METHODS

The constitutional NF1 gene mutations were identified in one laboratory using a battery of different mutation detection techniques: single strand conformational polymorphism analysis, heteroduplex analysis, the protein truncation test, dideoxy fingerprinting, denaturing high performance liquid chromatography, and direct DNA sequencing.^16–20 Clinical data were obtained from medical records of examinations within the previous 2 years, or if recent medical data were unavailable, the patients were re-examined. The examining physician was blinded to the patient’s mutation.

There were 113 NF1 patients with identified NF1 gene mutations. Three of these patients were excluded from the statistical analysis of genotype–phenotype correlations: two patients who did not meet the US National Institutes of Health (NIH) diagnostic criteria for NF1²¹ and a single patient with an identified NF1 large gene deletion. The common clinical abnormalities were coded as: Lisch nodules (present/absent), number of CAL macules (0, 1–6, >6), axillary freckling (present/absent), number of dermal and nodular neurofibromas (0, 1–10, 11–100, 101–500, >500), macrocephaly (present/absent), short stature (present/absent), the ratio of head circumference (cm) to height (m) (OFC/height ratio), pectus excavatum (present/absent), learning difficulties (present/absent), plexiform neurofibromas (present/absent), scoliosis (present/absent), and spinal neurofibromas (present/absent). Percentiles for the OFC/height ratio were calculated using charts.^22,23 The prevalence of other clinical abnormalities (glioma, sarcoma, phaeochromocytoma, gastrointestinal neurofibromas, pulmonary stenosis, renal artery stenosis, delayed puberty, aqueduct stenosis, Watson syndrome, Noonan syndrome, epilepsy, bone dysplasia) was too low (5% or less) for meaningful statistical analysis.

Logistic regression analysis was used to calculate relative risks (RR) and 95% confidence intervals (CI) for binary outcomes. Linear regression analysis was used for continuous outcomes. Each regression model had the covariates of the type of constitutional NF1 mutation and age at examination (as a continuous variable). The mutation covariate was coded as several binary variables (indicators of missense and splice site mutations, each compared to nonsense or frameshift mutations). Age at examination was included as a covariate because the penetrance of many NF1 disease features increases with age.^24,25 A logarithmic transformation was used for age at examination, which ranged from 2 to 70 years. There is a potential bias toward lower age at examination in non-probands relative to probands, but proband status did not contribute significantly to any of the models after including age at examination as a covariate, so probands and non-probands were combined in subsequent analyses. There were eight families with two affected relatives and two families with three affected relatives.

We evaluated genotype–phenotype correlations for each common clinical abnormality individually and for three groups of disease features:²⁶ (a) CAL macules, axillary freckling, and Lisch nodules, (b) dermal, nodular, and plexiform neurofibromas, and (c) macrocephaly, optic glioma, and other neoplasms. A patient was categorised as belonging to a group if at least two disease features in the group were present.

RESULTS

The identified constitutional NF1 gene lesions ranged from single base pair substitutions to gross deletions, but the microlesions appeared to be uniformly distributed across the gene (http://www.uwcm.ac.uk/study/medicine/medical_genetics/research/nf1/supplementary_data.html). The prevalence of common clinical abnormalities in this study was generally similar to those of other published studies (table 1). Delayed and precocious puberty occur in NF1,^27,28 but the exact prevalence is unknown. In this study, 6% of males and 5% of females had delayed puberty at16 years of age. One patient had an empty sella turcica without abnormal pituitary function and a second patient with very short stature had isolated growth hormone deficiency with otherwise normal pituitary function. There were no patients with precocious puberty.

Genotype–phenotype correlations

Eighty four patients had ocular examinations. As expected, the relative risk (RR) of Lisch nodules increased with increasing age at examination (table 2). When compared with people with nonsense or frameshift mutations, people with missense mutations had an RR of Lisch nodules that was lower than unity but of borderline statistical significance (RR=0.26, 95% CI 0.07 to 1.04, p=0.06, Cox and Snell R²=0.10). There were no significant genotype–phenotype correlations for any other individual abnormalities or for any of the three groups of disease features (data not shown).

Atypical cases

There were two patients who did not fulfil the NIH diagnostic criteria for NF1 and who were not included in the statistical analysis of genotype–phenotype correlations. One was a 10 year old child who, in common with his mother, had CAL macules but no other stigmata suggestive of NF1. The NF1 mutation in the family was a missense substitution (R1809L). The second patient was a 27 year old woman with no family history of NF1 and no CAL spots, freckling, or Lisch nodules. She had a small number (<10) of cutaneous and nodular neurofibromas, confirmed on biopsy. In addition, she had a vestibular schwannoma and a schwannoma of the posterior fossa involving the vagus nerve. A cervical cord ependymoma was resected, leaving the patient paraplegic. Although initially considered as a possible case of NF type 2, a single base pair insertion in exon 37 (6792 ins A) of the NF1 gene was subsequently identified in this patient.

There were three patients with NFNS. The first NFNS patient was a 6 year old boy from a large Ashkenazi kindred with NF1. He had only four CAL macules on examination but had been developing cutaneous neurofibromas since the age of 2 years, and had intracranial tumours and hydrocephalus secondary to aqueduct stenosis, which had required shunting. There were no Lisch nodules and he was not macrocephalic. He had facial dysmorphism suggestive of Noonan syndrome but had an inactivating splice site mutation in the NF1 gene (IVS 29+1 G→C).

The second NFNS patient was a 20 year old man who fulfilled the NIH diagnostic criteria for NF1, and had a pectus excavatum, learning difficulties, and mild hydrocephalus. He was also thought to have aqueduct stenosis causing hydrocephalus, which did not require shunting at the time of examination. The NF1 mutation was a microdeletion (2133 del CC). There was no evident cardiac abnormality in these two patients.

The third NFNS patient was a 10 year old boy who was born prematurely and had facial dysmorphism, widely spaced nipples, and an atrial–septal defect. His NF1 mutation was a microinsertion (4093 ins TG).

There were three patients with Watson syndrome. Two had mild to moderate pulmonary stenosis that was documented on echocardiography. Neither had a history of significant learning difficulties, but they did not complete their schooling. One patient had a novel 3 bp deletion (AAT) within exon 17 of the NF1 gene leading to a loss of ATG (methionine) in exon 17,²⁹ and the other had a paternally inherited nonsense mutation in exon 29 (S1745X). This patient’s father also had mild pulmonary stenosis with stigmata of NF1. The third patient was a 49 year old man with significant learning difficulties, behavioural problems associated with epilepsy, and pulmonary stenosis. He had an intragenic NF1 gene deletion of >20 bp.

Another atypical patient was a 31 year old woman who had stigmata of NF1 almost exclusively in the cranial region. There were four CAL macules on the head and neck only, and multiple large histologically confirmed neurofibromas on the scalp and behind the ears. There was also macrocephaly but no other manifestations of NF1 in the rest of the body. In addition, the proband was diagnosed with Van der Woude syndrome at the age of 6 months, with a cleft palate and lip pits, which were surgically treated. The NF1 mutation was a microdeletion in exon 22 (3731 del T).

DISCUSSION

NF1 patients with missense mutations had an RR of Lisch nodules that was less than 1, and of borderline statistical significance when compared to NF1 patients with nonsense or frameshift mutations and adjusted for age at examination. This exploratory finding merits investigation in other NF1 patient populations, although it is difficult to evaluate the pathogenicity of missense mutations without functional studies. In this study, multiple statistical comparisons were made for the different clinical abnormalities and different mutation types, but as these analyses were exploratory, not confirmatory, we chose to test associations at the α=0.05 level.

From the clinical viewpoint, our findings are notable for two reasons. The OFC/height ratio was at or above the 95th percentile in 97% of cases, but the prevalence of macrocephaly was 40% and the prevalence of short stature was 50%. This suggests that an increase in the OFC/height ratio is a useful clinical indicator of NF1, even in the absence of absolute macrocephaly or short stature. The finding of delayed puberty in 6% of males and 5% of females at 16 years of age confirms that this is an NF1 related phenomenon.

We did not find genotype–phenotype correlations for other common clinical features of NF1, similar to the findings of most other studies of genotype–phenotype correlations in NF1.^4,30–33

The only exception is the correlation of large NF1 gene deletions (up to 1.5 Mb genomic DNA) with an earlier age of onset of cutaneous neurofibromas, learning disability, dysmorphic features, and developmental delay.^18,34–38 However, not all patients with a gross NF1 gene deletion have this NF1 phenotype,^18,36 and flanking DNA sequence (or the lack of it) may affect the phenotype. Easton et al.³⁰ modelled the genetic contributions to eight traits in 175 people with NF1 (CAL macules, neurofibromas, head circumference, plexiform neurofibromas, optic gliomas, scoliosis, epilepsy, and remedial education). There were no significant correlations between the traits, indicating that the phenotypic expression of NF1 was largely determined by trait specific loci unlinked to the NF1 gene.

We were not able to evaluate genotype–phenotype correlations for some of the most clinically significant disease features of NF1, such as the various types of cancer. The RR of these complications is very high compared with the general population, but the prevalence in NF1 is relatively low, and there were too few patients with these complications in this study to support statistical analysis.

Identical NF1 gene mutations can occur in unrelated patients with very different phenotypes.^{11,19,38–41} Owing to the absence of strong genotype–phenotype correlations,³⁰NF1 mutation analysis has very limited predictive utility for specific sequelae. NF1 mutation analysis has greater diagnostic utility because mutation detection can confirm the aetiology of the disease in the relatively few families and individuals in whom the clinical phenotype does not fulfil the NIH diagnostic criteria. A provisional diagnosis of NF1 improves clinical management by triggering the initiation of routine blood pressure monitoring in patients from an early age and alerting medical care providers to the possibility of clinical complications. A firm diagnosis allows counselling regarding mode of inheritance, recurrence risk, and potentially, prenatal diagnosis.

Allelic heterogeneity is a possible cause of some of the multiple phenotypes in NF1, particularly Watson syndrome (multiple CAL macules, pulmonic stenosis, and dull intelligence),¹³ NFNS,¹⁵ and familial spinal neurofibromatosis.⁴² To date, however there is no evidence to support allelic heterogeneity as a cause for any of these three atypical phenotypes, nor did we find such evidence in the present study, as detailed below.

Allelic heterogeneity is not likely to be responsible for Watson syndrome because the three patients in the present study with Watson syndrome had NF1 mutations of differing types and locations. Divergent NF1 gene mutations have been identified: an 80 kb deletion in a Watson syndrome patient,¹⁶ and a 42 bp duplication that segregated in a family with a phenotype suggestive of NFNS.⁴³

About 12–13% of NF1 patients who are specifically examined for Noonan syndrome have a clinical phenotype that is similar to Noonan syndrome,^14,15 but Carey,⁴⁴ and Carey and Viskochil³² reviewed these patients, and concluded that they did not have typical Noonan syndrome, as features of Noonan syndrome without cardiovascular malformation have been observed in many NF1 patients. The three patients in the present study with features of Noonan syndrome had NF1 mutations of differing types and locations. As two of these patients had hydrocephalus associated with features of aqueduct stenosis, the dysmorphic features of Noonan syndrome could be a secondary effect, and NFNS may not be a distinct clinical entity. There is clinical overlap between Noonan syndrome and NF1, but the Noonan syndrome gene (PTPN11) has been mapped to 12q24.1 and there is no suggestion of locus heterogeneity for NF1.

Spinal neurofibromas have been reported in up to 38% of NF1 patients, but only 5% of these patients have spinal neurofibromatosis.⁴² All patients in our study who had familial spinal neurofibromatosis also fulfilled the NIH diagnostic criteria for NF1, and had Lisch nodules and dermal neurofibromas, as has been previously reported.^42,45 Ars et al⁴² described a frameshift mutation in exon 46 (8042 ins A) that cosegregated with spinal neurofibromatosis. In another study, two unrelated NF1 patients with spinal neurofibromatosis had an NF1 missense mutation in exon 33 (L2067P) and an NF1 splice site mutation (IVS31-5 A→G).⁴⁶ A recurrent NF1 splice site mutation (IVS19b-3 C→G) was identified in a patient with spinal neurofibromatosis.⁴⁷ In the present study, patients with spinal neurofibromatosis also had different types of mutations. These findings indicate that specific NF1 gene mutations are not associated with spinal neurofibromatosis.

One NF1 patient in this study also had Van der Woude syndrome, a dominantly inherited disorder with locus heterogeneity. The majority of Van der Woude patients have linkage to the 1q32–41 locus,⁴⁸ but there is a second locus on 1p34.⁴⁹ A modifier locus exists on 17p11.2,⁵⁰ but there is no evidence to suggest an association with the NF1 locus at 17q11.2. This patient may be segregating two distinct and unrelated autosomal dominant conditions.

The marked clinical variability between multiple affected relatives in NF1 families also could be due to the nature, timing, or location of the “second hit” mutations at the NF1 locus in dermal and plexiform neurofibromas,^51–55 MPSNTs,^56–58 pheochromocytomas,⁵⁹ pilocytic astrocytomas,⁶⁰ and juvenile chronic myelogenous leukaemia cells.⁶¹ Variable expressivity has also been observed in other tumour prone genetic conditions such as neurofibromatosis 2, tuberous sclerosis 1 and 2, and familial adenomatous polyposis.^62,63

Somatic mosaicism, which has been documented in over 60 monogenic disorders,⁶⁴ is another potential cause of inter-individual phenotypic variation. Somatic mosaicism can occur sufficiently early in embryonic life to involve both somatic and germline cells, and such individuals may also be at risk of having affected children.⁶⁵ Somatic mosaicism in NF1 is thought to be the cause of segmental neurofibromatosis.⁶⁶ There is inter-individual variation in mitotic recombination,⁶⁷ and the genes that control this process may be an important cause of clinical variability within NF1 families. Serra et al⁶⁸ reported that mitotic recombination was the causal mechanism for the loss of heterozygosity of an NF1 constitutional mutation.

Our study gives further support to the probable importance of modifying loci, as suggested by Easton et al³⁰ This hypothesis is supported by the observation of many investigators that identical NF1 gene mutations often give rise to very different phenotypes in unrelated patients. GDNF is a candidate modifier gene in NF1,⁶⁹ and preliminary work in a mouse model of NF1 has also suggested the existence of modifying genes.⁷⁰ The identification of modifying genes in humans will be difficult and will require large cohorts of patients identified through multicentre collaborations, but the study of modifying genes may eventually allow more accurate prediction of specific clinical features and complications of NF1 that could possibly lead to new therapeutic approaches.