The metalloproteinase inhibitor Reck is essential for zebrafish DRG development

During normal embryonic development, cells are often generated in locations far removed from their place of differentiation. Cell migration is important in the formation of many tissues, and the dysregulation of migration has drastic consequences for embryos and adult organisms alike. Neural crest is a transient cell lineage that gives rise to tissues including but not limited to: cartilage and bone of the craniofacial skeleton, secretory neuroendocrine cells, pigment cells, and many of the neurons and glia of the peripheral nervous system (Le Douarin and Kalcheim, 1999). Neural crest is specified as a limited cell population at the lateral border of neural and non-neural ectoderm; it must then migrate in regulated phases and expand to generate these derivatives. All vertebrate taxa possess neural crest, and the mechanisms that distinguish neural crest from all other cell types are considered to have been critical in the origin and evolution of the vertebrate clade (Gans and Northcutt, 1983; Knecht and Bronner-Fraser, 2002).

The relationship between neural crest differentiation and migration is poorly understood. Most studies have identified molecules crucial to early events in migration, such as delamination from the neuroectoderm or maintenance of neural crest segmentation. However, little attention has been paid to the cues that terminate migration, an essential condition for appropriate neural crest cell differentiation. Furthermore, although studies have defined a handful of guidance factors and extracellular matrix cues that direct neural crest to its targets, it is unclear how these migratory cues interact with specific neural crest fate decisions.

Dorsal root ganglia (DRG) are a crest-derived, segmentally arrayed series of sensory neurons and their associated glia that detect pain, temperature, mechanical and proprioceptive stimuli. Expression of the Neurogenins (Neurog), a family of basic helix-loop-helix (bHLH) transcription factors, is currently the earliest known indicator of DRG neuronal identity (Greenwood et al., 1999; Ma et al., 1999; Perez et al., 1999). In the zebrafish, neurog1 is essential for the formation of DRG neurons (Andermann et al., 2002; Cornell and Eisen, 2002); in its absence, cells differentiate as myelinating Schwann cells (McGraw et al., 2008). Although Neurog transcription factors clearly determine sensory neuron identity, the mechanism by which their expression is initiated is unclear.

To identify genes implicated in the initial cell fate decisions in DRG, we conducted a mutant screen in the zebrafish for mutants lacking DRG sensory neurons. We isolated a mutant, sensory deprived (sdp), in which neural crest derivatives form normally with the exception of the DRG. Subsequent work mapped the associated mutation to the gene encoding Reck (reversion-inducing, cysteine-rich protein containing Kazal motifs), an inhibitor of metalloproteinases (Takahashi et al., 1998). reck has previously been implicated in the development of the mammalian cortex (Muraguchi et al., 2007) and circulatory system (Oh et al., 2001). In cell culture studies, loss of reck function is associated with a hypermigratory phenotype (Morioka et al., 2009; Silveira Corrêa et al., 2010), presumably because upregulated metalloproteinase activity allows cells to permeate the extracellular matrix more easily. This cell migratory behavior is reflected in studies of human cancers, which are more aggressive following Reck inactivation (Clark et al., 2007). Our results suggest a mechanistic link between termination of migration and initial differentiation of sensory neuron precursors from the neural crest.

Zebrafish were maintained at 28.5°C on a 14-hour/10-hour light/dark cycle following established methods and Institutional Animal Care and Use Committee standards (Westerfield, 2007). Embryos were maintained in E2 medium, and staged according to Kimmel et al. (Kimmel et al., 1995). DRG were identified using Tg(neurog1:egfp)^w61 (McGraw et al., 2008).

The 4.9 kb sox10 promoter (Carney et al., 2006), fluorescent protein Eos (Wiedenmann et al., 2004) (Evrogen) or nls-Eos, and a polyadenylation sequence (Kwan et al., 2007) were Gateway cloned (Invitrogen) to generate pSox10:Eos/pSox10nls-Eos; clones were microinjected with Tol2 transposase to generate germline transgenics Tg(sox10:eos)^w9 and Tg(sox10:nlseos)^w18 as previously described (Fisher et al., 2006).

Embryos were stained with the antibodies mouse anti-Elavl1 (Invitrogen), 1:500; rabbit anti-GFP (Invitrogen) 1:1000; mouse anti-acetylated tubulin (Sigma; 1:5000), mouse anti-MF20 (Developmental Studies Hybridoma Bank, University of Iowa; 1:100), rabbit anti-Sox10 (gift of Sarah Kucenas, University of Virginia; 1:500), as previously described (Ungos et al., 2003) and imaged by confocal microscopy.

We treated ^*AB males with 3 mM ethylnitrosourea (Solnica-Krezel et al., 1994). F₃ progeny were screened by immunostaining for Elavl1. Sdp^w15 was identified in the ^*AB background during screening. Three additional alleles were identified: sdp^w13 by complementation screening against sdp^w15 carriers and sdp^w12 and sdp^w14 isolated as novel mutations and later established as sdp alleles by complementation.

sdp^w12 heterozygotes were outcrossed to the Wik strain and F₂ progeny were subjected to bulked segregant analysis using SSLP markers followed by analysis of individuals, as described (Bahary et al., 2004). Zebrafish reck sequence was generated by alignment of expressed sequence tags (ESTs) identified by BLAST search using Mus musculus RECK protein as the search string, and used as a template to clone the zebrafish cDNA by PCR. Staged embryos were homogenized in Trizol (Invitrogen), and total RNA was isolated by column purification (Qiagen). RNA was treated with DNAse I (Fermentas) to remove genomic DNA and reverse transcribed (Invitrogen). Amplicons were cloned into pCR4-TOPO (Invitrogen) and sequenced.

To identify sdp^w12 mutant embryos, tissue was homogenized by overnight incubation in 200 μg/μl proteinase K. A 233 bp amplicon was PCR-amplified from genomic DNA using primers (F: 5′-GAATCTCCACCCATCGCCAAGATG-3′; R: 5′-GAATGTTAGCAGCTGTGAGGTTTG-3′) and digested with AvaII (NEB), which cuts the wild-type product.

A splice-blocking morpholino oligonucleotide (MO) with the sequence CAGGTAGCAGCCGTCACTCACTCTC was generated based on prediction of efficacy by the manufacturer (Gene Tools). This MO targets the reck exon 7-intron 7 boundary. Injected embryos were processed for cDNA as above. We designed primers flanking intron 7 (F: 5′-CATCAACAACTCACTACCAGGAGC-3′, exon 7; R: 5′-GGGTGTAGTCTGGTGTAGTTCTCC-3′, exon 8) to detect appropriately spliced reck transcript.

Tg(fli1a:egfp)^y1 embryos (Lawson and Weinstein, 2002) were injected at 48 hours post-fertilization (hpf) in the sinus venosus with a 40% solution of Qdot fluorescent nanocrystals (Invitrogen) as previously described (Kamei et al., 2010; Weinstein et al., 1995) and imaged using an Olympus FluoView FV-1000 confocal microscope.

Wild-type (wt) and sdp^w12 reck coding sequences were ligated into pCS2p+, linearized with NotI and transcribed in vitro with SP6 polymerase to generate capped, polyadenylated mRNA (Ambion). A range of mRNA concentrations was injected into embryos derived from a cross between Tg(neurog1:egfp)/sdp^w12/+ parents. Data were analyzed using a two-factor ANOVA; post-hoc significance between conditions was determined using the Scheffé method.

Digoxigenin-labeled riboprobes for crestin (Rubinstein et al., 2000) and neurog1 (Korzh et al., 1998) were generated as previously described. A 2.1 kb fragment of reck was subcloned into pBluescript II SK, linearized with EcoRV and transcribed with T7 polymerase (Invitrogen) to generate digoxigenin-labeled riboprobe. In situ hybridization was carried out as previously described (Thisse and Thisse, 2008). Fluorescent in situ hybridization was carried out as previously described (Vize et al., 2009). Embryos were then embedded in 4% agarose and sectioned at 100 μm on a Leica VT1000S vibratome. Colocalization of reck and crestin was analyzed by confocal imaging in single optical planes sampled every 7 μm. The anterior-posterior position of each section was determined by alignment to the section containing the yolk constriction.

Tg(neurog1:egfp)/sdp^w12/+ parents were incrossed; the resulting embryos were sorted by phenotype at 48 hpf and homogenized in lysis buffer (150 mM NaCl, 10 mM HEPES, 2 mM DTT, 0.1% Triton X-100, pH 8.0). Protein concentration was normalized using the Bradford method. Lysate was incubated with DQ gelatin (Invitrogen) at room temperature; fluorescence was assayed using a fluorimeter (Tecan).

Donor Tg(neurog1:egfp) embryos were injected with 1 nl 2.5% tetramethyl rhodamine dextran, 0.1 M KCl and either 1.75 μg/μl reck MO or sterile water. Host nacre^w2/w2 embryos were injected at the one-cell stage with 1 nl 0.2% Phenol Red, 0.1 M KCl and either 1.75 μg/μl reck MO or sterile water. Donor and host embryos were dechorionated and mounted in 4% methylcellulose at sphere or dome stage (∼3.5-4.5 hpf). Cells were aspirated into a pulled capillary glass needle and deposited into host embryos. Transplants were allowed to recover in E2 media supplemented with 100 U penicillin/100 μg/ml streptomycin.

Tg(sox10:eos)/sdp^w12/+ fish were incrossed; the resulting clutch was fixed at 24 hpf, immunostained for MF20 and imaged by confocal microscopy. A threshold was applied to the images using Photoshop and images were then subjected to a MatLab (Mathworks) algorithm measuring the length of y-direction signal (i.e. first black pixel to last black pixel) for each x coordinate. This plot was smoothed by averaging over 10-point bins. Local maxima were detected by an iterative point-to-point comparison function (peakdet, http://billauer.co.il/peakdet.html).

Tg(sox10:nls-eos) embryos (18 somite-stage; 18s) and Tg(sox10:nls-eos)/Tg(neurod:tagrfp)^w69 embryos (30 hpf) were immobilized in 1.5% Type VII agarose (Sigma) and imaged every 10 minutes using a 3i Marianas spinning disk confocal microscope system. Image files were processed using Slidebook (3i) and ImageJ (NIH) software. Movies from the early interval were initiated at the beginning of neural crest migration. Movies from the late interval were synchronized so that the lateral line crosses somite 16 at the same frame. Divisions were identified by resolution of mitotic figures; cell death was determined by nuclear fragmentation.

Tg(neurog1:egfp)/sdp^w12 embryos were treated in drug dissolved in 1% DMSO in E2 medium from 16s to 3 days post-fertilization (dpf); drug solution was changed daily. Embryos were fixed and immunostained for Elavl1 and GFP. Drugs were administered at the following concentrations: GM6001 (Enzi Life Sciences), 100 μM; batimastat (Tocris Biosciences), 500 μM; marimastat (Tocris Biosciences), 100 μM; DAPT (Sigma), 100 μM.

Fish were fixed at 30 hpf and immunostained with rabbit anti-Sox10 antibody. Embryos were permeabilized with 20 μg/ml proteinase K (Sigma) and subjected to TUNEL labeling using fluorescein substrate (Roche). TUNEL⁺ and Sox10⁺ cells were counted from confocal stacks and the embryos were subsequently genotyped.

To identify mutations that alter DRG development, we screened for perturbed expression of Elavl1 in the peripheral nervous system (Henion et al., 1996; Marusich et al., 1994; Szabo et al., 1991). We identified four alleles of a recessive mutation, sensory deprived (sdp). All mutant embryos completely lack DRG up to and including 5 dpf (Fig. 1, supplementary material Fig. S1). However, for all mutations all other neural crest derivatives appear to develop normally (supplementary material Fig. S2), including pigment cells, enteric neurons, lateral line glia and sympathetic chain (supplementary material Fig. S2); although normal in morphology, the ceratohyal cartilage is displaced (supplementary material Fig. S2K-L′). Cranial ganglia, which have both neural crest and placode origin, appear normal (supplementary material Fig. S2) and have comparable cell counts (supplementary material Fig. S3). Fish homozygous for any of the sdp alleles die by day 12.

Mutant embryos display defects in DRG formation at the earliest stages of their development. sdp embryos exhibit no neurog1 expression in DRG precursors as assessed by a neurog1:egfp transgenic line (Fig. 1A-B′) and by in situ hybridization for neurog1 transcript (Fig. 1C-D′). In addition, no cells in the periphery ectopically express these markers, nor are any ectopic neurons revealed by Elavl1 staining, suggesting that the sdp phenotype is not caused by DRG displacement. By contrast, neurog1 expression in the central nervous system and cranial placodes appears normal. These results suggest that the sdp phenotype is a result of a defect occurring upstream of neurog1.

Neurons and glia of the zebrafish DRG are derived from a common subpopulation of neural crest (Raible and Eisen, 1994; McGraw et al., 2008). The sdp phenotype could be caused by a failure to specify this neuroglial lineage or by a later defect in neuronal differentiation. To address this question, we generated the transgenic line Tg(sox10:eos), in which the fluorophore Eos (Wiedenmann et al., 2004) is expressed under the control of the promoter for sox10, an SRY box-containing transcription factor expressed in neural crest and maintained in crest-derived glial cells (Carney et al., 2006). Schwann cells apposed to motor axons are retained in sdp embryos; however, satellite glia adjacent to DRG are notably absent (Fig. 1E,E′). Schwann cells are also found to be normally distributed along the lateral line nerve (supplementary material Fig. S2E,H). We conclude that in sdp mutants DRG neuron formation is compromised rather than specification of the neuroglial lineage.

To identify the genomic location of sdp alleles, we used recombination mapping in an sdp^w12/Wik hybrid background. We identified a critical region of ∼200 kb on chromosome 24 associated with sdp (Fig. 2A). Direct sequencing of sdp^w12 cDNA corresponding to transcripts in this region identified a mutation in the exon 9 splice donor of reck that caused a 510 bp intron to be retained (Fig. 2C). This intron contains a stop codon, which would truncate any translated protein to 309 amino acids. reck encodes a 956 amino acid protein containing an N-terminal signal sequence and two Kazal motifs (Fig. 2B); it is a glycosylphosphatidylinositol (GPI)-linked inhibitor of metalloproteinases (Takahashi et al., 1998). All sdp alleles possessed perturbations in the reck sequence (Fig. 2B-C′): a C38R missense mutation in a highly conserved residue (sdp^w13), a W252X premature stop mutation (sdp^w14) and a 0.5 Mb deletion of chromosome 24 that eliminates at least four other genes in addition to reck (sdp^w15). Although each sdp allele possesses a different lesion in reck, they are equivalent with respect to the DRG phenotype (supplementary material Fig. S1). Complementation analysis reveals that w12, w13 and w14 are likely to be null alleles as no additional phenotypes were observed when crossed to the w15 deficiency (not shown). The severe swelling phenotype of the w15 deficiency (supplementary material Fig. S1) is reduced by complementation, suggesting that this phenotype is affected by loss of additional genes found in the interval.

Reck inhibits the maturation and activity of the metalloproteinases Mmp2, Mmp9, Mmp14 and Adam10 (Muraguchi et al., 2007; Oh et al., 2001; Takahashi et al., 1998). We incubated protein extracts from 48 hpf sdp^w12/w12 and sibling embryos with a fluorescein/gelatin conjugate, and found that mutant embryos exhibited greater metalloproteinase activity than siblings (supplementary material Fig. S4). Reck therefore appears to be a major inhibitor of metalloproteinases in developing zebrafish embryos.

To confirm that reck loss of function is the cause of the DRG phenotype observed in sdp, we generated a splice-blocking MO directed against the exon7-intron7 boundary. Embryos injected with this MO fail to develop DRG (Fig. 3A-D) despite the appropriate formation of cranial ganglia (Fig. 3C), Schwann cells (Fig. 3G), lateral line glia (Fig. 3H) and pigment cells (Fig. 3I), matching the phenotypes of reck mutants. DRG counts declined in a dose-dependent fashion (Fig. 3E). We verified that reck expression was disrupted by this MO by subjecting cDNA to quantitative RT-PCR using primers flanking intron 7 (Fig. 3F). These primers generate an amplicon of 366 bp when applied to appropriately spliced transcript; we cannot detect this amplicon in cDNA from embryos injected with reck MO. These results confirm that DRG formation fails in the absence of reck function.

To show that reintroduction of reck was sufficient to rescue the sdp phenotype, we injected wt or sdp^w12 mutant mRNA into embryos derived from sdp^w12/+ incrosses (Fig. 3, Table 1). Injection of mutant mRNA has no effect on mutants or siblings (Fig. 3J,L). Injection of wt reck mRNA restores DRG to wild-type levels in sdp^w12/w12 embryos (Fig. 3M). Overexpression of wt reck mRNA fails to generate any observable phenotype on its own (Fig. 3K). We conclude that reck loss of function is solely responsible for the DRG defect observed in sdp.

We examined reck expression by in situ hybridization to identify tissues in which reck might act to influence DRG development. In situ hybridization for crestin was used to identify neural crest. In the trunk, crestin is expressed adjacent to the dorsal neural tube and in cells migrating between neural tube and somite (Fig. 4A,B, supplementary material Fig. S5C,D,I-I″′), whereas reck is expressed primarily in ventral mesoderm, consistent with a role in vasculature development (Fig. 4A,B, supplementary material Fig. S5G,H,J-J″′). However, a band of dorsal reck expression appears at 22 hpf (Fig. 4A,B, supplementary material Fig. S5G,H,J-J″′). This dorsal expression is transient, disappearing by 30 hpf (Fig. 4C,D)

To quantify the degree to which reck and crestin expression colocalize, embryos at 22 and 30 hpf were subjected to double fluorescent in situ hybridization followed by serial sectioning (Fig. 4B,D). Cells were counted as reck⁺, crestin⁺ or reck/crestin⁺ in each section. At 22 hpf, essentially all neural crest cells express reck. Although fewer neural crest cells are observed in posterior segments, we do not observe significant changes in reck/crestin colocalization (Fig. 4E,G). By 30 hpf, however, the expression of reck and crestin has largely separated into two separate cell populations (Fig. 4F,G); however, a small subset of cells (about three cells per hemisegment) continues to express both reck and crestin.

Mice with a targeted disruption of the reck locus display impaired vascular maturation and abdominal hemorrhage (Oh et al., 2001). To determine whether reck-deficient zebrafish embryos exhibit similar defects, we injected reck MO into Tg(fli1a:egfp) embryos, a transgenic reporter expressed in the endothelial cells of the vasculature (Lawson and Weinstein, 2002). Between 24 and 32 hpf, vasculature in MO-injected embryos is comparable to the vasculature of control embryos, although slightly delayed (Fig. 5A-C,G-I, supplementary material Table S1). However, by 48 hpf, moderate defects in circulation and vascular development are manifest. The parachordal vessel (PAV), a derivative of the caudal vein, often fails to form following reck knockdown (Fig. 5D,J, supplementary material Table S1). Additionally, although circulation initiates normally in MO-injected embryos, angiograms reveal poor perfusion in intersegmental vessels (Fig. 5E,K) as well as significantly impaired axial circulation (supplementary material Table S1). At 3 dpf, intracranial hemorrhage is apparent in reck-deficient embryos, suggesting that vascular integrity is also impaired (Fig. 5F,L, supplementary material Table S1). These results demonstrate that, as in mouse, loss of reck is associated with disruption of vasculature development.

Previous work has demonstrated that Reck is a GPI-linked protein that inhibits metalloproteinases in cis, blocking activity in the cell in which it is expressed (Muraguchi et al., 2007). reck is expressed in both neural crest and the environment through which it migrates, including vascular precursors that themselves have defective development. Therefore, the failure of DRG to form in the absence of reck function could be attributed either to cell environment or to incompetence in the neural crest itself. To differentiate between these possibilities, we performed transplants between MO- and mock-injected embryos.

Donor cells were derived from rhodamine-labeled Tg(neurog1:egfp) embryos and transplanted into nacre (mitfa – Zebrafish Information Network) host embryos (Lister et al., 1999); the resulting transplants were then immunostained for Elavl1. This allowed us to identify all donor cells incorporated into the transplant (rhodamine), donor cells which differentiated into sensory neurons (GFP), and all differentiated neurons in the transplant (Elavl1). Because nacre embryos lack trunk melanophores, we were able to verify whether donor cells had become neural crest by the appearance of donor-derived melanophores. Analysis was restricted to those embryos in which melanophores developed.

When cells from mock-injected embryos are deposited into mock-injected hosts, they are able to differentiate into DRG sensory neurons (Fig. 6A,B, Table 2). However, when cells from MO-injected embryos are transplanted into mock-injected embryos, they never form DRG neurons, despite the fact that donor cells frequently generate neural crest (Fig. 6C,D). Conversely, when cells from mock-injected donors are transplanted into MO-injected hosts, they form DRG neurons readily but are unable to induce host cells to form DRG neurons (Fig. 6E,F). Therefore, we conclude that reck is required cell autonomously in migrating neural crest for normal DRG formation.

Prior work has established that suppression of reck function leads to hypermigratory behavior in several cell lines (Chang et al., 2006; Liu et al., 2003; Morioka et al., 2009; Yoshida et al., 2008); conversely, forced expression of reck reduces migration (Hsu et al., 2006; Kang et al., 2007; Liu et al., 2003; Morioka et al., 2009; Silveira Corrêa et al., 2010; Yoshida et al., 2008). We hypothesized that reck loss of function might cause the neural crest to migrate excessively, causing the DRG deficit observed in sdp. We first measured the dorsoventral migration of neural crest streams at 24 hpf and found no significant difference in neural crest migration distance (supplementary material Fig. S6), suggesting that the overall extent of migration is not perturbed.

We next used time-lapse microscopy to search for more subtle differences in neural crest behavior at two different stages of development. We first imaged Tg(sox10:nls-eos) embryos between 18s and 30 hpf; this interval encompasses initiation of neural crest migration (Fig. 7A-F, supplementary material Movies 1, 2). Movies were analyzed as the first neural crest cell began to migrate ventrally. The position of each sox10⁺ cell was manually tracked and its velocity in the dorsoventral axis was measured. We also followed cell count, proliferation and apoptosis. During the initial migration of neural crest, we found cell velocity to be ∼40% faster in reck-depleted neural crest cells (2.66±0.25 μm/10 minutes, n=40 cells versus 1.88±0.24 μm/10 minutes, n=54 cells; mean ± s.e.m., P=0.028). The overall number of cells, the rate of proliferation and the rate of apoptosis are unchanged over this interval (Fig. 7C-F).

To examine migratory behavior in differentiating neural crest cells, we imaged uninjected and reck MO-injected Tg(sox10:nls-eos)/Tg(neurod:tagrfp) embryos between 30 and 60 hpf; this interval includes the overt expression of the first neuronal markers in DRG sensory neurons (Fig. 7G-L, supplementary material Movies 3, 4). The position of each sox10⁺ cell was manually tracked, and velocity (Fig. 7J), cell count (Fig. 7I), proliferation (Fig. 7K) and apoptosis (Fig. 7L) were measured as above. At this stage, we found cell velocity of reck-depleted neural crest cells to be half that of control cells (0.54±0.03 μm/10 minutes, n=72 cells versus 1.14±0.14 μm/10 minutes, n=163 cells; mean ± s.e.m., P=0.005). The rate of proliferation and the overall number of cells are not statistically different following reck depletion. We observed a small but significantly greater apoptotic rate in reck-depleted neural crest cells compared with that in uninjected embryos, although still too infrequent to significantly change overall cell number.

To determine whether neural crest cells are capable of inhabiting the appropriate microenvironment for DRG formation following reck depletion, we assessed the dorsoventral position of each cell at each time point (Fig. 7M). DRG precursors were previously noted to perform a dorsalward migration to the position of the future ganglion prior to differentiation (Raible and Eisen, 1994). This dorsal migration corresponds to the initiation of neurog1 expression (McGraw et al., 2008). We found that by 55 hours, 68% of crest-derived sox10⁺ cells observed in uninjected embryos are situated in this microenvironment, compared with only 20% in MO-injected embryos (Fig. 7M). Therefore, in the absence of reck, neural crest cells exhibit aberrant migratory behavior and fail to inhabit the appropriate position for differentiation into DRG.

The developmental phenotypes observed in our reck alleles reflect, in part, those seen in a previously published mouse knockout model (Oh et al., 2001). reck is expressed in the zebrafish vasculature up to and including 48 hpf (data not shown). We observe no overall gross deficit in trunk vasculogenesis; however, the parachordal chain fails to form in reck-deficient embryos. Additionally, reck MO-injected embryos exhibit hypoperfusion and cerebral hemorrhage, suggesting that vascular integrity is disrupted, concordant with the reported phenotypes in mouse. Zebrafish reck mutants might represent a model for studying the role of this inhibitor in vascular development, and additional phenotypes will be described elsewhere. Recently, other investigators uncovered a developmental interplay between motor neurons and the formation of the parachordal chain (Lim et al., 2011); further study of the relationship between RECK function, vasculogenesis and sensory neurogenesis might therefore be warranted. Nonetheless, expression and mosaic analysis suggest that reck acts within the neural crest population to regulate the formation of sensory neurons.

There are several possible explanations for what might happen in sdp mutants to the neural crest precursors normally fated to become DRG neurons. It is possible that cells fail to express neurog1 and are re-specified. We previously found that in neurog1 mutants sensory neuron precursors instead became glia by using the neurog1:egfp transgene to track their fates (McGraw et al., 2008). As reck acts upstream of the initiation of neurog1 expression, we have no way of tracking these precursors in reck mutants. Although it is possible to sample the fates of individual zebrafish neural crest cells by single-cell injection, the cells that generate DRG neurons are only a small fraction of the trunk neural crest population. A possible change in fate would therefore be difficult to observe with statistical reliability. Although it would be of interest to track reck-expressing neural crest cells in vivo using reporter constructs, we have been unable to identify genomic elements that direct reck expression owing to the poor characterization of sequence surrounding its position at the end of chromosome 24.

An alternative possibility is that proliferation of DRG precursors is reduced following loss of reck function. Mouse embryonic fibroblasts have reduced proliferation following Reck depletion (Kitajima et al., 2010). However, we observed no significant difference in proliferation in reck-depleted neural crest cells. A third mechanism that might contribute to the sdp phenotype involves the death of DRG progenitors. In our time-lapse experiments, we observed a small increase in neural crest cell apoptosis during the interval of DRG neuron differentiation. However, apoptosis was undetectable in neural crest cells in either wild-type or mutant animals using histological techniques despite an abundance of nearby apoptotic cells, confirming that neural crest cell death is extremely rare (supplementary material Fig. S7, Table S2). Regardless, we do not view apoptosis of progenitors and inappropriate progenitor migration as being mutually exclusive mechanisms for DRG defects.

reck was first identified in a large-scale screen for cDNAs capable of returning Ras-transformed cells to a flat morphology (Takahashi et al., 1998). Subsequent work has established Reck as a regulator of cell migration and invasion (Clark et al., 2007; Noda and Takahashi, 2007). Many cancers exhibit Reck downregulation via epigenetic modifications (Chang et al., 2004; Chang et al., 2006; Chang et al., 2007), transcriptional repression (Hsu et al., 2006; Liu et al., 2003; Sasahara et al., 1999), post-translational modifications (Simizu et al., 2005) or endogenous microRNAs (Gabriely et al., 2008; Liu et al., 2010). In general, RECK expression is negatively correlated with tumor aggressiveness (Clark et al., 2011; Gabriely et al., 2008; Li et al., 2007; Masui et al., 2003; Rabien et al., 2007; Rabien et al., 2011; Song et al., 2006; Span et al., 2003; Takemoto et al., 2007; Takenaka et al., 2004; Takenaka et al., 2005; Takeuchi et al., 2004; van der Jagt et al., 2006; Xu et al., 2010); presumably in the absence of RECK, metalloproteinase hyperactivity enables greater tumor cell migration and invasion.

Like cancer cells, neural crest cells migrate through the extracellular environment to reach their final targets. One possible explanation for the DRG phenotype caused by result of reck loss of function is that neurogenic neural crest migrates inappropriately. We observed broad reck expression in neural crest cells at 22 hpf followed by restriction to a subset of neural crest cells by 30 hpf, suggesting that winnowing of reck expression might confer DRG progenitor identity as development progresses. We found that the requirement for reck is cell autonomous, a result that was somewhat surprising given the extensive expression of reck transcript in cells situated in the path of neural crest migration and the large increase in metalloproteinase activity in sdp embryos, suggesting that it is broadly active in tissues besides neural crest. Reck has been shown to inhibit metalloproteinases in cis (Muraguchi et al., 2007), suggesting that a potential metalloproteinase, the inhibition of which is crucial for normal DRG development, might be expressed within the neural crest itself, consistent with a cell-autonomous function. However, we cannot rescue the sdp defect by treating with several broad-spectrum inhibitors of metalloproteinases (supplementary material Fig. S9), although the efficacy of these inhibitors against specific zebrafish metalloproteinases is unknown. Whether any of the >50 zebrafish metalloproteinases interact with Reck awaits further analysis.

When viewed by time-lapse microscopy, reck-depleted cells exhibit changes in migratory behavior. Initially, we observed neural crest to be hypermigratory in reck MO-injected embryos, showing a small but significant increase in cell velocity. We note that we are unable to determine whether these differences reflect changes in behavior of all neural crest cells or only a subset. When viewed at later times, reck-depleted cells appear hypomigratory, in contrast to their earlier behavior. These results can be reconciled by a model in which reck-deficient cells initially show increased migration, as predicted from the cancer literature. In the absence of reck function, DRG precursors would then be re-specified (or die), leaving only slow-moving cells on the medial pathway destined to become Schwann glia. A subset of neural crest-derived sox10⁺ cells are unambiguously mislocalized following reck depletion, failing to move dorsally. These include neuronal precursors and cells that will form satellite glia surrounding the DRG. However, dorsal migration occurs after initial expression of neurog1, and both the failure of dorsal movement to form DRG and subsequent accumulation of satellite cells might be caused indirectly by neurog1 absence in reck mutants. Whether the lack of dorsal movement at this later stage is due to altered Reck function or whether it is the consequence of failed specification will await further study.

An alternative possibility for the function of reck in DRG specification is that altered activity of metalloproteinases could affect signaling required for neuron specification (reviewed by Bai and Pfaff, 2011). In the mouse, Reck regulates cortical neurogenesis through Notch signaling (Muraguchi et al., 2007). When Reck is disrupted, hyperactive ADAM10 results in reduced Notch signaling and precocious neurogenesis with an ultimate decrease in neurons due to precursor depletion. Might a similar impairment be the mechanism by which sdp embryos fail to form DRG? Disruption in Notch signaling results in a decrease in DRG (Cornell and Eisen, 2000; Cornell and Eisen, 2002), but this phenotype is the result of prior depletion in all trunk neural crest with concomitant increase in Rohon-Béard sensory neurons. We observe no change in Rohon-Béard cells in mutant embryos (supplementary material Fig. S8) and normal crest formation and differentiation (supplementary material Fig. S1). Moreover, loss of zebrafish notch1a receptor causes an increase in DRG neurons (Gray et al., 2001), more consistent with the neurogenic phenotype expected with loss of Notch function. Does sdp instead behave as a mutation resulting in Notch gain of function? We are unable to rescue DRG formation by treating sdp embryos with the Notch inhibitor DAPT (Geling et al., 2002) although we see an increase in DRG neurons when treating wild-type animals (data not shown), consistent with the notch1a mutant phenotype. Alternatively, Reck might alter other metalloproteinases that affect signaling pathways needed for DRG differentiation.

A limited body of evidence provides clues as to how reck might fit into a genetic hierarchy controlling DRG development. erbb2 and erbb3b function have previously been shown to be essential to the formation of DRG neurons upstream of neurog1 expression; migrating neural crest in these mutants becomes progressively disorganized and DRG precursors apparently fail to pause and differentiate (Honjo et al., 2008). ErbB2 is known to regulate Reck expression via ERK signaling (Hsu et al., 2006), and the loss of DRG observed in both erbb2 and reck mutants suggests that the two genes might act within a common signaling cascade to direct DRG formation. However, mutations affecting Erbb signaling also disrupt glial development (Lyons et al., 2005), suggesting that this pathway might affect DRG development at an earlier step involving the neuroglial precursor.

Taken together, our results suggest that Reck cell autonomously regulates the initial specification of DRG precursors upstream of neurog1. reck expression is restricted to a subset of neural crest, suggesting that these cells might represent DRG precursors. DRG cell fate choice might, therefore, rely on regulation of reck expression. Understanding which metalloproteinases are inhibited by Reck during zebrafish neural crest development will be crucial for resolving the mechanisms underlying how it regulates neural crest cell fate.

The authors wish to thank Adrienne Mueller, William Wood and Andrew Millspaugh for their assistance in writing MatLab modules. We also wish to thank Sarah Kucenas for generously providing the anti-Sox10 antibody. We thank the Sanger Centre for help with assembly of the distal end of chromosome 24.