Zebrafish neurofibromatosis type 1 genes have redundant functions in tumorigenesis and embryonic development

Type 1 neurofibromatosis (NF1) is an autosomal dominant inherited genetic disorder characterized by pigmented birthmarks known as café-au-lait spots, cutaneous and plexiform neurofibromas arising in the glial cells of the peripheral nervous system (PNS), optic pathway gliomas, cardiovascular abnormalities and learning defects (Williams et al., 2009). The disease results from mutations in the NF1 gene, encoding the large protein neurofibromin, which contains a GTPase-activating protein-related domain (GRD) capable of inactivating the RAS proto-oncogene (Cawthon et al., 1990; Viskochil et al., 1990; Wallace et al., 1990). Thus, NF1 loss results in aberrant activation of Ras signaling, which may predispose NF1 patients to a variety of cancers (Cichowski and Jacks, 2001). Heterozygous Nf1 mutant mice develop pheochromocytoma and myeloid leukemia, whereas the conditional loss of Nf1 in a p53-deficient background results in highly penetrant malignant astrocytoma formation (Jacks et al., 1994; Zhu et al., 2005a; Powers et al., 2007). Furthermore, two recent reports have identified NF1 mutations in approximately 15–23% of human glioblastoma patients (Parsons et al., 2008; The Cancer Genome Atlas Research Network, 2008). Although these studies demonstrate a strong link between NF1 function and high-grade glioma, the crucial signaling pathways governing the development of tumorigenesis remain to be elucidated. An animal model facilitating the rapid interrogation of epistatic and functional relationships within signaling pathways would serve as a valuable tool for probing the pathology underlying NF1-induced cell transformation.

We recently developed a zebrafish model of NF1 deficiency using antisense morpholino oligonucleotides to produce transient gene knockdown (Padmanabhan et al., 2009; Lee et al., 2010). Two zebrafish orthologues were identified that are highly homologous to human NF1 at the amino acid level, sharing approximately 84% identity, including 91–93% identity within the GRD. Both genes maintain syntenic relationships with human NF1 on chromosome 17q11.2 and are probably the result of the well-described genomic duplication event that occurred early in the evolution of teleosts (Amores et al., 1998). In our previous work with nf1 morphants, we observed defects in both cardiovascular and nervous system development. However, due to the transient nature of morpholino gene knockdown, the analysis of nf1-deficient phenotypes beyond the first 3 days of life was not possible.

We report here the generation of stable mutant nf1 zebrafish lines, using both zinc finger nuclease (ZFN) and targeting induced local lesions in genomes (TILLING) strategies, and the detailed phenotypic analysis of this new animal model of human NF1. We have successfully generated several independent null alleles of nf1a and nf1b. Mutant larvae carrying at least one wild-type nf1a or nf1b allele are viable, fertile and show no obvious phenotypes during early development. By contrast, nf1a^−/−; nf1b^−/− larvae exhibit overt pigmentation defects as early as 6 days post fertilization (dpf) and do not survive beyond 10 dpf. Beginning at 4 dpf, nf1a^−/−; nf1b^−/−larvae exhibit hyperplasia of oligodendrocyte progenitor cells (OPCs) and Schwann cells, as well as melanophore hypoplasia. Defects resulting from the loss of nf1 in pigment cell and glial cell lineages mirror those often observed in the tissues of NF1 patients. In a p53 mutant background (p53^e7/e7), nf1a^+/−; nf1b^−/− fish develop high-grade gliomas and malignant peripheral nerve sheath tumors (MPNSTs), demonstrating a tumor-suppressor function for the zebrafish nf1 orthologues. Therefore, we have developed and characterized a heritable zebrafish model of NF1 that exhibits clinical hallmarks of the disorder, including nervous system defects and increased susceptibility to tumorigenesis. Furthermore, nf1 mutant zebrafish represent the first vertebrate model of the pathognomonic pigmentation lesions associated with NF1.

We previously identified two zebrafish orthologues of human NF1, nf1a and nf1b, and described the phenotypes that result from their loss of function during early development induced by antisense morpholino oligonucleotides (Padmanabhan et al., 2009; Lee et al., 2010). Although this technology readily permits transient knockdown of gene expression, its efficacy is limited to only the first few days of life. In an effort to gain a better understanding of the roles of nf1a and nf1b during development, as well as in cancer predisposition, we employed multiple approaches to develop stable lines of zebrafish harboring germline mutations in each of these genes. Using a modular approach (Zhu et al., 2011), zinc finger nucleases (ZFNs) were engineered with binding specificities directed to exon 26 of nf1a and exon 17 of nf1b (Fig. 1A,B). Paired ZFN mRNAs were injected into zebrafish embryos and independent target-specific mutant alleles for nf1a (nf1a^Δ5 and nf1a^Δ8) and nf1b (nf1b⁺¹⁰ and nf1b^Δ55) were identified in the F1 generation (Fig. 1C; supplementary material Fig. S1A–D). Each of these mutations included a deletion and/or insertion within a coding exon that resulted in a frameshift, introducing premature stop codons that would be expected to truncate the neurofibromin protein upstream of the GRD (Fig. 1D,E). In a separate effort, we screened a library of N-ethyl-N-nitrosourea (ENU)-mutagenized zebrafish by targeting induced local lesions in genomes (TILLING) (Wienholds et al., 2002) and identified a single founder harboring a nonsense mutation in exon 29 of nf1a (nf1a^L1247X) (Fig. 1C; supplementary material Fig. S1E–G). To confirm that the targeted alleles disrupted production of full-length protein, we performed western blots using an antibody that should recognize both nf1a and nf1b with extracts prepared from 3 dpf wild-type, nf1a^Δ5/Δ5; nf1b^+/+, nf1a^+/+; nf1b^+10/+10 and nf1a^Δ5/Δ5; nf1b^+10/+10 larvae (Fig. 1F). We observed a complete loss of Nf1 signal in the double-homozygous null extracts. We detected only low levels of protein expression in nf1a^Δ5/Δ5; nf1b^+/+ mutant extracts as compared with wild-type or nf1a^+/+; nf1b^+10/+10 mutant extracts, which might reflect differences in expression levels of the two orthologues at 3 dpf. However, we cannot rule out the possibility that the neurofibromin antibody we used recognizes the two proteins with different affinities. We generated separate zebrafish lines with distinct null alleles of both nf1a and nf1b to provide evidence that the observed phenotypes were in fact due to nf1 loss and did not involve any spurious passenger mutations specific to the isolation of any individual nf1 mutant line (supplementary material Fig. S2). Because our data indicate that these various null alleles are equivalent, we refer to them without individual allelic designations henceforth (nf1a^−/− and nf1b^−/−).

Mutants carrying at least one wild-type allele of either nf1a or nf1b are viable and fertile. However, when crossing parental genotypes that would be expected to yield nf1a^−/−; nf1b^−/− progeny, none were observed in the adult population. To investigate this further, we performed quantitative survival studies. At 7 dpf, nf1a^−/−; nf1b^−/− larvae began to die, with none surviving beyond 10 dpf, although 100% of wild-type larvae survived to 10 dpf (Fig. 1G). Furthermore, 100% survival at 10 dpf was also observed in nf1a^−/−; nf1b^+/+ (n=26), nf1a^+/+; nf1b^−/− (n=22), nf1a^−/−; nf1b^+/− (n=28) and nf1a^+/−; nf1b^−/− (n=24) larvae. The swim bladders of nf1a^−/−; nf1b^−/− larvae were frequently observed to be underinflated. However, nf1a^−/−; nf1b^−/− larvae maintained the ability to both consume and transit live paramecia, suggesting that premature death was not the result of starvation (supplementary material Fig. S3). Additionally, an incompletely penetrant valvular insufficiency phenotype was appreciated in nf1a^−/−; nf1b^−/− larvae, as well as in those harboring only a single wild-type nf1 allele (nf1a^+/−; nf1b^−/−and nf1a^−/−; nf1b^+/−) (supplementary material Movies 1–3).

We previously described OPC hyperplasia after nf1a and nf1b morpholino knockdown in the context of a homozygous p53 mutant background (Lee et al., 2010). To examine nf1a and nf1b function in OPCs and other tissues beyond the first few days of life, we crossed several cell-type-specific zebrafish reporter lines into nf1a/nf1b mutant backgrounds. At 2 dpf, olig2 expression appeared normal in Tg(olig2:GFP); nf1a^−/−; nf1b^−/− embryos, as assessed by both whole-mount in situ hybridization analysis of endogenous olig2 mRNA expression and GFP expression (supplementary material Fig. S4A–D). We also evaluated nf1 loss in Tg(sox10:GFP) embryos. This transgene drives GFP expression in specified ventral spinal cord OPCs, but not the neighboring motoneurons that arise from a common progenitor cell, as well as in Schwann cells of the posterior lateral line nerve (PLLn). At 2 dpf, similar numbers of sox10:GFP-positive OPCs were detected in the dorsal and ventral spinal cord of wild-type and nf1a^−/−; nf1b^−/− embryos (supplementary material Fig. S4E,F). Examination of nf1-null PNS Schwann cells at 2 dpf showed no effect on the number of sox10:GFP-expressing cells associated with the PLLn, which innervate skin mechanosensory neuromast cells (supplementary material Fig. S4G,H).

However, nf1a^−/−; nf1b^−/− larvae exhibited increased numbers of OPCs at 4 dpf compared with controls, as evidenced by an excess of dorsally migrated olig2:GFP-positive OPCs (Fig. 2A,B) along with increased numbers of both dorsally and ventrally positioned sox10:GFP-positive OPCs (Fig. 2C,D), consistent with our analyses of nf1 morphants at 3 dpf (Lee et al., 2010). In addition, we observed an increase in sox10:GFP-positive Schwann cells associated with the PLLn (Fig. 2E,F). This increase in PLLn Schwann cell number was not associated with altered proliferation of these cells (supplementary material Fig. S5A–F). To assess the roles of nf1a and nf1b in the developing radial glial cells of the spinal cord, nf1a/nf1b mutants were crossed into the Tg(gfap:GFP) line, which expresses GFP from the glial fibrillary acidic protein (gfap) promoter. At 4 dpf, nf1a^−/−; nf1b^−/− larvae harboring a gfap:GFP transgene demonstrated no readily discernible differences in gfap:GFP-positive spinal cord radial glial cells as compared with wild-type larvae (supplementary material Fig. S4I,J). However, gfap expression in the extensive processes of radial glial cells precludes precise quantification and might obscure subtle differences.

To determine whether neuronal numbers increased in concert with OPCs in nf1a/nf1b mutant larvae, we used anti-HuC/D and anti-SOX10 antibodies (see Methods) to discriminate between olig2:GFP-positive neurons and OPCs, respectively. No difference between the number of olig2:GFP-positive and HuC/D-positive neurons was appreciable in 4 dpf spinal cord sections from wild-type and nf1a^−/−; nf1b^−/− larvae (Fig. 2G,H, green and magenta; Fig. 2K). However, the numbers of olig2:GFP-positive and Sox10-positive OPCs (Fig. 2G,H, arrowheads; Fig. 2L) and PLLn Schwann cells (Fig. 2I,J,M) were significantly increased at 4 dpf in nf1a^−/−; nf1b^−/− larvae relative to wild-type controls. OPC cell numbers continued to increase in nf1a^−/−; nf1b^−/− larvae at 8 dpf, as reflected by increased numbers of dorsally localizing olig2:GFP-positive OPCs as well as both dorsally and ventrally localizing sox10:GFP-positive OPCs (supplementary material Fig. S4M–P). Increased numbers of sox10:GFP-expressing PLLn Schwann cells were also evident in nf1a^−/−; nf1b^−/− larvae at 8 dpf (supplementary material Fig. S4Q,R). Thus, loss of nf1a and nf1b does not affect the specification of OPCs at 2 dpf, but instead promotes the progressive expansion of OPCs without a concomitant increase in neuronal cell numbers. Furthermore, nf1a/nf1b loss triggers Schwann cell hyperplasia beginning at 4 dpf.

Immunohistochemical analysis using the Zrf1 antibody, which labels Gfap in zebrafish, showed coexpression with GFP expressed from the gfap:GFP transgene and revealed a similar pattern of expression in wild-type and nf1a^−/−; nf1b^−/− larvae at 4 dpf (supplementary material Fig. S6A,B). Zebrafish radial glial cells also express brain lipid-binding protein (Blbp), and immunohistochemical analysis of wild-type and nf1a^−/−; nf1b^−/−spinal cords with an anti-BLBP antibody at 4 dpf revealed an obvious decrease in Blbp expression in the gfap:GFP-positive radial glia of nf1a^−/−; nf1b^−/− larvae (supplementary material Fig. S6C,D). These results suggest that although gfap:GFP-positive radial glial cells in nf1a^−/−; nf1b^−/− larvae appear normal in number, they fail to express appropriate levels of Blbp indicating a defect in gliogenesis. An additional abnormality of gliogenesis was observed at 8 dpf as a disruption in the regular segmental pattern of glial process outgrowth in Tg(gfap:GFP); nf1a^−/−; nf1b^−/− larvae (supplementary material Fig. S4K,L).

In nf1a^−/−; nf1b^−/− larvae, OPC numbers were increased relative to control animals at 4 dpf, but not at 2 dpf, suggesting that nf1-null OPCs might proliferate faster during this time period. We assessed OPC proliferation by labeling larvae with BrdU for 12 hours starting at 3.5, 4.5 and 5.5 dpf followed by immunohistochemical analysis of transverse sections through the spinal cord. By 4 dpf and continuing through 6 dpf, the number of olig2:GFP-positive and Sox10-positive OPCs in nf1-null larvae was significantly increased in comparison with wild-type controls (Fig. 3A-H, arrows; Fig. 3I). Mutant larvae further exhibited significantly increased numbers of olig2:GFP-, Sox10- and BrdU-positive OPCs at 4 and 5 dpf as compared with controls, indicating that neurofibromin normally suppresses the proliferation of OPCs during this period of development (Fig. 3A-H, arrowheads; Fig. 3J). There was little detectable BrdU incorporation at 6 dpf in either mutant or control populations.

We evaluated the ability of nf1a^−/−; nf1b^−/− OPCs to differentiate appropriately by examining the gene expression levels of proteolipid protein 1a (plp1a) and myelin basic protein (mbp), markers of differentiated oligodendrocytes, in wild-type and mutant larvae at 5 dpf. In nf1-null larvae, fewer plp1a-positive cells were detected in the midbrain and hindbrain regions (Fig. 4A,B) as well as along the dorsal and ventral spinal cords (Fig. 4C,D) as compared with controls. Central nervous system (CNS) expression of mbp, on the other hand, was indistinguishable between wild-type and nf1a^−/−; nf1b^−/− larvae at 5 dpf (supplementary material Fig. S7). However, mbp expression was elevated in Schwann cells of the PLLn in nf1a^−/−; nf1b^−/− larvae as compared with controls (Fig. 4E,F, arrowheads). These data are consistent with perturbed oligodendrocyte differentiation in the CNS as well as in PNS Schwann cells associated with the PLLn of nf1a^−/−; nf1b^−/− larvae.

We went on to examine the ultrastructure of myelinated CNS axons in control and nf1-null larvae by transmission electron microscopy (TEM). At 8 dpf, oligodendritic myelin sheaths were tightly wrapped around CNS axons in the ventral spinal cord of wild-type larvae (Fig. 4G,I). By contrast, nf1a^−/−; nf1b^−/− axons were loosely encircled by multiple lamellae rather than by compact myelin sheaths, indicating that neurofibromin is required for the normal formation of the concentric layers of oligodendrocyte membranes that enwrap neuronal axons of the CNS to promote neural conduction (Fig. 4H,J).

Given the well-described role of neurofibromin as a negative regulator of Ras, we hypothesized that nf1 loss in our mutants would lead to activation of downstream effector pathways. Western blot analysis of whole larvae extracts revealed an upregulation of phosphorylated ERK1 and ERK2 (pERK1/2) in nf1a^−/−; nf1b^−/−larvae at 3 dpf, whereas levels in nf1a^−/−; nf1b^+/+ and nf1a^+/+; nf1b^−/−larvae remained unchanged (Fig. 5A). These data are consistent with the absence of functioning neurofibromin protein in nf1a^−/−; nf1b^−/− larvae and support functional redundancy between nf1a and nf1b. We next assessed the activation of Ras effector pathways in the spinal cords of wild-type, nf1a^+/−; nf1b^−/−, nf1a^−/−; nf1b^+/−and nf1a^−/−; nf1b^−/− animals by immunohistochemical analysis of transverse larval sections. Antibodies directed against HuC/D, pERK1/2, and phosphorylated S6 (pS6) were used to label neurons and assess activation of ERK and mTOR signaling pathways, respectively (Fig. 5B–Q). Although pERK1/2 staining was only minimally observed in a few neurons and portions of spinal cord white matter at 4 dpf in wild-type larvae (Fig. 5B,N), a striking upregulation of pERK1/2 was detected in nf1a^−/−; nf1b^−/− larvae (Fig. 5E,Q). Increased ERK signaling was also noted at 3 dpf in spinal cord neurons and white matter of nf1a^−/−; nf1b^−/− larvae, but was absent at 2 dpf (supplementary material Fig. S8). Although pS6 signaling was evident in multiple spinal cord neurons, we observed no differences in these cells between wild-type and nf1 mutant animals at 2, 3 or 4 dpf (supplementary material Fig. S8Ae–Ah,Be–Bh; Fig. 5F–I). These data suggest that activation of mTOR signaling (as assessed by S6 phosphorylation) is not altered, at least in the spinal cord following nf1 loss.

Mammalian NF1 has been shown to be a potent tumor suppressor; however, we did not identify any tumors over 18 months of observation in adult zebrafish homozygous for either mutant nf1 allele alone (nf1a^−/−; nf1b^+/+ or nf1a^+/+; nf1b^−/−) or in combination with heterozygous loss of the remaining allele (nf1a^+/−; nf1b^−/− or nf1a^−/−; nf1b^+/−). Loss of p53 has been shown to cooperate with NF1 (Cichowski et al., 1999; Vogel et al., 1999) as well as other mutations that activate Ras signaling (Eliyahu et al., 1984; Parada et al., 1984; Kemp et al., 1994; Tanaka et al., 1994; Hundley et al., 1997) in mammalian tumorigenesis, so we next bred p53 mutant zebrafish into an nf1-mutant background to generate nf1a^+/−; nf1b^−/−; p53^e7/e7 fish. These animals were incrossed to derive nf1a^+/+; nf1b^−/−; p53^e7/e7 and nf1a^+/−; nf1b^−/−; p53^e7/e7 fish, which were subsequently monitored carefully for tumorigenesis. At 31 weeks post fertilization (wpf), nf1a^+/−; nf1b^−/−; p53^e7/e7 fish began to develop tumors with high penetrance although only one nf1a^+/+; nf1b^−/−; p53^e7/e7 fish developed a tumor at 44 wpf (Fig. 6A). At 45 wpf, tumor penetrance was higher in nf1a^+/−; nf1b^−/−; p53^e7/e7 fish (24/39; 62%) than in nf1a^+/+; nf1b^−/−; p53^e7/e7 fish (1/14; 7%). We have previously reported that p53^e7/e7 fish with wild-type nf1 alleles develop MPNSTs. These tumors did not begin to develop until 41 weeks of age, which was similar to the results with nf1a^+/+; nf1b^−/−; p53^e7/e7 animals (Fig. 6A). Furthermore, the penetrance of tumors in p53-null animals was only 28% by 66 wpf (Berghmans et al., 2005). Thus, the combined loss of p53 and 3 of 4 nf1 alleles in zebrafish markedly accelerates the onset and increases the penetrance of tumors as compared with the loss of p53 alone or the concomitant loss of p53 and both alleles of nf1b, but with intact nf1a.

Tumors in nf1a^+/−; nf1b^−/−; p53^e7/e7 fish were observed in the brain (n=2), eye (n=8), gill (n=1), abdomen (n=8) and trunk (n=5). Brain tumors developed very early (31 and 33 wpf) (Fig. 6A, arrows) and demonstrated features of diffuse high-grade gliomas, whereas all other tumor types were most consistent with MPNSTs (Fig. 6B–G). Histopathologically, the brain tumors were highly cellular and composed of ovoid to rounded cells with marked nuclear pleomorphism and diffuse single cell infiltration of parenchyma, including pre-existing neurons (Fig. 7A–F). Occasional mitoses were identified, but no necrosis or vascular proliferation was detected (Fig. 7G–I). Assessment of tumor lineage by immunohistochemical analysis showed that approximately 80% of tumor cells stained positive for the oligodendroglial marker Sox10, with little staining in matched wild-type tissue (Fig. 7J–L). The presence of a Sox10-negative tumor cell subpopulation is consistent with the level of heterogeneity for oligodendroglial transcription factors, such as Sox10 and Olig2, and is characteristic of astrocytic or mixed gliomas as compared with pure oligodendroglial class tumors (Fig. 7K, arrowhead) (Ligon et al., 2004; Bannykh et al., 2006). Staining for the astrocytic marker Gfap (Fig. 7M–O) highlighted a subpopulation of cells within the tumor with coarse, irregular cytoplasmic processes also consistent with the presence of an astrocytic lineage component (Fig. 7N, arrowheads). Tumor cells did not express the mature neuronal markers HuC/D or synaptophysin (Sypa), consistent with their glial origin (Fig. 7P–U).

Immunohistochemical analysis for pERK1/2 and pS6 to assess activation of ERK and mTOR signaling pathways, respectively (supplementary material Fig. S9A–F) revealed increased pERK1/2 staining in the more malignant of these two brain tumors, with normal amounts of pS6 (supplementary material Fig. S9C,D). The more hyperplastic brain tumor showed increased pERK1/2 and pS6 staining (supplementary material Fig. S9E,F). These data demonstrate that some, but not all, brain tumors in nf1a^+/−; nf1b^−/−; p53^e7/e7 animals demonstrate hyperactivation of ERK and mTOR pathways, consistent with mouse and human NF1-derived MPNSTs and gliomas (Dasgupta et al., 2005; Zhu et al., 2005b). Collectively, these findings suggest that the tumors were high-grade gliomas most closely resembling human anaplastic astrocytoma or anaplastic oligoastrocytoma WHO grade III. Furthermore, MPNSTs (Fig. 6C,E,G) exhibited spindle-shaped tumor cells and extensive necrosis consistent with this tumor type (Ducatman et al., 1986; Wanebo et al., 1993; Hirose et al., 1998). Taken together, we conclude that nf1a and nf1b mutations cooperate with p53 loss to generate high-grade gliomas and MPNSTs.

Deficits in motor coordination and cognition, including learning and memory, are characteristic of NF1 patients and animal models. To examine motor behavior and cognition in nf1a^−/−; nf1b^−/− larvae, we performed kinematic analysis of the short-latency C-start (SLC), a highly stereotyped yet modifiable acoustic startle reflex in the zebrafish (Burgess and Granato, 2007a; Wolman et al., 2011). Unlike their siblings, nf1a^−/−; nf1b^−/− larvae showed a deficit in short-term SLC habituation when presented with repetitive acoustic stimulation (supplementary material Fig. S10A). Furthermore, nf1a^−/−; nf1b^−/− larvae performed kinematically weaker SLC responses, as indicated by decreased head turning angle, maximum angular velocity and distance traveled following delivery of an acoustic stimulus (supplementary material Fig. S10B–D). Taken together, these data support the hypothesis that nf1a and nf1b function redundantly during zebrafish development and that only one of the four wild-type nf1 alleles is required for phenotypically normal embryonic development, motor behavior and cognition.

Notably, nf1 null larvae displayed aberrant lateral stripe pigmentation as compared with wild-type controls at 6 dpf (Fig. 8A–D). This phenotype was first appreciable at 4 dpf and was manifested as a disruption in the uniform pattern of melanophores arranged along the lateral stripes (Fig. 8B,D, brackets). To further investigate this phenotype, we quantified melanophore numbers along the lateral stripes of 3 and 6 dpf wild-type and nf1a^−/−; nf1b^−/−larvae. No significant differences in the number of melanophores comprising the 3 dpf lateral stripes were appreciable in any combination of mutant nf1 alleles as compared with sibling wild-type controls (supplementary material Fig. S11). However, at 6 dpf, nf1a^−/−; nf1b^−/− larvae exhibited a significant reduction in the number of lateral stripe melanophores (Fig. 8E). Less severe, but still significant, decreases were also noted in larvae carrying two or three mutant nf1 alleles (Fig. 8E). No difference in the number of apoptotic cells between wild-type and mutant larvae was discernible at 3 dpf (68.1±11 cells/larva, n=12 wild-type versus 63.8±8.5, n=12 mutants; P=0.27) (supplementary material Fig. S12A–F). Serial observation of melanophore development in individual nf1a^−/−; nf1b^−/− and wild-type larvae revealed a defect of migration or differentiation of regeneration and metamorphic lineage melanophores. At 3 dpf, a time point at which embryonic melanophore development is complete (Hultman et al., 2009), no difference in melanophore number along the lateral stripe was observed (supplementary material Fig. S13A,B,G), indicating that the melanophores develop normally in nf1 mutants. To assess the regeneration and metamorphic lineage of melanophores, we suppressed melanin synthesis after 3 dpf by treatment with N-phenylthiourea (PTU) (Hultman and Johnson, 2010), which allowed us to identify newly formed melanophores by their melanin-negative and pale appearance (supplementary material Fig. S13C,D, red arrows). Removal of PTU at 5 dpf restored melanin synthesis and regeneration and metamorphic melanophores appeared melanin-positive by 6 dpf (supplementary material Fig. S13E,F, black arrows). The abnormal appearance of the lateral stripes in mutant larvae can be attributed to defects in patterning of regeneration and metamorphic melanophores (supplementary material Fig. S13F,G, brackets), suggesting abnormal migration or differentiation of this lineage. Head melanophore numbers at 6 dpf were not significantly changed in nf1 mutant larvae and sibling controls (supplementary material Fig. S14). Collectively, these data demonstrate a specific defect in lateral stripe melanophore numbers following nf1 allele loss, most prominent in the setting of biallelic nf1a/nf1b loss. Each nf1a^−/−; nf1b^−/− larvae showed a unique pattern of lateral stripe melanophore loss, suggesting a stochastic defect in melanophore differentiation from neural crest, proliferation, migration or survival, rather than a defect specific to any particular somite of the developing embryo and larvae.

NF1 is a common tumor-predisposing, autosomal dominant genetic disorder characterized by café-au-lait macules and cutaneous neurofibromas. In addition to these pigmentation defects and tumors of the peripheral nervous system (PNS), NF1 patients demonstrate highly diversified clinical features with multiple tissue types affected. Prominent lesions include optic pathway gliomas, Lisch nodules, skeletal dysplasia, cardiovascular abnormalities, learning defects and various cancers such as leukemia and intestinal tumors (Side et al., 1997; Side et al., 1998; Bahuau et al., 2000; Andersson et al., 2005; Williams et al., 2009). It remains unclear how NF1 mutations contribute to the diverse symptoms and tissue types affected in patients. Because neurofibromin is a very large protein that is highly conserved evolutionarily, it is likely to have activities related to functional domains other than those affecting Ras signaling. For example, in addition to Ras, neurofibromin can bind to microtubules, syndecan, phospholipids and amyloid precursor protein (Xu and Gutmann, 1997; Hsueh et al., 2001; D’Angelo et al., 2006; De Schepper et al., 2006). Intriguingly, recent studies indicate that neurofibromin might function as a positive regulator of adenylyl cyclase (Guo et al., 2000; Tong et al., 2002; Dasgupta et al., 2003). This function of neurofibromin modulates neuronal differentiation (Hegedus et al., 2007), suggesting the possibility that cognitive defects in NF1 patients might be related to defects in cAMP signaling rather than activated Ras. Furthermore, the tremendous clinical variability in the phenotypic spectrum seen among families with the same molecular NF1 lesion posits the role of unlinked modifier loci in regulating the expressivity of disease characteristics (Easton et al., 1993; Sabbagh et al., 2009). Nevertheless, the identification of specific modifier genes and the relative contributions of Ras signaling versus other neurofibromin-regulated pathways for specific phenotypes have yet to be fully elucidated. The zebrafish model that we have developed offers an attractive tool for furthering this analysis because it is amenable to small molecule screens, genetic modifier screens and genetic rescue experiments.

A hallmark feature of human NF1 is the presence of pigmentation defects known as café-au-lait spots. Little is known about the underlying mechanisms responsible for this abnormality, and pigmentation abnormalities in other animal models of NF1 have not been described. In this regard, it is of interest that zebrafish lacking neurofibromin exhibit abnormal patterning of the melanophores that compose the lateral stripes, a phenotype similar to that following pharmacologic inhibition of the upstream Ras effector ErbB (Hultman et al., 2009; Hultman and Johnson, 2010). This easily observable and quantifiable phenotype offers the opportunity to probe underlying molecular pathways modulated by nf1 in melanophores.

Several studies employing murine models have previously shown a role for neurofibromin in regulating cell numbers of CNS OPCs and astrocytes as well as PNS Schwann cells and sympathetic neurons (Brannan et al., 1994; Gutmann et al., 1999; Bennett et al., 2003; Zhu et al., 2005b; Hegedus et al., 2007; Zheng et al., 2008). These are consistent with our findings in nf1a/nf1b mutants and indicative of a strong evolutionary conservation of Nf1 function in neural development. Our observation of impaired compact myelin formation and reduced CNS plp1a expression with unperturbed CNS and increased PNS mbp expression suggests that the differentiation programs of oligodendrocytes and Schwann cells respond differently to Nf1 deficiency. Alternatively, the observed ultrastructural defects in myelinated CNS axons might arise secondary to neuronal defects. The accessibility of the zebrafish embryo to mosaic analysis offers the ability to differentiate between these possibilities. Unlike Nf1-deficient mice, however, nf1a/nf1b mutant radial glia failed to demonstrate an appreciable increase in Gfap expression. Instead, we observed a decrease in Blbp expression and irregularities in patterning of nf1a^−/−; nf1b^−/− radial glial cells. This discrepancy might reflect species-specific differences in neural tissue as opposed to Nf1 function, because it remains unclear whether zebrafish radial glia-like ependymal cells are functionally equivalent to mammalian astrocytes.

We have previously characterized cardiovascular defects resulting from morpholino knockdown of nf1a and nf1b in zebrafish (Padmanabhan et al., 2009; Lee et al., 2010). These defects were observed at 48 and 72 hpf and probably resulted from impairment of both maternal and zygotic gene expression. In our stable compound mutants, we observed partially penetrant insufficiency of the atrioventricular valve at 3 dpf along with significant edema and impaired blood circulation associated with irregular heart rates beginning at 5–6 dpf (data not shown). However, we could not determine whether these effects were primary or secondary. It is possible that impaired cardiac function is the cause of death of these larvae. The absence of earlier cardiovascular phenotypes is most probably due to the activity of maternal transcripts (Abrams and Mullins, 2009); confirmation of this interpretation awaits the creation of a maternal zygotic mutation.

Ubiquitous and conditional Nf1 knockout mice have been generated to investigate the role of neurofibromin in development and tumorigenesis (Cichowski and Jacks, 2001; Le and Parada, 2007). Conditional loss of Nf1 with p53 deficiency in mice results in the development of grade III and IV astrocytomas with full penetrance (Zhu et al., 2005a), indicating that Nf1 mutations are associated not only with low-grade but also high-grade astrocytoma. Indeed, two independent studies have demonstrated that NF1 mutations are found in about 15–23% of human glioblastoma multiformes (GBMs) (Parsons et al., 2008; The Cancer Genome Atlas Research Network, 2008). Interestingly, we also observed gliomas in zebrafish lacking nf1 and p53. Likewise, MPNSTs are observed in both mouse and zebrafish models. Thus, the relative advantages of murine and zebrafish systems can be leveraged for future studies aimed at developing therapeutics for these lethal complications of NF1.

In summary, we have developed and characterized zebrafish lines containing specific targeted mutations in nf1a and nf1b. Compound deficiency of nf1a and nf1b results in lethality and predisposes to tumor formation. These studies provide a powerful new tool for analysis of neurofibromin function and for the development of therapies for a common human disorder.

The nf1a^Δ5, nf1a^Δ8, nf1b⁺¹⁰ and nf1b^Δ55 mutant alleles were generated by application of modularly assembled ZFNs. The nf1a^L1247X mutant allele was generated by TILLING. Our nf1 mutant alleles were crossed into various transgenic lines, including Tg(gfap:GFP) (Lam et al., 2009), Tg(sox10:GFP) (Thermes et al., 2002; Carney et al., 2006) and Tg(olig2:GFP) (Shin et al., 2003), as well as the p53^e7/e7 mutant line (Berghmans et al., 2005). Zebrafish were maintained under standard conditions as previously described (Westerfield, 2000). All experiments involving animals were approved by the Institutional Animal Care and Use Committees of Harvard University and the University of Pennsylvania.

Individual samples from a preconstructed ‘live library’ of pooled genomic DNA from ENU-mutagenized F1 animals were used as a template for PCR with the following PCR primer pairs: nf1a_outer_F, 5′-TGGCAAATAAATGCTGACAGA-3′ and nf1a_inner_F, 5′-HEX-TTTTTATATCTCATGTTTAGCTCAC-AA-3′; nf1a_outer_R, 5′-AAGTCTTAAATGGCCTGAGTGG-3′ and nf1a_inner_R, 5′-6FAM-AAATGGCCTGAGTGGTAATAAA-3′. A nested PCR was performed first using the outer primer pair with the following PCR conditions: 94°C for 2 minutes; 30 cycles of 94°C for 30 seconds, 60°C for 40 seconds and 72°C for 1 minute; and 72°C for 5 minutes. Amplification was then performed using the inner primer pair with the following PCR conditions: 94°C for 2 minutes; 25 cycles of 94°C for 30 seconds, 60°C for 40 seconds and 72°C for 1 minute; and 72°C for 5 minutes. PCR products were denatured, allowed to re-anneal, subjected to CEL-I digestion and separated by acrylamide electrophoresis using a LI-COR DNA analyzer. Upon identification of a genomic DNA sample harboring a mutation in the analyzed region, the individual animals comprising that genomic DNA pool were rescreened to identify the appropriate F1 animal harboring the lesion of interest. This F1 animal was then outcrossed to wild-type fish and progeny were selected on the basis of the presence of the desired mutation. A genotyping strategy was developed to identify animals harboring the nonsense allele isolated by our TILLING strategy using the following PCR primers: nf1a_stop_PstI_F, 5′-CTCTCTTCGA-CTCTCGCCATCTGCTGTATCAGCTGC-3′ and nf1a_stop_R, 5′-GAAGCAGAAGGTCATAATCTTGCTGGCTAGGC-3′. PCR conditions were 95°C for 2 minutes; 32 cycles of 94°C for 30 seconds, 62°C for 30 seconds and 72°C for 30 seconds; and 72°C for 5 minutes. This generates a 134 bp PCR product. The wild-type allele is resistant to PCR digestion by PstI, but the mutant allele is not.

The desired three finger zinc finger proteins (ZFPs) were constructed by a splice overlap extension PCR strategy, with individual finger modules amplified from a archive of ZFPs with defined DNA-binding specificities (Zhu et al., 2011). Three individual fingers (F1, F2 and F3) were amplified using primers specific to their desired backbone position, followed by an overlapping PCR step to place them together into a single ZFP fragment which was cloned, sequence verified and subsequently subcloned in frame with a FokI nuclease variant in a pCS2 expression plasmid. The primers for the three backbone positions were as follows: F1 forward, 5′-GCGATGG-GTACCCGCCCATATGCTTGC-3′ and F1 reverse, 5′-CACTGGAAGGGCTTCTGGCCTGTGTGAATCCGGATGTG-3′; F2 forward, 5′-CATCCGGATTCACACAGGCCAGAA-GCCCTTCCAGTGTCGCATCTGC-3′ and F2 reverse, 5′-ATGTCGCATGCAAAAGGCTTCTCGCCTGTGTGGGTGCGG ATGTG-3′; F3 forward, 5′-CGAGAAGCCTTTTGCATGCGACA-3′ and F3 reverse, 5′-GCGTAGGATCCACCTGTGTGGATCTT-GGTGTG-3′. The PCR conditions for amplifying F1 were 98°C for 2 minutes and 15 cycles of 94°C for 30 seconds, 68°C for 30 seconds and 72°C for 20 seconds. The PCR conditions for F2 and F3 were 98°C for 2 minutes and 20 cycles of 94°C for 30 seconds, 57°C for 30 seconds and 72°C for 30 seconds. The three individual gel-purified PCR products for positions F1, F2 and F3 were combined (15 ng of each) in a PCR using Advantage 2 HiFi polymerase (Clontech) and subjected to the thermal cycling of 94°C for 2 minutes and five cycles of 94°C for 30 seconds, 55°C for 30 seconds and 72°C for 30 seconds. Following this cycling program, F1 forward and F3 reverse primers were added to the reaction and thermal cycling was continued as follows: 25 cycles of 94°C for 15 seconds and 68°C for 30 seconds. This PCR product of approximately 300 bp was gel-purified, cloned, sequence verified and subcloned in frame with the DD/RR or EL/KK FokI variants (Miller et al., 2007; Szczepek et al., 2007) with and without the 3′-UTR of nanos1 (Koprunner et al., 2001).

We utilized protocols similar to those recently described (Zhu et al., 2011). pCS2-based expression plasmids containing our constructed ZFNs were linearized downstream of the SV40 polyadenylation signal and used as templates for in vitro transcription of ZFN mRNAs (Ambion). One-cell fertilized zebrafish embryos were injected with varying amounts of DD/RR or EL/KK FokI variant mRNAs. Site-specific ZFN function was verified by a genotyping strategy wherein introduction of mutagenic lesions at the target site leads to loss of a unique endogenous restriction site. This same strategy was later used to genotype F1 animals. For nf1a and nf1b, PCR was performed using the following primer pairs and PCR conditions: nf1a F genotyping primer, 5′-GGTGTGTATGTAAATGGGCTCAATG-3′; nf1a R genotyping primer, 5′-TACAGTTTCCATAAAACCTGACATTTC-3′; nf1b F genotyping primer, 5′-TGCTACCTGCCGGCAGGCTCAG-3′; and nf1b R genotyping primer, 5′-ACCTGTGACCATCA-TGTTACTGACA-3′. PCR conditions for nf1a were 94°C for 2 minutes; 35 cycles of 94°C for 30 seconds, 54.8°C for 30 seconds and 72°C for 40 seconds; and 72°C for 5 minutes. PCR conditions for nf1b were 94°C for 2 minutes; 35 cycles of 94°C for 30 seconds, 62°C for 30 seconds and 72°C for 40 seconds; and 72°C for 5 minutes. For both nf1a and nf1b, a 223-bp PCR product was generated. Mutant alleles were resistant to subsequent digestion with BspCNI or DdeI (nf1a) and BslI or EcoNI (nf1b), whereas the wild-type alleles were completely digested. The molecular identities of the various mutant alleles were determined by cloning and sequencing the restriction enzyme-resistant PCR products from individual embryos derived from outcrosses of F1 animals to wild-type fish.

A modified fin clip genotyping strategy was utilized to identify larval genotypes prior to sacrifice. Briefly, 2-dpf larvae were anesthetized with tricaine and stereomicroscopic amputation was performed on the caudal fin with an angled dissecting knife (Fine Science Tools, 10056-12). Genomic DNA from fin-clipped tissue samples was prepared by collecting specimens in 1.5 μl of the surrounding medium and dispensing into PCR tubes containing 7.5 μl of 60 mM NaOH, which were incubated at 95°C for 20 minutes followed by 4°C for 5 minutes with the subsequent addition of 1 μl of 1 M Tris-HCl pH 8. Genotyping for nf1a or nf1b was performed as described with an increase in PCR cycle number to 40. Genomic DNA prepared by this strategy was sufficient for a single genotyping reaction. Thus, parental crosses were selected to ensure all progeny were homozygous for the non-genotyped nf1 allele (e.g. nf1a^+/−; nf1b^−/− incross). This was verified by genotyping the homozygous mutant allele in sibling clutchmates.

Antisense RNA probes were generated for plp1a (Park et al., 2002) and mbp (Lyons et al., 2005) using a digoxigenin RNA labeling kit (Roche). A previously published protocol (Thisse and Thisse, 2008) was followed with minor modifications. Terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) staining of larvae was performed using the ApopTag Peroxidase In Situ Apoptosis Detection Kit (EMD Millipore) with minor modifications. Larvae were subsequently blocked in 2% blocking reagent (Roche) for 4 hours at room temperature and incubated overnight in anti-digoxigenin-POD antibody (Roche; 1:500) at 4°C. TUNEL-positive cells were detected using the TSA fluorescein system (PerkinElmer). Apoptotic cells were quantified by counting TUNEL-positive cells between somites 6 and 15. For immunohistochemical analysis of zebrafish larvae, we used mouse anti-BrdU (Developmental Studies Hybridoma Bank G3G4; 1:1000), mouse anti-HuC/D (Invitrogen A-21271; 1:100), mouse anti-pERK1/2 (Sigma; 1:200), mouse anti-Zrf1 (Zebrafish International Resource Center; 1:1000), rabbit anti-BLBP (Millipore AB9558; 1:1000), rabbit anti-pS6 (Cell Signaling 2211; 1:200), and rabbit anti-SOX10 (Park et al., 2005) (a generous gift from Bruce Appel, University of Colorado, Denver, CO; 1:5000) as primary antibodies. For fluorescent detection of antibody labeling, we used anti-mouse and anti-rabbit IgG antibodies conjugated with Alexa Fluor 488, 568 and 647 (Invitrogen; 1:200). Immunohistochemistry of adult zebrafish tumor samples was performed according to a previously published protocol (Ligon et al., 2004). Primary antibodies included anti-pERK1/2 (Cell Signaling 4370; 1:200), anti-pS6 (Cell Signaling 2211; 1:50), anti-phosphorylated histone H3 (Cell Signaling 9706; 1:100), rabbit anti-SOX10 (Park et al., 2005) (1:3000), mouse anti-GFAP (Sigma G3893; 1:10,000), anti-HuC/D (Invitrogen A-21271; 1:200) and anti-synaptophys (Millipore MAB5258; 1:1000). Antibody binding was detected using a diaminobenzidine-peroxidase visualization system (EnVision+, Dako). Mayer’s hematoxylin was used for counterstaining. For BrdU labeling, embryos were incubated in BrdU solution (10 mM BrdU in 2% DMSO) for 12 hours. After BrdU incubation, embryos were fixed with 4% paraformaldehyde and embedded in 1.5% agarose. Sections from embedded frozen specimens were immersed in 2 M HCl for 15 minutes and processed for immunohistochemistry. Paraffin sectioning followed by hematoxylin and eosin (H&E) staining was performed at the Dana-Farber/Harvard Cancer Center Research Pathology Core.

Startle behavioral experiments were performed on 5-dpf larvae raised as previously described (Burgess and Granato, 2007b). Larvae with underinflated swim bladders were excluded from behavioral testing. Acoustic startle responses were elicited and measured as previously described (Burgess and Granato, 2007a; Wolman et al., 2011), such that larvae could be tracked and analyzed individually. All startle stimuli were 1000 Hz waveforms of 3 milliseconds duration at an intensity of approximately 150 m/second². Stimulus intensity was calculated by measuring the displacement of the testing arena due to vibration. To evaluate short latency C-start (SLC) behavior, images were recorded 30 milliseconds prior to and 90 milliseconds following the delivery of the 3 millisecond acoustic stimulus. To examine acoustic startle larval motor behaviors, we captured video recordings using a MotionPro high-speed camera (Redlake) at 1000 frames per second with 512×512 pixel resolution using a 50 mm macro lens. Behavioral analyses were carried out with the FLOTE software package (Burgess and Granato, 2007b; Burgess and Granato, 2007a). Startle short-term habituation was performed and analyzed as previously described (Wolman et al., 2011). Larvae were genotyped following behavioral testing.

Protein lysates were prepared from 3-dpf wild-type, nf1a^−/−; nf1b^+/+, nf1a^+/+; nf1b^−/− and nf1a^−/−; nf1b^−/− larvae. Briefly, groups of 10–20 larvae with identical genotype were anesthetized with tricaine, deyolked in a solution of ice-cold PBS with 0.1% Tween-20 (PBST), transferred to a pre-chilled microcentrifuge tube containing 5 μl of lysis buffer (150 mM NaCl, 50 mM Tris-HCl pH 8, 5 mM EDTA, 1 mM PMSF, 1% Igepal CA 630, 0.5% sodium deoxycholate, 0.1% SDS and 1× Halt protease inhibitor cocktail (Thermo Fisher Scientific) per larva, sonicated using a Bioruptor (Diagenode) and cleared by centrifugation. Larval (100 μg) and 293T (25 μg) protein lysates were separated by gel electrophoresis, transferred to PVDF membranes and probed overnight at 4°C with the following primary antibodies: anti-neurofibromin (Abcam ab17963; 1:1000), anti-α-tubulin (Sigma T6074; 1:1000), anti-pERK1/2 (Cell Signaling 4377; 1:1000), and anti-ERK1/2 (Cell Signaling 9102; 1:1000). Primary antibody binding was visualized on X-ray film using anti-mouse-HRP (Cell Signaling 7076; 1:10,000) or anti-rabbit-HRP (Cell Signaling 7074; 1:10,000) secondary antibodies along with LumiGLO (Cell Signaling) or SuperSignal West Femto (Thermo Fisher Scientific) chemiluminescent substrates. Each Western blot was performed in three independent replicates with a representative image of one presented.

Lateral stripe melanophores, including those observed along the horizontal myoseptum, were counted by a blinded observer in live progeny from nf1a^+/−; nf1b^+/− incrosses at 3 and 6 dpf. Each discrete melanized region was counted as a single melanophore. Following 6-dpf lateral stripe melanophore quantitation, larvae were immersed in E3 medium containing 5 mg/ml epinephrine (Sigma E4375) for 10 minutes to induce contraction of melanosomes around cell bodies and allow evaluation of head melanophore numbers. Larvae were fixed overnight in 4% paraformaldehyde in PBS at 4°C, washed twice with PBS for 5 minutes, and melanophores anterior to somite one across the crown of the head (supplementary material Fig. S14) were counted. Larvae were subsequently genotyped for nf1a and nf1b. Data analysis was performed by one-way ANOVA with Dunnett’s post-test (GraphPad InStat 3.1a, GraphPad Software).

Lateral stripe melanophores of live progeny from nf1a^+/−; nf1b^−/− or wild-type incrosses were individually imaged at 3 dpf followed by incubation with 0.2 mM N-phenylthiourea (PTU, Sigma) to prevent melanin synthesis. At 5 dpf, lateral stripe melanophores were again imaged, after which PTU was washed out. Larvae were reimaged at 6 dpf and subsequently genotyped for nf1a. Images were acquired using a Nikon SMZ1500 microscope and NIS-Elements F2.20 software with identical settings. Melanophores in a 200-μm region, corresponding to roughly 12 somites at 3 dpf, were counted at 3 and 6 dpf for each larva. Statistical analysis was performed using a one-tail, unpaired t-test (GraphPad Prism 5, GraphPad Software).

Groups of 5-dpf wild-type and nf1a^−/−; nf1b^−/− larvae were placed in individual wells of six-well plates containing feeding medium [4 ml E3, 2 ml paramecia culture and 5 μl of 2-μm yellow-green microspheres (Polysciences 18338)]. Larvae were incubated at 28.5°C for 1 hour followed by several E3 washes. Individual larvae were transferred to 24-well plates and visually assessed for the presence of fluorescent microspheres in the intestinal bulb. Extent of intestinal transit was observed at 2, 4, 6 and 24 hours. Transit was considered to be complete when fluorescence was no longer detectable in the intestinal tract.

TEM was carried out at the Harvard Medical School Electron Microscopy Facility. Briefly, embryos were fixed in 2% formaldehyde and 2.5% glutaraldehyde, post-fixed with 1% osmium tetroxide and embedded in epon. Sections were collected in the trunk region of embryos. Images were captured by a Tecnai G² Spirit BioTWIN electron microscope with an AMT 2k CCD camera.

nf1a^Δ8/+; nf1b^+10/+10; p53^e7/e7 fish were incrossed and progeny were manually evaluated weekly for 45 weeks. Animals identified as having tumors were separated and fin-clipped for genotyping purposes. These samples were subsequently subjected to histological and immunohistochemical analyses as described above for determination of tumor type. At the completion of 45 weeks, all tumor-free fish were genotyped for subsequent statistical analysis.

We thank members of the Epstein and Look laboratories for helpful discussions, especially Rajan Jain and Nikhil Singh. We are grateful to Jie He for her expert assistance with zebrafish husbandry; Michael Pack and Mary Mullins for thoughtful suggestions; Joshua Abrams, Jesse Isaacman-Beck and Allison Rosenberg for technical instruction; and Thomas Joseph for critical reading of the manuscript.