Neurofibromatosis type 1 (NF-1 [MIM 162200]) is a common autosomal dominant disorder associated with increased morbidity and mortality 1 . Neurofibromas are the hallmark of NF1. They are benign tumors that arise from connective tissue of nerve sheaths, especially the endoneurium. We used the classification proposed by Riccardi which defines four categories of neurofibromas 2 : (i) cutaneous neurofibromas, presenting as an exophytoc tumor moving with the skin on examination, (ii) subcutaneous neurofibromas that lie deeper in the skin, do not move with it, are firm to palpation and may be tender (iii) internal or deep neurofibromas may involve any nerve anywhere along its length and are not palpated. They are therefore identified later in their course of growth (iv) plexiform neurofibromas may involve multiple fascicles and branches, and extend into surrounding structures. Clinical investigation identifies thickened hypertrophic skin, hyperpigmentation of tissue and subcutaneous tumor 2 3 .

Subcutaneous neurofibromas (SC-NFs) were independently associated with mortality among adults with NF-1 in two independent populations from France 4 and North America 5 . The main causes of death were spinal cord compression by internal neurofibromas and malignant peripheral nerve-sheath tumours (MPNSTs) developed from pre-existing internal neurofibromas 4 5 . A recent study demonstrated that having at least two SC-NFs was independently associated with having internal neurofibromas 6 . However, this association needs to be characterised in detail. Furthermore, a strong association between peripheral neuropathy and SC-NFs has been reported 7 . We hypothesized that this association was related to an association between SC-NFs and internal neurofibromas located along the nerve roots.

The primary objective of this prospective matched case-control study was to characterise the association between SC-NFs and internal neurofibromas in patients routinely investigated with magnetic resonance imaging (MRI), the most sensitive method for detecting internal neurofibromas and assessing their features (e.g., type, distribution, location, and size). The secondary objective was to determine whether peripheral neuropathy was associated with having internal neurofibromas along the nerve roots. Peripheral neuropathy was detected using a routine electrophysiological study.

Between 2005 and 2008, 110 cases were matched for age, sex, and hospital to 110 NF-1 controls. Four cases and eight controls were excluded secondarily because their MRI scans were not interpretable. Table 1 reports the characteristics of the 106 cases and 102 controls. The cases had a mean age of 41 (± 13) years; 59% were females, 85 (80%) had two to nine SC-NFs and 21 (19%) at least ten SC-NFs. There were no significant differences between cases and controls for age and sex (matching variables) or clinical features (e.g., dermatological characteristics, orthopaedic complications, and neurological abnormalities) except the cutaneous and plexiform neurofibromas. Significantly greater proportions of cases than controls had no cutaneous neurofibromas (p = 0.03) and at least one plexiform neurofibroma (p = 0.04). Cases had fewer large café-au-lait spots compared to controls but the difference was not statistically significant (p = 0.07).

Table 2 summarises the results of the univariate analysis. The presence of SC-NFs was strongly associated with having paraspinal neurofibromas (aOR, 4.3; 95%CI, 2.2 to 8.2) and sciatic neurofibromas (aOR, 6.1; 95%CI, 2.9 to 13). The associations tended to be stronger for diffuse vs. focal paraspinal neurofibromas, intradural vs. extradural neurofibromas, and internal neurofibromas measuring at least 3 cm vs. less than 3 cm. Similarly, for sciatic neurofibromas, diffuse distribution and tumour size of at least 3 cm showed trends toward stronger associations with SC-NFs than focal distribution and size smaller than 3 cm, respectively. Finally, the presence of axonal neuropathy with SCVs was significantly associated with having SC-NFs (aOR, 7.7; 95%CI, 1.6 to 36.6) (Table 3).

Radiculopathies were not significantly associated with the presence of SC-NFs (Table 3).

The joint analysis identified no interactions between variables associated with the presence of SC-NFs. A strong association was found between the presence of sciatic neurofibromas and that of axonal neuropathy with SCVs (aOR = 22.1; 95% CI, 4.2 to 115.3). In the bivariate analysis including axonal neuropathy with SCVs and sciatic neurofibromas, axonal neuropathy with SCVs was no longer associated with SC-NFs (aOR, 3.0; 95%CI, 0.6 to 15.1; p = 0.18). The association between axonal neuropathy with SCVs and SC-NFs persisted only in the subgroup with sciatic neurofibromas (aOR, 2.00; 95%CI, 1.01 to 4.12; vs. aOR, 0.69; 95%CI, 0.24 to 4.12 in the subgroup without sciatic neurofibromas). Thus, the association between SC-NF and axonal neuropathy with SCVs was due to the presence of internal sciatic neurofibromas.

In the multinomial logistic regression analysis, patients with at least ten SC-NFs had a significantly higher risk of paraspinal neurofibromas (aOR, 82; 95%CI, 10.4 to 647.9) compared to patients with two to nine SC-NFs (aOR, 2.7; 95%CI, 1.4 to 5.3) (Figure 1). This dose-effect relationship was observed for distribution, location, and size (Figure 1, Table 4). Similar dose-effect relationships were found for sciatic neurofibromas, particularly regarding distribution, location, size (Figure 1, Table 4). Finally, axonal neuropathies with SCVs were more common in patients with at least ten SC-NFs (aOR, 29.9; 95%CI, 5.5 to 162.3) than in patients with two to nine SC-NFs (aOR, 4.4; 95%CI, 0.9 to 22.0) (Figure 1).

This study shows that sub-cutaneous neurofibromas were significantly associated with having paraspinal and sciatic neurofibromas detected by routine MRI. The strength of the association tended to be higher when the paraspinal and sciatic neurofibromas were diffuse, intradural, and at least 3 cm in diameter. Moreover, paraspinal and sciatic neurofibromas were significantly more common in patients having at least ten SC-NFs than in patients having two to nine SC-NFs. A dose-effect relationship was also found for distribution, location, and size of the internal neurofibromas. Finally, our data indicate that the strong association between SC-NFs and axonal neuropathy with SCVs is due to the association between SC-NFs and internal neurofibromas.

Neurofibromatosis-1 (NF1) is a common disease with an increased propensity for developing both benign and malignant tumors 10 . NF1 has been reported to be associated with a 15-year decrease in life expectancy 1 . One of the aims of the Réseau NF-France was to identify clinical predictive factors for mortality, in order to better adjust the follow up and propose new treatments for eligible patients. We previously demonstrated in two different NF1 populations, from France 4 and North America 5 , that the presence of at least 2 subcutaneous neurofibromas was associated with a higher risk of death. The main causes of death were compression of neighboring organs due to internal neurofibromas, and malignant peripheral nerve-sheath tumours developing from preexisting internal neurofibromas.

The main concern in the long-term clinical management of adults with NF-1 is the identification of patients at high risk for MPNSTs developed from pre-existing internal neurofibromas. At present, no effective treatment is available for MPNST untreatable by surgery. New treatments are usually approved based on proof of efficacy in a controlled trial. When conducting controlled trials, an important consideration is patient selection based on severity of illness. Controlled trials in adults with NF-1 should target patients who have a high-risk phenotype defined by the presence of SC-NF. Patients with a high risk of internal neurofibromas must be therefore identified as potential new therapies are being developed. For instance, there are at least seven phase II/III trials in which the internal neurofibromas are being treated by targeted therapy such as ranibizumab, niltotinib... (clinicaltrials.gov). In addition, we believe that targeted therapy will be of greater efficacy on non-transformed internal neurofibromas. At present time, an annual careful clinical examination is recommended for all patients with NF1 in order to detect clinical symptoms associated with MPNSTs such as pain, neurological deficit and enlargement of a pre-existing peripheral nerve sheath tumour 11 . Routine MRI screening, which is highly reliable for detecting internal neurofibromas, is not recommended in patients without the alarming clinical signs mentioned above 12 . It is therefore critical to precisely clinically identify the patients with the maximum likelihood of having internal neurofibromas and to closely follow them by it iterating MRI. The first step is to assess recognizable clinical features as potential predictors of morbidity and mortality in order to further improve the clinical management of NF1. We recently developed a clinical score for predicting the presence of internal neurofibromas in adults with NF1. The NF1Score, computed via a linear combination of four variables (age ≤ 30 years, fewer than six café-au-lait spots, no cutaneous neurofibromas, and two or more SC-NFs), can be used to accurately predict the presence of internal neurofibromas in NF-1 patients and help target patients for now emerging clinical trial 6 . The high point scoring of the NF1Score is the presence of SC-NFs. SC-NFs were highly associated with having internal neurofibromas (OR, 4.7; 95%CI, 2.1-10.5) 6 . The second step, our present work, was to confirm and explain a detailed and unbiased characterisation of the links between internal neurofibromas and SC-NFs. To our knowledge, this was the first controlled study involving a detailed characterization of the relationship between SC-NFs, internal neurofibromas, and peripheral neuropathies in adults with NF-1.

The internal validity of this study depends on many factors, including the unbiased recruitment of cases and controls and the accuracy of the information obtained about the presence of internal neurofibromas. MRI, an extremely reliable method for detecting internal neurofibromas, was performed routinely in all our cases and controls. Assessment bias was minimized by having the MRI data reviewed by senior radiologist (PB) who was unaware of the clinical data. Assessment bias was minimized by having the electrophysiological data reviewed by senior physiologist (JPL) who was unaware of the MRI data. Clinical information regarding the neuropathy was not available. However, according to our experience the diffuse symmetric peripheral neuropathies related to NF-1 are asymptomatic or pauci-symptomatic with minor sensory manifestations 7 .

As with all hospital-based case-control studies, selection bias cannot be excluded. However, the prevalence of the various clinical features in the cases were similar to those in the controls (except for the number of cutaneous neurofibromas) and to those reported previously in NF-1 patients, indicating that our study population was representative of NF-1 patients 13 . Regarding the number of cutaneous neurofibromas, the association linking the absence of cutaneous neurofibromas to the presence of SC-NF is in agreement with a previous study showing higher mortality among NF-1 patients who had no cutaneous neurofibromas 4 .

In summary, this study shows a strong association between the presence of internal neurofibromas and the presence of SC-NF. The increased risk in NF-1 patients with subcutaneous neurofibromas (high risk phenotype) can be ascribed in part to the associations between SC-NFs and internal neurofibromas (prone to transformation to MPNSTs) and the presence of a neuropathy with SCVs. The prevalence of neuropathy with SCVs is particularly high in patients with at least ten SC-NFs. To better understand the patho-physiological processes in the high risk phenotype of NF-1 patients, the next step would be to perform genotype-phenotype correlation studies. A more severe clinical phenotype has been reported in NF1 patients carrying genomic microdeletions involving NF1 and surrounding gene compare to patients with mutations restricted to the NF1 genes. A recent study confirmed in a large cohort that NF1 microdeletion patients have a significantly higher incidence of learning disabilities and facial dysmorphism than patients with intragenic NF1 mutation 14 . However no association were found between SC-NFs and MPNSTs and NF1 microdeletion patients.

Further studies are therefore needed to find others genes beside NF1 implied in the internal NF development. One of the major aims of such studies would be to identify key targets for novel drugs' development.

aOR: adjusted Odds Ratio; CI: Confident Interval; MPNST: Malignant peripheral Nerve sheath tumor; MRI: magnetic resonance imaging; NF-1: Neurofibromatosis 1; SC-NF(s): Sub-Cutaneous Neurofibroma(s); SCVs: Slow Conduction Velocities; SD: standard deviation

ES, SBG, LVA, SF, JPL, AD, PB, PC, SB, PW have non-financial interests that may be relevant to the submitted work.

PW conceived and designed the study, and is guarantor for the study. SBG contributed to the concept and design, and supervised analysis and interpretation of the data. ES performed the data analyses and wrote the initial draft of the article, to which all the authors contributed. All authors had full access to the data in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis. All authors have given final approval for the final version to be published. The NF-France network contributed to the patients' inclusion.