Abstract

Neurofibromatosis type 1 (NF1) is a common autosomal dominant disorder which displays considerable inter- and intra-familial variability in phenotypic expression. To evaluate the genetic component of variable expressivity in NF1, we examined the phenotypic correlations between affected relatives in 750 NF1 patients from 275 multiplex families collected through the NF-France Network. Twelve NF1-related clinical features, including five quantitative traits (number of café-au-lait spots of small size and of large size, and number of cutaneous, subcutaneous and plexiform neurofibromas) and seven binary ones, were scored. All clinical features studied, with the exception of neoplasms, showed significant familial aggregation after adjusting for age and sex. For most of them, patterns of familial correlations indicated a strong genetic component with no apparent influence of the constitutional NF1 mutation. Heritability estimates of the five quantitative traits ranged from 0.26 to 0.62. Moreover, we investigated for the first time the role of the normal NF1 allele in the variable expression of NF1 through a family-based association study. Nine tag SNPs in NF1 were genotyped in 1132 individuals from 313 NF1 families. No significant deviations of transmission of any of the NF1 variants to affected offspring was found for any of the 12 clinical features examined, based on single marker or haplotype analysis. Taken together, our results provided evidence that genetic modifiers, unlinked to the NF1 locus, contribute to the variable expressivity of the disease.

INTRODUCTION

Neurofibromatosis type 1 (NF1; MIM 162200), a common autosomal disorder with an estimated birth incidence of 1 in 3500 (1), is caused by dominant loss-of-function mutations of NF1, a tumour suppressor gene located at 17q11.2 and spanning over 280 kb of genomic DNA.

Although NF1 is a simply determined Mendelian disorder with complete penetrance, it is characterized by highly variable clinical expressivity with marked inter- and intra-familial variation in both the number of major features and the occurrence of complications (2). The major clinical features of the disease include multiple café-au-lait (CAL) spots, axillary freckling, Lisch nodules and neurofibromas, the occurrence and the number of which vary greatly from one patient to another (3). In addition, about one-third of patients develop one or more complications that can affect almost any organ and which seem to occur in a very unpredictable way. The mechanisms underlying NF1 clinical variability remain poorly understood, probably because of the involvement of complex pathophysiology and multiple factors.

The allelic heterogeneity of the constitutional NF1 mutation may be one of the factors explaining the disease phenotypic variability. Almost half of all NF1 cases are a result of sporadic mutations, and a huge number of different pathogenic mutations have been reported (4–6). Among them, 5–10% are large 17q11 deletions encompassing the entire NF1 locus and neighboring genes. These recurrent deletions have been commonly associated with more severe and atypical manifestations, the so-called ‘NF1 microdeletion syndrome’ (7,8). On the other hand, for patients with intragenic NF1 mutations (>90% of cases), no clear-cut allele–phenotype correlations have been established so far (9,10), with the exception of a 3-bp inframe deletion (c.2970-2972 delAAT) in exon 17 of the NF1 gene which has been associated with a particular clinical phenotype characterized by the absence of cutaneous neurofibromas (11). This recent result suggests that other NF1 genotype–phenotype correlations may exist and are still to be found. Nonetheless, since different affected members of the same NF1 family often have quite different disease phenotypes, it is clear that variation in the mutant NF1 allele itself does not account for all of the disease variability.

Up to now, only two studies have tried to evaluate the inherited component of variable expression in NF1 (12,13). By examining the phenotypic correlations between different types of relatives, both studies provided evidence for a strong genetic component and suggested the involvement of unlinked modifying genes, and perhaps also of the normal NF1 allele, in the variable expression of the disease. However, the study of Easton et al. (12) examined only a limited number of patients: 175 affected individuals from 48 extended families. The study of Szudek et al. (13), though dealing with a larger number of patients (904 affected individuals from 373 families), could not investigate many distant relatives due to the few number of extended families available. Moreover, the lack of accurate phenotypic information in this study did not enable the analysis of the main manifestations of NF1, such as CAL spots or dermal neurofibromas, as quantitative variables, and these features were instead treated as binary traits. Hence, the knowledge of the nature and extent of the genetic component of variable expressivity in NF1 is still incomplete and requires further in-depth analysis with the input of new data. Moreover, no study to date has formally tested the possible influence of the normal NF1 allele on the phenotypic expression of the disease.

The aim of the present study was to provide further insights into the contribution of genetic factors to the variable expressivity of NF1 by using the vast collection of well-phenotyped families with typical NF1 from the genotype–phenotype database mainly collected through the NF-France Network, a National Network of professionals devoted to neurofibromatoses. Variance components analysis based on maximum likelihood procedures was used to estimate the proportion of phenotypic variation that is attributable to genetic effects and patterns of familial correlations were examined for 12 clinical features of NF1. Moreover, we investigated the possibility that another variant in the NF1 gene, different from the primary mutation, may influence the disease expression variability. A family-based association study analyzing all common variants in the NF1 gene was used to test this hypothesis.

RESULTS

Heritability and familial correlations

We studied 750 NF1 patients from 275 French families of the PHRC NF1 database with at least two affected members. Twelve NF1-related clinical features, including five quantitative traits (number of CAL spots of small size and of large size, and number of cutaneous, subcutaneous and plexiform neurofibromas) and seven binary ones, were examined. The prevalence of each of the 12 clinical features of NF1 in the total study sample are shown in Table 1, together with the prevalences by age group. The prevalence of many NF1 features, notably Lisch nodules, blue-red macules and the number of neurofibromas, increases markedly with age, whereas the number of CAL spots of small size declines. Multiple regression analysis revealed a highly significant effect of age, measured as a quantitative variable, for CAL spots of small size, cutaneous neurofibromas, subcutaneous neurofibromas, plexiform neurofibromas, blue-red macules and Lisch nodules (P < 0.0001); a significant effect for skin-fold freckling (P = 0.0013), scoliosis (P = 0.0028) and facial dysmorphism (P = 0.0231); and a non-significant effect for CAL spots of large size, neoplasms and learning disabilities (P > 0.05). No significant effect of sex was observed on any trait (P > 0.05). Age and sex were included as covariates in all subsequent analyses to control for potential confounding.

Table 1.

Prevalences of NF1 clinical features by age category

Clinical feature No. and percentage of patients by age category (years)
 
 
 <10 10–19 20–39 ≥40 Total 
CAL spots of small size 
 <5 11 (8%) 21 (13%) 52 (21%) 76 (36%) 160 (21%) 
 5–9 38 (28%) 31 (20%) 54 (22%) 60 (28%) 183 (24%) 
 10–14 34 (25%) 33 (21%) 53 (22%) 25 (12%) 145 (19%) 
 15–19 19 (14%) 15 (10%) 25 (10%) 16 (7%) 75 (10%) 
 20–24 12 (9%) 19 (12%) 13 (5%) 14 (7%) 58 (8%) 
 ≥25 21 (16%) 37 (24%) 48 (20%) 23 (11%) 129 (17%) 
CAL spots of large size 
 <5 45 (33%) 33 (21%) 28 (11%) 59 (28%) 165 (22%) 
 5–9 61 (45%) 46 (29%) 110 (45%) 99 (46%) 316 (42%) 
 10–14 17 (13%) 41 (26%) 60 (24%) 35 (16%) 153 (20%) 
 15–19 6 (4%) 22 (14%) 33 (13%) 17 (8%) 78 (10%) 
 20–24 4 (3%) 8 (5%) 9 (4%) 2 (1%) 23 (3%) 
 ≥25 2 (1%) 6 (4%) 5 (2%) 2 (1%) 15 (2%) 
Cutaneous neurofibromas 
 0 126 (93%) 100 (64%) 43 (18%) 15 (7%) 284 (38%) 
 1–9 9 (7%) 44 (28%) 68 (28%) 32 (15%) 153 (20%) 
 10–99 12 (8%) 92 (38%) 80 (37%) 184 (25%) 
 ≥100 42 (17%) 87 (41%) 129 (17%) 
Subcutaneous neurofibromas 
 0 115 (85%) 95 (61%) 92 (38%) 104 (49%) 406 (54%) 
 1–9 19 (14%) 51 (33%) 112 (46%) 70 (33%) 252 (34%) 
 10–99 1 (1%) 8 (5%) 37 (15%) 34 (16%) 80 (11%) 
 ≥100 2 (1%) 4 (2%) 6 (3%) 12 (2%) 
Plexiform neurofibromas 
 0 120 (89%) 112 (72%) 143 (58%) 125 (58%) 500 (67%) 
 1 13 (10%) 31 (20%) 59 (24%) 52 (24%) 155 (21%) 
 ≥2 2 (1%) 13 (8%) 43 (18%) 37 (17%) 95 (13%) 
Skin-fold freckling 102 (76%) 140 (90%) 222 (91%) 192 (90%) 656 (87%) 
Blue-red macules 8 (6%) 13 (8%) 49 (20%) 46 (21%) 116 (15%) 
Lisch nodulesa 20 (24%) 55 (54%) 79 (68%) 77 (69%) 231 (56%) 
Facial dysmorphism 8 (6%) 24 (15%) 12 (5%) 9 (4%) 53 (7%) 
Scoliosis 15 (11%) 66 (42%) 86 (35%) 81 (38%) 248 (33%) 
Learning disabilities 53 (39%) 95 (61%) 119 (49%) 92 (43%) 359 (48%) 
Neoplasms 10 (7%) 19 (12%) 21 (9%) 23 (11%) 73 (10%) 
 No. of individuals examined 135 156 245 214 750 
Clinical feature No. and percentage of patients by age category (years)
 
 
 <10 10–19 20–39 ≥40 Total 
CAL spots of small size 
 <5 11 (8%) 21 (13%) 52 (21%) 76 (36%) 160 (21%) 
 5–9 38 (28%) 31 (20%) 54 (22%) 60 (28%) 183 (24%) 
 10–14 34 (25%) 33 (21%) 53 (22%) 25 (12%) 145 (19%) 
 15–19 19 (14%) 15 (10%) 25 (10%) 16 (7%) 75 (10%) 
 20–24 12 (9%) 19 (12%) 13 (5%) 14 (7%) 58 (8%) 
 ≥25 21 (16%) 37 (24%) 48 (20%) 23 (11%) 129 (17%) 
CAL spots of large size 
 <5 45 (33%) 33 (21%) 28 (11%) 59 (28%) 165 (22%) 
 5–9 61 (45%) 46 (29%) 110 (45%) 99 (46%) 316 (42%) 
 10–14 17 (13%) 41 (26%) 60 (24%) 35 (16%) 153 (20%) 
 15–19 6 (4%) 22 (14%) 33 (13%) 17 (8%) 78 (10%) 
 20–24 4 (3%) 8 (5%) 9 (4%) 2 (1%) 23 (3%) 
 ≥25 2 (1%) 6 (4%) 5 (2%) 2 (1%) 15 (2%) 
Cutaneous neurofibromas 
 0 126 (93%) 100 (64%) 43 (18%) 15 (7%) 284 (38%) 
 1–9 9 (7%) 44 (28%) 68 (28%) 32 (15%) 153 (20%) 
 10–99 12 (8%) 92 (38%) 80 (37%) 184 (25%) 
 ≥100 42 (17%) 87 (41%) 129 (17%) 
Subcutaneous neurofibromas 
 0 115 (85%) 95 (61%) 92 (38%) 104 (49%) 406 (54%) 
 1–9 19 (14%) 51 (33%) 112 (46%) 70 (33%) 252 (34%) 
 10–99 1 (1%) 8 (5%) 37 (15%) 34 (16%) 80 (11%) 
 ≥100 2 (1%) 4 (2%) 6 (3%) 12 (2%) 
Plexiform neurofibromas 
 0 120 (89%) 112 (72%) 143 (58%) 125 (58%) 500 (67%) 
 1 13 (10%) 31 (20%) 59 (24%) 52 (24%) 155 (21%) 
 ≥2 2 (1%) 13 (8%) 43 (18%) 37 (17%) 95 (13%) 
Skin-fold freckling 102 (76%) 140 (90%) 222 (91%) 192 (90%) 656 (87%) 
Blue-red macules 8 (6%) 13 (8%) 49 (20%) 46 (21%) 116 (15%) 
Lisch nodulesa 20 (24%) 55 (54%) 79 (68%) 77 (69%) 231 (56%) 
Facial dysmorphism 8 (6%) 24 (15%) 12 (5%) 9 (4%) 53 (7%) 
Scoliosis 15 (11%) 66 (42%) 86 (35%) 81 (38%) 248 (33%) 
Learning disabilities 53 (39%) 95 (61%) 119 (49%) 92 (43%) 359 (48%) 
Neoplasms 10 (7%) 19 (12%) 21 (9%) 23 (11%) 73 (10%) 
 No. of individuals examined 135 156 245 214 750 

aThere were 337 individuals with missing data for this trait (no examination by slit-lamp); the percentages were calculated by considering only patients with recorded data (n = 413).

The variance components analysis for the five quantitative traits of NF1 is shown in Table 2. All heritability estimates were significantly different from zero and ranged from 0.26 for cutaneous neurofibromas to 0.62 for CAL spots of small size. When we added an additional variance component to the model to estimate the contribution of the NF1 mutation to the total phenotypic variance, the variance component attributable to NF1 germ-line effects was low for CAL of small and large size (0.14 and 0.12, respectively) and it was null for the three types of neurofibromas. It was not significant in all cases (P > 0.05). We performed a similar analysis to investigate the influence of the type, instead of the nature, of the NF1 mutation. NF1 mutations were categorized in four distinct classes: truncating mutations (including non-sense and frameshift mutations, as well as splice mutations leading to frameshift), missense mutations, in-frame deletions and large deletions encompassing the entire NF1 gene. The variance component attributable to the NF1 mutation was then estimated to be zero for the five quantitative traits studied (data not shown).

Table 2.

Heritability and variance components estimates for the five quantitative traits

Clinical feature Heritabilitya (P-value) σ2A σ2G σ2E 
CAL spots of small size 0.62 ± 0.08 (4.0 × 10−140.37 ± 0.36 0.14 ± 0.19 0.49 ± 0.19 
CAL spots of large size 0.43 ± 0.08 (2.1 × 10−80.20 ± 0.42 0.12 ± 0.21 0.68 ± 0.22 
Cutaneous neurofibromas 0.26 ± 0.09 (0.0017) 0.26 ± 0.09 0.00 0.74 ± 0.09 
Subcutaneous neurofibromas 0.44 ± 0.09 (2.0 × 10−70.44 ± 0.09 0.00 0.56 ± 0.09 
Plexiform neurofibromas 0.45 ± 0.09 (2.7 × 10−80.45 ± 0.09 0.00 0.55 ± 0.09 
Clinical feature Heritabilitya (P-value) σ2A σ2G σ2E 
CAL spots of small size 0.62 ± 0.08 (4.0 × 10−140.37 ± 0.36 0.14 ± 0.19 0.49 ± 0.19 
CAL spots of large size 0.43 ± 0.08 (2.1 × 10−80.20 ± 0.42 0.12 ± 0.21 0.68 ± 0.22 
Cutaneous neurofibromas 0.26 ± 0.09 (0.0017) 0.26 ± 0.09 0.00 0.74 ± 0.09 
Subcutaneous neurofibromas 0.44 ± 0.09 (2.0 × 10−70.44 ± 0.09 0.00 0.56 ± 0.09 
Plexiform neurofibromas 0.45 ± 0.09 (2.7 × 10−80.45 ± 0.09 0.00 0.55 ± 0.09 

Data are presented as Mean ± SEM. All models include age and sex as covariates. σ2A, additive genetic component of variance; σ2G, NF1 mutation component of variance; σ2E, residual error (random environmental) component of variance.

aEstimated as the ratio of additive genetic variance to total phenotypic variance unexplained by covariates, in a model without the NF1 mutation component of variance.

The patterns of phenotypic correlations among different types of relative pairs, referred to as ‘patterns of familial correlations’, were examined for the 12 clinical features investigated (Table 3). Significant negative correlations were not observed for any of the features. For CAL spots of small size, subcutaneous neurofibromas, plexiform neurofibromas and learning disabilities, there were significant positive correlations among both sibling and parent–offspring pairs (ranging from 0.20 to 0.37; P < 0.05), whereas the correlations between more distant relatives were not significant. This suggests a minor role of the NF1 mutation since a high correlation even between distant relatives would be observed if variation in disease expression was primarily determined by differences in NF1 mutations. Lower phenotypic correlations between distant relatives than between close relatives are expected in a model with strong modifying genetic effects, but could also result from environmental factors that are more likely to be shared among first-degree relatives than among more distant relatives. However, the higher phenotypic correlations observed between monozygotic (MZ) twins compared with siblings for these four traits support the existence of a strong genetic component. For cutaneous neurofibromas, significant phenotypic correlations were observed among both first-degree relatives (0.95, 0.39 and 0.27 for MZ twin, parent–offspring and sibling pairs, respectively; P < 0.001) and more distant relatives (0.21 and 0.48 for second-degree and third-degree relatives, respectively; P < 0.05), which is suggestive of an effect of the constitutional NF1 mutation. A similar pattern of familial correlations was observed for blue-red macules, which showed a highly significant phenotypic correlation among third-degree relatives (0.65, P < 0.001); but no significant correlation was observed among second-degree relatives for this trait. CAL spots of large size, skin-fold freckling, Lisch nodules and facial dysmorphism had greater phenotypic correlations between siblings than between parent–offspring pairs, consistent with either a dominance component, a common sibling environment component, or an effect of the normal NF1 allele since affected sibs are expected to share the same normal NF1 allele more often than parent–offspring pairs do. The trait ‘neoplasms’ did not show any significant correlation among affected relatives, whatever the degree of relationship. In considering sex-specific pairs (data not shown), homogeneity tests did not reveal any significant difference between the father–son, father–daughter, mother–son and mother–daughter correlations for any of the features. In contrast, significant differences between brother–brother, sister–sister and brother–sister pair correlations were observed for blue-red macules (0.46, −0.14 and 0.21, respectively; P = 0.003) and facial dysmorphism (0.09, 0.72 and 0.07, respectively; P < 0.0001).

Table 3.

Adjusted intra-familial correlation coefficients for 12 clinical features of NF1 among 750 affected relatives of 275 families

Clinical feature rmz rss rpo r2nd r3rd 
CAL spots of small size 0.64 0.37*** 0.32*** 0.05 −0.22 
CAL spots of large size 0.13 0.45*** 0.11 0.04 −0.14 
Cutaneous neurofibromas 0.95*** 0.39*** 0.27*** 0.21* 0.48* 
Subcutaneous neurofibromas 0.52 0.24** 0.22*** −0.14 0.08 
Plexiform neurofibromas 0.90*** 0.29*** 0.25*** −0.14 −0.16 
Skin-fold freckling 1.0 0.23** 0.08 0.29** −0.09 
Blue-red macules 1.0 0.22* 0.25*** −0.13 0.65*** 
Lisch nodules 0.47 0.66*** 0.31* −0.15 −0.33 
Facial dysmorphism 1.0 0.28*** 0.10* 0.03 −0.09 
Scoliosis 0.17 0.26** 0.12* 0.24* −0.14 
Learning disabilities 1.0 0.20* 0.28*** 0.10 −0.32 
Neoplasms 1.0 0.11 −0.08 −0.01 −0.08 
Clinical feature rmz rss rpo r2nd r3rd 
CAL spots of small size 0.64 0.37*** 0.32*** 0.05 −0.22 
CAL spots of large size 0.13 0.45*** 0.11 0.04 −0.14 
Cutaneous neurofibromas 0.95*** 0.39*** 0.27*** 0.21* 0.48* 
Subcutaneous neurofibromas 0.52 0.24** 0.22*** −0.14 0.08 
Plexiform neurofibromas 0.90*** 0.29*** 0.25*** −0.14 −0.16 
Skin-fold freckling 1.0 0.23** 0.08 0.29** −0.09 
Blue-red macules 1.0 0.22* 0.25*** −0.13 0.65*** 
Lisch nodules 0.47 0.66*** 0.31* −0.15 −0.33 
Facial dysmorphism 1.0 0.28*** 0.10* 0.03 −0.09 
Scoliosis 0.17 0.26** 0.12* 0.24* −0.14 
Learning disabilities 1.0 0.20* 0.28*** 0.10 −0.32 
Neoplasms 1.0 0.11 −0.08 −0.01 −0.08 

Correlation coefficients were computed for 6 MZ twin pairs (rMZ), 233 sibling pairs (rSS), 411 parent–offspring pairs (rPO), 100 second-degree relatives pairs (r2nd) and 22 third-degree relative pairs (r3rd), except for Lisch nodules where these numbers were 5, 155, 95, 15 and 3, respectively, due to missing data for 337 individuals for this feature. All correlation coefficients were adjusted for age and sex.

*P < 0.05; **P < 0.01;***P < 0.001.

Family-based association analysis of NF1 SNPs

We tested the hypothesis that genetic variants in the NF1 gene, different from the disease mutation itself, could modulate the phenotypic expression of the mutation. We used SNP data of the European HapMap sample (CEU) to look at patterns of linkage disequilibrium (LD) around the NF1 gene and applied a tagging SNP approach to identify the SNP markers necessary to capture the common genetic variation across NF1 (see Material and Methods section for SNP selection). A subset of nine tag SNPs were selected and genotyped in 1132 individuals from 313 families with NF1. Figure 1 illustrates the organization of the NF1 gene, the location of the nine SNP markers we studied and the extent of LD in the region in our sample.

Figure 1.

Location and pairwise LD of the nine NF1 SNPs selected for family-based association study. The chromosomal location and gene structure were taken from NCBI's RefSeq resource with annotations from the UCSC genome browser database (March 2006 Build; NCBI Build 36.1). The locations of the nine tag SNPs in NF1 are indicated by arrows. The LD structure for the nine NF1 SNPs was determined using our own genotype data in the family sample. High pairwise LD (r2) between markers is illustrated with dark shading. The r2 values (×100) for the marker pairs are listed in the corresponding boxes.

Figure 1.

Location and pairwise LD of the nine NF1 SNPs selected for family-based association study. The chromosomal location and gene structure were taken from NCBI's RefSeq resource with annotations from the UCSC genome browser database (March 2006 Build; NCBI Build 36.1). The locations of the nine tag SNPs in NF1 are indicated by arrows. The LD structure for the nine NF1 SNPs was determined using our own genotype data in the family sample. High pairwise LD (r2) between markers is illustrated with dark shading. The r2 values (×100) for the marker pairs are listed in the corresponding boxes.

Genotyping success rates were 97.6% or greater for each SNP. A total of 25 Mendelian errors were detected in the entire sample for all the SNPs tested: the genotypes of these families were considered as missing data in the analysis. Individuals with missing genotypes at three or more SNP markers were removed from analyses. Thus, a total of 1105 subjects (740 NF1 patients and 365 non-affected relatives) from 306 families (corresponding to 371 nuclear families) were finally included in FBAT analysis. None of the SNPs gave a significant deviation from Hardy–Weinberg equilibrium when considering the unrelated founders of the sample (all P > 0.05). The minor allele frequency (MAF) of the nine SNPs was between 0.10 and 0.41, and was above 0.20 for six of them (Table 4). Heterozygosity ranged from 0.159 to 0.468.

Table 4.

Single-marker FBAT tests for association of NF1 SNPs with the number of plexiform neurofibromas

Marker Allele Frequencya Familiesb S-E(S)c Zd P-valuee 
rs2905788 0.684 28 −1.800 −0.696 0.562 
rs2905788 0.316 55 −0.667 −0.179 0.858 
rs2953016 0.753 18 0.450 0.220 0.826 
rs2953016 0.247 53 1.417 0.389 0.684 
rs2953014 0.799 18 −1.250 −0.611 0.625 
rs2953014 0.201 49 −10.050 −2.736 0.006 
rs2905804 0.720 24 −1.383 −0.582 0.674 
rs2905804 0.280 53 −1.417 −0.387 0.892 
rs7215555 0.732 26 −0.472 −0.189 0.850 
rs7215555 0.268 53 −7.772 −2.094 0.036 
rs11080149 0.882 – – – 
rs11080149 0.118 34 −4.617 −1.552 0.121 
rs7405740 0.898 – – – 
rs7405740 0.102 29 2.500 0.857 0.391 
rs964288 0.588 34 −3.300 −1.154 0.249 
rs964288 0.412 51 1.333 0.360 0.718 
rs3815156 0.821 11 – – – 
rs3815156 0.179 49 4.667 1.308 0.191 
Marker Allele Frequencya Familiesb S-E(S)c Zd P-valuee 
rs2905788 0.684 28 −1.800 −0.696 0.562 
rs2905788 0.316 55 −0.667 −0.179 0.858 
rs2953016 0.753 18 0.450 0.220 0.826 
rs2953016 0.247 53 1.417 0.389 0.684 
rs2953014 0.799 18 −1.250 −0.611 0.625 
rs2953014 0.201 49 −10.050 −2.736 0.006 
rs2905804 0.720 24 −1.383 −0.582 0.674 
rs2905804 0.280 53 −1.417 −0.387 0.892 
rs7215555 0.732 26 −0.472 −0.189 0.850 
rs7215555 0.268 53 −7.772 −2.094 0.036 
rs11080149 0.882 – – – 
rs11080149 0.118 34 −4.617 −1.552 0.121 
rs7405740 0.898 – – – 
rs7405740 0.102 29 2.500 0.857 0.391 
rs964288 0.588 34 −3.300 −1.154 0.249 
rs964288 0.412 51 1.333 0.360 0.718 
rs3815156 0.821 11 – – – 
rs3815156 0.179 49 4.667 1.308 0.191 

The univariate FBAT was performed by use of the dominant model for the minor allele. SNPs are listed in order (5′→3′) in NF1 gene. The FBAT statistic was not computed when the number of informative families available was fewer than 15. The sign of the Z-score (+/−) indicates whether the allele is associated with a larger or a fewer number of plexiform neurofibromas.

aFrequency represents the single allele frequency.

bNumber of informative families for the specific allelic test (i.e. families with at least one heterozygous parent).

cDifference between the test statistic S of FBAT for the observed number of transmitted alleles and the expected value of S under the null hypothesis (i.e. no linkage or association).

dZ-score of the test statistic S.

eExact P-values, uncorrected for multiple testing.

The genotypes for the nine SNPs were analyzed with FBAT using both an additive and a dominant genetic model. Due to the large number of tests performed, the issue of multiple testing should be addressed here. Since the nine SNPs of NF1 are correlated, as are the 12 phenotypes studied, accounting for the number of tests using a simple Bonferroni correction is expected to be conservative. Then an effective number of independent SNPs was calculated by the method proposed by Li and Ji (14) implemented in SNPSdP (15). We found that the nine correlated SNPs correspond to seven effective independent ones. Matrix spectral decomposition (matSpD), a variant of SNPSpD, was used to estimate the effective number of independent phenotypes being analyzed (n = 11). Hence, an estimate of the effective number of tests that we performed is roughly 154 (seven SNPs, 11 phenotypes and two genetic models), resulting in a robust Bonferroni-corrected significance threshold of 0.0003. We found no significant evidence for association of any single marker with any phenotype at this threshold, under either additive or dominant models. Nonetheless, suggestive evidence for association was observed for SNPs rs2953014 and rs7215555 under a dominant model (Table 4). The minor alleles of these two SNP were associated with a fewer number of plexiform neurofibromas in our sample (Z = −2.736, P = 0.006 for rs2953014; Z = −2.094, P = 0.036 for rs7215555).

Because of their ability to capture LD more informatively, testing associations with haplotypes may be more powerful than with individual SNPs. Therefore, we also used the HBAT command of FBAT to test for evidence of haplotypic association. By use of the expectation-maximization algorithm implemented in the haplotype FBAT, six major NF1 haplotypes (i.e. frequency ≥0.05) were revealed (Table 5). Haplotype-specific tests were used to analyze the effects on the 12 clinical features of NF1 for each common haplotype, compared with all other haplotypes grouped together. The TGGCGCGAA haplotype, bearing the variant alleles at both rs2953014 and rs7215555 (underscored), was found to be associated with the number of plexiform neurofibromas, under a dominant model (Z = −2.446, P = 0.009). This result is highly consistent with the univariate FBAT results showing significant association of these two SNPs with the same clinical feature at a level of 0.05 (Table 4). However, these associations were no longer statistically significant after Bonferroni correction for multiple comparisons.

Table 5.

Haplotype-specific tests of haplotype FBAT for association of NF1 SNPs with the number of plexiform neurofibromas

Haplotype Frequencya Familiesb S-E(S)c Zd P- valuee 
TGTCACGAA 0.316 57 0.551 0.139 0.958 
TGGCGCGAA 0.187 42 −8.330 −2.446 0.009 
CCTTACGGG 0.141 37 4.174 1.352 0.152 
TGTCACCG0.087 22 3.950 1.554 0.124 
TGTCGCGAA 0.073 23 1.322 0.556 0.484 
CGTTATGG0.069 19 −2.500 −1.127 0.319 
Haplotype Frequencya Familiesb S-E(S)c Zd P- valuee 
TGTCACGAA 0.316 57 0.551 0.139 0.958 
TGGCGCGAA 0.187 42 −8.330 −2.446 0.009 
CCTTACGGG 0.141 37 4.174 1.352 0.152 
TGTCACCG0.087 22 3.950 1.554 0.124 
TGTCGCGAA 0.073 23 1.322 0.556 0.484 
CGTTATGG0.069 19 −2.500 −1.127 0.319 

The haplotype-specific test of the haplotype FBAT was performed by use of the dominant model. Haplotype sequences are formed by the succession of SNPs rs2905788, rs2953016, rs2953014, rs2905804, rs7215555, rs11080149, rs7405740, rs964288 and rs3815156 along the NF1 gene (in the direction 5′→3′). Variant alleles compared to the consensus sequence TGTCACGAA are shown in bold. The nucleotides at rs2953014 and rs7215555 are underscored for comparisons with the single-marker FBAT results of the respective SNPs in Table 4. Only haplotypes with frequencies ≥0.05 are presented. The sign of the Z-score (+/−) indicates whether the allele is associated with a larger or a fewer number of plexiform neurofibromas.

aFrequency represents the single haplotype frequency.

bNumber of informative families for the specific haplotype test (i.e. families with at least one heterozygous parent).

cDifference between the test statistic S of FBAT for the observed number of transmitted alleles and the expected value of S under the null hypothesis (i.e. no linkage or association).

dZ-score of the test statistic S.

eExact P-values, uncorrected for multiple test.

DISCUSSION

A large number of Mendelian genetic disorders display considerable inter- and intra-familial variability in phenotypic expression. Differences in environmental factors and different mutations can easily be seen to underlie a proportion of inter-familial manifestations. However, intra-familial variability, especially in siblings, cannot intuitively be so readily accounted for by these types of mechanisms. It is now increasingly apparent that genetic modifiers, distinct from the disease locus itself, have a considerable role to play in phenotypic variations of single-gene disorders (16–18). Their effect on disease expression may vary from strong effects under a ‘monogenic-like’ model to much milder effects under a ‘multifactorial-like’ model. Identifying these genetic modifiers may be of great interest from the viewpoints of both treatment and genetic counselling, but it remains very challenging, despite the powerful genetic tools available today. Thus, before launching expensive and time-consuming genetic studies to identify these genetic modifiers, it is important to make sure that they really exist and that environmental factors or other mechanisms, such as allelic heterogeneity, do not suffice to explain this phenotypic variability. Toward this end, a first step is to examine phenotypic correlations among various family members. In this approach, larger pedigrees have a distinct advantage over nuclear families in that more pairwise combinations containing a broader array of genetic relationships are possible. First-degree relatives, such as siblings and parents and offspring, have 50% of their genes in common. Second-degree relatives, such as avuncular and grandparental relationships, share 25% of their genes, and third-degree relatives, such as first cousins, share half that again (12.5%). Hence if the phenotypic variation of the disease is determined primarily by alleles in genes unlinked to the disease locus, the phenotypic correlation is expected to decrease with the degree of relationship, being strongest between first-degree relatives and decreasing to zero for distant relatives in the same family. Such patterns of familial correlations may also be the result of a more shared environmental effect among closer relatives. This possibility can be examined specifically by comparing phenotypic correlations in MZ twins and siblings.

The objective of the present study was to evaluate the genetic component of variable expressivity in NF1, one of the most common autosomal dominant disorders in humans, and to assess the relative contributions of the NF1 locus and putative unlinked genetic modifiers. Only two previous studies have examined familial aggregation of NF1 features among different classes of affected relatives. Easton et al. (12) studied the expressivity of NF1 in 175 affected members of 48 extended families, including 6 MZ twin pairs, 76 sib pairs, 60 parent–offspring pairs, 54 second-degree relative pairs and 43 third-degree relative pairs. They examined seven NF1-related features: two quantitative traits (number of CAL spots of large size and number of cutaneous neurofibromas) and five binary traits (presence or absence of plexiform neurofibromas, optic gliomas, scoliosis, epilepsy and referral for remedial education). Szudek et al. (13) studied 904 individuals with NF1 from 373 families. Although this is the largest group of NF1 families ever studied, there were no pairs of MZ twins, nor pairs of third-degree relatives. They examined ten clinical features of NF1 but for four of them (scoliosis, seizures, optic glioma and other neoplasms), the multivariate probit regression model used failed to converge on correlation coefficients among second-degree relatives because of the low frequency of these features and insufficient sample size. Moreover, they did not have counts of CAL spots, cutaneous neurofibromas, subcutaneous neurofibromas and plexiform neurofibromas. These features were instead treated as binary variables. The present study has much strength. First, we examined familial correlations in a large number of affected individuals, including MZ twins and third-degree relatives. Second, analyses were performed on objective quantitative variables such as lesion counts for CAL spots and the different types of neurofibromas. This enabled a more detailed analysis of familial aggregation. The other features studied were binary by nature. Finally, as a mutational screening of NF1 at both RNA and genomic DNA levels was carried out for all affected individuals of the NF-France database, we were able to exclude from analysis all individuals with no identifiable mutation in NF1, thereby limiting the presence of some genetic heterogeneity [50 of 561 (9%) families excluded]. For instance, two French families fulfilling the National Institutes of Health (NIH) NF1 diagnostic criteria, but with no NF1 mutation identified, were actually found to carry loss-of-function mutations in the SPRED1 gene (19).

Evaluating the genetic component of variable expression in NF1

Data from this family study indicated that all of the 12 NF1 clinical features studied, with the exception of neoplasms, show significant familial aggregation, after adjusting for age and sex. Different patterns of familial correlations were observed (Table 3). For CAL spots of small size, subcutaneous neurofibromas, plexiform neurofibromas and learning disabilities, correlation was highest between MZ twins, lower but still significant between first-degree relatives, with similar correlation values between siblings and parent–offspring pairs, and it was lower still and not significant between more distant relatives. Such a pattern of familial correlations suggests a strong genetic component which is mainly additive and not related to the constitutional NF1 mutation. Of these four traits, only learning disabilities, referred to as ‘referral for remedial education’, and the presence or absence of plexiform neurofibromas were examined in the study of Easton et al. (12). These features showed a high concordance between MZ twins which diminished with increasing degree of relationship, consistent with our findings. Nonetheless, the results observed with learning disabilities have to be interpreted with caution: given a background frequency of ‘learning disability’ of 10% in the general population (20), it is likely that some of the individuals displaying this feature have learning disabilities that are actually not related to the NF1 disease. This may have biased the results if the causal factors underlying NF1-related learning disabilities significantly differ in nature from those involved in the development of learning disabilities in the general population. Szudek et al. (13) examined three of the four traits: presence or absence of CAL spots, subcutaneous neurofibromas and plexiform neurofibromas. It is, however, hard to compare their results to ours since the size of the CAL spots examined was not specified in their study, and the way of scoring these features differed between the two studies (presence/absence versus counts). Nonetheless, they observed significantly higher correlations among siblings than among parent–offspring pairs for subcutaneous and plexiform neurofibromas, contrary to our findings.

For the number of cutaneous neurofibromas, the six pairs of MZ twins were almost completely concordant for numbers of neurofibromas as categorized in this study (r = 0.95, P < 0.001). A complete concordance was also observed for this feature for the six pairs of MZ twins studied in Easton et al. (12), as well as for the nine pairs of MZ twins described in the literature (21–26). Correlations were found to be weaker but still highly significant among siblings (r = 0.39, P < 0.001) and parent–offspring pairs (r = 0.27, P < 0.001), and they were as high and still significant among second-degree and third-degree relatives. Such a strong correlation even between distant relatives suggests a role of the NF1 mutation in the onset of cutaneous neurofibromas. Interestingly, the sole genotype–phenotype correlation established so far for an intragenic mutation of the NF1 gene concerns a small in-frame deletion in exon 17 of NF1 and the absence of cutaneous neurofibromas (11). Although the biological mechanism that relates this specific mutation to the suppression of cutaneous neurofibroma development is still unknown, other correlations involving this feature and other specific mutations may exist. The finding of Szudek et al. (13) of a strong correlation among both first-degree and second-degree relatives (r ∼ 0.50; P < 0.05) for the presence/absence of cutaneous neurofibromas is consistent with this interpretation.

For CAL spots of large size, skin-fold freckling, Lisch nodules and facial dysmorphism, greater correlations among siblings than among parent–offspring pairs were observed. This is consistent with either a dominant genetic component unlinked to the NF1 gene, a common sibling environment component, or an effect of the normal NF1 allele. The few number of MZ twins included in this study provided limited power to distinguish between these different components, their effects being completely confounded in a family design lacking MZ twins. Moreover, although we minimized the confounding effect of age, it is still conceivable that a residual age effect contributed to the observed differences between sibling and parent–offspring pairs by producing a greater correlation among affected relatives of similar age, such as siblings. This is particularly true for Lisch nodules and skin-fold freckling for which the effect of age was shown to be highly significant.

The few number of malignant peripheral nerve sheath tumours (MPNSTs) in our series (n = 9) compelled us to pool this trait with the other types of malignant tumours and we were not able to reliably infer the existence of genetic modifiers for this particular trait. But we cannot exclude the possibility of an influence of genetic modifiers in the development of MPNSTs since such an effect may have been hindered by the lack of significant familial aggregation for the other malignant tumours.

The results from variance components analysis demonstrated only a limited contribution of the NF1 mutation to total phenotypic variance for the five quantitative traits studied, with a small effect for CAL spots and a null effect for neurofibromas (Table 2). Our findings are in total agreement with those of Easton et al. (12), who estimated the NF1 mutation component to 0.028 (P > 0.05) for the number of CAL spots of large size and to zero for the number of cutaneous neurofibromas. Conversely, the additive genetic variance component was significantly greater than zero for the five features, with an estimated heritability ranging from 0.26 to 0.62. The predominantly white population that makes up most of the PHRC NF1 cohort may, however, limit the universality of our findings. Significant differences in the prevalence of genes important to heritability and different magnitudes of genes by environment effects may lead to estimates of heritability that differ among ethnic groups. Moreover, since this study primarily focused on familial cases with at least two affected members per family, one should be aware that some of the most severe forms of the disease may have been discarded from the study. Indeed, given the high mutation rate of the NF1 gene and mutation-selection balance, a high selection coefficient is expected for this disease. This means that a vast proportion of NF1 individuals do not have offspring, most likely in the case of the more severe forms.

Influence of the normal NF1 allele

From the results of familial correlation analyses, we could not exclude the possibility of an influence of the normal NF1 allele on the development of particular NF1 clinical features, notably CAL spots of large size, skin-fold freckling, Lisch nodules and facial dysmorphism. Phenotypic manifestations of NF1 might not be simply a matter of haploinsufficiency and slight differences in the expression of the normal allele in trans to the mutated allele might have a major impact on the phenotypic expression of the disease. There are several examples of dominantly inherited disorders where phenotypic expression depends on the normal allele through trans effect. For instance, it has been demonstrated that patients with the dominant form of erythropoietic protoporphyria, an inherited disorder of heme biosynthesis with incomplete penetrance, usually share a hypomorphic allele that is common in the general population in trans to a rare loss-of-function allele (27–29). This is a striking example of a dominant disease where the penetrance of the deleterious mutation is modulated by a functional polymorphism occurring at the same locus. Similarly, the occurrence, in trans position to an alpha-allele responsible for hereditary elliptocytosis, of a low expression allele, allele alpha LELY, was shown to enhance the severity of elliptocytosis (30).

In the present study, we investigated for the first time the role of common NF1 polymorphisms in the variable expression of NF1. We adopted the intra-family design in order to overcome spurious associations due to population stratification effects, and a tagging SNP approach was applied to capture most of the genetic variation at the NF1 locus. The extensive long-range LD over 300 kb surrounding the NF1 gene (Supplementary Material, Fig. S1) allowed us to considerably reduce the number of common variants required to capture the full haplotype information. A set of nine tag SNPs was retained and genotyped in 1132 individuals from 313 families with NF1 (Fig. 1). No significant deviations of transmission of any of the NF1 variants to affected offspring was found for any of the clinical features examined, based on single marker or haplotype analysis. These results suggest that genetic variants in NF1, where either located in cis or in trans position to the primary mutation, are unlikely to contribute to the variable expressivity of the disease. We can therefore exclude the possibility of an influence of the normal NF1 allele. Although our results provided some suggestive evidence of an association of two SNPs in NF1 with the number of plexiform neurofibromas in single-marker and multiple-marker haplotype analysis, under a dominant model (Tables 4 and 5), these results were not significant after Bonferroni correction and therefore should be considered as principally negative. The replication of this finding in an independent data set would nonetheless be valuable to determine whether this association represents a spurious finding or a true causal relationship.

In conclusion, our results provided evidence for a strong genetic component in most NF1 clinical features with no apparent influence of the NF1 gene on disease variation since neither the constitutional NF1 mutation nor the normal NF1 allele seem to contribute significantly to the overall phenotypic variation for each trait. The remainder of this phenotypic variation could be due to unlinked genetic modifiers, environmental factors or some combination of these. In accordance with previous findings of Szudek et al. (31,32), we found several statistically significant associations between combinations of clinical features of NF1 (unpublished data), suggesting that some NF1 features may share common genetic determinants. It is noteworthy that the number of plexiform neurofibromas was significantly associated with the number of cutaneous neurofibromas (P < 0.0001), the number of subcutaneous neurofibromas (P < 0.0001) and the presence of blue-red macules (P = 0.0072). Similarly, a significant association between the number of CAL spots of small size, the number of CAL of large size and the presence of skin-fold freckling was found (P < 0.0001). Such results indicate a possible common repertoire of genetic modifiers for some combinations of traits and may thereby facilitate the identification of the underlying genetic influences. Further studies of genetic linkage and association are warranted to identify the specific genetic variants associated with variable expression in NF1. Understanding the genetic mechanisms that control phenotypic expression in NF1 will provide insight into the fundamental disease processes and provide a rational to develop new therapeutic strategies.

MATERIALS AND METHODS

Study samples

Families for this study were enrolled between 2002 and 2005 in a project of the French Clinical Research Program entitled ‘Study of expressivity of neurofibromatosis 1: constitution of a phenotype-genotype database’ (PHRC NF1). The study was approved by the local ethical committee and all participants gave their written informed consent. The PHRC NF1 database constituted through this program includes a collection of 561 families, consisting of 1697 individuals among whom 1083 fulfilled NIH diagnostic criteria for NF1 (33,34). For each patient, the full phenotypic information was recorded in a standardized way and was coupled with a comprehensive mutation screening of the NF1 gene including intragenic NF1 microsatellites study, NF1 sequencing at both RNA and DNA levels and real-time PCR-based gene-dosage. The NF1 mutation was successfully identified in 512 of 561 families (91%).

For heritability and familial correlation analyses, families with only one affected member and/or with no identifiable NF1 mutation were excluded. The final study sample consisted of 750 NF1 patients (336 males and 414 females) in 275 families, including 233 nuclear families and 42 extended families. The mean age at the time of the interview was 28.57 years (±17.58 SD), ranging from 6 months to 80 years (the age distribution of the individuals examined is given in Table 1). The family sizes ranged from 2 to 8 affected members. There were 146 families with two members, 86 families with three members, 26 families with four members, 10 families with five members and 7 families with at least six members. The sample included 6 MZ twin pairs, 233 sibling pairs, 411 parent–offspring pairs, 29 grandparent–grandchild pairs, 71 avuncular pairs and 22 cousin pairs.

For family-based association analysis, NF1 patients with no available parents or with one single parent and no siblings were excluded from the database. Thus, a total of 1132 individuals from 313 families were selected for genotyping. The dataset comprised 271 nuclear families with at least one affected offspring and 42 extended families, with a total of 758 NF1 patients and 374 non-affected relatives.

Phenotypes

From the 22 phenotypic traits examined in the 1083 NF1 patients of the PHRC NF1 database, we discarded ten traits whose prevalence was too low (5% or less) for meaningful statistical analysis. Among the remaining 12 clinical features of NF1, there were five quantitative traits: number of CAL spots of small size (0.5–1.5 cm in longest diameter) and of large size (>1.5 cm in longest diameter); number of plexiform neurofibromas and number of cutaneous and subcutaneous neurofibromas, each classified into quantitative categories of 0, 1–9, 10–99 and >100. There were seven binary traits: presence/absence of skin-fold freckling, blue-red macules, Lisch nodules, facial dysmorphism, scoliosis, learning disabilities and neoplasms. Most of the features were identified by physical examination. Lisch nodules were diagnosed or excluded by slit-lamp examination. Blue-red macules are a peculiar type of neurofibroma: histologically, they correspond to neurofibromatous tissue infiltrating capillary blood vessels and venules (35). A facial dysmorphism was diagnosed if the following signs were observed: coarse face, flat occiput/brachycephaly, facial asymmetry, prominent forehead, frontal bossing, ptosis, downslanting deep set eyes, eversion of the lateral eyelid, epicanthic folds, high and broad nasal bridge, bulbous nasal tip, large and low set ears, malar hypoplasia, wide and prominent philtrum, micrognathia, small pointed chin and low posterior hairline. For scoliosis, only scoliotic curves of >10° were taken into account in our analysis. The dystrophic form of scoliosis and lumbar scoliosis accounted for 8% (n = 21) and 31% (n = 77) of all scoliosis cases (n = 248), respectively. The diagnosis of learning disabilities was performed on specific testing of cognitive abilities and/or history of scholar difficulties leading to repeating at least one level. Learning deficits most frequently involved visual spatial, visual motor integration skills and language-based skills. We included in the item ‘neoplasms’ the following tumours: optic gliomas (n = 51), MPNST (n = 9), pheochromocytomas (n = 3), aggressive gliomas (n = 2), low-grade gliomas (n = 4), lymphoblastic leukaemia (n = 1), neuroblastoma (n = 1), carcinoid tumour (n = 1) and intestinal tumour (n = 1). As the small number of observations of each individual tumour would have resulted in a lack of power in statistical analyses, we pooled all types of tumours into a single item and performed analyses on this composed trait. Because of significant skewness, we applied a square-root transformation to the two variables quantifying the number of CAL spots prior to the analyses (skewness: 1.52 versus 0.20 and 1.35 versus −0.23; kurtosis: 2.14 versus 0.04 and 4.40 versus 0.92; before transformation versus after transformation for the number of CAL spots of small size and of large size, respectively). This was found to be the best-fitting Box-Cox transformation (36). For the two traits quantifying the number of cutaneous and subcutaneous neurofibromas, we assigned an appropriate value to each category: the values chosen were 0, 1.5, 3.5 and 5, which are approximately the average of the logarithms of the upper and lower limits of the appropriate interval.

Heritability and familial correlation analyses

Heritability was estimated using a variance components method as implemented in the SOLAR (Sequential Oligogenic Linkage Analysis Routines) computer package (37), which simultaneously utilizes data on all family relationships in pedigrees of arbitrary size and complexity. This method applies maximum likelihood estimation to a mixed effects model that incorporates fixed effects for known covariates (age and sex in the present study) and variance components for genetic effects. The variance-component model follows classical quantitative genetic principles, in which the phenotypic variance (σ2P) is decomposed into additive components for additive genetic (σ2A) and random environmental effects (σ2E). The heritability of a phenotype was estimated as the ratio of additive genetic variance to total phenotypic variance unexplained by covariates (h2 = σ2A/σ2P). We also performed the analysis by adding another variance component to the model, σ2G, so as to estimate the contribution of the constitutional NF1 mutation to the total phenotypic variance. This was modelized as a factor shared by all members of the same family, as described in Easton et al. (12). Estimates of the means and variances of components of the models were obtained by maximum likelihood methods and significance was determined by likelihood ratio tests. Although SOLAR is able to handle binary traits as well, through the use of a liability threshold model, these analyses were only performed on quantitative traits because of the difficulty of reaching convergence for binary traits.

Familial correlation coefficients were calculated using the program FCOR within the SAGE v5.4.2 (Statistical Analysis for Genetic Epidemiology) software package (38). Correlations were calculated between the residual trait values, after allowing for age and sex. Because the size of the families in the study varied, the uniform weighting scheme, in which the contribution to the sum of squares from each pedigree is the same regardless of the number of relative pairs in the pedigree, was used to calculate familial correlations. The asymptotic standard error of a given correlation was estimated by using a second-order Taylor series expansion and replacing all correlation parameters with their respective estimates. A test for homogeneity of correlations among subtypes (e.g. mother–offspring and father–offspring) within main type (e.g. parent–offspring) was also performed. Under the null hypothesis of homogeneity, the test statistic has an approximate χ2 distribution with degrees of freedom equal to the number of subtypes minus one.

LD and tagging analyses

The pattern of LD around the NF1 locus was studied in a 750-kb region encompassing the entire NF1 gene. Because almost all individuals in the NF-France database were of European origin, we used HapMap SNP data (the International HapMap Project (39)/Public Release #21) from the CEU HapMap sample (CEPH collection of Utah residents of northern and western European ancestry). LD block structure was examined by the program Haploview (40). The D′ for all pairs of SNPs was calculated and the haplotype blocks were estimated using the confidence interval method (41). All SNP markers in the NF1 gene were included in a single block of strong LD of 339 kb, which encompasses the entire NF1 gene and solely this gene (Supplementary Material, Fig. S1). We then used the Tagger program (42) to select SNPs that efficiently tagged all common variations in this block. We required the minimum estimated r2 between the tagged and tag SNP sets to be ≥0.95, implying only a slight loss of power in typing only tags. From the original set of 114 common SNPs in the CEU sample (MAF >0.05), a subset of nine tag SNPs were selected: rs2905788, rs2953016, rs2953014, rs2905804, rs7215555, rs11080149, rs7405740, rs964288 and rs3815156. The mean r2 between tagged and tag SNP sets was 0.993.

SNP genotyping

TaqMan® SNP Genotyping Assay-by-Design method (Applied Biosystems, Foster City, CA) was used to genotype the nine tag SNPs in NF1 with an allele-specific hybridization approach. Probes and primers were designed by Applied Biosystems and are available on demand. PCR amplification was carried out according to the manufacturer's instructions. A total reaction volume of 10 µl included: 10 ng of template DNA, 4.5 µl of TaqMan® Genotyping Master Mix (Applied Biosystems) and 0.5 µl of SNP Genotyping Assay Mix (including specific primers and fluorescently-labelled probes). All PCR reactions were performed with an ABI Prism 7900 Sequence Detection System (Applied Biosystems) under the following conditions: 95°C for 10 min followed by 40 cycles of 92°C for 15 s and 60°C for 1 min. The PCR was followed by allelic discrimination using the ABI Prism 7900 to perform plate reading. Automated allele calling was performed by allelic discrimination plots with ABI SDS software version 2.2 from Applied Biosystems. Genotypes were checked for Mendelian inheritance errors using FBAT software (43–45). Hardy–Weinberg equilibrium was tested in the parental data for each locus, by use of the χ2 goodness-of-fit test.

Family-based association analysis

The software package FBAT v2.0.2c (43–45) was used to test for association between the nine tag SNPs of the NF1 gene and each of the 12 clinical features of NF1 included in this study. This method is a generalized version of the original transmission disequilibrium test (46) which incorporates a set of statistical procedures to accommodate variable pedigree constellations, dichotomous or quantitative phenotypes, phenotype-unknown subjects, multiple loci and various genetic models. It decomposes pedigrees into individual nuclear families and treats them as independent. Haplotype analysis was performed using the HBAT command of FBAT, using both the multi-allelic and bi-allelic (each allele against all others) mode of testing. Only common haplotypes (frequency >0.05) were included in the global test. In all FBAT analyses, exact P-values were computed using 100 000 permutations under both additive and dominant genetic models and the minimum number of informative families necessary to perform the analyses was set to 15.

SUPPLEMENTARY MATERIAL

Supplementary Material is available at HMG online.

FUNDING

This work was supported in part by grants by Association Neurofibromatoses et Recklinghausen; Ligue Française Contre les Neurofibromatoses; the French Clinical Research programme (PHRC 2002); INSERM (Nf1GeneModif project) and Ministère de l'Enseignement Supérieur et de la Recherche. Some of the results of this paper were obtained by using the program package S.A.G.E., which is supported by a U.S. Public Health Service Resource Grant (1 P41 RR03655) from the National Center for Research Resources.

ACKNOWLEDGEMENTS

We express our gratitude to NF-France Network for its precious collaboration: H. Adamski, C. Baumann-Morel, S. Bastuji-Garin, C. Bellanne, E. Bieth, P. Bousquet, C. Brandt, X. Balguerie, L. Boudali, P. Berbis, P. Castelnau, Y. Chaix, J. Chevrant-Breton, E. Collet, J.F. Cuny, P. Chastagner, M.L. Chandeclerc, E. Cheuret, P. Cintas, H. Dollfus, C. Derancourt, V. Drouin-Garraud, M. D'Incan, H. De Leersnyder, O. Dereure, D. Doumar, N. Fabre, V. Ferraro, C. Francannet, L. Faivre, F. Fellmann, N. Feugier, D. Gaillard, A. Goldenberg, L. Guyant-Maréchal, J.S. Guillamo, S. Hadj-Rabia, D. Hamel Teillac, I. Kemlin, J.P. Lacour, V. Laithier, J.C. Leonard, N. Lesavre, S. Lyonnet, K. Maincent, S. Maradeix, L. Machet, E. Mansat, N. Meyer, M. Mozelle, J.C. Moreno, O. Montagne, C. Moret, E. Puzenat, S. Pinson, D. Rodriguez, S. Sportich, J.F. Stalder, E. Schweitzer, C. Thalamas, C. Thauvin, L. Taillandier, A. Verloes and J. Zeller.

Conflict of Interest statement. None declared.

REFERENCES

1
Carey
J.C.
Baty
B.J.
Johnson
J.P.
Morrison
T.
Skolnick
M.
Kivlin
J.
The genetic aspects of neurofibromatosis
Ann. N. Y. Acad. Sci.
 , 
1986
, vol. 
486
 (pg. 
45
-
56
)
2
Friedman
J.M.
Gutmann
D.H.
MacCollin
M.
Riccardi
V.M.
Neurofibromatosis: phenotype, natural history, and pathogenesis
 , 
1999
Baltimore
Johns Hopkins University Press
3
Friedman
J.M.
Riccardi
V.M.
Friedman
J.M.
Gutmann
D.H.
MacCollin
M.
Riccardi
V.M.
Clinical and epidemiological features
Neurofibromatosis: phenotype, natural history, and pathogenesis
 , 
1999
Baltimore
Johns Hopkins University Press
(pg. 
29
-
86
)
4
Fahsold
R.
Hoffmeyer
S.
Mischung
C.
Gille
C.
Ehlers
C.
Kücükceylan
N.
Abdel-Nour
M.
Gewies
A.
Peters
H.
Kaufmann
D.
, et al.  . 
Minor lesion mutational spectrum of the entire NF1 gene does not explain its high mutability but points to a functional domain upstream of the GAP-related domain
Am. J. Hum. Genet.
 , 
2000
, vol. 
66
 (pg. 
790
-
818
)
5
Ars
E.
Kruyer
H.
Morell
M.
Pros
E.
Serra
E.
Ravella
A.
Estivill
X.
Lázaro
C.
Recurrent mutations in the NF1 gene are common among neurofibromatosis type 1 patients
J. Med. Genet.
 , 
2003
, vol. 
40
 pg. 
e82
 
6
Mattocks
C.
Baralle
D.
Tarpey
P.
ffrench-Constant
C.
Bobrow
M.
Whittaker
J.
Automated comparative sequence analysis identifies mutations in 89% of NF1 patients and confirms a mutation cluster in exons 11–17 distinct from the GAP related domain
J. Med. Genet.
 , 
2004
, vol. 
41
 pg. 
e48
 
7
Kluwe
L.
Siebert
R.
Gesk
S.
Friedrich
R.E.
Tinschert
S.
Kehrer-Sawatzki
H.
Mautner
V.F.
Screening 500 unselected neurofibromatosis 1 patients for deletions of the NF1 gene
Hum. Mutat.
 , 
2004
, vol. 
23
 (pg. 
111
-
116
)
8
Venturin
M.
Guarnieri
P.
Natacci
F.
Stabile
M.
Tenconi
R.
Clementi
M.
Hernandez
C.
Thompson
P.
Upadhyaya
M.
Larizza
L.
, et al.  . 
Mental retardation and cardiovascular malformations in NF1 microdeleted patients point to candidate genes in 17q11.2
J. Med. Genet.
 , 
2004
, vol. 
41
 (pg. 
35
-
41
)
9
Easton
D.F.
Upadhyaya
M.
Cooper
D.N.
Genotype-phenotype relationships in neurofibromatosis type 1
Neurofibromatosis type 1: from genotype to phenotype
 , 
1998
Oxford
BIOS Publishers
(pg. 
167
-
171
)
10
Castle
B.
Baser
M.E.
Huson
S.M.
Cooper
D.N.
Upadhyaya
M.
Evaluation of genotype-phenotype correlations in neurofibromatosis type 1
J. Med. Genet.
 , 
2003
, vol. 
40
 pg. 
e109
 
11
Upadhyaya
M.
Huson
S.M.
Davies
M.
Thomas
N.
Chuzhanova
N.
Giovannini
S.
Evans
D.G.
Howard
E.
Kerr
B.
Griffiths
S.
, et al.  . 
An absence of cutaneous neurofibromas associated with a 3-bp inframe deletion in exon 17 of the NF1 gene (c.2970–2972 delAAT): evidence of a clinically significant NF1 genotype-phenotype correlation
Am. J. Hum. Genet.
 , 
2007
, vol. 
80
 (pg. 
140
-
151
)
12
Easton
D.F.
Ponder
M.A.
Huson
S.M.
Ponder
B.A.
An analysis of variation in expression of neurofibromatosis type 1 (NF1): evidence for modifying genes
Am. J. Hum. Genet.
 , 
1993
, vol. 
53
 (pg. 
305
-
313
)
13
Szudek
J.
Joe
H.
Friedman
J.M.
Analysis of intrafamilial phenotypic variation in neurofibromatosis type 1 (NF1)
Genet. Epidemiol.
 , 
2002
, vol. 
23
 (pg. 
150
-
164
)
14
Li
J.
Ji
L.
Adjusting multiple testing in multilocus analyses using the eigenvalues of a correlation matrix
Heredity
 , 
2005
, vol. 
95
 (pg. 
221
-
227
)
15
Nyholt
D.R.
A simple correction for multiple testing for single-nucleotide polymorphisms in linkage disequilibrium with each other
Am. J. Hum. Genet.
 , 
2004
, vol. 
74
 (pg. 
765
-
769
)
16
Houlston
R.S.
Tomlinson
I.P.
Modifier genes in humans: strategies for identification
Eur. J. Hum. Genet.
 , 
1998
, vol. 
6
 (pg. 
80
-
88
)
17
Agarwal
S.
Moorchung
N.
Modifier genes and oligogenic disease
J. Nippon. Med. Sch.
 , 
2005
, vol. 
72
 (pg. 
326
-
334
)
18
Génin
E.
Feingold
J.
Clerget-Darpoux
F.
Identifying modifier genes of monogenic disease: strategies and difficulties
Hum. Genet.
 , 
2008
, vol. 
124
 (pg. 
357
-
368
)
19
Pasmant
E.
Sabbagh
A.
Hanna
N.
Masliah
J.
Jolly
E.
Parfait
B.
Ballerini
P.
Cartault
F.
Barbarot
S.
Landman-Parker
J.
, et al.  . 
SPRED1 germline mutations caused a neurofibromatosis type 1 overlapping phenotype
J. Med. Genet.
 , 
2009
 
in press
20
Bloom
B.
Cohen
R.A.
Freeman
G.
Summary health statistics for U.S. children: National Health Interview Survey, 2007
Vital. Health Stat.
 , 
2009
, vol. 
239
 (pg. 
1
-
80
)
21
Loftis
E.L.
Recklinghausen's disease in identical twins
Arch. Dermatol.
 , 
1940
, vol. 
42
 pg. 
657
 
22
Dresner
E.
Montgomery
D.A.D.
Primary optic atrophy in Von Recklinghausen's disease (multiple neurofibromatosis)
Q. J. Med.
 , 
1949
, vol. 
18
 (pg. 
93
-
104
)
23
Riccardi
V.M.
Kleiner
B.
Lubs
L.
Neurofibromatosis: variable expression is not intrinsic to the mutant gene
Birth Defects
 , 
1979
, vol. 
15
 (pg. 
283
-
289
)
24
Vaughn
A.J.
Bachman
D.
Sommer
A.
Neurofibromatosis in monozygotic twins: a case report of spontaneous mutation
Am. J. Med. Genet.
 , 
1981
, vol. 
8
 (pg. 
155
-
158
)
25
Akesson
H.O.
Axelsson
R.
Samuelsson
B.
Neurofibromatosis in monozygotic twins: a case report
Acta Genet. Med. Gemellol.
 , 
1983
, vol. 
32
 (pg. 
245
-
249
)
26
Crawford
M.J.
Buckler
J.M.H.
Optic gliomata affecting twins with neurofibromatosis
Dev. Med. Child Neurol.
 , 
1983
, vol. 
25
 (pg. 
367
-
373
)
27
Gouya
L.
Deybach
J.C.
Lamoril
J.
Da Silva
V.
Beaumont
C.
Grandchamp
B.
Nordmann
Y.
Modulation of the phenotype in dominant erythropoietic protoporphyria by a low expression of the normal ferrochelatase allele
Am. J. Hum. Genet.
 , 
1996
, vol. 
58
 (pg. 
292
-
299
)
28
Gouya
L.
Puy
H.
Lamoril
J.
Da Silva
V.
Grandchamp
B.
Nordmann
Y.
Deybach
J.C.
Inheritance in erythropoietic protoporphyria: a common wild-type ferrochelatase allelic variant with low expression accounts for clinical manifestation
Blood
 , 
1999
, vol. 
93
 (pg. 
2105
-
2110
)
29
Gouya
L.
Puy
H.
Robreau
A.M.
Bourgeois
M.
Lamoril
J.
Da Silva
V.
Grandchamp
B.
Deybach
J.C.
The penetrance of dominant erythropoietic protoporphyria is modulated by expression of wildtype FECH
Nat. Genet.
 , 
2002
, vol. 
30
 (pg. 
27
-
28
)
30
Randon
J.
Boulanger
L.
Marechal
J.
Garbarz
M.
Vallier
A.
Ribeiro
L.
Tamagnini
G.
Dhermy
D.
Delaunay
J.
A variant of spectrin low-expression allele alpha LELY carrying a hereditary elliptocytosis mutation in codon 28
Br. J. Haematol.
 , 
1994
, vol. 
88
 (pg. 
534
-
540
)
31
Szudek
J.
Birch
P.
Riccardi
V.M.
Evans
D.G.
Friedman
J.M.
Associations of clinical features in neurofibromatosis 1 (NF1)
Genet. Epidemiol.
 , 
2000
, vol. 
19
 (pg. 
429
-
439
)
32
Szudek
J.
Evans
D.G.
Friedman
J.M.
Patterns of associations of clinical features in neurofibromatosis 1 (NF1)
Hum. Genet.
 , 
2003
, vol. 
112
 (pg. 
289
-
297
)
33
Ferner
R.E.
Huson
S.M.
Thomas
N.
Moss
C.
Willshaw
H.
Evans
D.G.
Upadhyaya
M.
Towers
R.
Gleeson
M.
Steiger
C.
, et al.  . 
Guidelines for the diagnosis and management of individuals with neurofibromatosis 1
J. Med. Genet.
 , 
2007
, vol. 
44
 (pg. 
81
-
88
)
34
National Institutes of Health Consensus Development Conference Statement
Neurofibromatosis
Arch. Neurol. Chicago.
 , 
1988
, vol. 
45
 (pg. 
575
-
578
)
35
Zeller
J.
Wechsler
J.
Revuz
J.
Wolkenstein
P.
Taches violines ou macules pseudo-atrophiques au cours de la neurofibromatose 1
Ann. Dermatol. Vénéréol.
 , 
2002
, vol. 
129
 (pg. 
180
-
181
)
36
Box
G.E.P.
Cox
D.R.
An analysis of transformations
J. R. Stat. Soc. B.
 , 
1964
, vol. 
26
 (pg. 
211
-
252
)
37
Almasy
L.
Blangero
J.
Multipoint quantitative-trait linkage analysis in general pedigrees
Am. J. Hum. Genet.
 , 
1998
, vol. 
62
 (pg. 
1198
-
1211
)
38
S.A.G.E
Statistical Analysis for Genetic Epidemiology 5.4.2 Computer program package available from the Department of Epidemiology and Biostatistics
2008
Cleveland
Case Western Reserve University
39
The International HapMap Consortium
The International HapMap Project
Nature
 , 
2003
, vol. 
426
 (pg. 
789
-
796
)
40
Barrett
J.C.
Fry
B.
Maller
J.
Daly
M.J.
Haploview: analysis and visualization of LD and haplotype maps
Bioinformatics
 , 
2005
, vol. 
21
 (pg. 
263
-
265
)
41
Gabriel
S.B.
Schaffner
S.F.
Nguyen
H.
Moore
J.M.
Roy
J.
Blumenstiel
B.
Higgens
J.
DeFelice
M.
Lochner
A.
Faggart
M.
, et al.  . 
The structure of haplotype blocks in the human genome
Science
 , 
2002
, vol. 
296
 (pg. 
2225
-
2229
)
42
de Bakker
P.I.W.
Yelensky
R.
Pe'er
I.
Gabriel
S.B.
Daly
M.J.
Altshuler
D.
Efficiency and power in genetic association studies
Nat. Genet.
 , 
2005
, vol. 
37
 (pg. 
1217
-
1223
)
43
Horvath
S.
Xu
X.
Laird
N.M.
The family based association test method: strategies for studying general genotype-phenotype associations
Eur. J. Hum. Genet.
 , 
2001
, vol. 
9
 (pg. 
301
-
306
)
44
Horvath
S.
Xu
X.
Lake
S.L.
Silverman
E.K.
Weiss
S.T.
Laird
N.M.
Family-based tests for associating haplotypes with general phenotype data: application to asthma genetics
Genet. Epidemiol.
 , 
2004
, vol. 
26
 (pg. 
61
-
69
)
45
Laird
N.M.
Horvath
S.
Xu
X.
Implementing a unified approach to family-based tests of association
Genet. Epidemiol.
 , 
2000
, vol. 
19
 (pg. 
S36
-
S42
)
46
Spielman
R.S.
McGinnis
R.E.
Ewens
W.J.
Transmission test for linkage disequilibrium: the insulin gene region and insulin-dependent diabetes mellitus (IDDM)
Am. J. Hum. Genet.
 , 
1993
, vol. 
52
 (pg. 
506
-
516
)