Neurofibromatosis type 1-associated tumours: Their somatic mutational spectrum and pathogenesis

Somatic gene mutations constitute key events in the malignant transformation of human cells. Somatic mutation can either actively speed up the growth of tumour cells or relax the growth constraints normally imposed upon them, thereby conferring a selective (proliferative) advantage at the cellular level. Neurofibromatosis type-1 (NF1) affects 1/3,000-4,000 individuals worldwide and is caused by the inactivation of the NF1 tumour suppressor gene, which encodes the protein neurofibromin. Consistent with Knudson's two-hit hypothesis, NF1 patients harbouring a heterozygous germline NF1 mutation develop neurofibromas upon somatic mutation of the second, wild-type, NF1 allele. While the identification of somatic mutations in NF1 patients has always been problematic on account of the extensive cellular heterogeneity manifested by neurofibromas, the classification of NF1 somatic mutations is a prerequisite for understanding the complex molecular mechanisms underlying NF1 tumorigenesis. Here, the known somatic mutational spectrum for the NF1 gene in a range of NF1-associated neoplasms --including peripheral nerve sheath tumours (neurofibromas), malignant peripheral nerve sheath tumours, gastrointestinal stromal tumours, gastric carcinoid, juvenile myelomonocytic leukaemia, glomus tumours, astrocytomas and phaeochromocytomas -- have been collated and analysed.


Introduction
Neurofibromatosis type 1 (NF1) is a common autosomal dominantly inherited tumour predisposition syndrome affecting 1/3,000 -4,000 individuals worldwide. 1,2 NF1 manifests a variety of characteristic features that include: hyperpigmentary abnormalities of the skin (café-au-lait macules and inguinal/axillary freckling), iris hamartomas (Lisch nodules) and the growth of benign peripheral nerve sheath tumours (neurofibromas) in the skin. Neurofibromas display many different subtypes and are associated with a variety of different clinical complications. Cutaneous neurofibromas are present in almost all adult NF1 patients. 3 Plexiform neurofibromas (PNFs), a more diffuse type of tumour, are present in 30-50 per cent of NF1 patients, and some 10 -15 per cent of these benign tumours are transformed to malignant peripheral nerve sheath tumours (MPNSTs), the main cause of morbidity in NF1. 4 Other NF1-associated clinical features include: skeletal abnormalities, such as tibial bowing or pseudoarthrosis; skeletal and orbital dysplasia; ostopenia/osteoporosis; aqueduct stenosis; macrocephaly; pectus excavatum; short stature; cardiovascular malformations; learning difficulties; and attention deficit disorder. 1,5 Cancer represents the transformation of a cell whose growth is normally tightly controlled into one that is no longer under strict regulation, allowing the cell to multiply uncontrollably and even metastasize. This dramatic alteration in cellular control arises as a consequence of the accumulation of genetic and epigenetic changes: activated oncogenes speed up cell growth through the acquisition of gain-of-function mutations, whereas tumour suppressor genes (TSGs) promote progression by acquiring loss-of-function mutations. TSGs typically encode proteins involved in growth regulation, apoptosis initiation, cellular adhesion and DNA repair. In accordance with Knudson's two-hit hypothesis, 6 both alleles of a TSG must be inactivated for cellular transformation to occur. Typically, a patient will inherit a germline mutation in one TSG allele; a second-hit or somatic mutation then occurs post-fertilisation, thereby inactivating the remaining wild-type allele. Somatic mutation is thus a key event in cancers associated with TSG inactivation. Upon transformation, a cell may acquire many additional somatic mutations elsewhere in the genome, a few of which actively encourage tumour progression, designated as 'driver mutations', while most occur simply because of the increased number of cell replications and are usually of unknown biological consequence and so are designated as 'passenger mutations'. 7 The NF1 gene encodes neurofibromin, a negative regulator of the Ras/mitogen-activated protein kinase (MAPK) pathway. NF1 is a TSG and, consistent with Knudson's two-hit hypothesis, most patients carry (in all their cells) both a normal and a dysfunctional NF1 gene copy -the latter harbouring the inherited (germline) mutation. It may be inferred that any tumours that arise will have acquired a second, somatic 'hit' that inactivates the normal NF1 allele, resulting in a complete loss of functional neurofibromin; a double hit (NF1 2/ -) is critical for NF1 tumorigenesis to occur. 8,9 The question as to why only a few of these benign tumours eventually go on to become malignant, however, is still puzzling. Consistent with a central role for neurofibromin in cellular function, recent cancer genome sequencing studies have found that somatic NF1 gene mutations occur not only in association with NF1, but also in a number of other common cancers. 10 -16 In the context of NF1, few genotype -phenotype correlations are evident. Indeed, marked intrafamilial variation in terms of the clinical phenotype is common. 5,17 The existence of such families is perhaps an indication of the importance of the second hit, since differences in the type and timing of somatic NF1 mutations may help to explain the variability in patient phenotype. 18 An appreciation of the spectrum of somatic mutations in NF1-associated tumours is therefore essential if we are to understand the molecular pathways involved -itself a prerequisite for improvements in clinical treatment and the development of new therapeutics. This paper attempts to collate and review the spectrum of somatic NF1 mutations so far reported in NF1-associated tumours, with a view to assessing how they may serve to induce tumour growth and whether or not any genotype -phenotype correlation may be discerned The NF1 gene: Structure and function The NF1 gene spans 283 kilobases (kb) of genomic DNA at 17q11.2 19 and contains 61 exons. 3,20 Neurofibromin, the 327 kDa protein encoded by the NF1 gene, is translated from a 12 kb messenger mRNA transcript, and has a number of alternative isoforms 21 -24 (reviewed by Upadhyaya 25 ). Neurofibromin contains 2,818 amino acids and is expressed at low levels in all cells, with higher levels in the nervous system. It functions as a negative regulator of active Ras, and of the associated Ras/MAPK signalling pathway. Neurofibromin contains a Ras-specific GTPase activating protein (GAP)-related domain which interacts directly with Ras, resulting in a conformational change that greatly stimulates the intrinsic GTPase activity of the Ras protein, thus significantly accelerating the conversion of the active GTP-bound form of Ras into its inactive GDP-bound form and effecting a net decrease in overall mitogenic signalling in the cell. As the Ras/MAPK cascade is critical for the control of cellular growth and differentiation, a lack of functional neurofibromin results in the constitutive activation of this central signalling pathway and in unregulated cell growth. 26

NF1 tumour biology
A variety of benign and malignant tumours are associated with NF1 and all involve tumorigenesis of neural crest-derived cells. Several murine models of neurofibromatosis have both successfully recapitulated much of the NF1 human phenotype and shown that NF1 is indeed a classical TSG. 27,28 Neurofibromas exhibit extensive cellular heterogeneity, being composed of hyperproliferative Schwann cells (SCs), fibroblasts, mast cells and perineural cells. The SCs have been identified as the initiating cell type in neurofibromas and it is only in these cells that the NF1 gene becomes biallelically inactivated. 29 SCs are also the target for various growth factors known to stimulate neurofibroma formation and growth. What is still not known, however, is the precise cell type within the SC cell lineage in which the somatic mutation occurs, the cell type which subsequently precipitates neurofibroma growth.
Cutaneous neurofibromas are thought to arise from skin-derived precursor cells (SKPs) 30 and these cells may well be under hormonal control, since most such tumours develop only during puberty. 31 Further, an increase in tumour size and number has also been noted during pregnancy, with some evidence for a postnatal decrease in tumour size. 32,33 Almost all PNFs appear congenitally and it is thought that they are induced by a somatic NF1 mutation in SC precursors within the embryonic gestational window of 12.5 -15.5 days. 34 It may be that this second hit does not render the SC precursor tumorigenic, but instead induces aberrant axonal segregation. 35 The extracellularly expressed transmembranal guidance protein, Sema4F, is strongly downregulated in neurofibromas and it has been suggested that this somehow indirectly promotes SC proliferation by rendering these cells more responsive to environmental signals, possibly through inhibition of axonal re-attachment. 36 In this way, the disruption of normal SC axonal interactions leads to neurofibroma development. An NF1 2/1 haploinsufficient cellular environment is also considered necessary, probably because of the growth advantage conferred by the signalling deficiency due to reduced neurofibromin levels. Indeed, Le et al. 30 found that NF1 inactivation is necessary, although not sufficient, for neurofibroma formation, highlighting the importance of the tumour microenvironment. There is some evidence to indicate that the haploinsufficiency (NF1 2/1 ) of the other supporting cells (fibroblasts, mast cells and perineurial cells) cooperates in neurofibroma development. 37 Additionally, it has been shown that NF1 2/1 haploinsufficient mast cells readily migrate into preneoplastic nerves, probably in response to Kit ligand, which exhibits four-fold increased levels in nullizygous SCs as compared to normal SCs. 38,39 The molecular mechanisms underlying both PNF and cutaneous neurofibroma formation are becoming clearer, although the major details are still lacking. It would appear that the key to understanding neurofibroma formation lies in the elucidation of the precise molecular interactions of the haploinsufficient tumour microenvironment within the initial cell type harbouring the biallelically inactivated NF1 gene.

Cutaneous neurofibromas and PNFs
Neurofibromas are a characteristic feature of NF1 and have a diverse clinical presentation. They are classified as grade 1 tumours by the World Health Organization; they have multiple forms and may affect nerves in any body location. Tumours derived from skin sensory nerves are designated dermal or cutaneous neurofibromas, and usually present as discrete tumours that remain associated with a single nerve ending. Approximately 20-50 per cent of cutaneous neurofibromas exhibit loss of heterozygosity (LOH) at the NF1 locus and the majority of these lesions appear to be due to mitotic recombination. 40 -42 Tumours associated with larger nerves within the skin may spread within the dermis and appear as a diffuse mass. PNFs are much larger tumours, usually associated with major nerve trunks and nerve plexi. They are generally slow growing, may develop at both internal and external body locations and can often result in major disfigurement. PNFs occur in some 30-50 per cent of patients with NF1 and, although these tumours generally remain benign, some neurological impairment may result from their growth. Approximately 10-15 per cent of PNFs may become malignant.
While the genetic basis of neurofibroma development is still not fully understood, biallelic NF1 inactivation does seem to be required, as all tumour cells harbour both a constitutional and a somatic NF1 gene mutation. 5 About 70 per cent of PNFs have been reported to display LOH at the NF1 locus; 20 however, there is no obvious correlation between the type or location of germline NF1 mutations in NF1 patients and those of their somatic counterparts arising in their tumours. 20 Another interesting, although as yet unexplained, observation is that a few patients mildly affected by NF1 who never develop any cutaneous neurofibromas or PNFs have been shown to carry the same germline NF1 mutation (c.2970-2972delAAT)namely, an in-frame 3-base pair (bp) deletion that leads to the loss of a methionine residue. 3 MPNSTs Cells derived from within some 10-15 per cent of PNFs may eventually undergo malignant transformation into an MPNST. MPNSTs are aggressive and highly invasive soft tissue sarcomas with an annual incidence of 0.16 per cent in NF1 patients, compared with only 0.001 per cent in the normal population, 43 and with a lifetime risk of 8-13 per cent in NF1 individuals 44,45 (reviewed by Upadhyaya 4 ). This form of malignancy represents a major cause of morbidity and mortality in NF1. Malignant transformation usually appears to evolve from within a pre-existing PNF. 46 The distinction between benign PNFs and MPNSTs has been sensitively visualised by noninvasive [ 18 F]-2-fluoro-2-deoxy-D-glucose positron emission tomography (FDG-PET) imaging, 47 suggesting a potential role for FDG-PET-based noninvasive imaging in future diagnostic tests. The aberrant molecular pathways that underlie this malignant transformation are still largely unknown, and considerable effort is being directed towards elucidating the molecular defects involved. NF1 patients carrying large (usually 1.4-megabase [Mb]) genomic deletions (which remove the entire NF1 gene plus a variable number of flanking genes) have an increased risk of MPNST development in certain patient groups. 48,49 Indeed, over 90 per cent of MPNSTs have been found to harbour large NF1 somatic deletions. 20 More recently, significantly increased frequencies (relative to the general NF1 population) of PNFs, subcutaneous neurofibromas, spinal neurofibromas and MPNSTs have also been reported in association with molecularly ascertained 1.4 Mb type-1 NF1 deletions. 50 The MPNSTassociated deletion breakpoints have been found not to involve the paralogous repetitive sequences that are involved in most germline NF1 deletions. 18 The smallest common region of somatic deletion overlap is, however, restricted to approximately the same 2.2 Mb interval that contains most of the genes deleted in recurrent constitutional NF1 deletions. 51 Although it is clear that biallelic NF1 gene inactivation is required for transformation to occur, mutations at the NF1 locus are insufficient to explain the process of tumorigenesis, as most benign neurofibromas also exhibit such biallelic NF1 inactivation. The best evidence for the involvement of other loci relates to the tumour protein 53 gene (TP53), for which several different mutations have been found in MPNSTs that have not been reported in benign neurofibromas. 4,20,52,53 Mice with heterozygous mutations in both their Nf1 and Tp53 genes developed malignancy, 27,54 an indication, perhaps, that TP53 loss is critical to transformation. The homozygous loss of the cyclin-dependent kinase inhibitor 2A gene (CDKN2A), which encodes p16INK4A and p14ARF, has also been associated with NF1 malignancy. 55 -57 Another recent report has indicated that phosphatase and tensin homologue deleted on chromosome 10 gene (PTEN) dosage, and/or phosphatidylinositol 3-kinase/AKT8 virus oncogene cellular homologue (PI3K/AKT) pathway activation, may be rate-limiting steps in NF1 malignant transformation. 58 As yet, however, no characteristic gene expression signature has been defined for MPNST development, although several cellcycle and signalling regulation genes: -cyclindependent kinase inhibitor (CDKN2A); tumour protein 53 (TP53); retinoblastoma 1 (RB1); epidermal growth factor receptor (EGFR); CD44 antigen (CD44); platelet-derived growth factor receptor alpha (PDGFRA); hepatocyte growth factor (HGF); proto-oncogene protein (C-MET) and transcription factor (SOX9) -are frequently deregulated. 4 Recent studies of the micro-RNA expression profile of MPNSTs have expanded the pathogenic spectrum associated with this tumour. For example, microRNA-34a (miR-34a) is downregulated in MPNSTs; this microRNA (miRNA) regulates many cell cycle genes and is also upregulated by p53, suggesting that TP53 loss would lead to downregulation of miR-34a and possibly several other miRNAs. This implies that this could be a critical event in malignant transformation. 59 In similar vein, miR-10b has been reported to be upregulated in SCs from NF1 tumours, while miR-10b inhibition reduced MPNST cell proliferation, migration and invasion. 60 NF1 mRNA is also a specific target for miR-10b, 60 indicating that these miRNAs represent potential therapeutic targets.

Spinal neurofibromas
About 40 per cent of NF1 patients present with tumours involving their spinal nerves. This is especially marked in individuals affected with familial spinal neurofibromatosis (FSNF), a variant form of NF1 in which bilateral tumours involving multiple spinal nerve roots are often the only manifestation of NF1. 61 -63 Patients with FSNF have been reported to be significantly more likely to harbour missense or splice-site germline mutations compared with patients with classical NF1. 64 A recent study of the NF1 locus found LOH in eight of 22 spinal tumours analysed, with most (75 per cent) of this LOH being due to mitotic recombination rather than genomic deletions. 64 Gastrointestinal stromal tumours (GISTs) GISTs are the most common mesenchymal tumours of the gastrointestinal tract. Although most GISTs harbour activating somatic mutations of KIT and PDGFRA, the absence of such mutations from NF1-associated GISTs (NF1-GISTs) is probably indicative of a different pathogenetic mechanism. In NF1, the majority (60 per cent) of GISTs develop in the small intestine, whereas sporadic non-NF1 GISTs mainly involve the stomach. 65 Somatic NF1 mutations have been identified in the interstitial cells of Cajal (ICC) throughout the gastrointestinal tract and in NF1-GISTs lacking KIT or PDGRA mutations. 66 Increased signalling through the Ras/MAPK pathway has also been shown to occur in NF1-GISTS, as opposed to sporadic GISTs. This would seem to indicate that a decrease in neurofibromin level, in the presence of normal c-KIT and PDGFRA levels, leads to tumour formation. It also suggests that NF1 haploinsufficiency is required for ICC hyperplasia, again demonstrating that, although a somatic NF1 mutation is absolutely necessary, it is not sufficient to permit tumorigenesis: additional genetic events required. These observations concur with Knudson's two-hit hypothesis. Somatic inactivation of the NF1 gene through gene deletion; intragenic deletion; and LOH through mitotic recombination have also been described. 66,67 Gastric carcinoid Gastric carcinoid tumours are associated with multiple endocrine neoplasia, atrophic gastritis and pernicious anaemia but are very rare in NF1. 17 LOH at the NF1 locus has been demonstrated in a gastric carcinoid tumour derived from an NF1 patient. 67 Juvenile myelomonocytic leukaemia (JMML) Young NF1 patients are at particular risk of developing JMML, 68 a clonal haematopoietic disorder characterised by hypersensitivity (at least in vitro) to granulocyte-macrophage colony-stimulating factor (GM-CSF). Moreover, some 15-20 per cent of JMML patients harbour a somatic NF1 inactivating mutation, even though most exhibit no other NF1 symptoms. 69 Patients may also carry inactivating mutations of other genes, with a recent study identifying that 70-80 per cent of mutations involve genes in the Ras/MAPK pathway, including one tyrosine-protein phosphatase non-receptor type 11 (PTPN11), neuroblastoma RAS viral oncogene homologue (NRAS), and v-Ki-ras2 kirsten rat sarcoma viral oncogene homologue (KRAS) as well as NF1 genes. 70 Additional somatic mutations have also been reported in the casitas B-lineage lymphoma (CBL) and additional sex combs-like 1 (ASXL1) genes. 71 In most cases, the NF1 gene is lost either via LOH or by compound heterozygous microlesions, 72 which lead to a complete loss of neurofibromin and hyperactive signalling through the Ras/MAPK pathway. LOH may occur through 1.2-1.4 Mb interstitial deletions mediated by low copy number repeat (LCR) elements that flank the NF1 gene. 73 LOH through uniparental interstitial isodisomy (50-52.7 Mb) of chromosome 17 through double mitotic recombination, in an as-yet-unknown initiator cell, has also been reported. 72 The rarity of such events may indicate the existence of a selective advantage, conferred upon the NF1 2/2 cells, which might explain the propensity of NF1 patients to develop leukaemia. 74

Astrocytomas (ACs)
Optic pathway tumours or ACs are found in 15 per cent of paediatric NF1 patients, 75 with the complete loss of neurofibromin evident in NF1-associated optic gliomas. 76 Approximately 84 per cent of NF1-associated ACs also exhibit LOH in the NF1 region, with many tumours also exhibiting LOH of 17p, suggesting the likely role of TP53 -or other 17p13-located genes -in AC formation. 77 As with MPNSTs, biallelic somatic NF1 mutation in ACs is, again, apparently insufficient to induce transformation.

Phaeochromocytomas (PCs)
PCs are extremely rare tumours, with only one to six cases observed per million individuals. PCs develop from neural crest-derived chromaffin cells, and the tumour cells produce and release catecholamines, which cause hypertension and flushing. These are tumours of the adrenal medulla and are primarily associated with mutations of the Ret proto oncogene (RET), von Hippel-Lindau (VHL), succinate dehydrogenase complex, subunit B (SDHB), succinate dehydrogenase complex, subunit C (SDHC), and succinate dehydrogenase complex, subunit D (SDHD) genes, although LOH in the NF1 region, as well as LOH of other loci on both 17q and 17p, have been observed. 78,79 Glomus tumours Glomus tumours are small (,5 mm), benign, but often very painful tumours that develop specifically within the highly innervated glomus body located at the end of each digit. These tumours appear to develop from a-smooth muscle actin-positive cells that have undergone biallelic NF1 inactivation, resulting in increased Ras/MAPK activity. 80 The somatic NF1 mutations often differ between glomus tumours, indicating highly specific tumorigenic events. Brems et al. 80 have suggested that glomus tumours, although rare, should now be recognised as an integral component of the NF1 spectrum of disease.
The somatic mutational spectrum of NF1-associated tumours A review of all published -and the authors' many unpublished -somatic NF1 alterations associated with NF1 tumours was undertaken to gain a better appreciation of NF1 tumorigenesis. As of July 2010, at least 577 different somatic NF1 gene changes had been reported in different NF1-associated tumours, with more than half (323/577; 56 per cent) corresponding to LOH in the NF1 gene region, some involving much larger regions of chromosome 17 (Table S1). The level of LOH detected also differs between cutaneous neurofibromas, PNFs and MPNSTs (40 per cent, 79 per cent and 85 per cent, respectively; Table 1). Tumour DNA analysis has also identified 254 somatic NF1 gene lesions, including nonsense, missense, splice site, microdeletion/microinsertions (,20 bp), indels (combined insertion-deletion events) and larger (.20 bp) deletions/insertions (Tables 3, S2). The consequences of all deletions and insertions for the reading frame were also determined, with five sequence changes being compound heterozygous NF1 mutations found in five haemopoietic tumours; however, with no other tissue available for analysis, it was not possible to differentiate between germline and somatic NF1 point mutations (Table S2). About 75 per cent (191/254) of the somatic mutations associated with NF1 tumours comprise mutations that are predicted to give rise to truncated proteins. Of these 191 changes, only 18 result from the insertion or duplication of bases; the remaining 173 truncations arise from deletion, nonsense mutation or frameshift events. Splice site mutations form a considerable proportion (39/254; 15.0 per cent) of the mutational spectrum, while missense changes only account for some 9.4 per cent (24/254) of the somatic NF1 mutations.
Any attempt to make direct comparisons between the various tumour types would be unwise at this stage, owing to the paucity of somatic mutation data, especially for the less commonly encountered tumours. Table 3 nevertheless attempts to summarise the available data. The bias inherent in the data is immediately evident, with 211/254 (83 per cent) mutational changes originating from the analysis of cutaneous neurofibroma DNA. Hence, the relative frequencies of the various mutation types in cutaneous neurofibromas are essentially comparable with the germline mutational spectrum, with nonsense mutations, splice site mutations and missense  Glomus tumour 1 (14%) 6 (86%) 7 Overall 323 (55%) 254 (44%) 577 * Compound heterozygous NF1 mutations were identified in five of six haemopoietic tumours analysed. As no other normal tissues were available in these five cases, it was not possible to distinguish between the associated germline and somatic NF1 mutations.
alterations found in cutaneous neurofibromas at frequencies of 28 per cent (59/211), 15 per cent (32/ 211) and 10 per cent (21/211), respectively (Table 3). Table 3 does, however, serve to highlight the high proportion of truncating mutations (191/ 254; 75 per cent) involved in the somatic inactivation of the NF1 gene in all tumour types, especially cutaneous neurofibromas. An additional comparison between the frequency distributions of somatic microlesions and LOH is made in Table 1. There appears to be a marked difference between cutaneous neurofibromas, PNFs and MPNSTs, with 40 per cent, 79 per cent and 85 per cent, respectively, of somatic mutation events represented by LOH. This may be explained in part by the extent of the molecular rearrangements in each tumour type; MPNSTs, for example, would be predicted to exhibit a greater extent of genetic aberration than a benign dermal neurofibroma. The types of analyses performed, however, will have a direct influence on such conclusions, in that either microlesions or LOH may not be screened for in some studies.
In summary, the more severe MPNSTs show a greater degree of genetic abnormality than other tumour types, with LOH constituting a much more frequent event in these tumours. Further comparison within and between the rarer tumour types would not be valid, however, owing to the relative paucity of mutation data currently available for analysis.
Mutational mechanisms underlying the known somatic NF1 gene lesions Somatic inactivation of the NF1 gene may result from different mutational mechanisms and may involve intragenic mutations, LOH and epigenetic modification of the promoter region. Among the 254 somatic NF1 mutations listed in Table S2, 72 nonsense mutations were found, of which 36 involved mutations in just 15 codons in different tumours (codons 192, 304, 426, 440, 816, 1241, 1306, 1362, 1513, 1569, 1604, 1748, 1939, 1976 and 2429), with many previously reported in different tumours or different studies. Ten of these 15 different recurrent nonsense mutations involve C . T or G . A transitions within CpG dinucleotides and are compatible with the endogenous mutational mechanism of methylation-mediated deamination of 5-methylcytosine (5mC). Of these 72 nonsense mutations, 28 have also been reported as germline mutations in NF1 patients (Human Gene Mutation Database [HGMD]), 81 indicating that the same mutational mechanism is operating in both the soma and . For the ten of these 15 nonsense mutations that correspond to C . T or G . A transitions within CpG dinucleotides, we may infer that the mutated cytosine must be methylated both in the soma and in the germline, thereby explaining the vulnerability of these sites to methylation-mediated deamination in both cell lineages. Among the somatic NF1 mutations listed in Table S2 are 21 different missense mutations. Of these, two (in codons 519 and 776) have been reported more than once in different tumours or different studies, although neither is compatible with methylation-mediated deamination of 5mC. Of the 21 missense mutations, only one (in codon 176) has also been reported in the germline (see HGMD). Since this Asp176Glu mutation has also been reported more than once in NF1-associated tumours, it may well be that this residue is of importance for the function of neurofibromin in both the soma and the germline. Furthermore, this residue is conserved in different species, including Drosophila and Fugu, and has not been identified in 250 unrelated normal individuals.
Nonsense mutations are not the only type of NF1 mutation to occur recurrently in the soma. Among the somatic NF1 microdeletions listed in Table S2 are five that have been reported more than once in different tumours (c.1888delG, c.2033delC, c.3058delG, c.4374_4375delCC and c.5731delT) with three microdeletions occurring in mononucleotide tracts (G 4 , C 7 and T 3 , respectively), suggestive of a model of slipped mispairing at the DNA replication fork. Importantly, c.2033delC has also been reported in the germline (see HGMD), indicating that this tetranucleotide stretch is a hotspot for mutation in both the germline and the soma. A microinsertion (c.1733insT, located within a T 6 tract) has also been found to occur recurrently in the soma but this has not so far been reported in the germline. The reader interested in a detailed comparative analysis of germline and somatic mutations in human TSGs is referred to Ivanov et al. 82

NF1 gene somatic mutations in non-NF1-associated tumours
Various studies have identified somatic NF1 gene mutations in non-NF1-associated cancers. Thus, somatic NF1 aberrations have been identified in glioblastoma multiforme (GBM) tumours, lung adenocarcinomas, malignant breast tumours, leukaemia, ovarian serous carcinomas (OSCs) and neuroblastoma. 10 -12,14 -16,83 Some of the NF1 gene changes are relatively frequent in these tumours and therefore have the potential to represent specific prognostic and diagnostic markers. For example, 23 per cent of sporadic GBM tumours harbour an inactivating NF1 somatic mutation, and this may enable such GBM tumours to differentiate into the mesenchymal molecular subclass. 13 Similarly, in 22 per cent (9/41) of primary OSCs, an NF1 mutation was detected, six of which exhibited biallelic inactivation. 12 Interestingly, all nine of these OSC samples also contained a TP53 mutation, highlighting the likely involvement of this TSG in OSC pathogenesis. 12 Given the pivotal role that neurofibromin plays in several cell signalling pathways, it is not surprising that its loss will affect distinct molecular subtypes in different cancers. Indeed, the efficacy of any future therapeutic intervention for many tumours will almost certainly hinge upon our ability successfully to identify such molecular subclasses of tumour.

Prospects for the development of new treatments/therapies
As the complex picture underlying the molecular nature of NF1 tumorigenesis becomes better defined, the treatment regimens available to patients should greatly improve. Although the future is encouraging, the optimal treatment for NF1 tumours currently rests with their surgical resection, in spite of the high chance of recurrent malignancy. Gottfried and colleagues 84 have suggested that the recruitment of supporting cells around the neurofibroma, coupled with aberrant Remak bundles, could explain how the neurofibroma integrates into the surrounding tissue, and it is this that may lead to the surgical difficulties that often lead to tumour recurrence. Moreover, it has been suggested that surgical interference may even increase the recruitment of surrounding cell types, thereby inadvertently increasing the growth of lesions leading to the formation of new neurofibromas. 84 Surgical biopsy is therefore inherently problematic, and novel therapeutics are urgently required. Clinical and preclinical trials targeting different components of the Ras/MAPK signalling pathway and related growth factor receptors appear to be more promising. It is likely, however, that treatment with multiple drugs may be more effective for NF1 tumours. 5

Concluding remarks
Biallelic inactivation of the NF1 gene, resulting in the complete loss of functional neurofibromin, initiates the pathogenic process that eventually results in the formation of nerve sheath tumours. NF1 gene inactivation may occur through relatively subtle lesions that affect just a few DNA bases, or may involve large genomic changes that affect large chromosomal regions, or even the entire chromosome 17. This review demonstrates that NF1-associated tumour types display a considerable degree of variation in terms of the level of LOH detected, with cutaneous neurofibromas, PNFs and MPNSTs. MPNSTs manifest increased levels of deletion-based LOH, whereas cutaneous neurofibromas appear to be associated with a localised deletion of the NF1 gene through mitotic recombination (the situation in PNFs being somewhat intermediate). In MPNSTs, additional mutations at different gene loci are almost certainly involved in the progression of the tumour.
In terms of the molecular mechanisms of mutagenesis, both methylation-mediated deamination of 5-methylcytosine and slipped mispairing within polynucleotide tracts appear to be responsible for the occurrence of mutation hotspots in both the germline and the soma. For some types of tumour, there is interplay between the soma and the germline, in that the location of the germline mutation can influence the nature, frequency and location of the subsequent somatic mutation. 85,86 As yet, however, there is no evidence for this phenomenon in the context of NF1 tumorigenesis.
Although our knowledge of the role of the NF1 gene in tumorigenesis is ever expanding, definitive markers of malignant transformation remain to be discovered. Mouse and other animal models, including zebrafish, 87 have provided new perspectives for research, with various knockout and mutagenesis studies potentiating functional studies. It is already clear that, in order to clarify the role of the NF1 gene in NF1-associated tumours, we must improve our understanding of the significance of the somatic (second-hit) mutations. The brief assessment of the compilation of somatic NF1 mutations in NF1-associated tumour types reported here failed to unearth any specific genotypic correlations. The limited size of the mutation dataset means that reliable conclusions are hard to draw, and that larger and better-defined patient groups will be needed, to allow more reliable comparisons to be made. Additionally, definitive prognostic markers should be identified that permit differentiation between benign neurofibromas that are likely to progress to malignancy and those that are not.
This review nevertheless emphasises that NF1 is a highly individual condition that exhibits extreme somatic mutational heterogeneity both within and between patients. These are the mutations which are ultimately responsible for the molecular changes that can lead to tumour formation. If we can come to understand how these changes bring about tumorigenesis, we shall be better placed not only with respect to the provision of genetic counselling, but also in terms of exploring new avenues for the development of new drug-based therapies.

Continued
Neurofibromatosis type 1-associated tumours: Their somatic mutational spectrum and pathogenesis