Mutations in the planar cell polarity genes CELSR1 and SCRIB are associated with the severe neural tube defect craniorachischisis

Authors


  • Communicated by Iain McIntosh

Abstract

Craniorachischisis (CRN) is a severe neural tube defect (NTD) resulting from failure to initiate closure, leaving the hindbrain and spinal neural tube entirely open. Clues to the genetic basis of this condition come from several mouse models, which harbor mutations in core members of the planar cell polarity (PCP) signaling pathway. Previous studies of humans with CRN failed to identify mutations in the core PCP genes, VANGL1 and VANGL2. Here, we analyzed other key PCP genes: CELSR1, PRICKLE1, PTK7, and SCRIB, with the finding of eight potentially causative mutations in both CELSR1 and SCRIB. Functional effects of these unique or rare human variants were evaluated using known protein–protein interactions as well as subcellular protein localization. While protein interactions were not affected, variants from five of the 36 patients exhibited a profound alteration in subcellular protein localization, with diminution or abolition of trafficking to the plasma membrane. Comparable effects were seen in the crash and spin cycle mouse Celsr1 mutants, and the line-90 mouse Scrib mutant. We conclude that missense variants in CELSR1 and SCRIB may represent a cause of CRN in humans, as in mice, with defective PCP protein trafficking to the plasma membrane a likely pathogenic mechanism. Hum Mutat 33:440–447, 2012. © 2011 Wiley Periodicals, Inc.

Introduction

Neural tube closure is an event of early embryogenesis that is required for normal development of the brain and spinal cord. Primary neurulation in mammals begins with a de novo closure event at the boundary of the future hindbrain and cervical spine (so-called Closure 1) on day 22 postfertilization in humans [O'Rahilly and Muller, 2002] and embryonic day 8.5 in mice [Copp et al., 2003]. From this site, the neural tube “zips up” in a double wave of closure that spreads rostrally into the hindbrain region and caudally along the spine. Subsequently, closure in human embryos initiates separately at the rostral edge of the forebrain, generating a caudally directed wave of closure that meets the rostrally directed (hindbrain) wave to complete brain closure at the anterior neuropore [O'Rahilly and Muller, 2002]. In mice, there is a slightly more complex sequence of cranial closure events, with a second initiation site at the forebrain–midbrain boundary (Closure 2) and then bidirectional closure between this site and the rostral edge of the forebrain, and also between this site and Closure 1 [Golden and Chernoff, 1993]. In the spinal region of both humans and mice, closure is completed when zipping down the body axis reaches the upper sacral level, where the posterior (caudal) neuropore closes.

Defective closure during neurulation results in severe malformations of the central nervous system, termed neural tube defects (NTDs). Failure to complete low spinal closure causes spina bifida, whereas incomplete cranial closure results in anencephaly. These are common birth defects, affecting 0.5–2 per 1,000 pregnancies, worldwide [Botto et al., 1999]. The most severe NTD, craniorachischisis (CRN), arises earlier in development as a failure of Closure 1, leaving the neural tube open from the midbrain or rostral hindbrain to the base of the spine [Copp et al., 2003]. CRN is considered rare, although estimates of prevalence vary from 1/100,000 in Atlanta [Johnson et al., 2004] to 1/1000 in Northern China [Moore et al., 1997].

Despite the high prevalence of NTDs, the genes responsible for their largely sporadic occurrence have proven elusive. This likely reflects a complex inheritance pattern and an important contribution of nongenetic factors. Indeed, many genes are known to be essential for neurulation in mice [Harris and Juriloff, 2010], with increasing evidence of phenotypic modulation through gene–gene and gene–environment interactions [Copp et al., 2003]. The first mouse gene recognized as a cause of CRN was Vangl2, which is mutated in the loop tail mouse [Kibar et al., 2001b; Murdoch et al., 2001a] and encodes a key component of a β-catenin-independent, Wnt/frizzled signaling cascade, called the planar cell polarity (PCP) pathway [Strutt, 2008]. Subsequently, other PCP components were found to be essential for initiation of neural tube closure (Closure 1) in mice, including Celsr1, Dsh1/2, Scrib, Fz3/6, Ptk7, and Sec2b [Greene et al., 2009; Merte et al., 2010]. The developmental basis of this association is the requirement for convergent extension cell movements, which shape the early neural plate. Disturbance of PCP gene function in Xenopus [Wallingford and Harland, 2002] and Vangl2 mutant mice [Ybot-Gonzalez et al., 2007] both abolish convergent extension, producing a short, wide neural plate in which the neural folds are spaced too far apart to initiate closure.

Although putative mutations in VANGL2 (MIM# 600533) and its closely related paralog VANGL1 (MIM# 610132) have been reported in patients with several other types of NTD [Kibar et al., 2007; Lei et al., 2010], no evidence has been presented to suggest a role in human CRN. Indeed, our preliminary investigation of CRN cases did not reveal mutations in either VANGL1 or VANGL2 [Doudney et al., 2005]. In the present study, we have increased the size of our cohort to 36 patients-–to our knowledge, the only such cohort of this rare NTD anywhere in the world-–and have used DNA from these patients to investigate four further genes in the PCP signaling pathway: the core PCP gene CELSR1 (MIM# 604523), and the PCP-associated genes SCRIB (SCRB1; MIM# 607733), PTK7 (MIM# 601890), and PRICKLE1 (MIM# 608500). CELSR1, the ortholog of flamingo in Drosophila, is mutated in crash and spin cycle mice in which homozygosity yields CRN [Curtin et al., 2003]. Of the three PCP-associated genes, SCRIB causes CRN in both the circle tail [Murdoch et al., 2003] and line-90 [Zarbalis et al., 2004] mouse mutants, and PTK7 causes CRN in mice homozygous for a gene trap mutation [Lu et al., 2004] and in the ENU-induced chuzhoi mutant [Paudyal et al., 2010]. Neither of these genes has previously been implicated in human NTDs. In contrast, PRICKLE1, which is the ortholog of a Drosophila PCP gene, is required for regulation of apico-basal polarity in mouse epiblast [Tao et al., 2009] and was recently found to be mutated in patients with epilepsy [Tao et al., 2011]. Interestingly, rare variants have been reported among patients with open or closed spina bifida, although CRN cases were not examined [Bosoi et al., 2011].

Like VANGL1 and VANGL2 [Doudney et al., 2005], in the present study we did not detect putative mutations in patients with CRN in the genes for PRICKLE1 or PTK7. However, we do find putative heterozygous missense mutations in CELSR1 and SCRIB, prompting screening and identification of further sequence variants in a second series of cases. Comparative functional analysis with known pathogenic mouse mutations in Celsr1 and Scrib was then used to evaluate the likely effect of these human variants. We demonstrate, by in vitro assay of reporter constructs, a profound effect for several missense mutations of both genes on the subcellular localization of these normally membrane-associated proteins, indicating a possible mechanism resulting in loss of function.

Materials and Methods

Human Subjects

Ethical approval for the collection and analysis of the human DNA samples used in this study was granted by the Hammersmith and Queen Charlotte's and Chelsea Hospitals and the Institute of Child Health/Great Ormond Street Hospital Research Ethics Committees. The study included 36 fetuses with CRN of which 17 were collected at the Greenwood Genetic Center, South Carolina and three were from the Hammersmith and Queen Charlotte's and Chelsea Hospitals, London as previously described [Doudney et al., 2005]. A further 16 fetal samples affected with CRN were obtained from Groupe Hôpitalier Necker-Enfants Malades, Paris. Control DNAs were randomly selected from a total of 350 consecutive consenting normal pregnancies of white European ancestry recruited at Queen Charlotte's and Chelsea Hospital [Apostolidou et al., 2007]. Details of a variety of parameters such as sex and birth weight were collected for these controls, and none of the patients or close relatives had an NTD.

DNA Sequencing

Coding exons and flanking exon–intron boundaries were amplified by PCR from genomic DNA, for CELSR1 (NM_014246), SCRIB (NM_182706), PTK7 (NC_000006.11), and PRICKLE1 (NG_012965.1). Primer sequences are available on request. The PCR products were subjected to exonuclease I/shrimp alkaline phosphatase (Exo/SAP) treatment before sequencing with BigDye terminator v3.1 cycle sequencing kit (Applied Biosystem, Carlsbad, CA) and loading onto the ABI Prism 3100 Genetic Analyzer. Sequence analysis was performed using Sequencher software (V4.8, GeneCodes Corporation, Ann Arbor, MI) and putative variants were confirmed by resequencing both DNA strands. Sequence variants were investigated for population frequency in control DNAs (described above) as well as dbSNP132 (http://www.ncbi.nlm.nih.gov/snp) and (release 6) of the 1000 genomes project (http://browser.1000genomes.org/). Nucleotide numbering of the coding sequence variants described (designated c.) is based on cDNA sequences with +1 corresponding to A of the ATG translation initiation codon of the respective reference sequences (CELSR1: NM_014246.1 and SCRB1: NM_015356). Peptide sequence conservation was investigated using the web-based programs BLAST (http://blast.ncbi.nlm.nih.gov/Blast.cgi) and CLUSTAL X (http://www.ebi.ac.uk/Tools/msa/clustalw2/).

Expression Constructs

The full-length hSCRIB cDNA cloned into pEGFP-C1 expression plasmid was obtained from Dr. Patrick Humbert (Peter MacCallum Cancer Centre, Melbourne, Australia). FLAG-SCRIB fusion proteins were generated from the same full-length clone. Constructs expressing the putative SCRIB interacting proteins HA-VANGL2, HA-VANGL2Δ4, and GFP-LGL2 were obtained from Dr. Lelita Braiterman (Johns Hopkins University School of Medicine, Baltimore). The pCDNA3-Celsr1-GFP construct expressing full-length mouse Celsr1 cDNA (NM_009886) was kindly provided by Dr. Caroline Formstone (MRC Centre for Developmental Neurobiology, Kings College, London). SCRIB and CELSR1 missense changes were introduced to the respective constructs by PCR-mediated site-directed mutagenesis using PfuTurbo DNA polymerase (Stratagene, La Jolla, CA). For each point mutation, a pair of complementary mutagenesis oligonucleotide primers was designed (sequences available on request). In addition to DNA sequencing to confirm the introduction of a unique and correct change, constructs were tested for expression following cell transfection by western blotting.

Cell Transfection

MDCK (strain II) cells were cultured in minimum essential medium (MEM) (Invitrogen, Carlsbad, CA) supplemented with 5% (v/v) fetal bovine serum (FBS) and 2 mM glutamine, as recommended by the supplier (Health Protection Agency Culture Collections; HPACC). Between 8 and 12 hr before transfection, cells were seeded in 24-well TC plates (0.5 × 10 cells/well). Transfection was performed using Lipofectamine 2000 reagent (Invitrogen), according to the manufacturer's instructions. Twenty hours prior to localization studies, transfected MDCK cells were replated onto poly-L-lysine (10 µg/ml) coated glass coverslips (1.5×105cells/coverslip). Cells were examined for localization studies at 24, 48, and 72 hr posttransfection.

Protein Extracts, Co-purifications, and Western Blotting

Protein extraction was performed at 4°C. Cells harvested 24 hr posttransfection were washed in phosphate-buffered saline (PBS) and lysed in 1% (v/v) Nonidet P-40 buffer (150 µM NaCl, 10 µM Tris-HCl, pH 8, 1× complete protease inhibitor cocktail, in PBS) for 1 hr. Cell extracts were cleared by centrifugation at 3,000 × g for 20 min. For co-purification of FLAG fusion proteins and associated proteins, lysates were incubated with anti-FLAG agarose (Sigma) for 2 hr. The bound agarose was washed three times with 0.1% (v/v) Nonidet P-40 wash buffer (150 µM NaCl, 10 µM Tris-HCl; pH 7.8) before eluting the bound FLAG protein complex by the addition of FLAG peptide (Sigma). Co-purification or total lysate samples were resolved by 10% SDS-PAGE and subjected to western blot analysis with mouse monoclonal anti-FLAG antibody (M2; Sigma), rabbit polyclonal anti-GFP antibody (Invitrogen), or rat monoclonal anti-HA antibody (3F10; Roche). Bands were visualized by enhanced chemiluminescence (Amersham Biosciences, United Kingdom) (see Supp. Figs. S1 and S2).

Cell Localization Studies

For localization of wild-type and mutant SCRIB and CELSR1 proteins, cells plated on poly-L-lysine coated coverslips were fixed with 4% (w/v) paraformaldehyde on ice and permeabilized with ice-cold 100% methanol before being rehydrated in PBS. Samples were blocked in 1% (w/v) bovine serum albumin (BSA) in PBS, incubated with primary antibody for 1 hr, rinsed in PBS, and incubated with secondary antibody for 1 hr. Samples were washed with DAPI (4′,6-diamidino-2-phenylindole) and mounted in Mowiol 4-88 mounting medium (Sigma-Aldrich, MO; prepared with glycerol and 0.2 M Tris pH 6.8). Primary and secondary antibodies were diluted in 1% (w/v) BSA in PBS and all incubations were performed at room temperature. Primary antibodies used in this study were against SCRIB (goat, 1:300, sc-11049, Santa Cruz Biotechnology, Inc., Santa Cruz, CA), Celsr1 (rabbit, 1:800, a kind gift from Dr. Caroline Formstone), GFP (rabbit, conjugated with Alexa Fluor 488, 1:1000, A21311, Invitrogen; chicken, 1:300, A10262, Invitrogen), and goat immunoglobulins (rabbit, biotinylated, 1:250, E0466; DAKO, Glostrup, Denmark). Secondary antibodies used were Alexa Fluor 568 goat anti-rabbit IgG (A21069, Invitrogen), rabbit anti-chicken FITC conjugated (613111; Invitrogen), and fluorescent streptavidin (conjugated with Alexa Fluor 546, S11225, Invitrogen). Labeled cells were examined and quantified by epifluorescence on an inverted LSM710 confocal system mounted on an AxioObserver Z1 microscope (Carl Zeiss Ltd, United Kingdom). The images were acquired using 63× oil immersion objective. The thickness of optical sections was set at 0.2- to 0.9-µm thickness. The AlexaFluor 488 dye was excited by a 488 nm line of an Argon laser and AlexaFluor 546 and AlexaFluor 568 by a 561-nm diode laser. Z-projections of confocal stacks were created in ImageJ. Images were further processed in Photoshop CS3 (Adobe).

Cells over-expressing SCRIB and CELSR1 were initially examined for localization studies at 24, 48, and 72 hr posttransfection (Supp. Fig. S3). The 48 hr time point was found to be optimal for investigating membrane localization and was therefore used in all localization experiments detailed in this study. The subcellular localization of GFP expression was classified for each cell into one of the following three different categories: (1) GFP exclusively at the membrane; (2) GFP exclusively in the cytoplasm; (3) GFP at the membrane and in the cytoplasm. For each transfection, at least 200 cells were counted and each transfection was performed in triplicate (i.e., at least 600 cells were counted in total for each treatment). Quantitative data were analyzed by two-way analysis of variance to determine the statistical significance of comparisons between treatments (wild-type and the various mutant proteins) (Sigmastat v 3.5).

Results

The coding sequence and exon–intron boundary regions of CELSR1, PRICKLE1, PTK7, and SCRIB were sequenced in 20 fetuses with CRN, identifying a variety of synonymous and nonsynonymous coding sequence variants. Data reporting a lack of likely causative mutations in VANGL1 and VANGL2 in the same set of patients have previously been published [Doudney et al., 2005]. For PRICKLE1 and PTK7, two coding variants each (rs3747563–F633F and rs3747562–S634S; rs6905948–G617G and rs34764696–A777V) were identified, all present in controls and dbSNP132 (http://www.ncbi.nlm.nih.gov/sites/entrez). Although there is a possibility of reduced penetrance for some mutations, we considered these alleles less likely to be causative and not investigated further in this study.

Missense Changes Identified in CELSR1 and SCRIB

For CELSR1 and SCRIB, a variety of single nucleotide coding variants were detected. Nonsynonymous variants were investigated for their presence in dbSNP and the 1000 Genomes Project (release 6) data (http://browser.1000genomes.org/index.html) and/or frequency in white European controls (Supp. Tables S1 and S2). In the initial patient set (n = 20), two unique missense substitutions were identified in CELSR1: CRN3-c.2318C>T (A773V) and CRN19-c.8216A>C (N2739T) (Fig. 1A and Supp. Fig. S4). Three patients were each found to carry two missense changes: CRN8 (R2312P + S2964L), CRN3 (A773V + P2983A), and CRN6 (S2964L + P2983A). These include the changes c.6935G>C (R2312P); c.8891C>T (S2964L); and c.8947C>G (P2983A) which are reported as rare alleles in dbSNP (rs7287089, rs6008777, and rs61741871, respectively). The c.8947C>G (P2983A) change was also found to be heterozygous in three of 199 unaffected individuals, although none of the others were present in our control cohort. In addition, all of these variant residues were evolutionarily conserved in most mammalian species (except for the arginine at position R2312, which is a glutamine in mouse) but were poorly conserved with the invertebrate Drosophila (Supp. Table S1 and Supp. Fig. S5). Screening the second set of patients (n = 16) revealed overlap with the frequent variants but also revealed a third unique missense change located at a conserved CELSR1 residue, R2438Q (c.7313G>A) (Supp. Table S1).

Figure 1.

Schematic diagrams of CELSR1 and SCRIB proteins. Location of missense changes and mouse mutations with frequency in patient and control chromosomes are shown for (A) CELSR1 and (B) SCRIB relative to their functional domains.

For SCRIB, two unique changes were identified in patients CRN5 and CRN20, respectively, comprising a C>T transition at c.1360 (P454S) and a G>A transition at c.4604 (R1535Q) (Fig. 1B and Supp. Fig. S4). The proline residue at P454 was not conserved between species (Supp. Fig. S5). In those cases where a parent or additional family member was available for screening, we found evidence for inheritance, indicating that if causal, the mutations have incomplete penetrance (Supp. Table S3).

Functional Analysis of Missense Mutations: Co-immunoprecipitation

Functional analysis of previously reported PCP missense variants has relied on two main strategies: perturbation of normal protein–protein interaction, as detected by co-immunoprecipitation, and mislocalization of GFP-tagged proteins within polarized cells [Iliescu et al., 2011; Kibar et al., 2007; Lei et al., 2010]. In the present study, the first of these approaches was investigated for SCRIB only, since no protein–protein interactions had yet been described for CELSR1.

SCRIB is known to interact with both VANGL2 and LGL2 [Kallay et al., 2006]. We found that wild-type FLAG-tagged SCRIB could immunoprecipitate HA-tagged VANGL2 but not a similar construct lacking the NH3-terminal amino acids Glu-Thr-Ser-Val, which correspond to the PDZ-binding domain of VANGL2 (Fig. 2A). This strongly implicates binding through one or more of the four PDZ-binding sites in SCRIB. A series of missense mutations were then introduced into the wild-type construct using in vitro mutagenesis. This included the pathogenic mutation found in the line-90 mouse (I285K), the putative human mutations P454S and R1535Q, and several variants that are also present in dbSNP (G855E, G980R, and H1217P). None of the variants were found to ablate co-immunoprecipitation (Fig. 2B).

Figure 2.

Effect of missense variants on SCRIB-VANGL2 interactions by Co-Ip. A: Specific interaction of SCRIB with VANGL2 is through the PDZ-binding domain. Western blotting of the lysate with anti-FLAG and anti-HA demonstrates the presence of FLAG-SCRIB and HA-VANGL2 proteins, respectively. Following immunoprecipitation with anti-FLAG, full-length VANGL2 is pulled down but VANGL2Δ (lacking PDZ-binding domain) is not. B: A similar experiment was conducted with full-length HA-VANGL2 and FLAG-SCRIB constructs with various missense changes or wild-type sequence. None of the variants, including the mouse pathogenic change I285K affect the ability of SCRIB to pull down VANGL2.

Next, we investigated the interaction between SCRIB and LGL2 (lethal giant larvae 2), previously shown to occur via the leucine-rich repeat region of SCRIB [Kallay et al., 2006]. A similar co-immunoprecipitation experiment was performed as for HA-VANGL2, this time using GFP-tagged LGL2. Wild-type FLAG-SCRIB was able to pull down GFP-LGL2, as was the same set of SCRIB missense variants, including the line-90 mouse mutant (Fig. 3). These assays confirm the specific binding of SCRIB to both VANGL2 and LGL2 and confirm that none of the missense changes in SCRIB affect these protein interactions. Notably, this includes the known pathogenic mouse mutation I285K (line-90), demonstrating that SCRIB pathogenicity, in terms of causing CRN, can occur in the absence of alteration to protein–protein interactions.

Figure 3.

Effect of missense variants on SCRIB-LGL2 interactions by Co-Ip. Western blotting of the lysate with anti-FLAG demonstrates the presence of wild-type FLAG-SCRIB protein and mutant proteins with different missense changes. Following FLAG immunoprecipitation, blotting with anti-GFP confirms interaction between SCRIB and LGL2 with all of the missense variants tested.

Functional Analysis of Missense Mutations: Subcellular Localization

To assess PCP protein function through a more general assay, we investigated whether subcellular localization might be disrupted by the known pathogenic, and putative, mutations in the CELSR1 and SCRIB genes.

CELSR1 is a cadherin EGF LAG seven-pass G-type receptor transmembrane protein while the cytoplasmic scaffold protein SCRIB, although intracellular, localizes tightly to the plasma membrane of polarized epithelial cells (Fig. 1A). Therefore, as a means of assessing the effect of missense changes on protein trafficking, we used site-directed mutagenesis to introduce individual variants into either full-length mouse Celsr1 cDNA (NM_009886) expressed from a pCDNA3-Celsr1-GFP construct, or into full-length SCRIB cDNA (NM_015356) cloned downstream of the EGFP open reading frame in a pEGFP-C1 expression vector.

Clones were transfected into polarized MDCK epithelial cells (strain II) and subcellular localization was studied at 24-, 48-, and 72-hr time points, with 48 hr chosen for quantitative analysis of CELSR1 and SCRIB variants (Supp. Fig. S1). Cells were dual stained with anti-GFP and either anti-CELSR1 or anti-SCRIB. The subcellular localization of GFP expression was classified for each cell into one of the following three different categories: (1) GFP exclusively at the membrane; (2) GFP both at the membrane and in the cytoplasm; (3) GFP exclusively in the cytoplasm.

A characteristic pattern of localization was observed at the plasma membrane for wild-type CELSR1, with very few cells showing diffuse signal in the cytosol (Fig. 4A and B). CELSR1 missense variants S2964L, R2438Q, and P2983A resulted in a significantly reduced frequency of membrane localization, closely parallel to the effects of the crash (D1040G) and spin cycle (N1110K) mouse pathogenic mutants [Curtin et al., 2003] (Fig. 4A and B). The A773V variant showed a less dramatic, although statistically significant, reduction in membrane localization, whereas the reduction observed in the R2312P and N2739T variants did not reach statistical significance.

Figure 4.

Effect of putative missense mutations in CELSR1 on subcellular localization. A: Protein localization in MDCK cells: Top = immunofluorescence with anti-GFP (green) and anti-CELSR1 (red); Bottom = anti-GFP alone. B: Quantitative analysis of protein localization for GFP +ve cells were scored as cytoplasmic only, plasma membrane (PM) and cytoplasmic combined, or PM only. **Statistically significant differences of percentage cytoplasmic and percentage PM compared with wild-type (P < 0.001) were determined by two-way analysis of variance.

Both wild-type SCRIB, as expressed from the construct, and endogenous SCRIB exhibited strong plasma membrane localization (Fig. 5A and B). With the R1535Q variant, a significant increase was observed in the proportion of cells with cytosolic localization; very few cells showed specific accumulation at the cell membrane. This was similar to findings with the mouse pathogenic line-90 mutant (I285K) [Zarbalis et al., 2004] (Fig. 5A and B). In contrast, distribution of the nonconserved P454S variant mimicked that of wild-type.

Figure 5.

Effect of putative missense mutations in SCRIB on subcellular localization. A: Protein localization in MDCK cells: Top = immunofluorescence with anti-GFP (green) and anti-SCRIB (red); Bottom = anti-GFP alone. B: Quantitative analysis of protein localization for GFP +ve cells were scored as cytoplasmic only, plasma membrane (PM) and cytoplasmic combined, or PM only. **Statistically significant differences of percentage cytoplasmic and percentage PM compared with wild-type (P < 0.001) were determined by two-way analysis of variance.

We conclude that all three known pathogenic mouse mutations, and some but not all of the human CELSR1 and SCRIB variants, abolish the membrane association, a feature that characterizes wild-type proteins and is known to be required for PCP pathway function [Strutt, 2008].

Discussion

Our findings provide the first evidence for a causal association between human PCP mutations and CRN, a relationship that has been strongly established through studies of mouse mutants [Doudney and Stanier, 2005]. Indeed, it is remarkable that, of the more than 150 different genes, which cause NTDs when mutated in mice, only genes of the PCP signaling pathway are associated with the CRN phenotype [Harris and Juriloff, 2010]. Hence, the link between noncanonical Wnt signaling and failure of Closure 1 at the onset of mouse neurulation appears highly specific.

We did not detect any effects of the SCRIB pathogenic or putative mutations on the interaction of SCRIB protein with VANGL2 or LGL2 [Kallay et al., 2006]. A similar assay of the CELSR1 missense changes was not possible, owing to a lack of current knowledge of CELSR1s binding partners. The lack of effect of missense genomic alterations on protein–protein interactions could reflect specificity for the particular site of protein contact, in this case SCRIBs PDZ domain in the interaction with VANGL2, and its LRR domain in the interaction with LGL2 [Kallay et al., 2006]. The SCRIB missense changes we detected in CRN patients were all outside these functional domains, hence perhaps explaining the lack of effect on protein partner binding.

In contrast, use of a protein subcellular localization assay in polarized MDCK cells, showed a clearly detrimental effect of many of the putative mutations in both CELSR1 and SCRIB. This suggests that one likely pathogenic mechanism of PCP mutation in mammals may be through impaired protein targeting to the plasma membrane. The disruption of membrane localization by the CELSR1 and SCRIB mutations is similar to that identified for loss-of-function missense mutations in mouse Vangl2, including the Lp mouse mutant S464N [Gravel et al., 2010] and two other mutant alleles, D255E and R259L [Iliescu et al., 2011]. Here, for the first time, we show that the mouse Celsr1 and Scrib mutations, found in the crash, spin cycle, and line-90 alleles, all behave similarly in disrupting membrane localization. This suggests a common mechanism for missense mutations in membrane-associated PCP genes and provides a plausible mode of action to explain the effect of the missense changes found in our human CRN cohort.

Work in Drosophila has shown that the “core” PCP proteins adopt distinctive asymmetric plasma membrane localization, during establishment of planar polarity. The two transmembrane proteins, Flamingo (Celsr) and Frizzled, the latter receiving external Wnt signals, localize specifically to one side of cell–cell junctions. Intracellularly, these proteins associate with the cytoplasmic protein Dishevelled, which is essential for transduction of the PCP signal. On the opposite side of the cell junction, but in contact with Frizzled-Celsr via the extracellular space, is a complex of Vangl and Celsr membrane-localized proteins, in association with cytoplasmic Prickle. Such asymmetric complexes between cells may enable polarized cell–cell communication, during the establishment of long-range planar polarity within a tissue [Strutt, 2008]. While there is accumulating evidence, as in the present study, for plasma membrane localization of PCP proteins also in mammalian cells, there is little indication that asymmetric complexes are formed as in Drosophila. This may point to a difference in functioning of the PCP pathways between insects and mammals, or could merely reflect a limitation on subcellular protein imaging in mammalian cells.

Initiation of mammalian neural tube closure appears uniquely sensitive to loss of function in the PCP pathway, owing to the need for this pathway in shaping the embryo prior to neurulation [Colas and Schoenwolf, 2001]. During the late stages of gastrulation, the elliptical gastrula is converted into the keyhole-shaped neurula largely as a result of the cell movements of convergent extension, which involve lateral to medial displacement of cells in the plane of the ectoderm, and in the underlying mesoderm [Keller, 2002]. Cell intercalation in the midline leads to rostro-caudal extension and narrowing of the body axis, which ensures the neural folds are sufficiently closely spaced for closure [Wallingford and Harland, 2002; Ybot-Gonzalez et al., 2007].

Heterozygous mutation of PCP genes diminishes pathway function in mice, as demonstrated by the predisposition of Vangl2Lp/+ embryos to develop CRN in the presence of concentrations of Rho kinase inhibitor that are too low to affect wild-type littermates [Ybot-Gonzalez et al., 2007]. Moreover, mice that are compound heterozygotes for two independent Vangl2 mutant alleles develop CRN in a manner indistinguishable from mice homozygous for a single mutation [Kibar et al., 2001a]. The present data suggest a similar dependence of neural tube closure initiation on PCP function in human embryogenesis, where 5 of 36 of our CRN cases carry putative mutations with deleterious effects on protein subcellular localization of CELSR1 or SCRIB. It therefore seems likely, that diminution of PCP pathway function, perhaps in association with other (as yet unknown) genetic defects, leads to failure of closure initiation resulting in CRN.

In this study, we used a resequencing strategy to identify putative mutations in a targeted patient cohort, based on the assumption that causative alleles will be extremely rare or unique to a given family. These alleles are therefore not comparable with typical association studies since they cannot be part of a common haplotype shared with other unrelated patients who are affected with the same phenotype. Our model is essentially an autosomal dominant mode of inheritance with reduced penetrance, where penetrance is modified by the genetic background (i.e., a second compounding allele) or other nonspecific environmental factors. Mouse studies demonstrate the potential for CRN inheritance to be strongly influenced by interactions at multiple loci. For example, all three pairwise combinations of Vangl2, Celsr1, and Scrib heterozygous mutations cause CRN [Iliescu et al., 2011; Murdoch et al., 2001b]. Since convergent extension cell movements appear defective in homozygotes of all three of these genes, it seems likely that partial diminution of PCP signaling at different levels of the pathway can have additive effects, leading to overall loss of function. It is, however, interesting to note that when Vangl2 interacts with the PCP gene Ptk7, or with the non-PCP genes Cthrc1, Grhl3, or Cobl, the resulting defect is spina bifida or anencephaly, but not CRN [Carroll et al., 2003; Lu et al., 2004; Stiefel et al., 2007; Yamamoto et al., 2008]. Clearly, there is significant diversity of genetic interactions, and embryonic mechanisms, in determining the precise pattern of neural tube closure disruption [Copp and Greene, 2010]. It will be interesting to determine whether CELSR1 and/or SCRIB mutations are found in non-CRN NTDs, as has previously been reported for VANGL1, VANGL2, and PRICKLE1 [Bosoi et al., 2011; Kibar et al., 2009, 2007; Lei et al., 2010].

Extrapolating from the mouse studies, it seems possible that human CRN cases may involve combined heterozygous mutation of two or more genes each affecting PCP signaling, while spina bifida and anencephaly are more likely to involve a combination of heterozygous PCP and non-PCP mutations (or entirely non-PCP gene defects). Use of high-throughput genomic or exomic deep sequencing will enable testing of this hypothesis by identifying combinatorial hits in interacting PCP and non-PCP genes in individuals with varying NTD phenotypes.

Acknowledgements

We thank Gudrun Moore for control DNA samples, Cindy Skinner and Sayeda Abu-Amero for assistance with sample preparation and sequencing, and Bertrand Vernay for technical support with confocal microscopy.

Ancillary