 488, 514 nm (blue excitation for GFP, Alexa488, CY2, and DiA), with a HeNe1 laser 543 nm (green excitation for Alexa 594, CY3, and DiI) and a Diode laser 405 nm (for Hoechst staining). Z-stacks of 10-15 plans were acquired for each CC coronal section in a multitrack mode avoiding crosstalk artifacts of the fluorochromes. Z-stacks of 40-50 sections were acquired for each CC section for the creation of isosurfaces with Imaris4.3 software.
Imaris images processing. All 3-D Z-stack reconstructions and image processing were performed with Imaris 4.3 software (Bitplane). Some image stacks contained approximately 40 sections each, giving an excellent z-axis resolution. To create real 3-D datasets, we used the mode ''Surpass'' of Imaris. Single sections of a Z-stack were displayed by using the ''Slice'' mode of Imaris. The generation of isosurfaces (object defining a surface surrounding voxels located between two threshold values) allowed us to visualize the contours of cells and to delimit the cell-free spaces between the neurons of the CC. Using a navigator function of IMARIS 4.3 software in the ''Animation mode,'' it was then possible to explore the organization of the cell-free spaces within the thickness of the CC. The Animation mode allows Key-Frame Animation of the exploration of CC slices to be saved in AVI file format. The colocalization between two fluorochromes was calculated and visualized by creating a yellow channel using Imaris. Figures were processed in AdobePhotoshop CS2, and schematic illustrations were produced using Adobe Illustrator CS2.

Ultrastructure
E16.5 and E18.5 embryos were killed by decapitation. Brains were dissected and fixed by immersion for 24 h at 4uC in a solution containing 2% glutaraldehyde and 4% paraformaldehyde in 0.1 M phosphate buffer (pH 7.4) with the addition of 2% sucrose. The brains were then rinsed in 0.1 M cacodylate buffer (pH 7.4), postfixed at room temperature for 2 h in 1% OsO 4 , dehydrated in graded ethanols, and embedded in Epon. The regions containing the CC of the embedded brains were trimmed and mounted on blocks to cut semithin and ultrathin sections. The ultrathin sections were mounted on Formvar-coated single-slot grids and contrasted with 2% uranyl acetate and 0.2% lead citrate.
For pre-embedding immunocytochemistry, embryonic brains were fixed by immersion in a solution containing 4% paraformaldehyde and 0.1% glutaraldehyde in 0.1 M phosphate buffer (pH 7.4) supplemented with 2% sucrose for 24 h. Fifty-micrometer-thick sections were cut with a vibratome and immunoreacted. Endogenous peroxidase reaction was quenched with 0.5% hydrogen peroxide in methanol, and unspecific binding was blocked by adding 2% normal horse serum during preincubation and incubations in Tris-buffered solutions. The primary antibodies were detected with biotinylated secondary antibodies (Jackson ImmunoResearch) and the Vector-Elite ABC kit (Vector Laboratories). Following the diaminobenzidene reaction, sections were dehydrated and embedded in Epon. The plastic-embedded specimens were prepared for ultrathin sectioning following the same protocol as above.

CC Cell Populations Analysis
In slices of WT mice, the CR + , Tbr1 + , and GAD67-GFP + neurons in the CC were counted at E16.5 and E18.5 as the number of cells in the CC region from at least five slices per condition. To study the total neuron number through the CC, the values were reported as a percentage of the total number of the cells encountered within the same region of the CC and labeled by Hoechst staining. To study the neuronal subpopulations repartition, the values were reported as a percentage of the total number of labeled neurons encountered within the same region of the CC.
In slices of WT and Mash1 2/2 mice, the GABAergic interneurons in CC were counted as the number of neurons labeled for GABA per surface unit from at least five slices per condition.

Statistical Analysis
For all analyses, values from at least three separate experiments were at first tested for normality. Values that followed a normal distribution were compared using Student t-test or one-way ANOVA and Fisher t-tests. Values that did not follow a normal distribution were compared using Mann-Whitney and Kolmogorov-Smirnov nonparametric tests.

Atlas and Nomenclature
The nomenclature for callosal development is based on the Atlas of the Prenatal Mouse Brain [76]. On the basis of our results, we considered that the CC is divided into two sectors: the medial part is bordered dorsally by the IG and the longitudinal fissure, and ventrally by the GW and the dorsal limit of the septal area. The lateral part comprises the white matter bordered by the CCi superficially, and by the ventricular zone between the GW and the mediodorsal angle of the lateral ventricle towards the ventricular side. DiA crystals, respectively, in the frontal (CFr) and parietal (CPa) cortex of WT (G i -G iv ) and Mash1 2/2 (H i -H iv ) mice. Whereas callosal axons of the dorsal path (DiI-labeled) and of the ventral path (DiA-labeled) are segregated in the WT CC, they intermix and are misrouted in the mutant (arrowheads in [H ii -H iv ]). Schematic drawings of callosal fiber trajectories in the WT (G v ) and Mash1 2/2 (H v ) brain, respectively. Bar indicates 435 mm in (A i , B i , G i , and H i ), 160 mm in (A ii and B ii ), 110 mm in (C i , C ii , D i , D ii , E i , F i , G ii , G iii , G iv , H ii , H iii , and H iv ), and 70 mm in (E ii and F ii ). LV, lateral ventricle; SVZ, subventricular zone. Found at: doi:10.1371/journal.pbio.1000230.s005 (9.31 MB TIF) Figure S6 Expression of guidance factors and receptors in WT and Mash1 2/2 mice. In situ hybridization for EphA4 (A i -A ii and B i -B ii ), for ephrinB2 (C i -C ii and D i -D ii ), for EphB1 (E i -E ii and F i -F ii ), for Npn1 (G i -G ii and H i -H ii ), for Sema3C (I i -I ii and J i -J ii ), and for Slit2 (K i -K ii and L i -L ii ) mRNAs on coronal sections from E18.5 WT (A i -A ii , C i -C ii , E i -E ii , G i -G ii , I i -I ii , and K i -K ii ) and Mash1 2/2 (B i -B ii , D i -D ii , F i -F ii , H i -H ii , J i -J ii , and L i -L ii ) mice. Ephrins, Sema3C, and its receptor Npn1, as well as Slit2, all known to play a role in guidance of callosal axons are normally expressed in the Mash1 2/2 embryos. Bar indicates 600 mm in (A ii , B ii , C i , C ii , D i , D ii , E i , E ii , F i , F ii , G ii , H ii , I ii , J ii , K i , K ii , L i , and L ii ) and 300 mm in (A i , B i , G i , H i , I i , and J i ). CFr, frontal cortex; LV, lateral ventricle. Found at: doi:10.1371/journal.pbio.1000230.s006 (9.83 MB DOC) Figure S7 GABAergic interneurons of the CC are required for correct callosal axon navigation. (Ai) Experimental paradigm used to confirm the growth of E16.5 GFP + WT callosal axons in slices from WT mice. To this end, GFP + WT frontal cortex (CFr) is transplanted in a WT slice. (Aii-Aiii) GFP immunocytochemistry showing that WT GFP + callosal axons grow normally and cross the midline when they are exposed to a WT environment. (Bi) Experimental paradigm used to test whether the presence of cortical GABAergic interneurons is necessary to direct the growth of callosal axons. To this end, GFP + Mash1 2/2 CFr is transplanted in a WT slice. (Bii-Biii) GFP immunocytochemistry showing that Mash1 2/2 GFP + callosal axons grow normally and cross the midline when they are exposed to a wild-type environment. (Ci) Experimental paradigm used to test whether the CC GABAergic interneurons are required for normal growth of callosal axons. To this end, GFP + WT cortex is transplanted in a Mash1 2/2 slice missing GABAergic interneurons. (Cii-Ciii) GFP immunocytochemistry showing that GFP + callosal axons of WT cortical explants do not cross the CC midline, but rather form Probst bundles (PB, open arrowhead). Bar indicates 435 mm in (Aii, Bii, and Cii) and 220 mm in (Aiii, Biii, and Ciii). and Npn-1 (Ci and Di) and for GFAP and L1 (C ii and D ii ) in coronal CC sections from E17.5 WT (C i -C ii ) and Sema3C 2/2 (D i -D ii ) mice. (C i -C ii ) At E17.5, the hemispheres of the WT brain have fused, allowing both callosal and hippocampal commissure (HIC) fibers labeled with Npn-1 and L1 to cross the midline and project into the contralateral cortex. (D i -D ii ) By contrast, in Sema3C 2/2 , the hemispheres do not fuse properly, and the callosal fibers do not cross the midline, but instead form ectopic bundles of axons on either side of it in the IG, reminiscent of Probst bundles (PB; arrowheads in [D i and D ii ]). (C i and D i ) CR-positive neurons distribute normally in the Sema3C 2/2 CC. (C ii and D ii ) In control and Sema3C 2/2 brains, glial cells distribute similarly in the IG and GW, and extend radial processes from the lateral ventricles towards the midline. Bar indicates 220 mm in (B i , C i , C ii , D i , and D ii ), 110 mm in (A i and B ii ), and 70 mm in (A ii ). LV, lateral ventricle; SVZ, subventricular zone; VZ, ventricular zone. Found at: doi:10.1371/journal.pbio.1000230.s008 (8.63 MB TIF)

Supporting Information
Video S1 The neurons of the CC cooperate to form complex neuronal network. Video S1 was coded using DivX codec. The DivX coded for Windows can be downloaded free on the following site: http://www.divx.com/divx/windows/. Downloading this allows the viewing of video S1. Downloading video S1 can take around 30 s. The geometry of the cell-free spaces and the cellular environment of the isosurface for the three stains (CR, Mash1-GFP, and Hoechst) are explored with the navigator function of IMARIS 4.3 software. The 3-D reconstruction is reoriented according to the direction of axonal growth within the CC. Exploring the thickness of the 3-D stack imaging of the CC reveals that the CR-positive neurons (red) and the GFP-positive GABAergic interneurons (green) form adjacent rows of neurons regularly distributed within the entire CC. Rows of neurons are separated by cell-free domains of small or large diameter, which likely segregate the callosal axons into bundles. This network of cell-free domain could define preferred pathways for the navigation of callosal axons. This analysis shows that the two types of neurons that transiently populate the CC form a complex neuronal network and are well positioned to interact with ingrowing callosal axons. Found at: doi:10.1371/journal.pbio.1000230.s009 (10.00 MB AVI)