Structure of the NF1 gene and mutational analysis

The NF1 gene contains 60 exons and generates multiple alternatively spliced isoforms³⁷. More than 1,400 mutations in the NF1 gene have been reported in the Human Gene Mutation Database, most of which are clearly loss-of-function alleles. These include splice site, nonsense and missense mutations, as well as deletions, insertions, frameshifts and translocations³⁸. Notably, several patient missense mutations that affect the neurofibromin GRD selectively diminish GAP activity, which supports the notion that the regulation of RAS has a crucial role in NF1 disease³⁹. Identification of NF1 mutations in patients remains difficult owing to the large gene size and structure, as well as the large range of mutations that have been identified⁴⁰. For many years, 95% of patient mutations were identified using a combination of complementary methods, including protein truncation, fluorescence in situ hybridization, heteroduplex, Southern blot and cytogenetic analyses³⁸. A preliminary report applied next-generation sequencing to samples from patients with NF1 (REF. 41), and DNA-based sequencing is now being offered as a clinical test for diagnostic purposes (see the website of University of Alabama at Birmingham Medical Genomics Laboratory). Thus, NF1 mutation analysis can assist in diagnosing cases of NF1 in which a clinical diagnosis cannot be established with certainty.

The variability in manifestations in patients from a single family with the same NF1 mutation does not support a major role for genotype–phenotype correlations in NF1 (REF. 42), although there are several important exceptions. Germline splice site mutations occur in 30% of patients with NF1, and these patients may have an increased overall tumour risk⁴³. Mutations that delete the NF1 gene, and several flanking genes, occur in up to 10% of patients with NF1. This class of mutations predisposes affected individuals to an increased risk of intellectual disability, to greater numbers of cutaneous neurofibromas and to MPNSTs⁴⁴. Another genotype–phenotype correlation is of a very rare 3-bp deletion in patients with NF1 who lack neurofibromas⁴⁵.

Modifier genes in NF1

Given the paucity of NF1 genotype–phenotype correlations, it was proposed that modifier genes underlie the variable penetrance of NF1. Monozygotic twins with NF1 showed a high degree of concordance for cutaneous neurofibroma tumour burden and numbers of café-au-lait macules, supporting the idea that modifier genes contribute to these features^46,47. The large polygenic deletions (mentioned above) indicate that modifier genes might be linked to NF1 (REF. 44). Recently, the gene encoding the chromatin remodelling complex Polycomb repressive complex 2 subunit SUZ12, which lies within this region, has been shown to be a cooperating tumour suppressor in mouse models and in human tumours^48–50. Studies in mouse models also support roles for modifier genes in NF1. Astrocytoma resistance alleles were recently identified as spinal cord resistance to astrocytoma modifier 1 (Scram1) and astrocytoma resistance locus in males 1 (Arlm1) loci^38,51,52. It is unclear whether disease modifier genes in NF1 are also relevant in NF1-mutant sporadic tumours.

In support of a role for modifier genes in NF1 is OPGs; these tumours have a decreased prevalence in the African-American population compared with other races⁵³. Sex-linked factors may modify prognosis; males are at an increased risk of sporadic high-grade glioma, but NF1 females with low-grade OPG have a worse prognosis than males with NF1-related OPG⁵⁴. In addition, male Nf1^−/− mouse astrocytes expressing dominant-negative p53 show increased tumorigenesis and inactivation of the RB protein compared with cells derived from females⁵⁵. An imprinting control region remotely interacts with an intergenic sequence between Nf1 and Wsb1 on chromosome 11 to regulate Nf1 transcription, and mutations in this intergenic sequence could also potentially modify NF1 disease⁵⁶.

NF1 mutations in sporadic cancers

The advancement of whole-genome sequencing has resulted in the identification of NF1 mutations in various non-NF1-associated sporadic cancers, including glioblastoma^33,57, neuroblastoma⁵⁸, AML³⁶, lung cancer³², ovarian cancer³⁴ and breast cancer⁵⁹. We anticipate that the identification of tumours that contain NF1 mutations will continue to increase with future sequencing efforts. A comprehensive study that analysed somatic mutation patterns in more than 1,500 cancer-related genes in a large panel of lung, breast, ovarian, pancreatic and prostate tumours identified NF1 (mutation frequency >5%) as one of ten genes that are mutated most often in these types of tumours. In comparison, the point mutation frequency was 33% for TP53, 7% for KRAS and 5% for cyclin-dependent kinase inhibitor 2A (CDKN2A)⁵⁹; however, deletions in CDKN2A are much more common than deletions in NF1. It is as yet unknown whether biallelic loss of NF1 is common or whether only hemizygous loss of NF1 contributes to tumour progression in sporadic disease. Consistent with the latter possibility, mouse cells hemizygous for Nf1 mutations show abnormal growth and invasion^60–62. Hemizygous NF1-mutant cells might show lower levels of GTP-bound RAS than cells with complete inactivation of NF1 and/or develop mutations in additional RAS–MAPK pathway genes to affect tumour properties. Although sporadic tumours with NF1 mutations are largely exclusive of those tumours that harbour mutations in MAPK kinase 1 (MAP2K1) or NRAS, on the basis of our analyses of somatic co-mutation patterns in The Cancer Genome Atlas data sets (cBio Portal for Cancer Genomics), 11 of 114 melanomas with NF1 mutations also show mutations in BRAF, NRAS or RAF1. Thus, there may be subcategories of tumours, and perhaps cells, in which BRAF, NRAS or RAF1 are co-mutated with NF1.

In tumours with NF1 mutations, the order of mutations seems to affect tumour grade in specific cell types; for example, initial loss of NF1 in nerve glial cells triggers neurofibroma. In this case, oncogene-induced senescence occurs, and inactivation of p53 bypasses this response for progression to MPNSTs⁶³. Similarly, benign grade 1 astrocytomas develop in genetically engineered mice (GEMs) when Nf1 is lost first⁶⁴. By contrast, aggressive gliomas form when Trp53 is co-mutated with Nf1 (REF. 65). Indeed, in human glioblastomas, half of the tumours with NF1 mutations also harbour TP53 mutations⁶⁶. Mutational order may explain why patients with NF1 are not predisposed to certain sporadic tumours, such as lung tumours, whereas >10% of sporadic lung cancers³² have NF1 mutations, which are probably acquired late in tumorigenesis.

Interactions of neurofibromin

In 1991, Bollag and McCormick⁶⁹ reported that the lipids arachidonate, phosphatidate and phosphatidylinositol-4,5-bisphosphate inhibit neurofibromin GAP activity, but no definitive in vivo role for lipid–neurofibromin interaction through the SEC14 and pleckstrin homology (PH) domains has so far been identified. Neurofibromin is also implicated in connecting RAS signalling to activation of RHO-family GTPases, which results in modulation of the cytoskeleton. Therefore, it is intriguing that the neurofibromin phospholipid-binding SEC14 and PH domains interact with LIM domain kinase 2 (LIMK2) and thereby inhibit activation of LIMK2 by RHO-associated protein kinase, which is known to modulate the actin cytoskeleton⁷⁰. Some neurofibromin domains may function as scaffolds that target neurofibromin to specific intracellular locations. Indeed, neurofibromin interacts with sprouty-related, EVH1 domain-containing protein 1 (SPRED1). This interaction results in localization of neurofibromin to membranes, which enables neurofibromin to down-regulate GTP-bound RAS⁷¹. Interestingly, SPRED1 is also a RASopathy gene. Unlike mutations in NF1, mutations in SPRED1 do not predispose affected individuals to neurofibromas, gliomas or MPNSTs but rather to lipomas. Distinct features of each disorder may arise from cell type-specific use of components of the MAPK pathway.

Neurofibromin is the main RASGAP in neuronal dendritic spines⁷², where the molecular chaperone valosin-containing protein (VCP) interacts with the leucine-rich repeat domain of neurofibromin. In mice, mutations in Vcp, like loss of Nf1, reduce spine density⁷³. Importantly, VCP functions downstream of neurofibromin, and expression of VCP rescues spine defects in Nf1^+/− neurons. VCP is now under investigation as a cancer target. Dimethylarginine dimethylaminohydrolase 1 (DDAH1), which degrades the endogenous nitric oxide inhibitor asymmetric dimethylarginine (ADMA), was identified as another neurofibromin interaction partner. Knockdown of Nf1 rescued the decrease in cell proliferation caused by knockdown of Ddah1 in mouse endothelial cells^74,75, and this is of special interest because the modulation of nitric oxide is an attractive therapeutic target.

Modulation of RAS signalling by neurofibromin

The absence of neurofibromin leads to slowed hydrolysis of GTP-bound RAS, which sustains RAS signalling. Nevertheless, upstream receptors are required to activate RAS, and several receptors have been implicated upstream of neurofibromin in particular cell types. For example, genetic and biochemical evidence support a necessary role of granulocyte–macrophage colony-stimulating factor (GM-CSF) and activation of the GM-CSF receptor to maintain JMML in Nf1-mutant mice⁸². This receptor may also have a role in neurofibroma formation after nerve injury⁸³. By contrast, activation of the KIT receptor by stem cell factor (also known as KIT ligand) has a key role in neurofibroma formation in mice⁶². In Drosophila melanogaster expressing mutant Nf1, the loss of the receptor tyrosine kinase Anaplastic lymphoma kinase (Alk) rescues the small size of the flies and ERK activation⁸⁴. Other receptors probably function as upstream regulators of neurofibromin in specific cell types.

Neurofibromin and downstream signalling

Of the many putative downstream effectors of RAS signalling — RAL guanine nucleotide dissociation stimulator (RALGDS), PI3K, phospholipase Cε, T lymphoma invasion and metastasis-inducing protein 1 (TIAM1), RAS association domain family protein (RASSF), RAS and RAB interactor 1 (RIN1), RIN2, RIN3, AF6 (also known as afadin and MLLT4), impedes mitogenic signal propagation (IMP; also known as BRAP2), RAS-associated and PH domains-containing protein 1 (RAPH1), growth factor receptor-bound protein 7 (GRB7), and PDZGEF1 (also known as RAPGEF2) (FIG. 2) — only a few have been studied in the context of NF1. The best studied is the neurofibromin–RAS–MAPK pathway. Loss of neurofibromin results in the activation of this pathway in multiple types of tumour^85,86. Heat shock response⁸⁷ and MAF regulation of mTOR signalling⁸⁸ also occur downstream of NF1 loss and subsequent activation of ERK. RAL guanine nucleotide exchange factors have also been implicated downstream of RAS in MPNSTs⁸⁹.

Several studies in mice support important roles for PI3K signalling when loss of Nf1 drives tumour formation, and several pathways have been implicated downstream of PI3K. For example, PI3K–mTOR signalling is enhanced in Nf1-mutant astrocytes and MPNSTs^90,91. PI3K–AKT signalling downstream of Rras2 has a role in the initiation of neurofibromas⁹²; activation of the PI3K pathway through loss of Pten in mice promotes the transformation of Nf1-driven neurofibromas to MPNSTs⁹³. Sustained activation of AKT in MPNST cells requires calcium and calmodulin⁹⁴. On the basis of these results, it will be important to consider cell type-specific neurofibromin–RAS effector pathways when attempting to identify therapeutic targets. In addition, simultaneously targeting multiple RAS effector pathways may provide enhanced effects.

NF1 and cAMP

In yeast, the NF1 homologues IRA1 (REF. 95) and IRA2 (REF. 96) regulate both Ras and cAMP signalling. Ira1 and Ira2 each simultaneously bind to Ras2 and adenylyl cyclase, thus regulating both pathways⁹⁷. It is accepted that increased levels of cAMP inhibit the proliferation of most cell types, and it has been proposed that altered levels of cAMP contribute to cancer. Mouse⁹⁸, D. melanogaster ⁹⁹ and zebrafish¹⁰⁰ cells expressing mutant Nf1 all show deregulated cAMP levels. In D. melanogaster lacking Nf1 cAMP levels are low, but it remains unclear whether this results from increased Ras activity or whether it occurs independently of Ras^98,99. By contrast, cAMP levels are increased in mouse Schwann cells (peripheral nerve support cells) or human MPNST cells lacking NF1 (REF. 101). Many different RAS–cAMP pathway interactions may occur in specific NF1-mutant cell types. In mouse neurons, atypical protein kinase C-mediated β-adrenergic receptor kinase 1 (ADRBK1)-driven inactivation of the G protein Gα_s modulated RAS activity^98,102 . In another system, the protein kinase A-activated transcription factor cAMP-responsive element-binding protein (CREB) bound to the mir-9 promoter, which repressed expression of NF1 and encouraged cell migration¹⁰³. Interfering with cAMP signalling may present a therapeutic opportunity in several manifestations of NF1, as was proposed for NF1-driven brain tumours¹⁰⁴. Blocking RAS–MAPK signalling and increasing cAMP levels may be useful therapeutically; for example, reversing one zebrafish brain defect required blockade of MEK, whereas reversing another defect required an increase in cAMP levels¹⁰⁰.

Low-grade astrocytoma

At least 15% of patients with NF1 develop OPGs, which are mainly grade I pilocytic astrocytomas⁷. These tumours are defined as benign, generally have a favourable prognosis and rarely progress¹⁰⁵. In contrast to NF1-related pilocytic astrocytomas, pilocytic astrocytomas in patients without NF1 are typically more aggressive, although they have RAS pathway mutations, including BRAF duplications or activating point mutations in BRAF or KRAS¹⁰⁶. In GEM models of grade 1 astrocytoma resulting from loss of Nf1, many neurons in the brain and most macroglial cells are Nf1^−/− owing to use of glial fibrillary acidic protein (Gfap)–Cre or brain lipid-binding protein (Blbp; also known as Fabp7)–Cre drivers, which excise Nf1 in most brain stem and progenitor cells^64,98,107. In this model, other cells of the body are Nf1^+/− (REF. 64). Evidence indicates that when progenitor cells of the developing third ventricle lack Nf1, they develop into aberrant astrocytes, which are characteristic of pilocytic astrocytoma¹⁰⁸. Consistent with this interpretation, loss of Nf1 in other brain cell types (such as oligodendrocytes or NG2 cells) failed to generate astrocytomas in zebrafish or mice^109,110.

Cutaneous neurofibroma

All neurofibromas contain nerve Schwann cells (with biallelic NF1 mutations¹¹¹), as well as NF1 wild-type or heterozygous fibroblasts, mast cells, macrophages, perineurial cells and endothelial cells. The percentage of each cell type varies from tumour to tumour. Neurofibromas in the human dermis or epidermis (known as cutaneous or dermal neurofibromas) are benign and do not transform; however, they can cause a substantial cosmetic burden in patients. It has been a source of confusion that some plexiform neurofibromas, which are more aggressive than cutaneous neurofibromas, also develop in the skin and subcutaneous tissue; alternative nomenclature has been discussed but not defined by consensus. There are no model organisms at present in which cutaneous neurofibromas spontaneously develop. However, after growth in vitro and transplantation into Nf1^+/− syngeneic hosts, skin hair follicle-derived Nf1^−/− precursors (SKPs) formed tumours that resembled dermal neurofibromas¹¹². Transplantation of Nf1^−/− SKPs into pregnant female mice increased growth of neurofibromas in the mouse model¹¹², and cutaneous neurofibromas can develop and grow during puberty⁵ and can increase in size and number during pregnancy¹¹³. However, it remains unclear whether and which hormones or other factors increase the growth of human cutaneous neurofibromas. It is also unclear which factors limit the growth of most dermal neurofibromas in humans, and why they are resistant to transformation.

Plexiform neurofibroma

Plexiform neurofibromas are complex tumours that can weigh kilograms and can compress vital structures. Therefore, they are of considerable interest as targets of therapy. Plexiform neurofibromas develop within peripheral nerves and their perineurial sheaths. However, plexiform neurofibromas can invade adjacent tissue by disrupting the perineurium, remaining non-metastatic and clinically ‘benign’ but accounting for substantial morbidity and an increased risk of mortality when symptomatic¹¹⁴.

Extensive modelling of plexiform neurofibromas has been carried out in mice. Plexiform neurofibroma models all have biallelic loss of Nf1 in the Schwann cell lineage^115–118, which is driven by P0 (also known as myelin protein zero (Mpz))–Cre, desert hedgehog (Dhh)–Cre, tamoxifen-inducible proteolipid protein (myelin) 1 (Plp1)–Cre, or Krox20 (also known as Egr2)–Cre. Each of these drivers knock out expression in Schwann cell progenitors, with differences in precise timing, location and the number of cells affected. Some models, probably those with fewer cells showing recombination, require additional hemizygous inactivation of Nf1 in haematopoietic cells to generate neurofibromas¹¹⁹. In all cases, the tumours (grade 1 neurofibromas) in GEM models resemble those of humans. In humans, plexiform neurofibromas primarily develop very early in life. Unexpectedly, plexiform neurofibromas can be induced even in adult mice, although in one model these are rare^116,117. Unfortunately, this result does not resolve the uncertainty around the plexiform neurofibroma cell of origin, as adult mouse dorsal root ganglia and dorsal roots contain stem-like cells, mature Schwann cells and satellite cells, all of which have been proposed as possible neurofibroma-initiating cells^{115–118,120}. Epidermal growth factor receptor (EGFR)-expressing Schwann cell precursor-like cells are present in mouse and human neurofibromas, and form neurofibroma-like lesions under the skin¹²⁰ or in peripheral nerves¹²¹ of immunocompromised mice. Extensively passaged Schwann cells from neurofibromas can also form neurofibromas in immunocompromised mice¹²². Thus, de-differentiated Schwann cells and/or Schwann cell precursors may each be cells of origin for neurofibromas.

MPNSTs

MPNSTs are nerve-associated sarcomas, most of which arise in pre-existing plexiform neurofibromas in patients with NF1 (REF. 123), and are aggressive tumours that typically metastasize to the brain, bone and other sites¹²⁴. Half of all MPNSTs arise sporadically, and the other half occur in patients with NF1 (REF. 125). Genetic alterations that affect the RAS pathway, including NF1, BRAF, NRAS or KRAS mutations, have been identified in some sporadic MPNSTs^126,127, and NF1 and sporadic MPNSTs share a gene signature¹²⁸. MPNSTs are also typically hyperdiploid, and their genomes — like genomes in most sarcomas — are highly rearranged¹²⁹. Mutations associated with the transformation from benign plexiform neurofibromas to malignant MPNSTs include early mutations in CDKN2A¹²⁹ and later mutations in TP53 (REFS ^130–132) and SUZ12 (REF. 133). Loss of the tumour suppressor gene RB1 (which encodes RB) is found in 25% of MPNSTs^134,135, and monosomy for the PTEN locus is observed in 50% of MPNSTs^{126,134,136,137}. Low-level amplification of growth factor receptor genes, including EGFR, is also common¹²⁶.

Human MPNSTs and MPNST cell lines contain CD133⁺ cells that may be stem-like cells^138,139. GEM models of MPNSTs include combined loss of Nf1 and Trp53 (REFS ^140,141) or Nf1 and Cdkn2a^142,143. MPNSTs from these models and human MPNST cell lines contain stem-like cells that propagate disease¹⁴⁴. Comparative oncogenomics and insertional mutagenesis screens have been used to identify candidate drivers of MPNST formation, including MEK, β-catenin and embryonic stem cell-expressed RAS (ERAS)^86,145,146. In zebrafish, ribosomal gene mutations predispose to the formation of MPNSTs, and resulting tumours show loss of p53 translation¹⁴⁷.

Other tumours in patients with NF1

Pheochromocytomas¹¹, rhabdomyosarcomas¹⁴, glomus tumours^20,148 and GISTs¹² are present at increased incidence in patients with NF1 and show biallelic inactivation of NF1. Of these, only pheochromocytoma has been successfully modelled to date. Pheochromocytomas form in 15% of Nf1^+/− mice¹⁴⁹; however, this model has not been used in preclinical tests, probably owing to low tumour incidence. JMML is a rare paediatric manifestation of NF1 (REF. 10) and a RAS-driven haematopoietic stem cell disorder. Thus, patients with NF1 and patients with RASopathies who have mutations in NRAS, KRAS, CBL, protein tyrosine phosphatase non-receptor type 11 (PTPN11; also known as SHP2) or son of sevenless homologue 1 (SOS1) are predisposed to this low-grade leukaemia¹⁵⁰, which is recapitulated in an Mx1–Cre;Nf1^fl/fl model¹⁵¹.

Preclinical testing

Importantly, the available mouse models of JMML, OPG, plexiform neurofibroma and MPNST are currently used for preclinical testing. Consistent with a major role for RAS–MAPK signalling in JMML, preclinical testing showed significant response to inhibition of MEK¹⁵¹. Disease burden was markedly reduced, although mutant stem cells persisted. The finding that the multi-receptor tyrosine kinase inhibitor imatinib or inhibition of MEK can shrink plexiform neurofibromas in mouse models led to clinical trials^86,119,151. In the imatinib trial, 6 of 36 patients, primarily young children with small plexiform neurofibromas, responded to treatment; the target kinase (or kinases) affected in these patients is not yet known¹⁵². In a Phase I trial of the MEK inhibitor selumetinib (presented at the 2014 American Society of Clinical Oncology meeting), 11 of 18 patients with plexiform neurofibromas, some up to kilograms in weight, shrank by ≥20% in response to therapy, and many showed prolonged response¹⁵³. These positive results not only support continued use of mouse models to guide clinical testing in human neurofibromas but also emphasize the finding that these benign tumours can respond to single-agent therapy for up to 3 years without showing resistance.

Complete surgical resection is required to cure MPNSTs, and no single-agent or combination tested to date has cured MPNSTs in any model system (reviewed in REF. 154). Consistent with the idea that combination therapy will be necessary to treat these aggressive cancers, the mTOR complex 1 (mTORC1) inhibitor rapamycin, together with agents that enhance oxidative stress, shrank MPNSTs¹³³. In MPNST xenografts, prolonged responses have been observed with Aurora kinase inhibition and bromodomain inhibitors, but these have not yet been tested in mouse models of MPNSTs or in patients with NF1 (REFS ^155,156). In some MPNSTs, autocrine chemokine (C-X-C motif) ligand 12 (CXCL12)–CXC receptor 4 (CXCR4) signalling activates β-catenin through the AKT–glycogen synthase kinase 3β–β-catenin stabilization pathway¹⁵⁷. Transposon mutagenesis confirmed WNT pathway signalling as a driver of the transformation to MPNSTs, and many β-catenin pathway genes are deregulated in neurofibroma and MPNSTs¹⁵⁸. Although other links between NF1 signalling and β-catenin pathway activation remain to be identified, these data and analysis of human NF1 Schwann cells support inhibition of the β-catenin pathway as a possible therapeutic target in NF1-driven tumours¹⁵⁹.

NF1 loss as a drug resistance marker

MPNSTs in patients with NF1 are notoriously resistant to chemotherapy and radiation therapy. Recently, a large-scale study showed almost no benefit of chemotherapy or radiotherapy for NF1 MPNSTs, and a worse outcome has been reported for NF1 than for sporadic MPNSTs¹²⁴. These data must be interpreted with caution, as late diagnosis of MPNSTs arising in plexiform neurofibromas in patients with NF1 may account for differences between the groups in treatment outcome¹⁶⁰. However, although reasons might differ as to what causes resistance in sporadic tumours, NF1 has been identified as a gene that confers resistance to targeted therapy, including inhibition of kinases and the RAS pathway, in sporadic neuroblastoma⁵⁸, lung carcinoma¹⁶¹ and melanoma¹⁶². In lung cancer models, resistance to EGFR therapy was mediated by NF1, and blocking MEK restored the response¹⁶¹. Melanoma develops in mice with mutant Braf, and loss of Nf1 blocked Braf-driven oncogene-induced senescence. Nf1-mutant and Braf-mutant tumours are resistant to BRAF inhibitors; however, they are sensitive to combined inhibition of MEK and mTOR¹⁶². Loss of NF1 was also identified as a crucial mediator of resistance to BRAF inhibitors in melanoma cells^163,164. These studies indicate that blocking NF1 pathways — for example, by targeting MEK — might enhance the therapeutic response in those tumours with NF1 mutations. There are probably also other resistance pathways downstream of NF1. In a neuroblastoma model, resistance was mediated by NF1 through ZNF423, which encodes a zinc-finger transcription factor⁵⁸.

It is increasingly clear that the recruitment of mast cells, macrophages and other stromal cells in the tumour microenvironment can elicit tumour cell resistance to therapy¹⁶⁵. NF1^−/− Schwann cells upregulate major histocompatibility complex class II mRNA and protein levels, which may influence tumour–immune cell interactions¹⁶⁶. Neurofibromas and MPNSTs contain blood-derived mast cells¹⁶⁷ and macrophages¹⁶⁸. The stroma in NF1 neurofibromas has been recently reviewed¹⁶⁹. These haematopoietic cells have a crucial role in the formation and growth of neurofibromas. In some GEM models, the haematopoietic cells must express mutant Nf1 for neurofibromas to form¹¹⁹. Treatment with PLX3397 — which targets both KIT signalling to prevent mast cell recruitment to tumours, and CSF1 receptor signalling to prevent macrophage recruitment to tumours — had two effects on neurofibromas. It increased their size when given during tumour initiation and enabled tumour regression in some mice when given after tumour establishment. Therefore, macrophages may protect against developing tumours and later become permissive to tumour formation¹⁶⁸. However, wounding peripheral nerves in Nf1-mutant mice, which recruits macrophages, facilitates neurofibroma formation, implying that macrophages may promote tumour formation in this context^83,170. OPGs that arise in Nf1-mutant mice also contain CX3CR1-expressing microglial cells (brain-resident macrophages). Cx3cr1^−/− mice show delayed optic glioma formation, which supports interference with the micro-environment as a possible therapy in patients with NF1 (REF. 171). In an MPNST xenograft, PLX3397 treatment resulted in macrophage depletion and substantially delayed MPNST growth. The effect was enhanced by the mTORC1 inhibitor rapamycin and correlated with enhanced depletion of macrophages¹⁷².

Discovering new drug targets for NF1-mutant cells

Screening a library of 200,000 small molecules on mouse Nf1-mutant MPNST cells identified compound 21, with selectivity towards NF1-mutant cells and efficacy in xenografts¹⁷³. Another library-screening study identified UC1 as a small molecule that targets NF1^−/− versus sporadic MPNST cell lines. Budding yeast in which IRA2 was deleted validated selectivity of UC1 for NF1-mutant cells¹⁷⁴. Direct targets of these compounds are not known. Gene expression, methylome and copy number changes on several sample sets are publically available for neurofibromas and MPNSTs, and are being used to identify pathways and targets for drug discovery¹⁷⁵. Preliminary assessments of miRNAs and serum biomarkers have also become available. Several proteins were identified either at increased levels in patients with NF1 but without neurofibromas (interferon-γ, interleukin-6 and tumour necrosis factor) or at increased levels in patients with NF1 and MPNST (insulin-like growth factor-binding protein 1, C-C motif chemokine 5 (CCL5) and adrenomedullin)^176,177. Expression of miR-801, miR-214 and miR-24 can distinguish patients with both NF1 and MPNST from patients with NF1 but without MPNST¹⁷⁸. However, no NF1 biomarker has yet been tested clinically.