A Translational Approach to Spinal Neurofibromatosis: Clinical and Molecular Insights from a Wide Italian Cohort

Simple Summary At present, no systematic study of the clinical spectrum and molecular characteristics of NF1 patients with spinal neurofibromatosis (SNF), a phenotypic subclass of neurofibromatosis 1 (NF1), has been carried out. Here, we provide evidence that SNF patients are at high risk of problematic neurofibromas, presenting not only bilateral neurofibromas involving all spinal roots, but also a higher incidence of internal neurofibromas and nerve root swelling. From a histopathological view, not only neurofibromas, but also neurogangliomas are present in SNF. The analysis of 19 families with at least 1 member affected by SNF showed a high phenotypic variability within the SNF families. Furthermore, we discovered a higher prevalence of missense mutations in SNF compared to classical NF1. Both clinical features and genetic testing can help in identifying cases at risk of SNF, and that are more likely to benefit from a spinal MRI scan. Abstract Spinal neurofibromatosis (SNF), a phenotypic subclass of neurofibromatosis 1 (NF1), is characterized by bilateral neurofibromas involving all spinal roots. In order to deepen the understanding of SNF’s clinical and genetic features, we identified 81 patients with SNF, 55 from unrelated families, and 26 belonging to 19 families with at least 1 member affected by SNF, and 106 NF1 patients aged >30 years without spinal tumors. A comprehensive NF1 mutation screening was performed using NGS panels, including NF1 and several RAS pathway genes. The main features of the SNF subjects were a higher number of internal neurofibromas (p < 0.001), nerve root swelling (p < 0.001), and subcutaneous neurofibromas (p = 0.03), while hyperpigmentation signs were significantly less frequent compared with the classical NF1-affected cohorts (p = 0.012). Fifteen patients underwent neurosurgical intervention. The histological findings revealed neurofibromas in 13 patients and ganglioneuromas in 2 patients. Phenotypic variability within SNF families was observed. The proportion of missense mutations was higher in the SNF cases than in the classical NF1 group (21.40% vs. 7.5%, p = 0.007), conferring an odds ratio (OR) of 3.34 (CI = 1.33–10.78). Two unrelated familial SNF cases harbored in trans double NF1 mutations that seemed to have a subclinical worsening effect on the clinical phenotype. Our study, with the largest series of SNF patients reported to date, better defines the clinical and genetic features of SNF, which could improve the management and genetic counseling of NF1.

In 2014, Ruggieri [11] reviewed all cases published up until the time of writing, and defined the clinical criteria of another subtype of the disease: spinal neurofibromatosis (SNF). It is characterized by bilateral neurofibromas involving all spinal roots and few, if any, cutaneous manifestations. This feature allows one to specifically distinguish SNF from neurofibromatosis type 1 (NF1) and multiple neurofibromas few spinal root (MNFSR), presenting a single or few isolated spinal neurofibromas. SNF entails greater morbidity than the classical NF1.
More recently, in a large descriptive study on spinal lesions in NF1, 37 new SNF cases were reported [12]. A high incidence of paraspinal plexiform tumors was observed in those patients, and 23.8% of them had a history of previous spinal surgery. In fact, spinal tumors can remain asymptomatic for years; however, once severe neurological deficits have developed, the likely success of any surgical treatment is greatly reduced. Nevertheless, the value of performing routine spinal MRI scanning at the time of the diagnosis is still controversial, and within the current guidelines, no specific surveillance is foreseen.
Furthermore, SNF is an example of the high intrafamilial variability observed in NF1: SNF patients can belong to families with classical NF1 or to families with all affected members with SNF. Even if some clinical features might be used to predict internal neurofibroma onset [13], so far, there are no reliable patterns to distinguish patients at risk of developing SNF. An association of SNF with missense or splicing mutations has been reported [11,[14][15][16], but further studies need to be carried out to confirm those correlations. In addition, in individuals with missense mutations affecting NF1 codons 844-848 or pArg1276, multiple spinal neurofibromas were more common [9,10].
The aim of this study was to describe the clinical and genetic features characteristic of sporadic and familial cases of SNF, and to also attempt comparisons between the SNF phenotype and classical NF1 groups.

Individuals and Phenotypic Data
A total of 74 individuals with a diagnosis of NF1, according to the revised diagnostic criteria for neurofibromatosis type 1 [17], and of the SNF phenotype, according to Ruggieri's criteria, were identified using the IRCSS C. Besta Neurological Institute, the IRCCS Ca 'Granda Foundation Ospedale Maggiore Policlinico, and the Azienda Ospedaliera Uni-versitaria dell'Università degli Studi della Campania "Luigi Vanvitelli" electronic databases. Forty-one individuals had sporadic NF1, while thirty-three had a familial form of the disease. All other affected members that were alive and available were clinically assessed, and a cohort of 27 affected relatives (11 MNFSR, 7 SNF, and 9 classical NF1) were added to the original patient set, making it possible to identify 19 NF1 families with at least 1 member affected by SNF, and a total of 81 SNF subjects. In Supplementary Materials Figure S1, the family pedigrees are reported. Two families had four affected members, five families had three affected relatives, and twelve families had two.
All medical records were surveyed. The data were collected at the time of the genetic screening and reverified for accuracy at the time of this study. We focused on symptoms and signs possibly related to spinal neurofibromas, such as pain that could be attributed to spinal abnormalities or tumors; neurological symptoms, including weakness, sensory deficit, and changes in tone and reflexes; in addition to the already well-known NF1 characteristics: CAL; skinfold freckling; cutaneous, subcutaneous, and plexiform neurofibromas; Lischnodules; visual impairment; epilepsy; cognitive impairment; optic nerve glioma (OPG); other neoplasms of the central nervous system and other organs; and skeletal and vascular abnormalities. In particular, a neurofibroma is defined as plexiform when it grows along the length of a nerve, and may involve multiple fascicles and branches. The date of birth, gender, age at the time of the last visit, and the mode of inheritance were also recorded. All patients underwent brain and spinal MRI scans with gadolinium.
Cases with missing data for a particular sign and/or symptom were classified as "unknown" and, consequently, excluded from that part of the genotype-phenotype analysis.
We assessed the clinical significance of the sequence variants according to the American College of Medical Genetics (ACMG)/Association of Molecular Pathology (AMP) guidelines [38].

Statistical Analysis
The X 2 test and two-tailed Fisher's exact probability test were used to compare categorical variables, with a p-value of < 0.05 considered as statistically significant. The odds ratios (ORs) and their 95% confidence intervals (CIs) were calculated. The genotype-phenotype associations were studied using multiple logistic regression. The Benjamini-Hochberg (B_H) method with false discovery rates of 0.25, 0.05, and 0.01 was used to correct p-values for multiple testing. The descriptive, frequency, and comparative statistical analyses were carried out using SPSS 22.0 software.

Demographics
On 30 June 2020, 768 subjects affected by NF1 and followed by the IRCSS C. Besta Neurological Institute, the IRCCS Ca 'Granda Foundation Ospedale Maggiore Policlinico, and the Azienda Ospedaliera Universitaria dell'Università degli Studi della Campania "Luigi Vanvitelli" underwent a spinal MRI scan. In 220 (28.6%) cases, the MRI was reviewed by a specialist neuroradiologist and showed spinal neurofibromas; in 81 (36.8%) cases, bilateral neurofibromas involving all spinal roots were present (SNF) ( Figure 1); and in 139 (63.2%) cases, a single or a few isolated spinal neurofibromas were detected.

Clinical Characteristics of the SNF Cohort
The demographic and clinical characteristics were analyzed in all 81 SNF ( male and 55 male) patients, 55 from unrelated families and 26 belonging to 19 fa with at least 1 member affected by SNF. The median age was 35, ranging from 15 years. The SNF patients' clinical features are reported in Table 1.

Clinical Characteristics of the SNF Cohort
The demographic and clinical characteristics were analyzed in all 81 SNF (26 female and 55 male) patients, 55 from unrelated families and 26 belonging to 19 families with at least 1 member affected by SNF. The median age was 35, ranging from 15 to 74 years. The SNF patients' clinical features are reported in Table 1.
The spinal neurofibromas were symptomatic in 44 out of 81 (54.3%) cases. In most cases, internal neurofibromas (45.7%) and nerve root swelling (32.1%) were also found. SNF patients underwent the spinal MRI scan at a median age of 28  years, and the median follow-up duration was 98 (6-341) months. In total, 15 out of 81 (18.5%) SNF patients underwent spinal surgery (10 at the cervical level, 2 at the cervical and lumbar levels, and 3 at the lumbar level) at a median age of 25  years, after a median time of 2 (1-86) months from the first spinal MRI scan. The reason for surgery was not only the tumor growth according to REINS criteria [40], but also the report of myelopathy. The histopathological diagnosis was neurofibromas in 13 cases and ganglioneurofibromas in 2 cases. A total of 15/81 (18.5%) cases had less than 6 CALS, and 50/81 (47.3%) had no freckling; 9/81 (11.1%) had neither more than 5 CALS nor freckling, but they fulfilled the revised diagnostic criteria for neurofibromatosis type 1 [17] due to the presence of other clinical signs, such as neurofibromas and Lisch nodules. Only 17/81 (21%) individuals had more than 10 CALS.
For 74 patients, data on the presence or absence of Lisch nodules were available: they were present in 59/74 (79.7%) cases.

Comparisons of the Clinical Characteristics
Observed in the SNF Cohort, the Cohort of the "Classical" NF1 Phenotype from Our Institutions, and in Previously Described Classical NF1 Cohorts from the Literature The clinical characteristics of our SNF cohort were compared with those observed in a cohort of classical NF1 patients (i.e., patients without spinal tumors) followed by the same institutions. There were 68 females and 38 males with a median age of 47 (31-75) years. Furthermore, when data were available, the clinical features were also correlated with those previously reported in large-scale NF1 classical cohorts that had already been used in other genotype-phenotype studies [1,[18][19][20][21][22][23][24][25][26][27][28][29][30][31]. All comparisons are reported in Table 2.
The main features of our SNF cases were a higher number of internal neurofibromas (45.7 vs. 6.6%; p < 0.001) as well as nerve root swelling (32.1 vs. 2.8; p < 0.001) compared to our classical NF1 cohort. In our classical NF1 cohort, patients with spinal NF were deliberately excluded; therefore, no symptoms such as low back pain or neurological deficit related to spinal involvement were reported. However, the frequency of symptomatic spinal neurofibromas reported in two previous studies in which patients with spinal tumors were included [18,24] were significantly lower (54.3% vs. 1.7 and 1.8; p < 0.001). In addition, within skeletal abnormalities in our cohort, scoliosis was more frequent, while other abnormalities were more rare than previously reported in large-scale cohorts of classical NF1 [18,23,24,29].
The SNF patients, as well as the cases with classical NF1 followed by our institutions, had a higher risk of being affected by tumors different from OPG and by cardiovascular abnormalities compared to the classical cohorts reported in the literature [24,26]. These findings probably reflect the differences in the case selection and not specific phenotype features.

Phenotypic Variability within SNF Families
We identified 19 NF1 families with at least 1 member affected by SNF. In Supplementary Materials Figure S1, the family pedigrees are reported. Two families had four affected members, five families had three affected relatives, and twelve families had two. Overall, 26 patients had SNF, 12 had MNSFR, and 9 had a classical form of the disease.
We observed a phenotypic variability within the SNF families. In most families (11/19), all NF1 individuals were affected by SNF or MNFSR.
As proposed by Ruggieri, we called families "pure SNF families" when all affected members had SNF, "partial SNF families" when all affected members had SNF or MNSFR, and "multiple phenotype families" when at least one member had SNF and the other affected members had MNSFR or classical NF1.
No phenotype differences were observed between SNF cases belonging to "pure", "partial", or "multiple phenotype" families, or between SNF patients included in the 19 families and the others SNF patients.

Mutation Analysis
Mutational analysis for NF1 was performed in 208 NF1 cases (81 SNF, 11 MNFSR, and 115 classical NF1). NF1 variants were observed in 204 cases. In four other cases, in all those affected by SNF and fulfilling the diagnostic criteria for NF1 [17], no NF1 causative variants were identified. We identified 160 different NF1 variants. The NF1 variants observed are reported in Tables 3-5 along with the molecular details (DNA, RNA, and protein change) and the classification of the variants by type, tertile [41], and domain [42][43][44]. In Table 3, 19 families with at least 1 member affected by SNF are reported; Table 4 presents 55 SNF cases; and in Table 5 106 classical NF cases are described.
Seventeen causative variants detected in the classical patients and twenty-eight in the SNF (thirteen belonging to the SNF families) were never reported; the others were already described (Tables 3-5).
Interestingly, five patients showed more than one NF1 variant. For three familial cases belonging to three families, it was possible to infer whether one or both NF1 alleles were affected. Family 1, family 17, and family 18 were informative in answering the above question.
Precisely, in family 1, the SNF patient 1136 (proband), who carried the c.62T > A (pLeu21His) NF1 missense variant inherited from his MNFSR-affected father (1139) and shared by his SNF-affected brother (1140), showed a second c.528T > A (p.Asp176Glu) NF1 missense mutation inherited from his mother, indicating that the two missense variants were in trans. Despite his mother (1141) not being affected by NF1, according to the revised diagnostic criteria for neurofibromatosis type 1 [17], the p.Asp176Glu substitution was predicted to be damaging by 9/20 predictors, and may have a subclinical significance. Accordingly, patient 1136 showed a more severe phenotype than his affected father and brother.
In family 17, the SNF patient N04 (proband) presented c.3314 + 2T > C splicing NF1 variants inherited from her MNFSR-affected mother N05, and showed a second c.7532C > T (p.Ala2511Val) NF1 missense variant, predicted to be damaging by 10/20 predictors inherited from her father (never clinically evaluated) and shared by her brother N06, displaying a cutaneous NF1 form. The other brother N03, harboring only the c.3314 + 2T > C splicing NF1 variants inherited from the mother, was also affected by SNF, as the proband, but the clinical phenotype was less severe: no internal or plexiform neurofibromas were present. In addition, in this case, the second mutation could have a subclinical effect that could worsen the clinical phenotype of the proband carrying mutations on both NF1 alleles.
In family 18, the SNF patient N07 (proband) showed the pathogenic c.1595T > G (p.Leu532Arg) and the uncertain c.3242C > G (p.Ala1081Gly) NF1 missense variants, both shared by his sister, patient N08, and inherited by his nephew, patient N09, indicating that the two variants were in cis. The evidence that both the sister and her child were affected by classical NF1 suggests that this double-mutated allele is not specifically associated with a specific NF1 form.       Two sporadic SNF patients were carriers of two concomitant variants in the NF1 gene, but we were not able to define their phase as in cis or in trans, because their parents were not available for a segregation study. Precisely, the patient 2207 (2207) presented a pathogenic stop variant and the missense variant NF1 c.1246 C > T (p.Arg4016*) and c.403C > T (p.Arg135Trp). Even if the second variant was reported as "uncertain" in Clinvar, this variant has been classified as potentially damaging by 19 predictors out of the 20 interrogated, and it is absent from controls in the GnomAD and in the 1000 genomes (1000g2015aug_eur) databases. Moreover, the variant replaces the conserved basic amino acid arginine at residue 135 to polar-neutral tryptophan. The patient 891 (891) had a pathogenic frameshift NF1 variant c.6346_6347insA (p.Ser2116Tyr*6) and the missense variant NF1 c.5221G > A (p.Val1741Ile), classified as potentially damaging by 8 out of 20 predictors questioned and not reported either in the GnomAD or in the 1000 genomes databases. The variant replaces the conserved hydrophobic and aliphatic amino acid valine with hydrophobic and aliphatic isoleucine at residue 1741 in the PH domain of NF1.

Comparison of the NF1 Variants between the SNF and the Classical NF1 Cohorts
We chose one case for each family, the proband, and we compared the NF1 causative mutations, excluding uncertain variants, recorded in the SNF and classical NF1 patients. The numbers of SNF patients with large deletion (microdeletion type 1 and atypical microdeletion), frameshift, missense, nonsense, splicing, and small deletion/insertion mutations were 7 (10%), 17 (24.3%), 15 (21.4%), 10 (14.2%), 19 (27%), and 2 (2.9%), respectively. The proportion of missense pathogenic variants was higher in the SNF cohort than in the classical NF1 group (p = 0.007; OR 3.34; CI 1. 33-8.38), while the proportion of nonsense mutations was lower (p = 0.055; OR 0.46; CI 0.20-1.03). Furthermore, after applying the Benjamini-Hochberg correction for multiple testing with a false discovery rate of 0.05, the first differences remained statistically significant (Table 6). In bold statistically significant p-value.* Statistically significant p-values with a false discovery rate of 0.05 after correction for multiple testing using the Benjamini-Hochberg procedure.
Furthermore, we added our analysis the data concerning the distribution of NF1 causative variants in 49 SNF patients already published in the literature and reported by Ruggieri [11]. Among the 49 patients, we chose 1 case (the proband) for each family and all of the reported sporadic patients for a total of 25 SNF patients. The combined analysis with our data (Table 7) showed a statistically significant increase in missense mutations (25.3% vs. 7.5%; p = 0.001; OR 4.14; CI 1.76-9.75) in the SNF cohort compared to our classical patient cohort; the p-value remains statistically significant after correcting with the Benjamini-Hochberg correction method for multiple testing with a false discovery rate at 0.025 and 0.01.
The proportion of truncating variants was lower (p = 0.03; OR 0.44; CI 0.21-0.93) and the proportion of nontruncating variants was higher (p = 0.036; OR 2.24; CI 1.04-4.81) in the SNF patients when compared to those observed in the classical NF1 cases (Table 8).
We also investigated whether the risk of developing SNF was associated with causative variants in one particular NF1 domain (Table 9). In bold statistically significant p-values. § Statistically significant p-value with a false discovery rate of 0.05, 0.025, and 0.01 after correction for multiple testing using the Benjamini-Hochberg method. In bold statistically significant p-values. * Statistically significant p-values with a false discovery rate of 0.05 after correction for multiple testing using the Benjamini-Hochberg procedure. n = number of NF1 mutations from the analyses; whole gene deletions were excluded. Some mutation locations may overlap different regions. In bold statistically significant p-value. p-values of 0.025 after correction for multiple testing using the Benjamini-Hochberg procedure with a false discovery rate of 0.25 did not remain significant.
The patients with harboring variants in the HLR domain had a mild to higher risk of developing SNF (p = 0.025; OR 2.5; CI 1.1-5.64) (Table 9). Furthermore, after placing the HLR domain and the HLR-CTD domain together, the risk of developing SNF correlated with variants in those domains (34.3% vs. 16%; p = 0.006; OR 2.74; CI 1.31-5.72). The difference remained significant after Benjamini-Hochberg correction with an FDR of 0.25. As concerns the causative variant types within the nineteen SNF families, of the six families with at least two SNF cases, in two families, missense variants were observed; in another two, exon deletions were observed; in one, a splicing mutation was observed; and in another one, a frameshift mutation was observed. Conversely, in the six multiple phenotype families, five out of six mutations were truncating.

Discussion
To our knowledge, this is the largest series of SNF patients reported to date. All cases were followed by three different Italian institutions conforming to the same clinical protocol. In fact, at present, no systematic study of the clinical spectrum and molecular characteristics of NF1 patients with the SNF phenotype has been carried out, even though spinal neurofibromatosis was first described by Pulst in 1991 [45] and by Poyhonen in 1997 [46]. In 2015, Ruggeri established precise criteria to identify this distinct phenotype, and enucleated 49 cases with a true SNF phenotype out of 98 spinal NF cases already described in the literature [11]. Recently, Curtis-Lopez described 37 cases that met the criteria for SNF in a retrospective review of 303 NF1 patients with different types of spinal lesions; unfortunately, no data on NF1 gene mutations were reported [12].
The reported frequency of spinal neurofibromas in NF1 ranges from 15.9% [47] to 65% [25]. In our series, the prevalence of spinal neurofibromas of any type (i.e., single, few isolated neurofibromas, and SNF) and the prevalence of the SNF phenotype alone was 28.6% and 10.5%, respectively. The prevalence of the SNF phenotype was similar to those reported by Curtis-Lopez (12.2%) [12], confirming that, if rigorous diagnostic criteria are applied, SNF is a rare distinct phenotype of NF1. On the contrary, the frequency of spinal neurofibromas in our study was lower than those previously reported in [25] and by Curtis-Lopez [12], at 65% and 58.1%, respectively; however, this is in line with Well (39.6%), who performed whole-body MRI scans on the studied patients [31]. This discrepancy may reflect the differences in selecting patients to undergo a spinal MRI scan.
In our cohort, spinal neurofibromas were symptomatic in 54.3% of the cases. A varied prevalence of the symptomatic spinal tumors in NF1 patients has been described in the literature, ranging from to 2% [25] to 24% [12]; however, both patients with SNF and MNFSR were included by the authors. These findings further demonstrate the effect of spinal abnormalities on affected patients, and emphasize the necessity to evaluate routinely performed MRI scans regarding these abnormalities.
Only a minority of those patients undergo surgery: 18.5% in our study and 23.85% in that by Curtis-Lopez [12]. In 15 cases in our cohort, spinal tumors were removed; the diagnosis was neurofibromas in 13 cases and ganglioneuroma in 2 cases. In the literature, approximately 15 SNF patients [14,25] underwent spinal surgery; the histopathological diagnosis was neurofibroma in all, including one who had multiple cervical and dorsal ganglioneurinomas [48]. Furthermore, 30 NF1 patients with gangliomas and NF1 have been described, but only in 5 were ganglioneurinomas located in the spine, and only 1 case had the SNF phenotype [49]. Ganglioneuromas originate from neural crest cells in the sympathetic ganglia or adrenal medulla, can be present at a young age, are sporadic or in association with NF1, and are mostly located in the posterior mediastinum, retroperitoneum or, very rarely, heterotopic areas, including sensory ganglia and nerves [50]. Our results show that not only neurofibromas, but also a rare histological type of neoplasia, ganglioneurinoma, are found in the spinal phenotype.
At present, the appearance of spinal neurofibromas seems to be an age-dependent process. The risk of developing internal neurofibromas, including spinal neurofibromas, could occur between adolescence and the age of 30 years [13]. Furthermore, we observed that some patients with spinal neurofibromas localized only in some roots developed new neurofibromas during the course of the disease and evolved to SNF. Therefore, we chose to compare our SNF patients to the patients aged >30 years without spinal tumors, because the likelihood that they could develop a spinal NF was the lowest.
Our study better defines the cutaneous features of the SNF phenotype. As already reported by Ruggieri, hyperpigmentation signs (CAL spots and freckling) were rarer in our SNF patients [11]. Conversely, our cohort showed that subcutaneous neurofibromas and, above all, internal neurofibromas were more frequently observed in SNF patients. One limitation of the research is that we did not perform a whole-body MRI scan in the cases studied; therefore, it is possible to speculate that the frequency of the internal neurofibromas reported is likely underestimated. Plotkin, who assessed the internal tumor burden using whole-body MRI, reported a statistically significant correlation between the presence of internal sheath tumors and the decreasing number of CAL spots and the presence of subcutaneous neurofibromas in NF1 patients [29]. The presence of subcutaneous neurofi-bromas has been associated with a higher risk of mortality [28]. Because they are not a cause of mortality, per se, they could be an indicator of a more aggressive form of the disease. The co-occurrence of spinal neurofibromas and subcutaneous neurofibromas and internal neurofibromas may suggest the presence of shared pathogenic factors. All neurofibromas arise from Schwan cells, but the Schwan cells' interactions with axons and mostly other cells, fibroblast, endothelial cells and several components of the microenvironment, such as mast cells, macrophages, and lymphocytes, are needed for development [51][52][53][54]. A better knowledge of the tumor microenvironment and of the interactions within and between the cells that compose the different subtypes of NF is needed to understand why some body regions are particularly affected by neurofibromas.
Plexiform neurofibromas were more frequent in the SNF cases, and the difference reached statistical significance only when they were compared to the general NF cohorts reported in the literature. While in all cases studied by us, a spinal MRI scan was performed, in the previously published papers [18,24], only externally visible plexiform NF cases were counted.
The risk of developing malignancies appears to be higher in SNF cases only if they were compared to cases previously described in the NF1 cohorts from the literature [24] but not when they were compared to our classical patient cohort. All included cases were followed by highly specialized hospitals where more severe cases are referred; the difference is likely related to a biased selection.
Strong positive relationships with any type of spinal neurofibromas (i.e., single, few isolated neurofibromas, and SNF) and scoliosis were reported in a study by the Children's Tumor Foundation, in which 2051 adult NF1 cases self-reported phenotypic traits [55]. The cooccurrence of scoliosis and spinal tumors was observed in 45% of cases by Koczkowska [9] in patients harboring missense mutations affecting NF1 codons 844-848. An association of scoliosis with the scalloping of the vertebral bodies or meningoceles but not with intraforaminal tumors were reported by Well [31], while a mild to moderate degree of scoliosis was observed only in 18% of the SNF cohort [11]. In fact, bony remodeling due the presence of tumors but also other factors, such as dural ectasias, and abnormal bony metabolism can contribute to the development of scoliosis.
Another limitation of our study is that we did not assess quality of life. Symptoms and features of NF1 are very heterogenous, some patients may experience minimal effect on their life, while others struggle with disfigurement, neurological disfunction and disability. Most previous studies assessed the quality of life in the NF1 patients in comparison with a general population, reporting a lower quality of life in the NF1 cases [56]. In order to explore whether SNF could cause a significant decrease in the quality of life, we should use suitable and specific measures such as PlexiQol [57]. Unfortunately, the test is not yet translated and validated in the Italian language, and to limit the assessment to Englishspeaking patients could cause an important bias. Future research should assess the quality of life and the psychosocial factors of this population.
A high heterogeneity of symptoms and features characterize NF1, and the presence of both inter-and intrafamilial variability is well known [58][59][60]. As in the SNF patients already reported, only in a minority (3 out of 19, 15.5%) of our families with all affected individuals had SNF, confirming the possible co-occurrence of SNF, MNFSR, and the classical NF1 phenotype of the same family. Both clinical features and genetic testing can help in identifying cases who are at risk of SNF and are more likely to benefit from a spinal MRI scan.
The presence of a missense mutation is associated with the occurrence of the SNF subtype, conferring an odds ratio (OR) of 3.34 (CI = 1.33-8.38); when only cases from our clinics were considered, with an OR of 4.14 (CI = 1.76-9.75); and also when SNF cases previously reported in the literature were added. Truncating and frameshift mutations, proportionally more frequent in the classical patients, lead to a loss in protein functionality, while missense mutations, observed more frequently in the SNF patients, could also lead to residual activity and a gain in the function of neurofibromin, which may have an impact on other interactors and pathways, specifically implicated in SNF, yet to be identified.
Here, we also report a higher prevalence of mutations in the SNF patients compared to the classical NF1 patients, in the C-terminal domain of the neurofibromin, containing the nuclear localization signal (NLS) and the syndecan-binding domain (SBR). The NLS domain is necessary for the nuclear localization of neurofibromin, while the function of SBR is to bind syndecans. The interaction between neurofibromin and syndecan is important for cell differentiation and proliferation, and for synaptic plasticity [61]. Functional studies are needed to confirm the possible role of C-terminal NF1 mutations in the development of the spinal form of the disease.
The mutation rate of the NF1 gene is one of the highest known among human genes. The occurrence of germline double NF1 gene mutations in the same subject is a very rare phenomenon when it affects both the NF1 alleles. In fact, the mutation of both copies of the NF1 gene is generally a lethal condition. We here report two unrelated probands with in trans double NF1 mutations. In family 1, the proband 1136 showed a more severe phenotype with respect to the relatives carrying one of the two NF1 variants. Precisely, the mutation c.62T > A (pLeu21His), inherited from the father and shared with the brother, is a missense mutation, classified as a pathogenic variant, and it was already reported in [62]. The missense variant c.528T > A (p.Asp176Glu) was inherited by the mother, and it is reported as benign in LOVD [62][63][64]. In fact, the mother, at 55 years of age, had no tumor-related reports in her medical history, and upon physical examination, only two CAL spots, one on the left arm and one on the chest, were observed; no Lisch nodules were detected during the eye examination. An MRI scan with gadolinium showed several small (diameter of less than 1 cm) nodular-enhancing lesions in the laterocervical soft tissues, suggestive of neurofibromas, but no further sign of NF1. Thus, she did not meet the diagnostic criteria for NF1. The p.Asp176Glu substitution, predicted to be damaging by 9/20 predictors, may have a subclinical significance. Accordingly, patient 1136 showed a more severe phenotype than his affected father and brother. Similar to family 1, in family 17, the proband showed a more severe phenotype with respect to the relatives carrying one or two of the NF1 mutations. Precisely, the mother and one of the two brothers harboring only the splicing mutation had, respectively, an MNSFR phenotype and an SNF phenotype without internal or plexiform neurofibromas. The other brother, harboring only the missense variant inherited from his father (referred to as healthy but never clinically evaluated), displayed a cutaneous NF1 form. In addition, in this case, the second variant could have a subclinical effect that could worsen the clinical phenotype of the proband carrying variants on both of the NF1 alleles.
Moreover, in the two here-described cases, only another patient has been reported, by Fauth [65], with double in trans NF1 mutations. The NF1 mutational screening showed two mutations: the missense c3046T > C in exon 18 and a 3 bp deletion c8131-8133delGTT in exon 48 of the NF1 gene. The patient showed a severe phenotype with mild dermal features, paraparesis, spinal neurofibromas, and MPNSTs at the spinal level.
These few described cases suggest that one of the two NF1 variants maintains a residual function. In the future, expression studies should be carried out on SNF patients with two NF1 variants, aimed at determining the possible residual activity of each of the two mutated NF1 alleles.

Conclusions
NF1 clinical manifestations are highly variable and scarcely predictable. However, the delineation of a distinct phenotype is possible when cases are investigated using suitable diagnostic tools. Both spinal MRI and NF1 mutational screening are useful to identify SNF cases often characterized by a few pigmentary manifestations but at high risk of problematic neurofibromas, presenting not only bilateral neurofibromas involving all spinal roots, but also a higher incidence of internal neurofibromas and nerve root swelling. Based on our results, the present clinical guidelines for the management of NF1 cases could be implemented, allowing for an earlier diagnosis, establishing an appropriate radiological surveillance, and significantly improving surgical intervention. Furthermore, new insights into the genetics of spinal NF1 could improve molecular diagnosis and counselling, opening a path for pharmacological studies focusing on the identification of new molecules targeting the pathway(s) specifically affected in this disease.