Ccm3, a gene associated with cerebral cavernous malformations, is required for neuronal migration

Neuronal migration is one of the fundamental processes governing central nervous system development. In the mammalian cerebral neocortex, projection neurons are generated by progenitor cells residing in the ventricular and subventricular zones (VZ and SVZ, respectively) of the dorsal telencephalon and segregate by radial migration, eventually settling into distinct cell layers according to their birthdates (Angevine and Sidman, 1961). To migrate, early-born neurons send their leading processes directly to the marginal zone (MZ) whereas late-born neurons depend on radial glia, a transient cell type whose cell soma is in the VZ and whose long basal process spans the cerebral wall. Radial glia cells serve as both neural progenitors and a dynamic scaffold for migrating neurons (Rakic, 1972; Miyata et al., 2001; Malatesta et al., 2003; Noctor et al., 2004; Yokota et al., 2010).

Impaired neuronal migration invariably leads to cortical lamination defects and has been associated with mutations in several genes (e.g. Deuel et al., 2006; Koizumi et al., 2006; Bilgüvar et al., 2010), many of which encode components of two well-investigated pathways. The first consists of reelin, its receptors Vldlr and Lrp8 (also known as ApoER2), and its effector Dab1 (Rice and Curran, 2001); the second comprises Cdk5 and its neural-specific activators p35 (Cdk5r1 - Mouse Genome Informatics) and p39 (Cdk5r2 - Mouse Genome Informatics) (Dhavan and Tsai, 2001; Su and Tsai, 2011) and appear to regulate, respectively, radial glia-independent somal translocation and radial glia-guided locomotion (Gupta et al., 2003; Ohshima et al., 2007; Franco et al., 2011).

Cerebral cavernous malformation 3 [CCM3; also known as programmed cell death 10 (PDCD10)], is one of three genes [krev1 interaction trapped gene 1 (KRIT1) (also known as CCM1), CCM2 and PDCD10] (Riant et al., 2010) that are mutated in familial CCM, a common neurovascular disorder characterized by the presence of cerebrovascular lesions (Russell and Rubinstein, 1989; Ozturk et al., 2011). CCM3 is a cytoplasmic protein that can be found in a complex with CCM1 and CCM2 (Hilder et al., 2007; Voss et al., 2007), and is thought to promote assembly of the Golgi apparatus in non-neuronal cell lines (Fidalgo et al., 2010; Kean et al., 2011). CCM1, CCM2 and CCM3 are expressed in endothelial cells and are essential for vascular development (Whitehead et al., 2004; Boulday et al., 2009; Kleaveland et al., 2009; Whitehead et al., 2009; He et al., 2010; Zheng et al., 2010; Chan et al., 2011; Yoruk et al., 2012). They are also expressed in pyramidal neurons and astrocytes (Guzeloglu-Kayisli et al., 2004; Seker et al., 2006; Tanriover et al., 2008); in cultured neurons, CCM3 is found throughout the cell body and processes, but is not enriched in the Golgi (Lin et al., 2010). Loss of CCM3 in neural progenitors has cell-autonomous (astrocyte activation) and non-autonomous effects (diffusely dilated cerebrovasculature and lesion formation) in postnatal brain (Louvi et al., 2011). CCM3 downregulation in cell lines and primary astrocytes enhances proliferation and cell survival (Chen et al., 2009; Louvi et al., 2011). For an overview of current models of CCM protein function, see Draheim et al. (Draheim et al., 2014).

We examine the role of CCM3 in cortical development by analyzing conditional mutants that either lack CCM3 in neural progenitors and their progeny (hGfap-Cre;Ccm3^lox/lox and Emx1-Cre;Ccm3^lox/lox) or retain CCM3 in radial glia and SVZ progenitors but not in postmitotic neurons in the IZ and cortical plate (NEX-Cre;Ccm3^lox/lox). We show that CCM3 is specifically required in neural progenitors and nascent pyramidal neurons that migrate through the SVZ and uncover cell-autonomous and non-autonomous functions in neuronal migration. Cortical lamination defects are underlain by radial glia abnormalities and resemble those of Cdk5 and p35 mutants, suggesting a possible involvement of CCM3 in radial glia-dependent locomotion. We further demonstrate that CCM3 is necessary for cytoskeletal remodeling of postmitotic neurons, and that loss of CCM3 in neural progenitors results in RhoA activation in the brain.

In embryonic forebrain, Ccm3 mRNA is expressed in the VZ and SVZ, which contain progenitor cells and newly postmitotic neurons migrating through the SVZ towards the cortical plate (Fig. 1A,B; supplementary material Fig. S1A-E,K,M,N). We crossed Ccm3^lox/lox animals (He et al., 2010; Louvi et al., 2011) with Emx1-Cre (Gorski et al., 2002) and hGfap-Cre (Zhuo et al., 2001; Malatesta et al., 2003) mice to target neural progenitors and their progeny, respectively, in dorsal telencephalon at E9.5, or broadly in the CNS starting at E13.5. Loss of CCM3 expression in the conditional mutant animals was confirmed by in situ hybridization, immunostaining and western blotting (supplementary material Fig. S1E-G and data not shown) (Louvi et al., 2011).

We previously reported large brains or neocortices in hGfap-Cre;Ccm3^lox/lox and Emx1-Cre;Ccm3^lox/lox postnatal and adult mice (henceforth hGfap/Ccm3 cKO and Emx1/Ccm3 cKO) (Louvi et al., 2011). Brain weights of cKO animals were indistinguishable from controls in the first postnatal week [weight average±s.d.: control, 0.168±0.01789 g versus hGfap/Ccm3 cKO, 0.155±0.00707 g (P>0.05, P0)]; control, 0.284±0.01819 g versus hGfap/Ccm3 cKO, 0.29±0.01732 g (P>0.05, P6)]. No differences in cortical thickness between cKO and control animals were noticed at birth (P>0.05), which significantly increased from P7 onwards (Emx1/Ccm3 cKO, P7; P=0.01034, n=3; hGfap/Ccm3 cKO; P=0.01332, n=3). In Nissl-stained preparations, total neuron number in matched areas of somatosensory cortex (SSC) was similar in control and Emx1/Ccm3 cKO (P7 and P15, n=2 per age, P>0.05). To estimate neuronal cell size, we measured cell density in a defined area of SSC and calculated the mean cell area (field area per number of neurons); in Emx1/Ccm3 cKO brains, the number of neurons was indistinguishable from controls (P7 and P15; n=2 per age) (e.g. mean neuronal cell area±s.d.: P7 control, 361.28291±53.37218 μm² versus cKO, 346.97748±42.51494 μm²; P=0.51575).

Cortical organization was visualized using Nissl staining (supplementary material Fig. S1O-R) and in situ hybridization with layer-specific markers [Ctgf (subplate); Tle4 (subplate and L6); Er81 (Etv1 - Mouse Genome Informatics) (L5); Rorb (L4) and Cux2 (L2-4)] indicated neuronal migration defects. Cortical lamination was severely disrupted in Emx1/Ccm3 cKO mutants: subplate neurons occupied a diffuse band near the middle of the neocortex (Fig. 1C,D,M); L5 and L6 neurons were displaced to the upper half (Fig. 1E-H, Fig. 2B,E), and those of L2-4 were dispersed either across almost its entire depth (L4) (Fig. 1I,J) or in the bottom two-thirds (L2/3) (Fig. 1K,L, Fig. 2H,K). These abnormalities were exacerbated in the significantly thickened cingulate cortex, persisting in mature animals (supplementary material Fig. S2A-J), suggesting that they were not due to developmental delay. Notably, the positions of deep layer neurons relative to each other were as in wild type, with L5 neurons located superficially to those of L6 (Fig. 1E-H); however, L2-4 neurons failed to migrate past them. Although layer order was maintained in hGfap/Ccm3 cKO mutants, where recombination is initiated after formation of the cortical plate and generation of early-born neurons from Ccm3-expressing progenitors, many late-born L2-3 neurons were broadly distributed and, consequently, layer borders were blurred (Fig. 2A,D,G,J; supplementary material Fig. S3A-J).

To examine the molecular properties of the Emx1/Ccm3 cKO neocortex, we analyzed whether Satb2 and Ctip2 (Bcl11b - Mouse Genome Informatics), which mark non-overlapping populations of upper layer and L5 neurons (Arlotta et al., 2005; Alcamo et al., 2008; Britanova et al., 2008), became abnormally co-expressed. The number of cells positive for either marker was comparable in control and cKO at P0, P3 and P8 (n=2; P>0.05 for each age); however, their laminar distribution varied significantly [e.g. the upper one-third of the mutant neocortex at P3 contained fewer Satb2⁺ cells (P=0.02242) but more Ctip2⁺ cells (P=0.00801) than the control]. Nevertheless, in mutants, as in controls, only a small fraction of cells co-expressed both [respectively, 3.68% versus 5.02% (P3); 0.7% versus 0.33% (P8)], indicating that, despite ectopic positioning, mutant neurons were not molecularly respecified. Reelin-expressing Cajal-Retzius cells that could potentially contribute to cortical lamination abnormalities were normal (not shown).

We also generated hGfap/Ccm3^Delta/lox and Emx1/Ccm3^Delta/lox animals, carrying one null (Ccm3^Delta) and one conditional allele, by crossing Ccm3^Delta/lox (obtained by using ACTB-Cre) with hGfap- or Emx1-Cre;Ccm3^lox/+ mice. Cre is expected to target more efficiently the single floxed allele, decreasing mosaicism. Lamination defects were more pronounced in hGfap/Ccm3^Delta/lox and Emx1/Ccm3^Delta/lox, respectively, compared with hGfap/Ccm3 cKO and Emx1/Ccm3 cKO mutants (supplementary material Fig. S2K-T, Fig. S3K-T), representing an allelic series of different strengths (hGfap/Ccm3 cKO <hGfap/Ccm3^Delta/lox <Emx1/Ccm3 cKO <Exm1-Ccm3^Delta/lox). These observations suggest that CCM3 is necessary for neuronal migration and that the severity of cortical lamination defects depends on developmental timing and extent of Ccm3 inactivation in neural progenitors.

Our findings do not address whether CCM3 is independently required in radial glia and/or postmitotic neurons because, differences in onset of recombination notwithstanding, hGfap-Cre and Emx1-Cre lead to gene inactivation in both. We used NEX-Cre (Goebbels et al., 2006) to target postmitotic neurons in dorsal telencephalon. NEX-Cre;Ccm3^lox/lox (NEX/Ccm3 cKO) embryos maintained Ccm3 expression in VZ/SVZ progenitors (supplementary material Fig. S1H-N); the NEX-Cre and Ccm3 expression domains overlap in the outer SVZ at E13.5 suggesting that in early phases of NEX-Cre action, at least some Cre-positive, presumably nascent, neurons retain residual Ccm3 expression (supplementary material Fig. S1H,K,L). From E14.5 onwards, Ccm3 mRNA levels were clearly reduced away from the VZ (supplementary material Fig. S1M,N). NEX/Ccm3 cKO and NEX/Ccm3^Delta/lox mice were born at expected Mendelian ratios; brain size and cortical thickness (NEX/Ccm3 cKO, P14, n=3) were indistinguishable from controls (P>0.05) and lamination was normal (Fig. 2C,F,I,L; supplementary material Fig. S4A-H), suggesting that CCM3 is required in radial glia progenitors but is dispensable in postmitotic neurons in the IZ and cortical plate for migration and positioning (Fig. 2). To further test this, we inactivated Ccm3 in migrating neurons by in utero electroporation (IUE) using NeuroD-CreER^T2 to drive expression selectively in neurons (supplementary material Fig. S5A,B). IUE of the neocortical wall of floxed Ccm3 (Ccm3^lox/lox) embryos at E13.5 with NeuroD-CreER^T2 and a Cre-responsive GFP-expressing construct (Stop-GFP) (Shim et al., 2012) indicated that co-electroporated neurons migrated normally, similar to those co-electroporated with pCAG-Cre (Matsuda and Cepko, 2007), driving expression in all cells, and with Stop-GFP, which served as controls (supplementary material Fig. S5C,D), consistent with observations in NEX/Ccm3 cKO animals. Therefore, CCM3 deficiency in radial glia proper and likely nascent neurons in the VZ/SVZ is largely responsible for the migrational defects in hGfap/ and Emx1/Ccm3 cKO mutants.

In Emx1/Ccm3 cKO mutants positioning of late-born neurons, which depend on radial glia for migration, was disrupted, prompting us to examine the radial glia scaffold. Wild-type radial glia fibers labeled with rat-401 (nestin) spanned the cortical wall and were aligned perpendicular to the surface where end-feet anchored, but were irregular, wavy, less dense (often crisscrossing each other) and had fewer surface contacts in cKO mutants (Fig. 3A-D). DiI application onto the surface to label radial glia processes and their somata in the VZ (Malatesta et al., 2000) showed that fibers spanned the cortical wall at E13.5 in hGfap/Ccm3 cKO mutants (onset of recombination), but were disorganized at E15.5; hardly any cell bodies were labeled, suggesting that not many processes contacted the MZ (Fig. 3E-G); in the VZ, radial glia appeared normal (Fig. 3J,K). Staining of filamentous actin (F-actin) with falloidin revealed the characteristic honeycomb pattern at the ventricular surface in hGfap/Ccm3 cKO and Emx1/Ccm3 cKO. Electron microscopy showed that the basal lamina was intact in Emx1/Ccm3 cKO (supplementary material Fig. S6A,B); however, radial glia fibers oriented towards the surface were difficult to identify in upper cortical plate (Fig. 3H,I; supplementary material Fig. S6G,H), where neuronal cell bodies were irregularly arranged (supplementary material Fig. S6C,D,I,J). Mitotic figures in the VZ/SVZ and adherens junctions between VZ progenitors were normal (supplementary material Fig. S6E,F and Fig. S7A-C). These observations suggest that CCM3 regulates the morphology of radial glia basal processes.

To further test that CCM3 acts cell-autonomously in neural progenitors, we performed IUE of Ccm3^lox/lox mice at E14.5 with pCAG-Cre and pCAG-GFP. In electroporated, GFP-expressing, cells and their descendants, Ccm3 is inactivated as a consequence of Cre expression; non-electroporated cells retain Ccm3 expression. Analyses of mice at E16.5 (2 days post-IUE) revealed many GFP⁺ cells still in the VZ, suggesting that CCM3 is required for proper migration of newborn neurons; in embryos electroporated with pCAG-GFP, the majority of GFP⁺ cells migrated away from the VZ (Fig. 4A-C). Furthermore, IUE with pCAG-Cre and Stop-GFP at E13.5 showed that fewer neurons migrated into the cortical plate 3 days post-IUE compared with those electroporated with pCAG-GFP (Fig. 4D-F), likely due to morphological defects in surrounding Cre-expressing radial glia. The majority of cells electroporated only with pCAG-GFP developed leading processes directed towards the surface, whereas Ccm3 mutant cells had short irregular processes (Fig. 4G-J). These observations suggest that CCM3 is cell-autonomously required for development of radial glia processes, and that migration of neurons associated with defective radial glia is cell non-autonomously affected.

We examined whether Ccm3 deletion affected division of radial glia by labeling dividing progenitors in late G2/M or S phase for phosphorylated histone H3 (PH3) or BrdU incorporation, respectively, at E12.5 to E15.5. At all stages, PH3⁺ cells were lining the ventricle, and, from E13.5 onwards, were also in the neocortical wall. No statistically significant differences in number of progenitors dividing in the VZ or at basal positions were found between control and Emx1/Ccm3 cKO or hGfap/Ccm3 cKO brains (n=3 per genotype and stage) (supplementary material Fig. S7A-C). The total number of BrdU⁺ cells (1-hour pulse) was similar in control and Emx1/Ccm3 cKO embryos at E13.5 (n=5 per genotype) and E15.5 (n=3) (Student’s t-test, P>0.05), and in control and hGfap/Ccm3 cKO embryos at E13.5 (1 or 3 hours, n=3), and their radial distribution was similar across all littermates analyzed (P>0.05, Mann-Whitney U test) (supplementary material Fig. S7D,E,H,I,L). Neuronal birthdating [hGfap/Ccm3 cKO: E14.5 to P10 (n=2), E14.5 to 7 months (n=1), E14.5 to P7 (n=2), E16.5 to P2 (n=2); Emx1/Ccm3 cKO: E13.5 to P7 (n=2), E14.5 to P0 (n=2), E17.5 to P7 (n=4), E18.5 to P7 (n=2)] revealed comparable numbers of BrdU⁺ cells in control and cKO brains (P>0.05 for all comparisons between groups), but significantly altered laminar distribution, in agreement with in situ hybridization analysis (supplementary material Fig. S7F,G,J-L). Staining with Satb2 or Ctip2 and BrdU demonstrated that the percentage of BrdU⁺ cells co-expressing either marker did not differ significantly between brains from control and cKO littermates, suggesting that production of different neuronal types is not altered in the Emx1/cKO mutants [E13.5 to P0 (n=4); E15.5 to P4 (n=5); P>0.05 for all comparisons] (supplementary material Fig. S7M). Staining with β-tubulin type III indicated no delays in neurogenesis (not shown). In cKO neocortex, therefore, progenitor proliferation and neurogenesis are not disrupted, neurons with the same birthdate occupy different relative positions compared with normal, and the cytoarchitectonic defects are independent of the time of neuron origin. Our observations, which are indeed consistent with classic studies of the reeler mouse, the archetype of cortical laminar defects (Caviness and Sidman, 1973), further suggest that increased neocortical (Emx1/Ccm3 cKO) or overall brain (hGfap/Ccm3 cKO) size - and the absence of such effects in NEX/Ccm3 cKO mutants - is likely to be due to overproliferation of astrocytes postnatally, a notion supported by significant and progressive increase of Gfap mRNA and protein levels (Louvi et al., 2011).

Defects in radial glia morphology, adhesion or polarity may or may not be associated with changes in neuronal and glial differentiation (Rasin et al., 2007; Cappello et al., 2012). We used IUE to introduce pCAG-GFP in the neocortical wall of littermates from crosses between Ccm3^Delta/lox and Emx1/Ccm3^lox/+ animals at E13.5 and analyzed neuronal morphology 3 days post-IUE by fluorescence microscopy. GFP⁺ cells were localized in the IZ and cortical plate in control embryos; however, in cKO mice, most arrested in the SVZ/IZ and only a few reached the lower cortical plate with leading processes extending toward the pia (Fig. 5A-D). Control neurons were multipolar at the pre-migratory zone, and bipolar at the upper IZ and cortical plate (Tabata and Nakajima, 2003; Noctor et al., 2004). A large proportion of Ccm3 cKO neurons (56.36±8.83%, n=3) in the migratory zone were abnormal, many failing to transition into bipolar morphology, displaying multiple very thin processes (Fig. 5E,F). GFP-expressing axons extended tangentially within the IZ in controls, but were stunted in cKO lacking defined trajectories (Fig. 5G,H).

We established primary cortical neuronal cultures from hGfap/Ccm3 cKO and Emx1/Ccm3 cKO at E13.5 to E15.5 and examined at 3, 5 and 7 days in vitro (DIV) (Fig. 5I-P; supplementary material Fig. S8). cKO neurons had multiple short neurites that resembled filopodia, revealed by Tuj1 or falloidin staining; many expressed MAP2, identifying them as dendrites, and were significantly shorter in the cKO neurons (Fig. 5I-N; supplementary material Fig. S8A-G); doublecortin (DCX) staining indicated defects in the microtubule cytoskeleton (Fig. 5O,P). Golgi morphology and deployment, in part regulated by the CCM3 interactor STK25 (Matsuki et al., 2010; Ceccarelli et al., 2011; Xu et al., 2012), were normal (not shown).

Selectively labeled (Golgi-Cox impregnation) adult control neurons were polarized, with a long, thick apical dendrite extending toward the pia and multiple smaller dendrites arising from the soma, whereas neuronal arrangement lacked a recognizable pattern and dendritic development was abnormal in Emx1/Ccm3 cKO and to a lesser degree in hGfap/Ccm3 cKO mutants (Fig. 5Q-V and not shown). Finally, SMI32 and MAP2 staining revealed abnormal dendrites in NEX/Ccm3 cKO neocortex (not shown).

These observations indicate a defect of multipolar to bipolar transition in mutant neurons in vivo, consistent with morphological abnormalities in primary cultures. They also suggest that CCM3 function is required for dendrite formation and that its loss disrupts the actin and microtubule cytoskeletons. This further indicates that, in addition to being essential in radial glia for neuronal migration, CCM3 is required cell-autonomously for neurite outgrowth.

Staining of axonal projections with 2H3 (Fig. 6A-H) revealed thin, ectopic and abnormally bifurcated bundles in Emx1/Ccm3 cKO (Fig. 6D,G,H) compared with controls (Fig. 6B,E,F). Cytochrome oxidase histochemistry (Fig. 6I-K) and Nissl staining (Fig. 6L-N) showed a disorganized (hGfap/Ccm3 cKO) or nearly absent (Emx1/Ccm3 cKO) barrel field (Fig. 6J,K,M,N), suggesting that post-synaptic L4 neurons failed to segregate. Thalamocortical axons immunostained with vGlut2 formed barrel-shaped bundles in control, were irregular in hGfap/Ccm3 cKO and failed to cluster in Emx1/Ccm3 cKO (Fig. 6O-Q). These observations suggest abnormal axonal projections.

CCM3 loss results in defects in neuronal migration and in actin and microtubule cytoskeletal remodeling, processes partly regulated by Rho GTPases (Heng et al., 2010; Govek et al., 2011; Kawauchi, 2011). Several observations suggest a possible interaction between RhoA and CCM (Crose et al., 2009; Whitehead et al., 2009; Borikova et al., 2010; Stockton et al., 2010; Zheng et al., 2010; Louvi et al., 2011; McDonald et al., 2011; Cappello et al., 2012; McDonald et al., 2012). RhoA and Ccm3 have similar expression in neocortex and are required in radial glia for neuronal migration (this study) (Cappello et al., 2012). Active, GTP-bound Rho was elevated 2.7-fold in hGfap/Ccm3 cKO mutants (E15.5, n=3) and 1.4-fold in hGfap/Ccm3^lox/+ heterozygotes compared with controls (n=2) (Fig. 7A,B); whole-genome microarray analysis indicated a 1.43553-fold (P=0.00103957) RhoA upregulation in hGfap/Ccm3 cKO over control neocortex (n=3; P2), suggesting that RhoA activation may underlie the migrational and cytoskeletal defects in cKO neocortex.

Emx1/Ccm3 cKO, Cdk5 and p35 mutants (this study) (Ohshima et al., 1996; Chae et al., 1997; Gilmore et al., 1998; Kwon and Tsai, 1998; Ohshima et al., 2007; Hoerder-Suabedissen et al., 2009) have nearly identical cortical lamination defects and strikingly similar Ctgf, Er81 and Cux2 profiles (Fig. 1D,H,L) (Ohshima et al., 2007; Hoerder-Suabedissen et al., 2009) raising the possibility of CCM3 interaction with the Cdk5/p35 pathway, which regulates actin and microtubule dynamics during neuronal migration, as well as dendritic development of pyramidal neurons (Xie et al., 2003; Kawauchi et al., 2006; Ohshima et al., 2007), processes also disrupted in the Ccm3 cKO mutants. Cdk5 modulates actin reorganization via RhoA suppression and regulation of cofilin activity (Kawauchi et al., 2006), and controls microtubule organization via FAK phosphorylation (Xie et al., 2003). A principal target of FAK is paxillin, to which CCM3 has been shown to bind (Li et al., 2010), and we previously detected highly abundant cofilin in lesions by RNA-Seq (FPMK average±s.d.: 101.6±7.3) (Louvi et al., 2011). Cofilin activity is regulated through phosphorylation (Ser 3); we observed an increase in inactive, phosphorylated cofilin in neocortical lysates from Emx1/Ccm3 cKO (1.2-fold at E13.5, n=3; 1.6-fold at P3, n=5) compared with controls, consistent with RhoA activation (Fig. 7C,D). Immunostaining showed increased p-cofilin immunoreactivity in the Emx1/Ccm3 cKO neocortex compared with control (Fig. 7E,F). By contrast, paxillin levels or phosphorylation were unaffected (not shown). These findings suggest that the Cdk5/RhoA/cofilin pathway is disturbed in the Ccm3 cKO mutants.

Here, we demonstrate that CCM3 is a crucial regulator of neuronal migration and that its inactivation in radial glia and nascent neurons produces severe neocortical malformations in a developmental time-dependent manner. CCM3 impacts on the actin and microtubule cytoskeletons, and is required for neurite outgrowth in vivo and in vitro. These functions appear to be mediated, at least partly, by RhoA activation.

The allelic series of Ccm3 cKO mutants display a spectrum of neuronal migration and positioning abnormalities (Fig. 8A-F). The normal cortical lamination of NEX/Ccm3 cKO and NEX/Ccm3^Delta/lox mutants implies that CCM3 is not required in late-migrating neurons in the IZ and cortical plate for radial glia-independent migration, or association with and detachment from CCM3⁺ radial glia. In agreement with genetic findings, NeuroD-Cre-mediated deletion of Ccm3 in postmitotic neurons has no obvious consequences on migration. Because at early stages of NEX-Cre action, Cre⁺ (presumably newborn) neurons retain some CCM3 due to the delay in the effects of recombination on protein levels, whether CCM3 action is required in early-migrating neurons in the SVZ remains unknown. However, our finding that neuronal morphology is cell-autonomously affected suggests a role for CCM3 in neurite outgrowth, a process indeed initiated in early-migrating pyramidal neurons in the SVZ (Shoukimas and Hinds, 1978; Polleux and Snider, 2010), lending support to the notion that they require CCM3.

Despite broad defects in Emx1/Ccm3 cKO and Emx1/Ccm3^Delta/lox mutants, L5 neurons still migrate past their predecessors, implying that migration of early-born neurons is not disturbed, even though both radial glia and neurons lack CCM3. By contrast, most L2-4 neurons, which largely employ glia-guided locomotion, fail to migrate past earlier-born neurons, pointing to CCM3-deficient radial glia as the underlying cause of migration defects. Considering that CCM3 is not required in late-migrating neurons, our findings suggest that CCM3 also has a cell non-autonomous function in radial glia, affecting interactions with postmitotic neurons during radial glia-guided migration, which are reminiscent of its cell non-autonomous functions in the neurovascular unit (Louvi et al., 2011). However, CCM3 also acts cell-autonomously, regulating the morphology of embryonic radial glia (this study) and of radial glia-derived postnatal astrocytes (Louvi et al., 2011). These cell-autonomous functions warrant further investigation, taking into consideration that CCM3 overexpression in vitro and in vivo causes cell death (Chen et al., 2009; Lin et al., 2010). Genetic rescue experiments that control for cell-type specific conditional re-expression of CCM3 at levels comparable to endogenous levels in mutant neocortex would need to be contrasted to the effects of focal and controlled overexpression of floxed CCM3 constructs introduced into the Ccm3 cKO neocortex by IUE. The latter approach will also allow us to observe the behavior of sparse ‘rescued’ cells in an otherwise mutant environment. Conversely, by overexpressing Cre recombinase in Ccm3^lox/lox embryos, the morphology and behavior of sparse Ccm3-mutant cells can be analyzed within the wild-type cortex, an extension of experiments reported herein. Performing IUE at various embryonic stages and analyzing the outcome at different times thereafter, will address whether the rescued neurons eventually settle at normal positions, thus allowing us to evaluate cell-autonomous versus cell non-autonomous functions of CCM3 in greater detail. Finally, it would also be informative to examine using mouse chimeras the relative contributions of CCM3-positive and -negative neurons in cortical layer development and positioning.

In hGfap/Ccm3 cKO and hGfap/Ccm3^Delta/lox mutants, Ccm3 is initially maintained in progenitors and early corticogenesis is normal. At the onset of recombination, and presumably briefly, radial glia become transiently mosaic for Ccm3 expression; as the exact timing of recombination (and, therefore, Ccm3 inactivation) for any given cell is unknown, a neuron will stochastically attach to a radial glia fiber that might, or might not, still express CCM3, and do this until Ccm3 has been inactivated in all progenitors. With the caveat of CCM3 protein stability unknown, the hGfap/Ccm3 cKO animals will be useful for addressing interactions between newborn neurons and mosaic radial glia. Finally, we conclude that CCM3-dependent migration is required for laminar positioning per se, because in hGfap/Ccm3^Delta/lox mutants, even though the preplate is properly split and early-born neurons migrate normally, late-born neurons are mislocalized, displacing upwards those born earlier.

The dual cell-autonomous and cell non-autonomous function of a cytoplasmic protein in radial glia is counter-intuitive. Intracellular mediators may translate extracellular cues to modifications of the actin and microtubule networks, as well as cell-cell adhesion, both of which are modulated during neuronal migration (Marín et al., 2010). Therefore, CCM3 could control radial-glia-dependent neuronal migration non-autonomously by transducing extracellular signals to the intracellular migration machinery. Notably, LIS1, NDEL1 and their signaling partner 14-3-3ε, cytoplasmic proteins that regulate positioning of late-born neurons, have significant cell non-autonomous functions in neuronal migration revealed via mosaic analysis with double markers (Hippenmeyer et al., 2010).

Ccm3 is one of few genes known to be required exclusively in neural progenitors for migration, notwithstanding that only select key players have been tested independently in radial glia versus postmitotic neurons. For example, Cdk5 is required cell-autonomously in both (Hirasawa et al., 2004; Ohshima et al., 2007), Dab1 is required only in migrating neurons (Franco et al., 2011), whereas β1 integrins (Belvindrah et al., 2007) and RhoA (Cappello et al., 2012) are radial glia specific. Ccm3 (this study) (Louvi et al., 2011) and Itgb1 mutants display neuronal migration and cortical lamination defects (Graus-Porta et al., 2001; Schmid et al., 2004; Huang et al., 2006; Marchetti et al., 2010), pial detachment of endfeet (Graus-Porta et al., 2001; Kwon et al., 2011) and reactive gliosis (Robel et al., 2009), suggesting a - possibly indirect - interaction that warrants investigation, especially given the association of CCM1 with ICAP1α, which binds the β1 integrin cytoplasmic tail (Zawistowski et al., 2002; Faurobert et al., 2013). On the other hand, RhoA is necessary in radial glia to stabilize the actin and microtubule cytoskeletons, and to maintain the radial glia scaffold (Cappello et al., 2012); thus, our findings that RhoA is activated in Ccm3 conditional mutants could partially explain the migration defects.

In future studies it will be important to test the hypothesis that CCM3 may generate signals in radial glia, which are then transmitted to neurons to facilitate migration, thus accounting for the cell non-autonomous effects we uncovered, to determine whether extracellular cues and signaling cascades regulate CCM3 activity to control radial glia-dependent locomotion and to elucidate mechanistic relationships with the Cdk5/p35 pathway.

Mice were maintained in compliance with National Institutes of Health guidelines and approval of the Yale University Institutional Animal Care and Use Committee. Ccm3^lox mutants were reported previously (He et al., 2010; Louvi et al., 2011). hGfap-Cre (Zhuo et al., 2001; Malatesta et al., 2003), Emx1-IRES-Cre (Gorski et al., 2002) and ACTB-Cre (Lewandoski et al., 1997) mice were purchased from JAX. NEX-Cre (Goebbels et al., 2006; Belvindrah et al., 2007) mice were a gift from Drs Schwab and Nave (Max Planck Institute for Experimental Medicine, Goettingen, Germany).

Embryonic and postnatal brains were fixed, respectively, by immersion in or intracardial perfusion with 4% paraformaldehyde (PFA), post-fixed in 30% sucrose in 4% PFA and sectioned on a cryomicrotome (Leica Microsystems, Wetzlar, Germany). Sections were processed for in situ hybridization as described previously (Tanriover et al., 2008). RNA probes complementary to mouse Ccm3 (Louvi et al., 2011), Pax6 (Götz et al., 1998), Tbr1 and Tbr2 (Englund et al., 2005), Tle4 (Yao et al., 1998), Etv1 (Weimann et al., 1999), Rorb (Nakagawa and O’Leary, 2003), Cux2 (Nieto et al., 2004; Zimmer et al., 2004) and Ctgf (Heuer et al., 2003) were labeled with digoxigenin-11-UTP. Sections were analyzed using a Stemi stereomicroscope or AxioImager (Zeiss, Oberkochen, Germany) fitted with an AxioCam MRc5 digital camera. Images were captured using AxioVision software (Zeiss) and assembled in Adobe Photoshop.

Fixed brains were cryoprotected in 30% sucrose in PBS, sectioned and processed free-floating. For 3,3′-diaminobenzidine (DAB) staining, sections were treated with 1% H₂O₂, washed in PBS and pre-incubated in blocking solution containing 5% normal donkey serum (Jackson ImmunoResearch Laboratories, West Grove, PA), 1% bovine serum albumin, 0.1% glycine, 0.1% L-lysine and 0.03% Triton X-100. Primary antibodies were added for 48 hours at 4°C, and secondary antibodies [raised in donkey (Jackson ImmunoResearch Laboratories; 1:250 dilution)] for 2 hours at room temperature; sections were processed with Vectastain ABC Elite (Vector Laboratories, Burlingame, CA) for 2 hours and developed using the DAB substrate kit for Peroxidase (Vector). For immunofluorescence, the H₂O₂ step was omitted; secondary antibodies were Alexa-Fluor conjugated (Molecular Probes; 1:500 dilution).

Primary neurons were fixed with 4% PFA for 15 minutes, washed with PBS, blocked, incubated with primary antibodies for 1 hour at room temperature, washed and incubated with Alexa-Fluor-conjugated secondary antibodies.

Rat-401 (1:100; developed by S. Hockfield, MIT, Cambridge, MA, USA) and 2H3 (1:300; developed by T. Jessell and J. Dodd, Columbia University, New York, USA) were both obtained from DSHB developed under the auspices of the NICHD and maintained by the University of Iowa, Department of Biological Sciences, Iowa City, IA. Other antibodies used were TUJ1 (1:500; Covance, Emeryville, CA; #MMS-435), MAP2 (1:1000; Sigma-Aldrich, St Louis, MO; #M4403), PDCD10/CCM3 (1:200; Sigma; #027095), DCX (C-18) (1:100; Santa Cruz Biotechnology, Santa Cruz, CA; #sc-8066), vGlut2 (1:300; Millipore, Billerica, MA; ab2251), Satb2 (1:300; Santa Cruz; #sc-81376), Bcl11b/Ctip2 (1:500; Abcam, Cambridge, MA; #ab18465), Cux1 (1:150; Santa Cruz; #sc-13024), phospho-cofilin (Ser-3) (1:1000; Cell Signaling Technology, Danvers, MA; #3311), phospho-Histone H3 (Ser-10) (1:2000; Cell Signaling) and BrdU (1:200; Beckton Dickinson, San Jose, CA; #347580).

Neocortices were dissected at P2 (n=3) and RNA was isolated using the RNeasy lipid tissue kit (Qiagen, Valencia, CA) and amplified using TotalPrep RNA Amplification kit (Applied Biosystems). Biotin-labeled cRNA for hybridization onto Illumina arrays was used according to Illumina protocols; arrays were scanned on the Illumina Iscan.

PFA-fixed neocortices were flattened and sectioned tangentially, and incubated with cytochrome C type III in the presence of DAB.

Golgi-Cox staining was performed using the FD Rapid GolgiStain Kit (FD Neuro Technologies, Ellicott City, MD) according to the manufacturer’s instructions.

Western blot was performed by standard methods. Antibodies used were cofilin (1:1000; Cell Signaling; #3312), phospho-cofilin (Ser-3) (1:1000; Cell Signaling; #3311) and PDCD10/CCM3 (1:300; Proteintech, Chicago, IL; #10294). ImageJ was used for quantification of bands.

Pregnant females were injected intraperitoneally with a solution of 5-bromo-2′-deoxyuridine (BrdU; 15 mg/ml in saline) at 20 mg/g of body weight. For embryonic stages (n=3 litters per genotype), litters were analyzed only if they contained a minimum of two control and two cKO littermates. At least three (and up to six) matched sections were analyzed per embryo.

To quantify the distribution of neurons, the postnatal neocortex was divided radially into 10 equal-sized bins from the pia to the upper edge of the white matter. For embryonic analyses, the pia to the VZ (E13.5 and E14.5) or the upper edge of the intermediate zone to the ventricular zone (E15.5 and older) was divided radially into ten bins. The cells in each bin were quantified and reported as the percentage of total cells counted.

Data were analyzed by two-tailed Student’s t-tests with a significance level of at least P<0.05 for all statistical comparisons. Numbers of replicates are given in the main text or figure legends. ImageJ/NeurphologyJ (Ho et al., 2011) was used for quantification of neuronal morphology.

To label radial glia, crystals of the lipophilic dye DiI (Molecular Probes/Invitrogen, Carlsbad, CA) were applied on the pial surface of fixed brains. The brains were stored in 4% PFA in the dark for 2-4 weeks to allow diffusion of the dye, then sectioned at a thickness of 100 μm using a vibratome.

Active GTP-bound Rho was determined with a pull-down assay (Millipore, Temecula, CA; #17-294) according to the manufacturer’s instructions.

Primary cultures were established as described (Šestan et al., 1999) at plating density of 75×10³/cm² on laminin/poly-L-ornithine-coated glass coverslips in 24-well plates.

All surgeries were performed using sterile conditions as described previously (Saito and Nakatsuji, 2001) and in accordance with an IACUC approved protocol. Plasmids used were NeuroD-CreER^T2 (a gift from N. Šestan, Yale University, New Haven, CT, USA), an expression vector that consists of a fragment of the mouse NeuroD promoter driving Cre expression in postmitotic premigratory neurons; Stop-GFP, a Cre-responsive GFP-expressing construct (Shim et al., 2012); and pCAG-Cre (Matsuda and Cepko, 2007).

This article is dedicated to Marion Wassef (Régionalisation Nerveuse, Institut de Biologie de l’École Normale Supérieure, Paris, France) on the occasion of her recent retirement. We are grateful to N. Šestan for insightful discussions and comments, a critical suggestion, and for sharing the template for Fig. 8. We thank K. Kwan for discussions and comments on the manuscript; W. Han, K. Yasuno and S. Assimacopoulos for discussions; anonymous reviewers for suggestions; M. Schwab and K.-A. Nave for the NEX-Cre mice; K. Ishigame for excellent technical assistance; and M. Graham at the Center for Cell and Molecular Imaging (Yale School of Medicine) for help with electron microscopy.