Mice with missense and nonsense NF1 mutations display divergent phenotypes compared with human neurofibromatosis type I

ABSTRACT Neurofibromatosis type 1 (NF1) is a common genetic disorder characterized by the occurrence of nerve sheath tumors and considerable clinical heterogeneity. Some translational studies have been limited by the lack of animal models available for assessing patient-specific mutations. In order to test therapeutic approaches that might restore function to the mutated gene or gene product, we developed mice harboring NF1 patient-specific mutations including a nonsense mutation (c.2041C>T; p.Arg681*) and a missense mutation (c.2542G>C; p.Gly848Arg). The latter is associated with the development of multiple plexiform neurofibromas along spinal nerve roots. We demonstrate that the human nonsense NF1Arg681* and missense NF1Gly848Arg mutations have different effects on neurofibromin expression in the mouse and each recapitulates unique aspects of the NF1 phenotype, depending upon the genetic context when assessed in the homozygous state or when paired with a conditional knockout allele. Whereas the missense Nf1Gly848Arg mutation fails to produce an overt phenotype in the mouse, animals homozygous for the nonsense Nf1Arg681* mutation are not viable. Mice with one Nf1Arg681* allele in combination with a conditional floxed Nf1 allele and the DhhCre transgene (Nf14F/Arg681*; DhhCre) display disorganized nonmyelinating axons and neurofibromas along the spinal column, which leads to compression of the spinal cord and paralysis. This model will be valuable for preclinical testing of novel nonsense suppression therapies using drugs to target in-frame point mutations that create premature termination codons in individuals with NF1. Summary: Two mouse models carrying NF1 patient-specific mutations have distinct effects on embryonic development, neurofibromin function and plexiform neurofibroma formation, with implications for therapeutic studies.


INTRODUCTION
Neurofibromatosis type 1 (NF1) is a common genetic disorder affecting ∼1 in 3000 individuals and over two million people worldwide (Friedman, 1999;Evans et al., 2010). NF1 is characterized principally by the occurrence of neurofibromas and café-au-lait macules (CALMs) (Boyd et al., 2009). Both children and adults with NF1 are at an increased risk for developing certain malignancies and display an assortment of benign and malignant lesions including neurofibromas (dermal and plexiform), pheochromocytomas, pilocytic astrocytomas, optic gliomas, bone abnormalities, leukemias and malignant peripheral nerve sheath tumors (Boyd et al., 2009). NF1 is caused by inherited or de novo pathogenic variants in the NF1 gene, whereas tumor development requires additional mutation of the other NF1 allele (loss of heterozygosity) (Garcia-Linares et al., 2011;Steinmann et al., 2009). To date, no strong genotype-phenotype correlation has been established for NF1 because of the varied disease expression observed in humans. Many individuals with NF1, even from the same family, show differences in the presentation or severity of the NF1 phenotype (Pasmant et al., 2012;Alkindy et al., 2012;Sabbagh et al., 2009). This is exemplified by the National Institutes of Health (NIH) diagnostic criteria for NF1, which require a minimum of two of the following: six or more CALMs, two or more cutaneous neurofibromas or one plexiform neurofibroma, axillary or inguinal freckling, optic pathway glioma, two or more Lisch nodules, bony dysplasia, or a first degree relative with NF1 (Ferner et al., 2007). Furthermore, there are few in vitro assays available to examine neurofibromin function, which is especially problematic when trying to assess the potential pathogenicity of missense variants.
Several mouse models have been developed that display some of the phenotypes seen in individuals with NF1, including learning difficulties, as well as the development of plexiform neurofibromas, optic gliomas and malignant peripheral nerve sheath tumors (Boyd et al., 2009). These existing animal models generally employ conditional knockout strategies that lead to loss of function of one or both Nf1 alleles, which have proven to be valuable tools to investigate the pathogenesis of NF1, and to evaluate several therapeutic strategies such as inhibitors of RAS or intercellular signaling (Yang et al., 2008;Robertson et al., 2012;Jessen et al., 2013). However, the underlying mutations used in current animal models are not sufficient to evaluate novel therapeutic approaches to target the mutated gene or gene product, including nonsense suppression and exon-skipping therapies. These therapies have been established in Duchenne muscular dystrophy and cystic fibrosis (Du et al., 2002(Du et al., , 2006De Luca et al., 2008;Loufrani et al., 2004;Barton-Davis et al., 1999;Zhang et al., 2014;Xue et al., 2014;Mann et al., 2001;Lu et al., 2005;van Deutekom et al., 2001;Aartsma-Rus et al., 2003;Igreja et al., 2016). Thus, the use of NF1 mouse models harboring mutations identified in individuals with NF1 provides the ability to perform prospective investigations targeting a specific underlying mutation or mutation type.
To this end, we created and characterized two novel mouse models harboring mutations found in individuals with NF1, including a common nonsense mutation (nmNF1, exon 18 c.2041C>T; p.Arg681*) expected to produce complete loss of function, and a recurring missense mutation found in individuals with NF1 associated with multiple plexiform neurofibromas along spinal nerve roots (exon 21 c.2542G>C; p.Gly848Arg). We demonstrate that the nonsense c.2041C>T; p.Arg681* (hereafter called Nf1 Arg681* ) and missense c.2542G>C; p.Gly848Arg (hereafter called Nf1 Gly848Arg ) mutations in the mouse have different effects on neurofibromin expression, and each recapitulates unique aspects of NF1 depending upon the genetic context when assessed in the homozygous state, or when paired with a conditional knockout allele.

Establishing mouse models with NF1 patient mutations
Nonsense mutations make up ∼20% of characterized mutations (Messiaen and Wimmer, 2008). The Medical Genomics Laboratory at the University of Alabama at Birmingham has identified 19 of the most recurrent nonsense mutations found in individuals with NF1 (each responsible for at least 0.3% of NF1 cases). Among those is a nonsense mutation in exon 18, c.2041C>T; p.Arg681* observed in 49/8000 unrelated NF1-mutation-positive probands (0.61%). Of 39 individuals with phenotypic data available, 15 individuals were 0-2 years of age at the time of testing (mostly too young to fulfill clinical diagnostic criteria). Within these individuals, 2/15 fulfill NIH criteria with 2/15 individuals having <6 CALMs. The remaining 24 individuals are ≥3 years of age and all fulfill NIH criteria for NF1 diagnosis; although none possess symptomatic spinal neurofibromas (including half who were ≥19 years old), 8/24 individuals did have an externally visible plexiform neurofibroma.
In contrast, a missense mutation located in exon 21 (c.2542G>C; p.Gly848Arg), observed in 11/7800 unrelated NF1-mutationpositive probands (0.14%), is associated with symptomatic spinal nerve root tumors. Phenotypic data were obtained from 20 individuals from four different families with a p.Gly848Arg mutation. Three of these individuals are 0-2 years of age, with 1/3 fulfilling the NIH criteria and 2/3 having <6 CALMs, whereas 7/17 of the individuals aged ≥3 years fulfill the NIH criteria, with 4/17 having an externally visible plexiform neurofibroma. Symptomatic spinal nerve root tumors were found in 6/17 individuals, including 5/8 individuals who were ≥19 years old. For individuals 19 years or older, there is a significant difference in the frequency of spinal nerve root tumors between the c.2041C>T (0/12) vs c.2542G>A (5/8) mutation (2-tailed Fisher exact: P=0.0036).
To model each human mutation in the mouse, knock-in animals harboring the nonsense c.2041C>T; p.Arg681* mutation allele in exon 18 (Nf1 Arg681* ) and the missense c.2542G>C; p.Gly848Arg mutation allele in exon 21 (Nf1 Gly848Arg ) were created using conventional gene-targeting vectors and C57BL/6 embryonic stem cells, with chimeric mice and germ line transmission being confirmed (Table S1). Since the Nf1-floxed animals that have been used in previous studies were generated on a mixed genetic background (Zhu et al., 2002), we obtained mice from the International Knockout Mouse Consortium (IKMC) harboring a 'knockout first' allele for the Nf1 gene (Nf1 tm1a(KOMP)Wtsi ) in order to have all Nf1 alleles on the inbred C57BL/6 background. We generated a new conditional knockout 'floxed' allele of the Nf1 gene where exon 4 is flanked by loxP sites (hereafter referred to as Nf1 4F ) after removal of the FRT-flanked LacZ/Neo cassettes by breeding to a FLP-recombinase line. Matings between heterozygous Nf1 +/4F mice yielded homozygous floxed Nf1 4F/4F mice at the expected Mendelian ratio that were viable, fertile and did not show gross or histological abnormalities (data not shown). Heterozygous Nf1 +/Δ4 mice harboring a null allele resulting from exon 4 deletion were created by breeding Nf1 +/4F mice to an EIIaCre driver line [B6. FVB-Tg(EIIa-cre)C5379Lmgd/J strain #003724], thereby deleting the floxed sequence in all cells within the developing embryos.
Nonsense mutation NF1 Arg681* recapitulated in the mouse is non-functional Previous studies have demonstrated that mouse embryos fail to develop beyond embryonic day (E)13.5 in the absence of Nf1 expression as a result of cardiovascular defects (Jacks et al., 1994;Brannan et al., 1994). To determine the functionality of each allele, we performed intercross breedings of Nf1 +/Arg681* and Nf1 +/Δ4 mice and examined embryos from E9.5 to E13.5. Embryos heterozygous for either the Nf1 Arg681* or the Nf1 Δ4 allele were viable and developed to term with no observable defects. At E9.5, all homozygous Nf1 Arg681*/Arg681* and Nf1 Δ4/Δ4 embryos were indistinguishable from their littermate controls; however, these homozygous embryos fail to develop past this point (Fig. S2). Although Nf1 Arg681*/Arg681* mice display pericardial effusions at E10.5 (Fig. 1A,B), we were unable to ascertain the exact cause of death because of extensive tissue degradation.
Nonsense mutations create premature termination codons (PTCs) in genes that in turn trigger nonsense-mediated mRNA decay (NMD) to selectively and rapidly degrade PTC-bearing aberrant transcripts. To determine if Nf1 Arg681* mutations induce NMD, we compared levels of wild-type versus mutant mRNA in Nf1 +/Arg681* mice. It is expected that in the absence of NMD, the percentage ratio of mutant to wild-type mRNA would be 50:50; however, if NMD were occurring, the distribution of mRNA would be skewed. RNA was extracted from brain tissues of Nf1 +/Arg681* mice and reverse transcription polymerase chain reaction (RT-PCR) was performed using primer sets flanking the nonsense mutation. The amplicons were then cloned into a TA vector for sequencing. Of twenty-two clones sequenced, only two (9.1%) were identified with the Nf1 Arg681* mutation whereas the remaining twenty clones (90.9%) contained wild-type sequences. This skewed and unequal representation of wild-type versus mutant clones (2-tailed Fisher exact; P=0.0068) indicates the occurrence of NMD of the Nf1 Arg681* allele transcripts. These findings indicate that both Nf1 Arg681* and the novel Nf1 Δ4 alleles significantly ablate Nf1 gene expression and function.
In the appropriate genetic context, the Nf1 Arg681* mutation induces neurofibroma formation in mice As described above, individuals with NF1 harboring nonsense NF1 Arg681* mutations may develop plexiform neurofibromas; therefore, we hypothesized that we could recapitulate this phenotype in Nf1 Arg681* mice in the appropriate genetic context. Using conditional knockout models, it has been established that neurofibroma development can be stimulated through Nf1 loss of heterozygosity within the Schwann cell lineage when driven by Cre recombinase expression (Wu et al., 2008). To investigate whether the Nf1 Arg681* and the Nf1 4F allele can induce neurofibroma development in vivo, we generated two conditional knockout NF1 mouse models employing a DhhCre transgenic mouse line. In this model system, Cre recombinase is expressed from desert hedgehog regulatory sequences at E12.5, with previous demonstration that inactivation of the Nf1 gene in Nf1 31&32F/31&32F ; DhhCre mice (with exons 31 and 32 flanked by loxP sites) elicits the formation of plexiform and subcutaneous neurofibromas (Wu et al., 2008).
To determine the source of the paralysis, we performed gross dissection of each Nf1 4F/Arg681* ; DhhCre or Nf1 4F/4F ; DhhCre mouse. Tumors were observed in 81.5% (44/54) of Nf1 4F/Arg681* ; DhhCre and 79.2% (19/24) of Nf1 4F/4F ; DhhCre mice. All paralyzed mice were found to have visible tumors arising at the dorsal root ganglia resulting in compression of the spinal cord. Gross tumor numbers ranged from one to thirteen per mouse, with 90.9% detected in the cervical or thoracic spine and 9.1% detected in the lumbar spine (Fig. 2B, Fig. 3A). These tumors seemed to arise around the dorsal root ganglia, with invasion of the neural foramen, leading to compression of the spinal cord. In addition to these lesions, we also observed enlarged spinal nerve roots and peripheral nerves in the mice examined (data not shown). Thus, Nf1 4F/Arg681* ; DhhCre mice develop neurofibromas with high efficiency.
Upon histological analysis, neurofibromas are characterized by disorganized Remak bundles (the non-myelinated axon-Schwann cell unit) and enriched deposition of collagen. Similar lesions were seen from tumor preparations of both models at the ultrastructural level ( Fig. 2C-F, Fig. 3B-D). These data indicate that Nf1 Arg681* functions as a null allele, and recapitulates key characteristics of human disease in the proper genetic context.

Missense mutation NF1 Gly848Arg recapitulated in the mouse retains function
Mice homozygous for the Nf1 Gly848Arg missense allele (Nf1 Gly848Arg/Gly848Arg ) are viable and are indistinguishable from their heterozygous (Nf1 +/Gly848Arg ) siblings. A cohort of 19 Nf1 Gly848Arg/Gly848Arg mice were monitored throughout their lifetimes for tumor development. No overt phenotype was observed. All animals underwent gross dissection, and we were unable to detect tumors in the spinal columns or the peripheral nerves. The distinct manifestations between Nf1 Gly848Arg and Nf1 Arg681* might suggest different underlying Nf1 gene expression and function. We therefore derived MEFs from E13.5 embryos and assessed neurofibromin expression by western blot (Fig. 4A). The Nf1 Gly848Arg allele impairs neurofibromin expression, as Nf1 Gly848Arg/Gly848Arg MEFs show ∼50% less protein than cells from wild-type controls, which is similar to that observed in Nf1 +/Δ4 samples (Fig. 4B). In response to neurofibromin reduction, pERK is hyperactivated in Nf1 Gly848Arg/Gly848Arg MEFs to a level comparable with Nf1 +/Δ4 samples, suggestive of increased Ras activity (Fig. 4C,D). Therefore, Nf1 Gly848Arg only partially decreases Nf1 gene expression in mouse. Given that spinal neurofibromas form the primary manifestation of this mutation in humans, we examined the spinal nerve roots of Nf1 Gly848Arg/Gly848Arg mice by electron microscopy for ultrastructural abnormalities. Neither myelinated nor non-myelinated axons were disrupted when compared with controls. (Fig. 4E,F).
Individuals with NF1 carry second (somatic) mutations that are distinct from germline mutations. We hypothesized that Nf1 Gly848Arg mice carrying a second null mutation in Nf1 might develop neurofibromas, which we tested by generating Nf1 Δ4/Gly848Arg mice. Similar to Nf1 Gly848Arg/Gly848Arg mice, these animals were viable and displayed no overt phenotype. We monitored ten Nf1 Δ4/Gly848Arg mice for up to twenty-four months and still were unable to find tumors. Therefore, we conclude that the missense mutation Nf1 Gly848Arg only partially impacts Nf1 gene function, and is unable to induce neurofibroma development in mice.

DISCUSSION
The factors that underlie the clinical heterogeneity seen in NF1 are still unclear. Assessment of the functional and physiological consequences of putative mutations is hampered by the lack of robust in vitro and in vivo assays, and limited knowledge of neurofibromin protein domains. In this study, two animal models were engineered to harbor NF1 mutations found in human patients, specifically nmNF1 (nonsense c.2041C>T; p.Arg681*) and a missense mutation of NF1 (missense c.2542G>C; p.Gly848Arg). This approach is possible given the greater than 98% amino acid sequence homology between the mouse and human neurofibromin protein (Bernards et al., 1993). Within the regions encoded by exons 18 and 21, the protein sequence homology is 98% and 96%, respectively (Fig. S3). The clinical features associated with these mutations in humans are distinct. Individuals with a p.Gly848Arg mutation suffer from multiple plexiform neurofibromas along the spinal cord but a paucity of dermal neurofibromas, whereas individuals with the c.2041C>T; p.Arg681* premature stop Examining the effects of these two different NF1 mutations in mouse models, we would like to make four points. Firstly, the two mutations have distinct effects on embryonic development and neurofibromin expression in the mouse. The Nf1 Arg681* allele functions as a null allele with complete loss of Nf1 function, whereas Nf1 Gly848Arg appears to retain some function. The Nf1 Arg681* allele is embryonic lethal when homozygous, whereas Nf1 Gly848Arg homozygous animals are viable. In accordance with embryonic lethality, complete ablation of neurofibromin expression was observed in Nf1 Arg681*/Arg681* cells whereas neurofibromin expression in the Nf1 Gly848Arg/Gly848Arg MEFs was reduced by ∼50%. Our data are consistent with a previous study on iPS cells and fibroblasts from individuals with NF1, which demonstrated that different germline NF1 mutations result in different degrees of neurofibromin expression (Anastasaki et al., 2015). However, in vitro analyses showed increases in pERK in MEFs from both homozygotes when compared with wild-type. This is comparable with what has been observed in neural progenitor cells from individuals with NF1, as individual NF1 mutations did not differentially affect Ras activation (Anastasaki et al., 2015). The observed changes in pERK suggest that the Nf1 Gly848Arg allele could potentially function as a hypomorphic allele, though further studies would be necessary to confirm this hypothesis.
Secondly, the Nf1 Gly848Arg allele failed to recapitulate the NF1 phenotype; whereas humans with NF1 suffer from multiple spinal plexiform neurofibromas, the mouse did not display an overt tumorigenic phenotype. A possible explanation for this discrepancy is the presence of modifier genes in the mouse impeding neurofibroma formation. In contrast, the Nf1 Arg681* allele stimulates tumorigenesis when placed in context with a loss-offunction allele as demonstrated with the Nf1 4F/Arg681* ; DhhCre mice. Similar observations were made in a parallel study where the same Nf1 Arg681* allele in a different mouse model led to development of optic gliomas, but these tumors were not seen in Nf1 Gly848Arg mice (Toonen et al., 2016).
Thirdly, the Nf1 4F/Arg681* ; DhhCre model displays a robust tumor phenotype, with 81.5% of animals grossly developing neurofibromas throughout their lifetime. Previous studies using the Nf1 31&32F/31&32F ; DhhCre model indicated that mice became paralyzed in one or both hind limbs in 28.6% of animals by six months of age (Wu et al., 2008); however, our data with the Nf1 4F/Arg681* ; DhhCre model reveals an earlier onset of paralysis with tumorigenesis at 5 months of age in 55.6% of mice. These differences may either stem from the Nf1 Arg681* allele, which provides a suitable environment for neurofibroma development, or from the conditional knockout of exon four versus the previously described exon 31 and 32, as the Nf1 4F/4F ; DhhCre mice also appear to exhibit higher penetrance than the existing Nf1 31&32F/31&32F ; DhhCre model. This hypothesis is further supported by the notable differences in embryonic lethality between the Nf1 Δ4 allele, Nf1 Arg681* and the previously published Nf1 31&32F allele. As previously reported, embryonic death of Nf1 Δ31&32/Δ31&32 animals occurred from E11.5 to E13.5 as a result of defects in cardiac development (Jacks et al., 1994;Brannan et al., 1994). However, all embryos carrying the newly developed Nf1 Δ4/Δ4 or Nf1 Arg681*/Arg681* died around E10. Additionally, a parallel study observed that optic gliomas from mice harboring the Nf1 Arg681* allele are more aggressive than those developing in mice with the traditional Nf1 Δ31&32/Δ31&32 mutation, with more microglia and increased proliferative indices (Toonen et al., 2016). While it is difficult to make direct comparisons and draw conclusions across these studies because of the difference in null alleles utilized, collectively these data suggest the Nf1 Δ4 and Nf1 Arg681* alleles might yield a more severe phenotype. Another possibility for the observed differences between Nf1 4F/4F ; DhhCre and Nf1 31&32F/31&32F ; DhhCre mice lies in the strains utilized in these studies; all mice in our experiments are on an inbred C57BL/6 background whereas previous studies used mice on a mixed 129/C57BL/6 background (Wu et al., 2008). Furthermore, Nf1 4F/Arg681* ; DhhCre mice possess a different tumor microenvironment resulting from the nonsense mutation present in all somatic cells. Of note, the neurofibromas arising in Nf1 4F/Arg681* ; DhhCre mice occur in a divergent location when compared with their human counterparts, as these animals develop tumors along the spinal column, whereas symptomatic spinal neurofibromas were absent in humans. This is likely due to the DhhCre driver line, which specifically promotes tumorigenesis along the spinal column per previous models (Wu et al., 2008).
Fourthly, there is a great need to develop non-surgical treatments for NF1 patients. To date, most therapeutic interventions that have been tested in preclinical models or clinical trials have been designed to regulate the downstream targets in the Ras signaling pathway or intercellular communication (Yang et al., 2008;Robertson et al., 2012;Jessen et al., 2013). Few studies have focused on restoring function of the defective NF1 alleles. The development of mutation-guided therapeutics provides new opportunities for treatment of NF1 caused by specific mutations. The robust phenotype of the nmNF1 mouse model and its parallels to human phenotype indicate that Nf1 4F/Arg681* ; DhhCre mice will provide a valuable system for evaluating new therapies targeting the causative genetic mutation, such as nonsense suppression therapies. Nonsense mutations comprise ∼20% of mutations in individuals with NF1 (Messiaen and Wimmer, 2008). For other diseases, therapies (gentamicin, PTC124 and other compounds) have been developed that can suppress nonsense mutations to restore partial protein activity, with some now being used clinically (Du et al., 2002;Keeling et al., 2013;Sangkuhl et al., 2004;Yang et al., 2007;Guerin et al., 2008;Zilberberg et al., 2010). Assessment of nonsense suppression therapy in vivo is absolutely required prior to its assessment in individuals with nmNF1. Currently, premature stop mutation mouse models for other genes are being used to evaluate nonsense suppression therapy, including the Idua-W392X mouse model of human Hurler syndrome (mucopolysaccharidosis type I-H) and a cystic fibrosis transmembrane conductance regulator model (Zhang et al., 2014;Gunn et al., 2014;Wang et al., 2012). A fundamental question to address for any preclinical nonsense mutation model, including our newly established nmNF1 mouse, is whether sufficient protein activity can be rescued using nonsense suppression therapy approaches.
In summary, we have engineered and characterized the first NF1 patient-based germ line NF1 mutations modeled in mice, establishing that the Nf1 Arg681* allele profoundly reduces neurofibromin expression and leads to plexiform neurofibroma formation. This study provides clear evidence that different mutations in NF1 can have unique consequences, and the germline mutation is an important factor underlying NF1 clinical heterogeneity. Further preclinical and translational studies of nonsense suppression therapy are underway utilizing the Nf1 Arg681* allele. Future investigations of additional NF1 models harboring other human patient mutations will be used to better understand the NF1 phenotype, and will serve as translational models for therapeutic studies.

Study approval
The patient data was collected after approval of a human subjects protocol (F080926009, Further Research on Samples Submitted for Molecular Genetic Testing) by the Institutional Review Board of the University of Alabama at Birmingham. All animal experiments were conducted after approval of an animal protocol (APN 130109835, Translational NF1 Mouse Models) by the Institutional Animal Care and Use Committee of the University of Alabama at Birmingham.

Mouse breeding
All animals were maintained on an inbred C57BL/6 background using a temperature-and humidity-controlled vivarium on a 12 h dark-light cycle with free access to food and water. All homozygous Nf1 Gly848Arg/Gly848Arg and Nf1 4F/4F mice were viable, fertile and did not display any detectable embryonic and postnatal phenotypes. Equal numbers of male and female animals were analyzed throughout the study. Nf1 Arg681*/Arg681* and Nf1 Δ4/Δ4 embryos failed to develop beyond E9.5. A DhhCre transgenic mouse line [FVB(Cg)-Tg(Dhh-cre)1Mejr/J] was kindly provided by Dr Dies Meijer (Erasmus University Medical Center, Netherlands) to breed with the Nf1 4F/4F and Nf1 +/Arg681* strains. Cre-negative Nf1 4F/Arg681* or Nf1 4F/4F littermates were used for controls. For analysis of embryos, we established timed matings, E0.5 was designated as noon of the day a vaginal plug was detected.

MEF generation & culture
To generate MEFs, embryos were isolated in cold PBS and the heads and visceral organs were removed. Each embryo was minced into small pieces in trypsin solution and placed at 37°C for 5-10 min. The trypsin solution was then neutralized with two volumes of MEF medium (10% heat-inactivated fetal bovine serum in DMEM). After centrifugation, the supernatant was discarded and the pellet resuspended in fresh MEF medium. The cell suspensions were seeded onto gelatin-coated culture vessels, and incubated at 37°C and 5% CO 2 in a culture incubator until the cells were confluent and ready for passage. These cells were designated as passage zero (P0), only those P4 or less were used in the study.

Nonsense-mediated mRNA decay assay
Total RNA was extracted from brain tissues of adult Nf1 +/Arg681* mice using TRIzol reagent (Thermo Scientific). Polyadenylated RNA was reversed transcribed into cDNA in a 20 µl reaction containing 1 μg of total RNA using SuperScript ® III First-Strand Synthesis kit (Thermo Scientific). Nf1 transcripts were amplified around the region of interest using 5 PRIME MasterMix (5 PRIME, Gaithersburg, MD) with 2 µl of the transcribed cDNA as the PCR template. A fragment of 417 bp was amplified with primers forward: 5′-GTTCCTGCGTGCCTGTACTC and reverse: 5′-TGTTCCCACTTTGCATGTGT. Three standard PCR steps were carried out for 30 cycles with 94°C denaturing, 55°C annealing and 72°C extension. The resulting PCR amplicons were cloned into pCR4-TOP vector from the TOP TA cloning kit (Thermo Scientific). Twenty-two individual clones were sequenced using the T7 primer.

Electron microscopy
Mice of indicated genotypes were intracardially perfused with Karnovsky's fixation solution (3% paraformaldehyde and 3% glutaraldehyde in 0.1 mol/l phosphate buffer, pH 7.4 to 7.6). Sciatic nerves and spinal nerve roots were dissected and post-fixed overnight. Tissues were then transferred to 0.175 mol/l cacodylate buffer, osmicated, dehydrated and embedded. Semi-thin sections were stained in uranyl acetate and lead citrate, and viewed on a FEI Tecnai T12 microscope.

Histological and immunological analyses
Mice were intracardially perfused with 4% paraformaldehyde (PFA, w/v) in PBS. Neurofibromas, sciatic nerves, spinal nerve roots and other tissues were dissected and post-fixed overnight, followed by cryoprotection and frozen sectioning. Serial sections of 12 μm thickness were prepared using a cryostat (CM3050 S; Leica). Hematoxylin and Eosin (H&E) staining was performed according to standard protocols.

Statistical analysis
Statistical analyses were performed using GraphPad Prism 5 software (GraphPad Software, Inc., La Jolla, CA). Statistical comparisons were analyzed with one-way ANOVA followed by a Bonferroni post-test. A value of P<0.05 was considered statistically significant. Data are presented as mean±s.e.m.