Risk of Optic Pathway Glioma in Neurofibromatosis Type 1: No Evidence of Genotype–Phenotype Correlations in a Large Independent Cohort

The occurrence of optic pathway gliomas (OPGs) in children with neurofibromatosis type 1 (NF1) still raises many questions regarding screening and surveillance because of the lack of robust prognostic factors. Recent studies of an overall cohort of 381 patients have suggested that the genotype may be the main determinant of the development of OPG, with the risk being higher in patients harbouring NF1 mutations in the 5’ tertile and the cysteine/serine-rich domain. In an attempt to confirm this hypothesis, we used strict criteria to select a large independent cohort of 309 NF1 patients with defined constitutional NF1 mutations and appropriate brain images (255 directly enrolled and 54 as a result of a literature search). One hundred and thirty-two patients had OPG and 177 did not. The association of the position (tertiles and functional domains) and type of NF1 mutation with the development of OPG was analysed using the χ2 test and Fisher’s exact probability test; odds ratios (ORs) with 95% confidence intervals were calculated, and Bonferroni’s correction for multiple comparisons was applied; multiple logistic regression was also used to study genotype–phenotype associations further. Our findings show no significant correlation between the site/type of NF1 mutation and the risk of OPG, and thus do not support the hypothesis that certain constitutional mutations provide prognostic information in this regard. In addition, we combined our cohort with a previously described cohort of 381 patients for a total of 690 patients and statistically re-analysed the results. The re-analysis confirmed that there were no correlations between the site (tertile and domain) and the risk of OPG, thus further strengthening our conclusions.

OPG in NF1 patients still raises many questions regarding screening and surveillance [21]. Contrast-enhanced brain magnetic resonance imaging (MRI) is the standard imaging technique used to identify and monitor OPG [2,22], but it is very costly and its use for screening purposes is controversial. The Optic Pathway Task Force has recommended against it because radiological and clinical findings do not match sufficiently, asymptomatic OPGs do not require treatment, and anesthesia is a risk for small children. However, an early OPG diagnosis in young children may improve the outcome [15,21,23].
There is a general consensus regarding the usefulness of an ophthalmological evaluation at least once a year up to 10 years of age, but visual acuity measurements are not always reliable in small children [15,24].
In this context, the challenge is to predict which children may develop OPG and which OPG may reduce visual acuity and/or lead to symptoms and signs.
Tumour localisation in the optic chiasm and/or posterior optic tracts, an early age at the time of tumour onset, and being a female all seem to lead to a worse prognosis, but the evidence is not sufficient to propose different surveillance strategies [15,24].
Over the last few years, many authors have focused on genotype-phenotype correlations in patients with NF1, and the findings of some studies suggest that the genotype may be the main determinant of the development of OPG in NF1 children.
The aim of this collaborative retrospective study was to confirm these correlations by gathering information concerning a large independent cohort of unrelated NF1 patients with defined constitutional NF1 mutations who have undergone brain MRI.
Identifying an association between the risk of developing OPG and specific NF1 mutations is an attractive prospect as it would be very helpful in genetic counselling, as well as in guiding screening and surveillance during childhood. In the long term, it may also be useful in further clarifying the pathogenesis of OPGs and developing targeted treatments.

Study Subjects
The study involved patients from four regional NF1 referral centres: Fondazione IRCCS Istituto Neurologico Carlo Besta and Fondazione IRCCS Ca' Granda-Ospedale Maggiore Policlinico in Milan, Istituto Giannina Gaslini in Genoa, and Spedali Civili in Brescia.
The study inclusion criteria were as follows: -a clinical diagnosis of NF1 based on the 1988 NIH diagnostic criteria [1]; -a defined constitutional NF1 mutation; -brain MRI scans available in the institutional picture archiving and communication system (PACS).
We excluded the following: -the youngest patients in familial cases in order to include patients with a more expressed phenotype and to avoid any confounding effects due to shared polymorphisms in possible modifying genes in family members; -patients in whom the entire NF1 gene and flanking genes were deleted (microdeletion) because it is not possible to study the correlation between the localisation of the mutation and phenotype in such patients; -patients harbouring more than one variant of the NF1 gene because the role of each variant needs to be tested separately; -patients with increased optic nerve tortuosity and nerve or sheath thickening without any clear signs of OPG because of the uncertain significance of these characteristics in terms of the subsequent development of OPG [30]; and -patients aged <10 years without an MRI detected OPG because they are still at risk of developing OPG.
The patients were selected using the clinical data in their electronic medical records. The patients or their parents signed an informed consent form before MRI and genetic testing were carried out, and gave their consent to the use of the anonymous results for retrospective research studies. The investigations were carried out in accordance with the principles laid down in the 2013 revision of the Declaration of Helsinki. This retrospective study is part of an NF1 research line approved by the Fondazione IRCCS Istituto Neurologico Carlo Besta Scientific Board and, in accordance with Italian regulations, does not require specific ethical approval because it only uses anonymous data collected during routine patient care.
The indications for brain MRI were neurological and/or visual signs and symptoms, baseline screening at the time of the first evaluation, and radiological surveillance of OPG or other brain lesions.
The diagnosis of OPG in patients with NF1 is currently based on the pathognomonic characteristics of brain MRI; an OPG biopsy is not usually used because of the high risk of vision loss [31].
The MRI examinations were performed using a 1.5 T or 3 T machine, and concentrated on the chiasmatic and orbital regions (differences in MRI resolution do not interfere with the detection of OPG). Two senior neuroradiologists (L.C. and S.A.) with more than 20 years' experience reviewed the neuroimages and confirmed the diagnosis of OPG on the basis of the pathognomonic characteristics of the brain MRI scans. Although there is no universally accepted radiological definition of NF1-associated OPG, the standard is an MRI scan of the brain and orbits using thin slices through the optic nerves and chiasm, including T1-weighted sequences with and without gadolinium, and T2-weighted sequences. As contrast enhancement may be heterogeneous, T2weighted sequences often define the tumour borders more accurately [31,32]. OPGs may show enlargement of the optic nerve, chiasm, optic tracts, and optic radiations. Gliomas of the optic nerve usually have a tubular/fusiform appearance with an often downward kink, whereas chiasmal gliomas can appear as an enlargement of the chiasm or a suprasellar mass. The radiographical interpretation of OPGs can be complicated by T2-hyperintense lesions (which are frequently detected in the brain of NF1 children) overlapping the borders of the tumour. In addition, OPGs are extra-axial and intra-dural and remain within the nerve sheath, which means they may resemble cerebrospinal fluid in T2-weighted sequences (the "pseudo-CSF" sign) [24,31,32]. Tortuous optic nerves and optic nerve or sheath enlargements are quite frequently observed in MRI scans of NF1 children, but their significance in terms of the subsequent development of OPG is debated [30]. In order to avoid missing or over-diagnosing OPGs, any doubtful cases without clear signs of an OPG (i.e., nerve tortuosity, suspected nerve, or sheath thickening) were excluded.
The patients were divided into two groups: NF1 patients of any age with an MRI confirmed diagnosis of OPG were included in the NF1 OPG group, whereas NF1 patients aged ≥10 years without any optic pathway tumour, as confirmed by MRI, were included in the NF1 non-OPG group (patients who have not developed an OPG by the age of 10 years are extremely unlikely to do so in later life) [21]. A total of 255 patients were selected and included in the study: 92 with OPG and 163 without. The MRI data and ages reported herein are those of the latest clinical evaluation by the referring physicians co-authoring this paper.

Literature Review
We systematically reviewed the recent scientific literature with the aim of increasing the number of cases to a minimum of 307 patients, because Anastasaki et al. have calculated that this number is needed to detect an effect size of 0.16 with 80% power using a chi-squared test to detect differences in proportions with an alpha level of 0.05 [27].
We looked for NF1 patients who had not been described in previous studies of genotypephenotype correlations in relation to the development of OPG [27][28][29]. The studies were identified by searching the Pubmed (2017-present) and Google scholar (2017-present) medical databases using the following terms: NF1, neurofibromatosis, and genotype-phenotype or genotype or phenotype.
We only selected patients fulfilling the inclusion and exclusion criteria described above. In particular, we excluded all reports without enough information to allow us to assign patients to the OPG or non-OPG group. In order to be included in the non-OPG group, the patients had to have undergone negative MRI after the age of 10 years. The OPG group only included patients with a reported neuroradiological diagnosis of OPG; we did not have access to the radiological images, but assumed that the reported data were correct. We also excluded known familial cases (we checked whether there was consanguineity in patients sharing the same mutation). The only patients with the same mutation were patient number 144 in the series described by Tsipi et al. [33] and patient number 10 in the series described by Ulusal et al. [34] and, as these patients belonged to different cohorts and their mutation is known to be frequent, there is little likelihood of consanguineity. The two patients harbouring the same NF1 mutation and retrieved from the paper by Trevisson et al. belonged to unrelated families [35].

Molecular Testing and Analysis of NF1 Mutations
Blood samples from 238 of the 255 patients attending the study centres were originally sent to the laboratory of the Molecular Neuro-oncology Unit of Fondazione IRCCS Istituto Neurologico Carlo Besta for molecular NF1 genetic testing in order to establish or confirm the diagnosis of NF1.
Constitutional intragenic NF1 mutations were determined by means of standard techniques using gDNA and, more recently, cDNA; that is, Sanger sequencing and NGS to test for point mutations, and MLPA to test for intragenic deletion/duplication [11,38].
Genomic DNA was obtained from blood EDTA samples, and RNA samples were collected in Tempus blood RNA tubes and reverse-transcribed into cDNA. NGS was carried out using the Ion Sequencing Kit v2.0. Sanger sequencing was used for the oldest samples, for the positive amplicons and the cDNA of negative amplicons obtained by means of NGS Ion Torrent sequencing. MLPA was carried out using NF1 MLPA salsa P081 and P082. The variants were defined in accordance with the reference sequence NM_000267.3 and human reference genome hg19.
The mutations were classified as nonsense (NS), frameshift (FS), missense (MS), inframe (ID), or splicing (SS) mutations; or large deletion (LD) consisting of at least an entire exon. Care was taken when interpreting the effects of the mutations on proteins as some putative MS mutations may also cause splicing abnormalities [10]; the exonic variations with effects on splicing were included in the splicing group. The novel mutations identified in this study were deposited with the LOVD database (http://www.LOVD.nl/NF1), and are described in accordance with HGSV recommendations.

Statistical Analysis
The χ2 test and Fisher's exact probability test were used to compare the frequencies of the independent variables (the locations of mutations by tertile and domain, and the type of mutations) between the NF1 OPG and the NF1 non-OPG group. Odds ratios (ORs) and their 95% confidence intervals (CIs) were calculated. Genotype-phenotype associations were studied using multiple logistic regression. Bonferroni's method was used to correct p-values for multiple testing. p-values of <0.05 were considered statistically significant. The descriptive, frequency, and comparative statistical analyses were carried out using SPSS 22.0 software.

Results
A total of 309 NF1 patients were selected on the basis of the inclusion and exclusion criteria and included in the study; 255 were recruited from the authors' institutions (92 with OPG and 163 without OPG) and 54 were retrieved from the recent literature (40 with OPG and 14 without OPG) [33][34][35][36][37][38][39][40][41]. One hundred and thirty-two patients were included in the NF1 OPG group (61 females, 57 males; the sex of 14 patients retrieved from the literature was not reported) and 177 in the NF1 non-OPG group (91 females, 76 males; 10 unknown). Table 1 shows their clinical details (age, sex, OPG diagnosis); the molecular details (DNA, RNA, protein change); and the classification of the variants by type, tertile, and domain.

Distribution of NF1 Mutations by Domain and the Risk of Developing OPG
The analysis was extended to all of the NF1 domains in order to evaluate whether the risk of developing OPG was associated with mutations in the CSRD and HLR, as previously reported. There were no statistically significant differences between the OPG and the non-OPG group. In particular, there were 20 mutations (15.2%) in the CSRD in the OPG group, and 30 (16.9%) in the non-OPG group (p = 0.67); and 25 mutations (18.9%) in the HLR domain in the OPG group, and 29 (16.4%) in the non-OPG group (p = 0.55) ( Table 3).
The proportion of nonsense mutations was higher in the OPG group than in the non-OPG group (OR 1.77; CI = 1.07-2.93). However, when the p-value was adjusted for multiple comparisons, the difference was not statistically significant (p = 0.15) ( Table 4).

Discussion
The phenotypic expression of NF1 varies widely from mild cutaneous manifestations to serious complications, and is largely unpredictable because of the lack of robust prognostic risk factors. This unpredictability is one of the most difficult aspects of disease management for both patients and clinicians as it makes the genetic counselling and screening and surveillance of NF1 patients extremely challenging [13,14,48].
In this context, any suggestion of a possible genotype-phenotype correlation arouses considerable interest as it might aid clinical care by offering an opportunity to better inform patients about what to expect in the future, and allowing personalised screening and monitoring based on changes in the NF1 gene. NF1 genotype-phenotype correlation studies began to be developed after the identification of the NF1 gene in 1990 [49,50], but they were limited by the difficulty of detecting NF1 mutations until the recent development and spread of new genetic diagnostic investigational techniques gave them renewed impulse [51].
However, identifying specific genotype-phenotype correlations is still very challenging because of the variability of the clinical presentation of the disease, the age-dependence of most of its features, the wide range of NF1 gene variants (often considered private variations), and the timing and number of second hits in specific cells [52]. Furthermore, given the phenotypic variability even among patients with familial NF1 who share the same germline mutation, it has been suggested that there are likely to be other influences such as modifier genes, as well as epigenetic and environmental factors, although their contribution to determining the phenotype is still unknown and may vary depending on disease traits [25].
Despite all of these limitations, at least four genotype-phenotype correlations in NF1 have recently emerged, and others are still being studied.
Firstly, patients with large deletions of the NF1 gene region and flanking genes (NF1 microdeletions) tend to present a more severe phenotype in terms of cognitive impairment and/or learning disabilities, facial dysmorphisms and cardiovascular malformations, and are at increased risk of malignant peripheral nerve sheath tumours (MPNSTs) [53].
Secondly, two specific intragenic mutations (a 3-bp in-frame deletion NF1 c.2970_2972 del p.Met992del, and missense mutations affecting p.Arg1809) have been associated with distinct, but partially concordant mild clinical phenotypes, with the first being characterised by the absence of cutaneous neurofibromas and other serious complications [51,54], and the second by the absence of cutaneous or plexiform neurofibromas with Noonan-like features [55,56].
Finally, it has recently been found that patients with missense mutations involving one of the codons 844-848 within the CSRD have a more severe phenotype and are at a higher risk of developing MPNSTs, OPGs, and malignant neoplasms [52].
In addition to these four genotype-phenotype correlations, the findings of recent studies suggest that the type and the position of NF1 mutations may be the main determinant of the risk of developing spinal neurofibromas (patients more likely to have MS or SS mutations) [57], pulmonary stenosis (a significantly higher prevalence of non-truncating NF1 mutations) [58], breast cancers (a higher proportion of both nonsense and missense mutations) [12], and OPGs.
In 2011, Sharif et al. were the first to find a trend toward the clustering of pathogenic changes in the 5′ tertile (exons 1-21) of the NF1 gene in 29 NF1-OPG patients from the United Kingdom [25], and the same trend was found in another cohort of 20 NF1-OPG patients with 12 different mutations described by Ars [59]. In order to determine whether the risk of developing OPG was associated with mutation location, Sharif et al. combined their finding with those of Ars and a larger series of Castle et al. [60] to give a total of 125 NF1 patients (predominantly from the United Kingdom), 36 of whom had OPGs. Statistical analysis showed a significant distribution of mutations in the 5' tertile of the NF1 gene in NF1 patients with OPGs when compared with NF1 patients without OPGs.
The clustering of mutations in the 5' tertile of NF1 gene in patients with OPG was subsequently supported by Bolcekova et al., who observed a similar NF1 mutation pattern in their series of 25 Slovakian NF1-OPG patients in comparison with 27 NF1 controls [26]. They found that 71% of the mutations in OPG patients were located in the first tertile as against 29% in the non-OPG group. They believed that the CSRD, which spans exons 11-17 in the first tertile of the gene, may be as important a functional domain as the RAS-GAP domain, and may play an important role in the development of OPG.
These associations were not confirmed by a subsequent study by Hutter et al. [61], who used stricter inclusion criteria to select 77 NF1 patients from Germany and Canada (thus reducing regional genetic differences), and found a point mutation in 37 OPG patients and 32 controls. They observed a greater proportion of mutations in the 5' region in both the OPG (17/37, 46%) and non-OPG patients (15/32, 47%). In order to avoid misinterpreting data from small cohorts, they combined their series with that of Bolcekova et al., thus increasing the cohort to 129 NF1 patients (66 OPG and 63 non-OPG), but found no significant correlation between mutations in the 5' tertile of NF1 and the development of OPG.
In 2017, Anastasaki et al. analysed 37 NF1 patients at Washington University (14 with OPG and 13 without) in order to ascertain the relationship between the type and the location of the germline NF1 mutation and the presence of OPG, and did not find any statistically significant correlation [27]. However, power calculations revealed that a sample size of 307 patients is required to determine the predictive value of the position or type of NF1 gene mutations, and so they combined their data set with those of the four previous studies by Ars, Sharif, Bolcekova, and Hutter to reach a sample size of 310 patients. It was found that children with OPG were more likely to harbour 5' gene mutations. In addition, the association was stronger in subjects with 5' tertile truncating or nonsense mutations, thus suggesting that the NF1 mutation may be a predictive factor [27].
However, using stricter inclusion and exclusion criteria, Xu et al. found no significant statistical association between mutations clustering in the 5' tertile and the risk of developing OPG when they analysed 215 NF1 patients (5 from Sun Yatsen University in China and 210 selected from the literature) [28]. Nevertheless, it is interesting to note that patients with mutations involving the cysteine/serine-rich domain of NF1 were at a higher risk of developing OPG than those with mutations in other regions, whereas those with mutations in the HEAT-like repeat region were at a lower risk. As different mutation types have different effects on protein structure and function, the authors also compared the distribution of the types in the two regions, and found that nonsense mutations were more frequent in the CSRD in the OPG group and splicing site mutations were more frequent in the HLR in the non-OPG group, but these between-region differences were not significant [28].
As recently as 2019, Anastasaki et al. [29] re-examined their previously reported data concerning patients from their institution and retrieved from the literature (a total of 310 patients) and the dataset of Xu et al. (215 patients, 144 of whom were the same as those in the dataset of Anastasaki et al.), and found that mutation clustering in the 5' tertile was significantly different between the patients with and without OPG when their dataset was considered alone (310 patients) or combined with that of Xu (making a total of 381 patients), but not when Xu's dataset was considered alone. In addition, an analysis of their own cohort of 310 patients confirmed the findings of Xu et al., that mutations in the CSRD were significantly positively associated with OPG, whereas mutations in the HLR were negatively associated with OPG, and this difference persisted when the combined dataset of 381 patients (127 with OPG and 254 without) was analysed.
This succession of papers with partially concordant findings about a correlation between OPG and NF1 mutations located in the first tertile of the gene or in the CSRD prompted us to contribute to the current debate concerning NF1 genotypes and the risk of developing OPG by validating these findings in a large independent series of patients.
Unfortunately, and unexpectedly, our data do not confirm the association reported by Xu et al. and Anastasaki et al., or the original observations of Sharif et al., as we did not find any correlation between the development of OPG and the site (tertile and domain) or type of mutation [25,[27][28][29].
The difference may be because of what we consider to be the most important limitation of the previous studies: patient enrolment. The studies by Sharif, Hutter, and Bolcekova were limited by their small sample sizes; in addition, some included familial cases and children under 10 years of age in the non-OPG group (even though the former can clearly carry the same modifier genes and the latter are still at risk of developing an OPG), and patients with whole gene deletions [25,26]. Furthermore, with the aim of reaching a large series with statistical power, Xu and Anastasaki progressively added a limited number of patients to the original cohorts of Sharif, Hutter, and Bolcekova [27][28][29] and, although they filtered the patients retrieved from the literature using stricter inclusion and exclusion criteria before the statistical re-analyses, a bias was generated because the same patients were repeatedly analysed in the various studies.
The strengths of the present study include the fact that it involved a large and completely independent sample of NF1 patients selected on the basis of strict inclusion and exclusion criteria and did not include family cases, patients harbouring a microdeletion, or patients aged <10 years in the non-OPG group. In addition, our series mainly consists of patients who were genetically analysed by a single laboratory and selected using precise radiological criteria for a diagnosis of OPG. Cases with increased optic nerve tortuosity and nerve or sheath thickening were excluded because it is unclear whether these characteristics are significant in terms of the subsequent development of OPG [30].
In order to achieve the statistical power described by Anastasaki et al., we searched the literature for patients meeting the same inclusion/exclusion criteria as those used to select our institutional cohort. We decided not to select any patients previously included in the studies of Anastasaki, Xu, and the others mentioned above in order to ensure a completely independent cohort.
However, to provide additional data, we combined our cohort with the most recent cohort described by Anastasaki [29] to create a large sample of 690 patients, and performed the same statistical analysis. The results confirmed the absence of a correlation between the position of the NF1 mutation (by tertile and functional domain) and the risk of OPG, thus further strengthening our conclusions (Tables S1 and S2).
It is worth noting that some of the patients analysed in our study (those in our institutional cohort and those retrieved from the literature) had different ethnic origins. We think that is another strength of the study as it limits the potential confounding effect of polymorphisms shared by the same ethnic group.
However, our study also has some limitations. The first is that the 30.3% of our OPG patients were retrieved from the literature and we cannot be sure that the same radiological criteria were used in their diagnosis.
In addition, when selecting the patients in the non-OPG group, unlike Hutter et al. [61], we did not consider non-optic gliomas among the exclusion criteria. Hutter et al. probably assumed that gliomas have the same biological basis in NF1 patients regardless of their location, but we believe that the biological basis of optic and non-optic gliomas is probably different because they occur at different patient ages and are histologically different (non-optic gliomas are more aggressive and OPGs are more stable). Moreover, adult patients are at risk of developing non-optic gliomas and, if we had used the same exclusion criterion as Hutter et al., we would only have been able to include patients at low risk; that is, very old patients.
Lastly, it would be very interesting to examine the correlations between site/type of NF1 mutations and other factors such as tumour size; position in the optic pathway; and, above all, symptomatology, because this could help to clarify why some OPGs behave aggressively and others have a stable and benign course over time; however, this would require a much larger sample of NF1 patients with OPG.

Conclusions
Knowing the type and location of NF1 mutations does not give patients or clinicians any valuable prognostic information regarding the development of OPG.
In order to identify other genotype-phenotype correlations, it is necessary to select extremely large and homogeneous groups of patients on the basis of their genotype or phenotype. Previous examples of recognised genotype-phenotype correlations in NF1 have involved patients with a single recurrent mutation or the deletion of the entire NF1 gene, who make much more homogeneous groups.
Our findings seem to support the hypothesis that other modifying influences such as modifier genes, as well as epigenetic and environmental factors, are likely to be involved in determining the NF1 phenotype.
Supplementary Materials: The following are available online at www.mdpi.com/xxx/s1, Table S1: Mutations in NF1 gene tertiles and the risk of developing OPG in the combined cohort of our and Anastasaki's patients; Table S2: Mutations in different NF1 gene regions and the risk of developing OPG in the combined cohort of our and Anastasaki's patients.